本篇博文主要内容为 2026-10-20 从Arxiv.org论文网站获取的最新论文列表，自动更新，按照NLP、CV、ML、AI、IR、MA六个大方向区分。
说明：每日论文数据从Arxiv.org获取，每天早上12:30左右定时自动更新。
提示: 当天未及时更新，有可能是Arxiv当日未有新的论文发布，也有可能是脚本出错。尽可能会在当天修复。
目录
概览 (2026-10-20)
今日共更新1716篇论文,其中:
- 自然语言处理共285篇(Computation and Language (cs.CL))
- 人工智能共487篇(Artificial Intelligence (cs.AI))
- 计算机视觉共346篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共514篇(Machine Learning (cs.LG))
- 多智能体系统共35篇(Multiagent Systems (cs.MA))
- 信息检索共41篇(Information Retrieval (cs.IR))
- 人机交互共42篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] opological Sheaves of Irreducibility: A Theoretical Framework for Multi-Agent Task Obstructions
【速读】：该论文旨在解决多智能体系统（multi-agent systems）在相同智能体数量与通信预算下，因支持不同形式的协同交互而导致系统能力差异的问题。其核心挑战在于区分任务层面的兼容性障碍（compatibility obstructions）与系统实际能力之间的界限。为此，作者提出拓扑不可约性层（Topological Sheaves of Irreducibility, TSI）这一层论（sheaf-theoretic）框架，将任务结构与系统容量分离建模：任务由交互复形（interaction complex）$ K $ 和容量层（capacity sheaf）$ \mathcal{F} $ 表示，非零上同调群 $ H^n(K; \mathcal{F}) $ 构成结构性签名（structural signature），用于刻画任务本身的拓扑约束；而通过注册干预（registered interventions）测试系统对这些约束的实际响应能力。关键在于，对于上链数据而言，上链实现（cochain realization）可解当且仅当其所代表的上同调类为零。该方法揭示了结构性签名具有层相关性（sheaf-relative），且一阶降维可导出循环空间的见证（cycle-space witnesses）与观测需求。实验验证表明，在15,625个一阶输入中，全智能体加法原语可精确求解决策与修复任务；最优成对系统在决策任务上达到88.80%准确率（仅高于88.79%多数类基线一个输入），修复任务为26.63%；在531,441个二阶四面体边界任务中，保持相同2-骨架结构的三智能体原语系统虽保留全部四个三智能体原语，却始终与仅使用成对原语的最优解性能一致，而引入类别填充型四智能体原语后，两类任务均实现精确求解。状态保持型鲁棒性检验维持了精确性能，但未在成本上匹配类别恢复型消融实验。综上，基于注册原语与成本模型，TSI提供了一种任务相关的、可证伪的高阶交互能力评估机制。
链接: https://arxiv.org/abs/2610.26777
作者: Steven Motta,Davide Mambelli,Alessandro Rozza
机构: distributed cognition
类目: Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: 31 pages, 1 figure, 3 tables
Abstract:Multi-agent systems can differ under identical agents and communication budgets when they support different forms of joint interaction. We introduce Topological Sheaves of Irreducibility (TSI), a sheaf-theoretic framework separating task-level compatibility obstructions from system capability. A task is represented by an interaction complex K and a capacity sheaf \mathcalF ; nonzero H^n(K;\mathcalF) defines a structural signature, while registered interventions test capability. For cocycle data, cochain realization is solvable exactly when the represented cohomology class vanishes. Examples show that the signature is sheaf-relative, and the degree-1 reduction yields cycle-space witnesses and observation requirements. We evaluate two finite one-round models. Across 15,625 degree-1 inputs, an all-agent additive primitive solves decision and repair exactly; the optimal pairwise system reaches 88.80% on decision, one input above the 88.79% majority-class baseline, and 26.63% on repair. On a degree-2 tetrahedral-boundary task over 531,441 inputs, the same-2-skeleton system retains all four three-agent primitives yet matches the pairwise-only optimum throughout the registered separation window, whereas adding the class-filling four-agent primitive makes both tasks exact. Status-preserving robustness checks retain exact performance but are not cost-matched to the class-restoring ablation. Conditional on the registered primitives and cost models, TSI provides task-relative, falsifiable tests of higher-order interaction.
[MA-1] Autonomous intelligence for mining productivity: a multi-agent AI approach to operational value creation
【速读】：该论文旨在解决集成远程操作中心（IROCs）在迈向全自主化过程中面临的两大核心问题：一是缺乏能够协调监督控制与数据采集系统、数字孪生、预测分析及人机交互界面的集成式智能体架构，尤其在安全关键约束下的协同机制缺失；二是现有评估框架无法兼顾实时性、健康-安全-环境（HSE）、合规审计性以及决策成本等工业级智能体部署的关键要求。其解决方案的关键在于提出EI2ROC——一个三层多智能体架构，通过大型语言模型（LLM）作为调度器，结合专用监控与产能预测智能体，并借助类模型上下文协议（Model Context Protocol）的工具接口实现高效通信。该架构不仅实现了跨系统的协同优化，还配套提出了一个多维度基准测试框架，涵盖场景库、测试工具集、评估指标体系及人因校准协议，系统性地填补了当前工业智能体评估的盲区。论文明确区分了原型系统中已实现组件、参考设计部分及未来工作，并基于实际运行站点提供了初步的成本与延迟观测数据，为资本密集型行业（如采矿、油气、能源）的智能化升级提供了可迁移的方法论蓝图。
链接: https://arxiv.org/abs/2610.26369
作者: N Guerin(CGS i3),A Dehoux(ERAMET),M Nakhla(CGS i3)
机构: 未知
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Integrated Remote Operation Centres (IROCs) are emerging as the architectural frontier of Industry 4.0 for capital-intensive sectors such as mining, oil and gas, and energy. Yet two intertwined gaps constrain their evolution towards full autonomy. First, no integrated agentic architecture has been proposed for mining IROCs that orchestrates Supervisory Control and Data Acquisition systems, digital twins, predictive analytics and human-machine interfaces under safety-critical constraints. Second, no evaluation framework jointly addresses the real-time, Health-Safety-Environment, regulatory-auditability and cost-per-decision requirements that govern industrial agentic deployments. This paper makes four contributions. It synthesises industry needs and quantified gains documented in analogous industries that motivate an agentic approach to mining IROCs. It introduces EI2ROC, a three-tier multi-agent architecture combining a Large Language Model orchestrator with specialised monitoring and production-capacityprediction agents communicating through a Model Context Protocol-style tool interface. It provides a critical review of agentic evaluation along six orthogonal dimensions and identifies the industrial blind spot of current benchmarks. Finally, it proposes a multi-dimensional benchmarking framework for industrial agentic AI comprising scenario library, harness, metrics and humancalibration protocol. The paper states explicitly which components are implemented in the current prototype, which belong to the proposed reference design and which remain future work, and reports preliminary cost and latency observations from an operating site. Together, EI2ROC and the benchmarking framework constitute a methodological blueprint transferable across CAPEX intensive industries.
[MA-2] Discovery-Probability Graph Planning for Embodied Agents in Tasks with Hidden Cooperation Requirements
【速读】：该论文旨在解决在动态环境中，智能体面对隐性协作任务（hidden-cooperation tasks）时的协同能力不足问题。这类任务的特点是协作需求仅在智能体观察并交互于任务相关目标后才显现，与传统假设协作任务预先已知的情况形成对比。其解决方案的关键在于提出一种名为发现概率图规划（Discovery-Probability Graph Planning, DPGP）的新方法，该方法构建了一个发现概率图（discovery-probability graph），用于编码任务相关目标的数量与位置的概率分布，从而支持智能体间高效的信息共享与协同决策。在此基础上，结合搜索算法生成最小化期望成本的动作规划，显著提升了协作成功率与执行效率。实验表明，DPGP在C-WAH和TDW-MAT基准上分别达到99.3%和58.3%的协作成功率，优于所有基线方法，并有效降低了整体实体执行时间与大语言模型（LLM）的词元（token）消耗。
链接: https://arxiv.org/abs/2610.26359
作者: Zihuang Cai,Changjian Chen,Zhuo Tang,Chang Tang,Kenli Li
机构: Hunan University(湖南大学); Geely Holding Group(吉利控股集团)
类目: Multiagent Systems (cs.MA)
备注: 25 pages, 6 figures, including appendices
Abstract:Embodied agents are increasingly capable of perceiving, planning, and acting autonomously in dynamic environments, demonstrating strong performance on complex cooperative tasks. Recent advances typically assume that cooperative tasks are known in advance. In real-world applications, however, hidden-cooperation tasks are more common: the need for cooperation only becomes apparent after an agent observes and interacts with task-related targets. In this paper, we first present HiddenCoop, a benchmark for effectively evaluating agent cooperation in hidden-cooperation tasks. We further propose Discovery-Probability Graph Planning (DPGP), a method for enabling effective agent cooperation in such tasks. At the core of DPGP is a discovery-probability graph that encodes the probability distribution over the number and locations of task-related targets, facilitating effective and efficient information exchange among agents. A search algorithm is then used to derive action plans that minimize expected costs. Experiments show that DPGP achieves cooperative-task success rates of 99.3% on C-WAH and 58.3% on TDW-MAT, surpassing all compared baselines, while also reducing overall embodied execution time and substantially lowering LLM token usage. Our code is available at this https URL.
[MA-3] he Control Layer Behind Rule-Following Agents Trading Across Market Conditions
【速读】：该论文旨在解决量化交易中交易纪律（trading discipline）难以持续执行的问题，尤其关注生成式 AI（Generative AI）驱动的交易代理在实际市场环境下的规则遵循能力。研究通过分析2026年2月12日至8月27日期间同一资产池与执行条件下的2,627个已关闭交易，评估四个AI交易代理在真实价格环境中的策略执行与风险控制表现。其解决方案的关键在于构建一个可复现的交易日志测试框架与代理评估工具包，能够有效区分交易结果是源于严格遵守预设规则，还是偶然性收益（luck）。研究发现，尽管部分代理未盈利，但所有代理均高度遵循止损与止盈规则，97.5%的亏损被控制在声明的止损水平内，且持仓数量限制在多数运行中得以严格执行。此外，系统在经历29%的市场回撤后仍保持风险控制有效性，表明其纪律性具备跨市场周期的稳定性。该工具包的核心价值在于为资本分配前提供“纪律准入”检验标准——即在赋予代理自主权或投入资金前，必须先通过规则遵循性的实证验证，而非仅依赖账户层面的收益表现。
链接: https://arxiv.org/abs/2610.26324
作者: Saad Naja,Nasser Odtallah,Jake Van-Praag,Sarah May
机构: 未知
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Multiagent Systems (cs.MA); Risk Management (q-fin.RM); Trading and Market Microstructure (q-fin.TR)
备注: 47 pages, 0 figures, 16 numbered tables. Includes a trading-agent evaluation toolkit and checklist
Abstract:Trading discipline often fails people. Our study tested whether four AI trading agents followed their strategies and risk rules. Offering traders the evidence and an agent evaluation toolkit to separate luck from following the rules, before giving agents meaningful autonomy or capital. Our study analysed 2,627 closed positions from 12 February to 27 August 2026 across the same asset universe and execution conditions. Most positions were simulated against live prices with a 0.09% round-trip fee; 18% involved real Hyperliquid orders. Each position declared a strategy, stop and target, with a limit of five open positions per agent. The fleet finished net positive after fees. Rule adherence was high across all four agents, including the unprofitable one. Profitability tracked how often agents won relative to their break-even win rates. Institutional-style controls ran by default and contained losses through a 29% market drawdown and recovery, with documented exceptions. The discipline held for all four agents. Agents kept 97.5% of gross loss within declared stops, and typical losses closed at declared levels. Declared risk remained stable after losses and broadly stable across calm and volatile months. Twenty-three of 24 runs respected the position limit; logs identify the exception’s size and duration. The same live data produced different results across assigned strategies. The trade-log tests can be rerun from the published trade snapshot; regime tests also require the published Bitcoin prices and regime labels. Section 3.5 specifies the stop-breach tolerance. These trade-level results are not account-level returns or proof of a lasting edge. Our study provides an evaluation toolkit to help traders distinguish rule-following from luck before committing capital. For anyone considering allocating capital to agents to trade, discipline should be the entry test, not the only test.
[MA-4] Bidirectionality is Necessary in Cellular Machines for Exact-Arrival Multi-Agent Path Finding
【速读】：该论文旨在解决在存在顶点冲突的有向无环图（Directed Acyclic Graph, DAG）上，基于精确出发与到达时间约束的多智能体路径规划（Exact-Arrival Multi-Agent Path Finding, EAMAPF）问题，将其建模为一个分布式输出任务：每个单元格需独立输出自身对有效联合路径掩码的贡献位。核心挑战在于设计一种完全分布式的、仅依赖局部通信的算法架构，避免全局解码器，并确保在任意拓扑结构下对所有可行查询均正确响应。其解决方案的关键在于证明：在无全局标志（k=0）的情况下，任何仅支持单一时间方向（前向或后向）的局部规则系统都无法在所有可行查询上保持正确性——这揭示了时间方向性的“必要性”；而引入有限个全局聚合标志（k≥1）后，错误性被转化为带宽与通信轮次之间的权衡。通过构造特定拓扑（如十二单元、双智能体、深度三的实例）和反例分析，论文进一步证明了该下界在数量级上的紧性：当将多个副本并列构成复合拓扑时，前向型机器至少需要满足 $ J_R + kR \geq p $ 轮通信，其中 $ J_R $ 表示查询信息传递至目标锥区域的程度，而 $ R $ 为通信轮数。该结果通过形式化验证工具 Lean 4 对构件的有限引理及配对不变性矛盾进行了验证，其余部分采用纸面证明。研究未报告基准测试结果。
链接: https://arxiv.org/abs/2610.26272
作者: Masashi Aono
机构: Amoeba Energy Co., Ltd.(阿米巴能源有限公司); Keio University(庆应义塾大学)
类目: Emerging Technologies (cs.ET); Computational Complexity (cs.CC); Distributed, Parallel, and Cluster Computing (cs.DC); Multiagent Systems (cs.MA)
备注: 33 pages. Ancillary files: Lean 4 development for the finite gadget ( this http URL ), exhaustive check scripts with frozen expected outputs, reproduction README and clean-run log. Companion paper: “Stability Certifies Validity: Amoeba-Inspired Rule Systems for SAT and Single-Agent Trajectory Constraints, Machine-Checked in Lean 4”
Abstract:On ranked directed acyclic graphs coupled by vertex conflicts we study exact-arrival multi-agent path finding (departure and arrival pinned in place and time) posed as a distributed output task: every cell must emit its own bit of a valid joint path-support mask. The machines are cellular: one automaton per cell, state factored over cells, communication only along a fixed local fabric, synchronous rounds, a query-independent output round, no central decoder, and a per-topology transition code. “Necessary” here means: on an explicit hard topology, over all its feasible queries, no universally correct local-only architecture can renounce either temporal orientation; with global flags, impossibility becomes a bandwidth-round trade-off. The class is not vacuous: a flooding machine solves every instance in as many rounds as the diameter of the local fabric. The bound survives two concessions: we supply every agent’s exact individual cavity, which alone solves the single-agent problem in zero rounds and makes the single-agent direction theorem vacuous; and we allow k one-bit global aggregate flags. On a twelve-cell, two-agent, depth-three topology no forward-only machine with k=0 is correct on two feasible queries, for any number of rounds; the time-reversed topology does the same to backward-only machines. Placing p copies side by side gives, for forward-only machines, J_R+kR\ge p at constant depth, where J_R measures how much of the query reaches the targets’ cones and is 0 here. A backward-only machine solves that family in constant rounds and a flag-streaming one in \lceil p/k\rceil+O(1) , so the bound is tight in order; on the composite family neither orientation can be renounced. The finite lemmas of the gadget and the pair-invariance contradictions are checked in Lean 4; the semantic bridge, the cone lemma and the counting are paper proofs. We report no benchmarks.
[MA-5] A GPU-Resident Pipeline for Multi-Robot 3D Inspection with Limited Perception
【速读】：该论文旨在解决多机器人在水下结构物三维巡检中如何高效规划协同轨迹的问题，核心挑战在于：在有限视场、短程相机约束下，如何为一组机器人（K台）生成联合覆盖目标表面一定比例的6-自由度（6-DOF）运动轨迹，同时最小化线性插值后的完工时间/总成本目标。由于该联合优化问题属于NP-hard问题，其解决方案的关键在于将复杂问题分解为五个可并行处理的核心阶段：候选视角生成、可见性计算、集合覆盖、基于混合整数规划（MIP）与提升型Miller-Tucker-Zemlin子环消除约束的最小最大车辆路径规划，以及带连续时间机器人冲突检测的优先级驱动多智能体路径规划；前后分别辅以网格加载与表面采样预处理，以及轨迹平滑后处理。关键创新包括：端到端的GPU加速集成框架及其内核设计；针对Lien的ε-可见性算法的四核定制化CUDA批处理实现；基于插值目标导出的β感知单车路径上界，结合Desrochers-Laporte系数提升与禁止对剪枝策略；对Zhou和Zeng的GPU并行前沿A*算法的4D时空扩展；并通过14个大规模网格（最多720万面）的实验验证了路由剪枝与并行前沿搜索的有效性，审计了轨迹平滑后的无碰撞性与航点可达性，并量化了端到端解与解析解及线性规划松弛下界之间的差距。
链接: https://arxiv.org/abs/2610.26106
作者: Kryštof Pšenička,Stefan Edelkamp
机构: 未知
类目: Robotics (cs.RO); Multiagent Systems (cs.MA)
备注:
Abstract:We present a complete, GPU-resident pipeline for multi-robot 3D inspection of underwater structures. Given a triangle mesh and a fleet of K robots, each carrying a short-range, limited field-of-view camera, the pipeline computes 6-DOF trajectories that jointly cover a target fraction of the surface while minimizing a linearly interpolated makespan/total-cost objective. The joint problem is NP-hard, so we decompose it into five core stages - candidate viewpoint generation, visibility computation, set cover, min-max vehicle routing solved as a mixed integer program (MIP) with lifted Miller-Tucker-Zemlin subtour elimination, and priority-based multi-agent path finding with continuous-time inter-robot conflict checking - flanked by mesh loading and surface sampling as preprocessing and by trajectory smoothing as postprocessing. Our contributions are: an end-to-end GPU-accelerated integration of the five stages, with its kernel design documented; a GPU batch adaptation of Lien’s \varepsilon -visibility with four custom CUDA kernels; a \beta -aware per-vehicle tour upper bound derived from the interpolated objective, supplementing the Desrochers-Laporte coefficient lift with forbidden-pair cuts; a 4D space-time extension of Zhou and Zeng’s GPU parallel-frontier A*; and an experimental evaluation on 14 meshes (up to 7.2 M faces) that ablates the routing cuts and the parallel-frontier search, audits collision freedom and waypoint attainment \emphafter trajectory smoothing, and measures the end-to-end gap against analytical and LP-relaxation lower bounds. Code and raw results are publicly available.
[MA-6] Behavioral Probes for Information Flow in LLM Swarms NEURIPS2026
【速读】：该论文旨在解决现有大语言模型（LLM）群体优化方法中过度依赖最终任务效用而忽视个体交互过程中信息处理、整合与依赖关系隐含性的问题。其核心解决方案是基于信息流视角，引入行为探针（behavioral probes）以量化分析信息流动模式。关键创新在于重新利用原本用于幻觉分析的外部上下文得分（External Context Score, ECS）和参数化知识得分（Parametric Knowledge Score, PKS），分别衡量节点对外部上下文的依赖程度和对自身参数化知识的依赖程度。实验表明，在受控与真实场景下，ECS随输入相关性增加并能区分多源信息贡献的有效性，而PKS则随有效上下文积累而下降。基于此启发式规律，研究进一步验证了探针引导的节点剪枝可减少冗余通信与令牌消耗，探针引导的推理路径搜索则显著加速收敛。结果表明，行为探针能够使LLM群体内部的信息流动过程可观测，并作为结构优化的有效辅助信号。
链接: https://arxiv.org/abs/2610.25966
作者: Junhui Chen,Yu Zhang,Lantian Li,Dong Wang
机构: Beijing University of Posts and Telecommunications (北京邮电大学); Tsinghua University (清华大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026
Abstract:Most existing approaches to LLM swarm optimization are utility-driven, focusing on final task utility while leaving the processing, integration, and reliance of information within individual interactions largely implicit. To address this, we study LLM swarm interactions from an information-flow-based view and introduce behavioral probes to quantify information flow patterns. Specifically, we repurpose External Context Score (ECS) and Parametric Knowledge Score (PKS)—originally designed for hallucination analysis—to measure a node’s external-context dependence and parametric-knowledge dependence, respectively. Across controlled and realistic swarm settings, we observe consistent patterns: ECS increases with input relevance and distinguishes useful multi-source contributions, while PKS decreases as informative context accumulates. Building on this heuristic interpretation, we demonstrate how these probes can complement utility-driven structure optimization in two controlled applications. Probe-guided node pruning reduces redundant communication and token consumption, and probe-guided reasoning-path search accelerates convergence compared to utility-only baselines. Our empirical results suggest that behavioral probes can make the internal information flow of LLM swarms observable and serve as effective auxiliary signals for structure optimization.
[MA-7] Learning Who Talks to Whom: Role-Aware Communication Topology Scheduling for Efficient LLM Multi-Agent Collaboration
【速读】：该论文旨在解决基于大语言模型（LLM）的多智能体系统在协作推理过程中，如何高效动态调度通信拓扑以平衡集体推理性能与指数级增长的令牌开销这一核心问题。现有方法要么采用静态通信图无法适应动态求解状态，要么忽视智能体角色互补性，且未将生成的拓扑结构显式融入团队层面的决策评估中。其解决方案的关键在于提出一种面向角色的自适应通信拓扑调度框架ACTS，该框架在集中式训练、分散式执行（CTDE）范式下引入一种新颖的双注意力机制：一是角色条件关系注意力（Role-conditioned Relational Attention）模块，根据智能体角色和演化执行状态动态选择变基数的交互伙伴；二是拓扑注意力混合器（Topological Attention Mixer），将有向通信拓扑显式嵌入团队价值分解中，实现精确的信用分配。实验在涵盖代码生成、多步推理、通用知识和竞赛数学等9个挑战性基准上验证了ACTS的有效性，相比现有最优基线平均准确率提升最高达6.19个百分点，同时令牌消耗降低超过15%。
链接: https://arxiv.org/abs/2610.25851
作者: Haibin Li,Wenya Yu,Xuefeng Jiang,Jia Li,Xiao Han,Shuai Su,Wei Wei
机构: Beihang University(北京航空航天大学); Nanyang Technological University(南洋理工大学); University of Chinese Academy of Sciences(中国科学院大学)
类目: Multiagent Systems (cs.MA)
备注: Project website: this https URL
Abstract:Multi-agent systems driven by large language models (LLMs) rely on structured communication topologies to facilitate information exchange and collaborative reasoning. However, effectively scheduling these topologies to dynamically determine who talks to whom and when to communicate remains a fundamental challenge in balancing collective reasoning performance against exponential token overhead. Existing topology scheduling approaches either enforce static communication graphs that cannot adapt to dynamic problem-solving states, or overlook agent role complementarity while failing to explicitly incorporate the synthesized graph structure into team-level decision evaluation. To tackle these limitations, we propose ACTS, a role-aware adaptive communication topology scheduling framework for efficient LLM multi-agent collaboration. Orchestrated within a Centralized Training with Decentralized Execution (CTDE) paradigm, ACTS introduces a novel dual-attention mechanism. Specifically, a Role-conditioned Relational Attention module dynamically selects variable-cardinality interaction partners based on agent roles and evolving execution states, while a Topological Attention Mixer explicitly embeds the directed communication topology into team value decomposition for precise credit assignment. Comprehensive evaluations across 9 challenging benchmarks covering code generation, multi-step reasoning, general knowledge, and competitive mathematics demonstrate that ACTS consistently outperforms state-of-the-art baselines, boosting average accuracy by up to 6.19 percentage points while reducing token consumption by over 15%. The project website is available at this https URL.
[MA-8] MAPS: A Multi-Agent Personalized Script-Driven Video Advertising System Enhanced by Reinforcement Learning
【速读】：该论文旨在解决现有基于检索的个性化视频广告系统在创意生成上的粒度不足与用户群体间表现不均的问题。此类系统依赖预生成的广告素材进行匹配，难以实现细粒度个性化，且在面对多样化用户偏好时存在性能差异。为此，论文提出一种多智能体驱动的个性化脚本生成式视频广告系统（MAPS），其核心在于构建一个闭环协同框架，包含用户画像建模、脚本生成、视频合成及反馈驱动的迭代优化四个智能体。该方案的关键创新在于为脚本生成智能体（Scripter Agent）设计了一种基于多维度奖励函数的强化学习训练框架，能够有效引导生成既符合用户内容偏好又精准传达产品信息的广告脚本。实验结果表明，该方法在离线数据集上显著提升了个性化创意生成与融合效果，而在快手广告系统的在线A/B测试中，相较生成基线模型，实现了点击率（CTR）提升2.87%、转化率（CVR）提升2.74%以及每用户平均收入（ARPU）提升9.85%，验证了其在实际应用中的有效性。
链接: https://arxiv.org/abs/2610.25748
作者: Zhongyu Wang,Chen Zuo,Qingxuan Sun,Jiayang Qiao,Qiuhao Qian,Songyan Cui,Guocheng Chen
机构: Beihang University (北京航空航天大学); Kuaishou Technology (快手科技)
类目: Information Retrieval (cs.IR); Computer Vision and Pattern Recognition (cs.CV); Multiagent Systems (cs.MA)
备注:
Abstract:With the rapid growth of online video consumption, video advertising has become a dominant component of digital advertising. Most personalized video advertising systems follow a retrieval-based paradigm, which matches ad content with user preferences in a feature space and recommends the most similar candidates. However, since the candidate creatives are pre-produced, this paradigm limits the granularity of personalization and leads to performance disparities across diverse user groups. To address this limitation, we propose a multi-agent personalized script-driven video advertising system (MAPS). We design four collaborative agents that form a closed-loop pipeline of user profile modeling, script generation, video synthesis, and feedback-driven iterative refinement. To train the Scripter Agent, we develop a reinforcement learning framework equipped with a multi-dimensional reward function, which steers the agent toward producing scripts that simultaneously align with user content preferences and target product information. Extensive offline experiments on representative datasets demonstrate that our approach achieves encouraging performance in generating and integrating personalized creatives. Online A/B experiments in a specific scenario in the Kuaishou advertising system show that MAPS improves CTR by 2.87%, CVR by 2.74%, and ARPU by 9.85% over the generative baseline.
[MA-9] How Far Can an Agent Push? Trust-Safety Margins in Live Language-Based Economic Games
【速读】：该论文旨在解决在多智能体系统中，如何在动态、语言驱动的交互场景下，持续评估并合理执行策略变更的问题。传统博弈论可判断某一策略性行动是否应被接受，但无法确定其在实时语言交互中的持续接受时长。本文通过构建一个包含1个受保护智能体Milton与4个工具型智能体的系统，在博弈论竞赛（GLEE）的最后阶段进行实验：Milton仅通过预设规则或经过回滚保护的生产规则接收策略更新，而其他智能体则通过每局独立的随机试验测试策略变化。关键解决方案在于引入“预注册实验”与“基于单元格的剂量路由规则”，其中针对说服任务的五剂量预注册实验表明，推荐所有单位可使销售额提升至理论最大值的0.194倍（95%置信区间[0.147, 0.241]），且其行为模式显示出早期购买率下降幅度大于诚实推荐，揭示了用户对推荐强度的敏感性。此外，通过对比设计与实际行为的偏差，识别出三处调度缺陷，验证了系统可解释性与行为一致性的重要性。整体而言，该研究的核心突破在于建立了一种可验证、可追溯、具备时间延续性的策略部署框架，弥补了传统博弈论在动态语言交互中对“持续接受性”建模的不足。
链接: https://arxiv.org/abs/2610.25508
作者: Dima Galat,Marian-Andrei Rizoiu
机构: University of Technology Sydney(悉尼科技大学)
类目: Multiagent Systems (cs.MA); Computer Science and Game Theory (cs.GT)
备注:
Abstract:The previous version contained Unicode em dashes. Paste this ASCII-only version: Game theory identifies when a strategic action should be accepted, but not how long a live, language-based counterpart will continue to accept it. We entered the GLEE competition in its final week with one protected agent, Milton, and four instrument agents that ran deterministic, per-game randomized trials of policy changes; Milton received changes only through pre-specified gates or registered, rollback-guarded production rules. In negotiation, learned pricing improves adjusted percentile by 0.071 (n=21,168); in bargaining, a continuation model conditioned on the rejected offer did not improve on a recency-weighted incumbent and was retired. In persuasion, a preregistered five-dose experiment (7,391 assigned games, blinded until a fixed stop) shows that recommending every unit raises sales by 0.194 of the maximum (95% CI [0.147, 0.241]) where the receiver’s slack makes obedience rational (52% of the Bayesian benchmark), lowers them where it does not (-0.067), and produces a larger early-to-late purchase decline than honest recommendations (-0.123). A rule fixed before unblinding now routes Milton’s dose per cell. Milton stood 11th of 460 agents in bargaining at the fixed stop; we also report the close, where the final day’s pool shift moved all five agents’ bargaining ratings together while their play was unchanged, and three dispatch defects were found by checking delivered behavior against the design. Subjects: Multiagent Systems (cs.MA); Computer Science and Game Theory (cs.GT) Cite as: arXiv:2610.25508 [cs.MA] (or arXiv:2610.25508v1 [cs.MA] for this version) https://doi.org/10.48550/arXiv.2610.25508 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-10] Explore or Confirm? LLM Agents Misallocate Measurements Under Noise
【速读】：该论文旨在解决在生成式模型推理成本降低的背景下，物理实验成为发现过程中的瓶颈问题——即如何在有限预算下合理分配资源，在探索新候选物与对已有潜在优质候选物进行重复验证（确认）之间做出最优决策。其核心挑战在于：每次实验结果具有噪声，高值可能源于真实优势或偶然性，导致模型易误判。研究提出Wellocate基准环境，基于生成数据景观及真实的GB1和PhoQ蛋白数据集构建，通过打乱数据防止模型记忆干扰，模拟真实实验流程。测试五种无外部工具、无长期记忆的语言模型作为最小代理，发现它们在决策上存在双向误配：三类模型过度重复验证（over-confirm），耗尽预算于再测；一类则因轻信单次读数而低估确认的重要性（under-confirm）。前者因候选列表枯竭导致探索不足，引入未试候选列表可显著改善表现；后者即便在终轮重测最优候选，也仅能挽回约25%的损失，表明确认行为需贯穿整个实验周期。研究强调，有效策略必须全局性地平衡探索与确认，并公开提供包含冻结数据划分、脚本基线和强化学习环境的Wellocate平台，以评估和训练智能体更优的预算分配策略。
链接: https://arxiv.org/abs/2610.25111
作者: Abdul Karim,Matt Keon
机构: Xstendyl
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 38 pages, 12 figures. Code, data and environment: this https URL
Abstract:As model reasoning gets cheaper, the physical experiment becomes the slow and costly step in discovery. Each experiment returns a noisy reading: a high value can mean a good candidate or a lucky draw. With a fixed budget, an AI agent must keep choosing between exploring a new candidate and confirming a promising one by re-testing it. We measure that choice in Wellocate, a simulated assay campaign with known true values. It is built on a generated landscape and on two real protein datasets, GB1 and PhoQ, scrambled so no model can answer from memory. We test five language models as minimal agents, with no tools and no memory beyond the history in each prompt. They misallocate in both directions. Three over-confirm, spending most of the budget re-testing. On the generated landscape one under-confirms, trusting single readings, and there every model ends with a worse final pick than a basic scripted strategy. Each failure has its own cause. The over-confirmers run short of new candidates worth trying: a list of untried candidates cuts their re-testing and, on real protein data, makes two of them pick much better. A running summary of their readings or a description of the landscape changes it by at most five percentage points. For the under-confirmer, re-testing its top candidates in a final round, replayed on its own campaigns, recovers only about a quarter of what it loses by picking the wrong candidate. This suggests confirmation has to run through the whole campaign. We release Wellocate as a benchmark and reinforcement-learning environment with frozen splits and scripted baselines. With it, anyone can measure how an agent splits its budget before trusting it with real experiments, and train agents to split it better. Code, data and the environment are available at this https URL.
[MA-11] Scam Town: Can Scam Education Spread Like a Vaccine?
【速读】：该论文旨在解决老年人群体在面对诈骗时缺乏有效集体防护机制的问题，即现有诈骗防范研究多聚焦个体层面，忽视了信息传播在社区中的潜在协同保护作用。其核心问题是：能否通过社区内部的口碑传播实现类似疫苗接种的群体免疫效应，从而系统性降低诈骗受害率。解决方案的关键在于验证“预警信息是否可通过人际传播形成扩散效应”，并探索最优传播路径。研究构建了名为Scam Town的模拟社区，利用大语言模型模拟居民行为，发现仅靠自发的口头传播难以形成广泛影响（每条警告平均仅引发少于一次额外传播），但若结合官方机构（如银行、老年中心）发出的权威预警，并由居民进一步传递，则可显著提升防护效果——在模拟中使诈骗发生率降低约一半。关键成功因素在于：确保预警信息首先触达高风险人群（如年长者），并设计易于传播的清晰叙事内容，使公众能够主动复述与分享。由于实验基于语言模型代理而非真实人类，研究结果被视为对现实干预策略的预测，提示未来应设计以可传播性为核心、面向高危群体的诈骗教育方案。
链接: https://arxiv.org/abs/2610.25040
作者: Simiao Ren,Tommy Duong,Xingyu Shen,Jiaqi Gan,Tsang Ng,Ankit Raj,Kidus Zewde,Huaiyu Guo,Siyu Zhang
机构: Scam.ai
类目: ocial and Information Networks (cs.SI); Computers and Society (cs.CY); Multiagent Systems (cs.MA)
备注: 28 pages, 12 figures, 11 tables
Abstract:Most older adults who hear a warning about scams hear it from someone they know: a relative, a neighbour, or a friend who was targeted. Yet research on scam prevention almost always studies one person at a time. We ask whether scam awareness can spread through a community and protect it the way a vaccine does, since a warning protects whoever hears it and can be passed on. To find out, we built Scam Town, a simulated community in which a large language model plays each resident. Scam centres phone the residents, and after a call a resident can warn the people they know. Word of mouth helped, but it did not take off on its own: each warning led to fewer than one more, so a community cannot count on warnings “going viral”. Across 80 simulated towns, everyday talk cut the number of people scammed by roughly a fifth. The largest gains came when an official warning, from a bank, a senior centre, or an agency, reached most of a town and residents then talked about it: in simulation (a combination we did not test live) the two together cut scams by about half. Protection was greatest when warnings reached the people scammers actually call, who here were older residents. Because our residents are language-model stand-ins rather than real people, we treat these findings as predictions to test with people, not as proof. They point to a simple recipe for scam education: give people a clear story they can retell, design campaigns to be passed along, and reach first those most likely to be called.
[MA-12] Same Data Different Bytes: Serialization-Invariant Integrity Checks for Agent -Published Datasets
【速读】：该论文旨在解决智能体网络中数据可信验证的核心问题，即请求智能体如何可靠地确认数据的位置、访问方式以及接收到的数据是否与数据所有者提供的原始数据一致。现有方法（如基于校验和的机制）在面对数据序列化形式变化但语义等价的情况下，容易产生误判，无法区分内容被篡改与合法重编码之间的差异。其关键解决方案是提出一种序列化无关的数据标识方法（serialization-invariant identification method），该方法通过定义显式的等价规则，为在特定规则下语义等价的数据表示分配相同的唯一标识符，从而实现对数据等价性的稳定识别。研究以DataFacts系统为案例，保留原有校验和的同时引入新标识符，在包含两个提供商共六种模型层级的受控实验中验证了该方法的有效性：在等价重编码场景下，该方案相较于普通工具基线和校验和基线分别提升了80.4和70.2个百分点的预期结果达成率；在检测内容篡改方面，也较校验和基线提升了3.5个百分点的证据支持拒收率。研究贡献包括：一套具有明确等价规则的序列化无关标识机制、基于DataFacts的集成实现方案，以及全面的实证评估。
链接: https://arxiv.org/abs/2610.24916
作者: Jin Gao,Maura Clark
机构: Independent Researcher(独立研究员); UOR Foundation (UOR基金会)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 4 pages. Accepted at the NANDA Workshop, IEEE TPS-ISA 2026
Abstract:Agent networks require reliable ways for requesting agents to verify where data is located, how it can be accessed, and whether the data received matches what the data owner made available. Checksum mismatches can reflect either altered content or serialization differences that preserve data equivalence under specified rules. Data identification mechanisms for agents, such as DataFacts, describe where data can be accessed and provide checksums for checking received data against published references. We propose a serialization-invariant identification method that assigns the same identifier to data representations equivalent under explicit rules. We use DataFacts as a case study, retaining its existing checksum alongside the proposed identifier. In a controlled agent study, we compare the proposed configuration with ordinary-tool and checksum baselines across equivalent re-encodings, altered content, and unavailable references. The study includes three model tiers from each of two providers. For equivalent re-encodings, the proposed configuration increased expected-outcome rates by 80.4 percentage points over the ordinary-tool baseline and by 70.2 percentage points over the checksum baseline. The proposed configuration also increased evidence-supported rejection of altered content by 3.5 percentage points over the checksum baseline. We contribute a serialization-invariant identification method with explicit equivalence rules, an integration approach illustrated through DataFacts, and an empirical evaluation against ordinary-tool and checksum baselines.
[MA-13] he One-Pixel of Reasoning : One Truthful Relation Is Enough to Mislead LLM Agents
【速读】：该论文旨在解决如何通过单一语句操控智能体信念的问题，具体聚焦于针对证据与结论之间隐含的推理纽带（implicit warrant）进行精准攻击。其核心解决方案是提出“单关系攻击”（one-relation attack, ORA），该方法首先定位关键的隐含推理纽带，并利用一句基于源信息的、能够对这一纽带进行资格限定（qualification）的句子实施干预，从而引导智能体形成错误信念。实验表明，在18个生物医学推理与漏洞检测模型上，最佳10次采样攻击在专有模型和开源模型上的平均成功率分别达到54.9%和80.0%，且攻击具备无需目标反馈即可迁移的能力；对照实验进一步验证了攻击中“目标定位”与“资格限定”两个关键机制的作用。此外，成功攻击语句可通过源支持检验并规避检测，多智能体辩论测试中团队错误率从50.0%上升至59.5%，凸显了隐含推理纽带作为认知攻击面的重要性。
链接: https://arxiv.org/abs/2610.24873
作者: Jinwei Hu,Youcheng Sun,Yi Dong,Xiaowei Huang
机构: University of Liverpool(利物浦大学); MBZUAI(MBZUAI)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR); Multiagent Systems (cs.MA)
备注: Under Review
Abstract:Can one sentence manipulate an agent’s belief by targeting a single implicit warrant linking evidence to a conclusion? In this paper, we introduce the one-relation attack (ORA), a novel cognitive attack on this inferential relation. ORA localizes a critical implicit warrant and performs a one-relation intervention using a single source-grounded sentence that qualifies this warrant and steers the agent toward an incorrect belief. Across eighteen models on biomedical inference and vulnerability detection, best-of- 10 attacks overturn initially correct answers at mean rates of 54.9% for proprietary and 80.0% for open-weight models. Attacks also transfer without victim feedback, and paired controls support the roles of targeting and qualification. Moreover, successful sentences can pass source-support checks and evade identification, and multi-agent debate raises team error from 50.0% to 59.5% in our balanced stress test. These findings establish implicit warrants as a cognitive attack surface for manipulating agents’ beliefs.
[MA-14] AgentS oV: Virtual Sensing for LLM Multi-Agent Systems via Stream of Variation
【速读】：该论文旨在解决大规模语言模型多智能体系统（Large Language Model Multi-Agent Systems, LLM-MAS）在执行复杂任务时可靠性难以分析的问题。由于系统故障不仅源于单个智能体的失效，还受信息与错误在系统中传播路径的影响，现有方法多采用事后故障归因，缺乏对异构多智能体架构的通用建模能力，且无法支持对信息影响和可观测性的全面分析。为此，论文提出AgentSoV——一种基于变异流（Stream of Variation, SoV）的虚拟传感框架，其核心在于将多智能体系统的执行过程建模为随机动态系统，通过将异构的执行轨迹映射至诊断阶段、从可观测输出构造虚拟测量，并利用状态空间模型学习跨阶段的变异传播规律。该方法在阶段依赖关系上具有架构感知性，但在分析语言上保持架构无关性，可统一适用于链式结构、编排型及有向无环图（DAG）结构的多智能体系统。无需重新执行系统，AgentSoV支持三项系统级分析：故障归因以定位异常变异的进入点；阶段输入影响度量以量化中间输入对下游结果的作用强度；以及在监控预算约束下优化虚拟传感器部署以获取互补观测。实验表明，AgentSoV实现了故障定位、信息影响分析与可观测性评估的统一框架，将智能体执行过程从单纯的文本审视重构为动态系统建模，为多智能体系统结构设计与组件优化提供了深层次洞察。
链接: https://arxiv.org/abs/2610.24845
作者: Congjing Zhang,Siyuan Xu,Yunfei Bai,Ke Ren,Xiaoning Qian,Shuai Huang
机构: University of Washington (华盛顿大学); Texas A&M University (德克萨斯农工大学); Amazon(亚马逊)
类目: Multiagent Systems (cs.MA); Systems and Control (eess.SY)
备注:
Abstract:Large language model (LLM) multi-agent systems (MASs) solve complex tasks through interacting agents, yet their reliability is difficult to analyze because failures depend on both individual agents and how information and errors propagate across the system. Existing approaches largely treat this as post-hoc failure attribution, but do not provide a common model across heterogeneous MAS architectures or support broader analysis of information influence and observability. We introduce AgentSoV, a virtual-sensing framework that models MAS execution as a stochastic dynamic system through Stream of Variation (SoV). AgentSoV maps heterogeneous trajectories into diagnostic stages, constructs virtual measurements from observable outputs, and learns cross-stage variation propagation with a state-space model. Its representation is architecture-aware in stage dependencies yet architecture-agnostic in analytical language, supporting chain-structured, orchestration-based, and directed acyclic graph (DAG)-structured MASs. Without re-executing the MAS, AgentSoV supports three system-level analyses: failure attribution identifies where abnormal variation enters; stage-input influence quantifies how strongly intermediate inputs affect downstream outcomes; and virtual-sensor placement selects complementary observations under a monitoring budget. Experiments across multiple MAS architectures demonstrate a unified approach to fault localization, information influence, and observability. AgentSoV reframes agentic execution as a dynamic system to model rather than merely text to inspect, providing insights for MAS structure and component design.
[MA-15] Measuring Consensus in AI Agent Networks
【速读】：该论文旨在解决多智能体系统在复杂协作任务中因信息分散、观点分歧导致的共识形成机制不明确的问题，尤其关注智能体在网络化环境中如何通过协商达成一致决策。其核心挑战在于缺乏对智能体间观点演化过程及支撑决策信息的系统性评估方法。解决方案的关键在于提出一个基于通信历史演化的评估框架，将共识形成建模为智能体公开立场随时间向共享区域收敛的过程；该框架通过追踪立场轨迹（position trajectories）、群体离散度（group dispersion）以及公开信息表达内容，量化分析智能体间的认知演化与信息交互质量。实验结果表明，共享解释性信息显著提升群体决策准确性，而即使整体答案准确，仍存在持续的个体分歧，揭示了决策准确性与有限分歧共存的现象。此外，通过引入成员变更实验，框架进一步区分了智能体间观点收敛程度与任务最优解逼近程度之间的差异，从而将决策准确性与有限分歧与交流过程中可获取的信息量关联起来，为评估生成式智能体系统的协作能力提供了可量化的分析工具。
链接: https://arxiv.org/abs/2610.24833
作者: Jin Gao
机构: Independent Researcher(独立研究员)
类目: Multiagent Systems (cs.MA)
备注: 4 pages. Accepted at the NANDA Workshop, IEEE TPS-ISA 2026
Abstract:As AI agents become increasingly connected through networks and take on tasks that require negotiation and joint decision-making, a systematic approach is needed to assess how their views converge and what information supports their decisions. We present an evaluation framework that treats consensus formation as the evolution of publicly stated positions toward a shared region, using accumulated communication history as context. The framework tracks position trajectories, group dispersion, and information expressed in public messages. Two case studies examine staffing decisions based on distributed facts and decisions requiring agents to reconcile duplicate information and a cancelled contract. Results from 1,872 completed runs associate sharing explanations with more accurate group decisions, while the second task reveals persistent disagreement despite accurate aggregate answers. Membership-change experiments provide a complementary evaluation. By separating convergence among agents from progress toward a task’s full-information answer, the framework relates decision accuracy and bounded disagreement to the information available during the exchange.
[MA-16] Cooperation Without Shared Rewards via Reciprocal Q-Augmentation
【速读】：该论文旨在解决非零和博弈（general-sum games）中因个体间存在欺骗或剥削行为而导致的协作困难问题。在社会困境中，即使单一背叛行为也可能使系统陷入次优的纳什均衡，从而损害整体社会福利。传统方法常依赖对其他智能体奖励或内部状态的访问，这在现实中往往不切实际，且隐含引入了额外的合作假设。为此，本文提出互惠式Q值增强（Reciprocal Q-Augmentation, RQA），其核心在于构建一个潜在变量对手模型，用于估计自身行动对其他智能体的影响，并将该估计信息融入自身的Q函数。该方法仅需利用自身Q函数即可学习合作与背叛策略，无需共享奖励或状态信息。实验结果表明，在迭代囚徒困境、旅行者困境、猎鹿博弈及CleanUp-Lite等任务中，所有采用RQA的智能体能更快达成更高水平的合作，且单个RQA智能体可通过适时切换合作与背叛策略，有效引导标准强化学习算法实现合作，显著提升社会福利。RQA的关键创新在于通过内在的反向影响建模实现无需外部协作信号的高效协同。
链接: https://arxiv.org/abs/2610.24793
作者: Francesco Morri,Adel Abusitta,Quentin Cappart,Margarida Carvalho
机构: Polytechnique Montréal(蒙特利尔工程学院); Mila(蒙特利尔人工智能研究所); Université de Montréal(蒙特利尔大学); UCLouvain(鲁汶大学)
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:General-sum games are more challenging than fully cooperative or adversarial settings because agents may deceive or exploit one another. Achieving high social welfare requires sustained cooperation, while a single defection can drive agents toward a welfare suboptimal Nash equilibrium. Moreover, assuming access to other agents’ rewards or internal states is often unrealistic and implicitly introduces additional cooperation into the setting. To address this issue, we introduce \textitReciprocal Q-Augmentation (RQA), a method that uses a latent-variable opponent model to augment the Q-function with estimates of how one’s actions affects other agents. RQA allows agents to learn both cooperative and defective policies using only their own Q-function. We show that, in social dilemmas, agents that all follow the cooperative policy of RQA learn to cooperate faster and reach higher levels of cooperation than fully cooperative algorithms. Moreover, a single RQA agent that switches appropriately between its cooperative and defecting policies can induce cooperation in agents trained with standard RL algorithms, which would otherwise end up defecting. We evaluate our approach on the Iterated Prisoner’s Dilemma, Iterated Traveler’s Dilemma, Iterated Stag Hunt, and \textttCleanUp-Lite. Our experiments show that the algorithms implementing RQA outperforms fully cooperative methods with shared rewards and enables cooperation with algorithms that otherwise fail to reach high social welfare.
[MA-17] Neuro-Inspired Inhibitory Message Passing for Accurate and Token-Efficient Multi-Agent Reasoning
【速读】：该论文旨在解决多智能体系统（multi-agent systems）在协作过程中因信息冗余导致关键证据被淹没的问题。当多个智能体之间持续交换消息时，重复信息会占据上下文空间，从而稀释稀有但决定性的证据，影响推理准确性。其解决方案的关键在于提出“抑制性消息传递”（inhibitory message passing）机制，该机制受皮层神经回路中抑制作用的启发：通过选择性抑制共享输入，同时保留对弱信号的敏感性。该方法无需训练即可实现，能够在每轮通信中仅向尚未拥有特定证据的接收者转发一次该证据，并排除发送者的推理过程与答案，有效避免冗余传播。实验表明，在三个基准测试中，相较于标准消息传递方式，准确率提升39–77个百分点，单次运行的令牌消耗降低10–46%。该方法在涵盖4亿至320亿参数的七种开源模型及两种商用模型、最多达64个智能体的群体中均表现优异，适用于点对点、编排-工作模式以及多层次层级等多种架构，验证了其在多智能体推理中的可扩展性与高效性。
链接: https://arxiv.org/abs/2610.24718
作者: Nima H. Siboni,Magdalena Springer,Vahid Rostami
机构: University of Cologne (科隆大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Multi-agent systems promise to solve problems by bringing together information beyond the reach of any individual agent. Retrieval and tool use can distribute this information across agents, making communication essential to collaboration. A central challenge is keeping crucial evidence visible in each agent’s context as messages accumulate. Repeated information can crowd out rare, decisive evidence. Models of cortical circuits suggest a principle for addressing this imbalance: inhibition can selectively suppress shared inputs while preserving sensitivity to weaker signals. We adapt this principle into inhibitory message passing, a training-free communication router that cancels redundant evidence separately for each evidence sentence and receiver. Each round, it forwards evidence sentences restated in agents’ messages, delivering each at most once to receivers whose own evidence lacks it and excluding the senders’ reasoning and answers. Against standard message passing, accuracy in peer groups rose by 39-77 percentage points across three benchmarks, and tokens per run fell by 10-46%. Gains held across seven open models spanning four to thirty-two billion parameters and two commercial models, in groups up to sixty-four agents, and across multi-agent architectures including peer-to-peer, orchestrator-worker, and multilevel hierarchical organizations. These findings support a scalable design pattern for more accurate and token-efficient multi-agent reasoning.
[MA-18] CARVS: A Congestion-Aware Algorithm for Online Multi-Agent Pickup-and-Delivery in Human-Collaborative Warehouses
【速读】：该论文旨在解决智能仓储系统中人机协作带来的动态不确定性问题，特别是人类行为不可预测性导致的路径阻塞与实时响应挑战。传统多智能体取送货（MAPD）算法在面对人类干扰时表现不佳，难以适应动态环境。为此，论文提出了多智能体取送货与人机协作（MAPDwHC）问题，引入了因机器人与人类交互导致的临时路径阻塞等关键约束，并强调实时响应能力。解决方案的核心是提出CARVS算法，其通过任务交换、有效性验证及随机初始化机制，协同优化任务分配与路径规划，在降低拥堵、提升对人机交互的适应性的同时，保障系统实时性。仿真结果表明，相较于传统MAPD算法，CARVS在大规模场景下显著减少路径规划超时，有效提升了系统的效率与可扩展性。
链接: https://arxiv.org/abs/2610.24549
作者: Chaoran Li,Zeyuan Zhao,Shao Zhang
机构: Shanghai Jiao Tong University (上海交通大学); Shanghai Artificial Intelligence Laboratory (上海人工智能实验室)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: 13 pages, 8 figures. Accepted at DAI 2026 (Oral)
Abstract:Intelligent warehouse systems have transformed logistics by integrating robots with minimal infrastructure changes, boosting efficiency and optimizing workflows compared to traditional human-only operations. However, human-robot collaboration introduces new challenges, particularly due to human uncertainty, which leads to dynamic interactions and unpredictable behavior. To address these complexities, this paper presents the Multi-Agent Pickup-and-Delivery with Human Collaboration (MAPDwHC) problem, an extension of the Multi-Agent Pickup-and-Delivery (MAPD) problem that incorporates human unpredictability. This problem adds key constraints, such as dynamic obstacles caused by robots temporarily blocking paths when interacting with humans, and the need for real-time responsiveness. We formalize the MAPDwHC problem with additional constraints and properties. Based on this problem setting, we propose the CARVS algorithm, which improves task assignment and path planning by integrating task swapping, validation checks, and random initialization. CARVS reduces congestion, enhances adaptability to human-robot interactions, and ensures real-time responsiveness. Simulation results show that CARVS significantly reduces timeouts of path planning in large-scale scenarios where traditional MAPD algorithms fail, improving both efficiency and scalability.
[MA-19] Low-Bit Distributed Constrained Optimization with Application to Optimal Scheduling and Allocation
【速读】：该论文旨在解决分布式约束优化中的通信效率与算法收敛性之间的矛盾问题，特别是在资源分配和调度场景中，如何在保证全局约束满足的前提下，实现极低比特通信开销的高效协同优化。其核心挑战在于：传统分布式优化算法通常需要多比特信息交换（如完整的梯度值），导致通信成本高昂；而现有单比特协议往往难以保证收敛性或最优性。本文提出的关键解决方案是基于拉普拉斯-梯度动力学设计一种仅需每条有向边每轮传输单比特（即本地代价梯度符号）的分布式算法，该方案通过引入非线性映射机制，在保持等式约束始终满足的前提下，实现了任意时刻的可行性。其关键创新点在于：第一，采用李雅普诺夫分析证明了固定步长下迭代序列在 O(1/ε2) 次迭代内进入以最优解为中心、半径为 O(τ)（决策变量）和 O(τ2)（代价残差）的邻域，且步长上界为仅依赖于图拉普拉斯谱和代价曲率的时不变网络常数，无需自适应调整；第二，放宽了对通信图的假设，首次在仅要求B-连通且权衡平衡的有向图下刻画了系统的不变集，并揭示任意奇数性、保号非线性函数均具有相同的不变集与下降结构，而单比特协议正是这一更广泛非线性类中通信最小化的极端实例；第三，通过光滑惩罚松弛处理局部盒约束，明确量化了惩罚权重与步长上界及最优性间隙之间的关系。仿真结果表明，在经济调度（1000台发电机）和CPU调度（500台服务器）任务中，该单比特协议相较多比特基线可降低5–60倍的总通信成本，仅以可调且显式界定的最优性间隙为代价；实验还发现递减步长策略可消除该间隙，但其严格收敛性分析仍为开放问题。
链接: https://arxiv.org/abs/2610.24431
作者: Mohammadreza Doostmohammadian,Sergio Pequito
机构: 未知
类目: ystems and Control (eess.SY); Distributed, Parallel, and Cluster Computing (cs.DC); Multiagent Systems (cs.MA); Optimization and Control (math.OC)
备注: Nonlinear Analysis: Hybrid Systems
Abstract:This paper proposes a distributed resource allocation algorithm in which each agent communicates only the sign of its local cost gradient, a single bit per directed link per iteration. The algorithm is based on Laplacian-gradient dynamics and preserves the equality (demand) constraint at every iterate, guaranteeing anytime feasibility irrespective of termination time. First, we provide a complete Lyapunov-based convergence analysis for the fixed-step-size regime: the iterates are shown to enter, in at most O(1/\varepsilon^2) iterations, a neighborhood of the optimizer whose radius scales as O(\tau) in the decision variable and O(\tau^2) in the cost residual, where \tau is the step-size. Crucially, the step-size bound is a time-invariant network constant (depending only on the graph Laplacian spectrum and the cost curvature bounds), so no adaptive tuning is required. Second, we characterize the invariant set under B-connected, weight-balanced digraphs, strictly weaker than the all-time connectivity and doubly-stochastic weights assumed in most prior work, and show that any odd, sign-preserving nonlinearity yields the same invariant-set and descent structure, with the single-bit protocol being the extreme communication-minimizing instance of a broader nonlinear class. Third, we address local box constraints via smooth penalty relaxation and quantify explicitly how the penalty weight interacts with the step-size bound and the optimality gap. Simulations on economic dispatch (1000 generators) and CPU scheduling (500 servers) demonstrate that the single-bit protocol achieves 5-60 times reduction in total communication cost relative to multi-bit baselines, at the price of a tunable and explicitly bounded optimality gap. A diminishing step-size schedule is shown empirically to eliminate this gap; its formal convergence analysis is identified as an open problem. Comments: Nonlinear Analysis: Hybrid Systems Subjects: Systems and Control (eess.SY); Distributed, Parallel, and Cluster Computing (cs.DC); Multiagent Systems (cs.MA); Optimization and Control (math.OC) Cite as: arXiv:2610.24431 [eess.SY] (or arXiv:2610.24431v1 [eess.SY] for this version) https://doi.org/10.48550/arXiv.2610.24431 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Mohammadreza Doostmohammadian [view email] [v1] Sat, 17 Oct 2026 03:49:52 UTC (746 KB) Full-text links: Access Paper: View a PDF of the paper titled Low-Bit Distributed Constrained Optimization with Application to Optimal Scheduling and Allocation, by Mohammadreza Doostmohammadian and 1 other authorsView PDFHTML (experimental)TeX Source view license Current browse context: eess.SY prev | next new | recent | 2026-10 Change to browse by: cs cs.DC cs.MA cs.SY eess math math.OC References Citations NASA ADSGoogle Scholar Semantic Scholar export BibTeX citation Loading… BibTeX formatted citation loading… Data provided by: Bookmark checked="checked"class=“labs-tab-input”> Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv’s community? Learn more about arXivLabs. Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?) mathjaxToggle(); We gratefully acknowledge support from our major funders, member institutions, , and all contributors. About Help Contact Subscribe Copyright Privacy Accessibility Operational Status (opens in new tab) Major funding support from
[MA-20] Adding Heavy-Ball Momentum to Distributed Constrained Optimization for Energy Resource Management Subject to Time-Delays and Quantization
【速读】：该论文旨在解决分布式优化中存在通信延迟、固定对数量化及非连续量化解耦误差时，如何保证系统在迭代过程中维持供需平衡（aggregate supply-demand equality）并实现稳定收敛的问题。其核心挑战在于处理由量化与延迟共同引入的非连续扰动，同时确保优化算法在受限通信条件下仍具备可分析的稳定性与收敛性。解决方案的关键在于采用基于李雅普诺夫-克拉索夫斯基（Lyapunov-Krasovskii）的分析框架，将离散量化误差建模为可叠加的扰动项，并通过构造一个可计算的不变子水平集（invariant-sublevel argument）来证明系统的实用稳定性（practical stability）。该方法在静态连通无向图和有界延迟条件下，实现了关于量化分辨率呈二次关系的遍历残差界（ergodic residual bound），并在无量化情形下恢复对平滑惩罚型等式约束最优解的精确收敛。此外，针对无延迟、无量化且代价函数为二次的情形，通过精确的模态分析（modal analysis）确定了重球（heavy-ball）参数的稳定区域及最优配置，其预测的收敛速率在固定稀疏网络上得到验证。研究还提供可复现的延迟实验，报告了所有常数及一个保证稳态水平低于初始值的证书，进一步增强了结果的可信度。最后，通过混合发电机组-储能系统实验，直接验证了符号化更新（signed update）的实际可行性。值得注意的是，有限光滑惩罚项仅近似满足局部盒约束（box constraints），但通常无法严格强制执行。
链接: https://arxiv.org/abs/2610.24421
作者: Mohammadreza Doostmohammadian,Sergio Pequito
机构: 未知
类目: ystems and Control (eess.SY); Distributed, Parallel, and Cluster Computing (cs.DC); Multiagent Systems (cs.MA); Dynamical Systems (math.DS); Optimization and Control (math.OC)
备注: Nonlinear Analysis: Hybrid Systems
Abstract:This paper studies a hybrid discrete-time signed gradient-consensus method with heavy-ball momentum, paired communication delays, and fixed logarithmic quantization. The method preserves the aggregate supply-demand equality at every iteration. For a static connected undirected graph and bounded delays, a Lyapunov-Krasovskii analysis establishes practical stability by isolating the discontinuous quantization error as an additive disturbance and closing a computable invariant-sublevel argument. The resulting ergodic residual bound is quadratic in the quantization resolution, and exact convergence to the smooth penalized equality-constrained optimizer is recovered in the unquantized case. For delay-free unquantized quadratic costs, an exact modal analysis identifies the heavy-ball stability region and optimal parameters, with the predicted rate validated on a fixed sparse network. A reproducible delayed experiment reports all constants and a certificate whose guaranteed steady-state level lies below the reported initialization. A separate mixed generator-battery experiment exercises the signed update directly. Finite smooth penalties approximate, but do not generally enforce, the local box constraints exactly.
[MA-21] IF-MAP: Influence-Function-Inspired Agent Prioritization in Multi-Agent Systems
【速读】：该论文旨在解决大语言模型（LLM）驱动的多智能体系统中，受限于资源条件下如何高效识别最具改进潜力的智能体以实现最大系统级性能提升的问题。现有方法依赖失败归因来判断智能体责任，但其责任程度并不能直接反映改进后所能带来的系统级增益。为此，本文提出一种受影响函数（influence function）启发的方法——IF-MAP，其核心在于通过在低维提示嵌入空间中对受控语义提示扰动下的端到端损失变化进行建模，构建正则化的局部响应模型，并利用估计的响应范数作为优先级评分。该方法为黑盒操作，无需访问模型参数或梯度，可有效生成智能体优先级序列，指导后续的提示优化或模型升级。实验在涵盖代码生成、问答和数学推理的六类多智能体工作流上验证了IF-MAP的有效性：基于其优先级进行提示优化，相较均匀选择平均提升3.81个百分点，而均匀选择仅达1.85；针对五种目标模型的智能体升级，平均提升达7.62–16.42个百分点，显著优于均匀选择的3.81–8.62。此外，相较于全量优化或升级，IF-MAP分别将Qwen3-8B和DeepSeek-V3.2的令牌消耗降低74.2%和35.2%，目标模型令牌使用减少63.6%–64.9%，证明其在保持高性能提升的同时大幅降低计算开销。因此，IF-MAP的关键在于通过端到端损失敏感性分析，精准量化各智能体对系统整体表现的影响，从而实现资源最优配置。
链接: https://arxiv.org/abs/2610.24383
作者: Jie Mao,Pingchen Lu,Zikun Qu,Jiahang Sun,Xiangyi Wang,Zhiwei Shang,Zhiyong Wang,Guojun Xiong,Zhongxiang Dai
机构: The Chinese University of Hong Kong, Shenzhen; Harbin Institute of Technology; School of Engineering and Applied Sciences, Harvard University
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Improving every agent in an LLM-based multi-agent system can be prohibitively costly, making it important to identify which agent should be prioritized to achieve the greatest system-level gain with limited resources. Failure attribution can inform this decision, but an agent’s responsibility for failures does not directly reveal the system-level gains achievable by improving it. We therefore introduce IF-MAP, an influence-function-inspired method for identifying agents with high potential for system-level improvement. For each agent, IF-MAP measures end-to-end loss changes under controlled semantic prompt perturbations, fits a regularized local response model in a low-dimensional prompt-embedding subspace, and uses the estimated response norm as a priority score. This black-box procedure requires no access to model parameters or gradients and produces agent priorities that can guide subsequent improvement effort. We evaluate IF-MAP across six multi-agent workflows spanning code generation, question answering, and mathematical reasoning, using the same priorities for prompt optimization and model upgrades. Optimizing the selected agents’ prompts achieves a macro-average gain of 3.81 percentage points, compared with 1.85 for uniform agent selection. Across five target models, upgrading the selected agents yields macro-average gains of 7.62-16.42 percentage points, compared with 3.81-8.62 for uniform agent selection. Despite its probing overhead, IF-MAP reduces Qwen3-8B and DeepSeek-V3.2 token use by 74.2% and 35.2%, respectively, relative to optimizing every agent, and reduces target-model tokens by 63.6-64.9% relative to upgrading every agent. These results support the use of IF-MAP to identify promising agents and direct improvement effort toward greater end-to-end performance gains.
[MA-22] ROAR: Benchmarking LMM Agents under Real-Time Constraints in Multiplayer Action-Shooter Games NEURIPS2026
【速读】：该论文旨在解决当前大型多模态模型（Large Multimodal Models, LMMs）在多智能体环境评估中普遍存在的“回合制”基准测试所导致的实时性缺陷问题。现有评估框架通常要求环境与其他智能体在智能体进行推理时暂停，这忽略了真实世界中环境持续动态演化的现实，使得延迟决策可能在执行前已过时，从而无法有效衡量LMM在真实时间约束下的决策能力。为应对这一挑战，论文提出ROAR——一个面向多智能体场景的实时动作射击游戏环境，涵盖合作、竞争、混合型及单智能体等多种情境。其核心解决方案在于设计了一种多智能体异步Gym接口（Multi-Agent Asynchronous Gym interface），通过将环境演化与智能体推理分别在独立线程中运行，实现了环境以固定时间间隔持续推进，且不依赖于LMM的推理耗时，同时各智能体可独立异步推理而互不阻塞。该机制突破了传统同步框架的瓶颈，使评估能够真实反映LMM在动态、实时多智能体系统中的表现，为推动LMM从静态回合制评估迈向实时多智能体决策研究提供了可扩展、可定制且具备多模态感知能力的实践平台。
链接: https://arxiv.org/abs/2610.24282
作者: Dawei Wang,Di Zhao,Jianran Yu,Chengming Zhou,Marci Chi Ma,Xiaoyang Liu,Beining Zhang,Shuliang Chu,Yun Sing Koh,Gary Ushaw,Richard Davison
机构: Newcastle University (纽卡斯尔大学); University of Auckland (奥克兰大学); CTW (CTW)
类目: Multiagent Systems (cs.MA)
备注: NeurIPS 2026
Abstract:Large multimodal models (LMMs) are increasingly deployed as agents that perceive, reason, and act in interactive multi-agent environments. Yet most existing benchmarks for LMM-based agents in multi-agent settings remain turn-based: the environment and other agents pause while an agent performs inference. Such protocols obscure a fundamental challenge of real-time constraints, where the world continues to evolve and delayed decisions may become outdated before execution. Motivated by this limitation, we introduce ROAR, a multiplayer action-shooter game environment for evaluating LMM-based agents in multi-agent settings under real-time constraints. ROAR covers cooperative, competitive, mixed cooperative-competitive, and basic single-agent scenarios. We further propose a Multi-Agent Asynchronous Gym interface that decouples environment evolution from agent reasoning by running the environment and each agent in separate threads, enabling the environment to advance at fixed time intervals without waiting for LMM inference and allowing agents to reason independently without blocking one another. ROAR also features convenient deployment, multimodal observations, and customizable levels. ROAR offers a practical testbed for advancing LMM-based agents beyond static, turn-based evaluation toward real-time multi-agent decision-making.
[MA-23] Model Capacity Frontier: A Development Framework for LLM -Supported Social Simulation
【速读】：该论文旨在解决生成式 AI（Generative AI）支持的基于主体模型（Agent-Based Models, ABMs）在模拟人类集体行为时面临的配置敏感性与架构局限性问题。由于大语言模型（Large Language Model, LLM）引入了开放式的建模设计空间，微小的配置选择（如提示工程和环境耦合方式）可能显著影响模拟结果，导致模型性能不稳定且难以评估。为此，本文提出“模型能力边界”（Model Capacity Frontier, MCF）框架，通过系统性地探索特定模型架构在可接受配置范围内的多目标性能表现，以估计其帕累托前沿（Pareto frontier），从而刻画该架构所能达到的最优实证性能边界。研究以约12,000个主体的大学毕业典礼人群仿真为例，利用时间占用率、空间分布与流动方向等实证运动数据进行验证。结果显示，初始阶段的LLM集成虽提升了流动方向的拟合度，但恶化了时间占用与空间分布表现；经对LLM层的后续架构优化后，时间占用性能得以恢复，空间分布性能亦被重建，同时保留了对流动方向的改进优势。MCF框架因此为在高度开放、配置敏感的设计空间中系统性推进LLM支持型ABM的架构演进提供了可量化的实证目标与评估路径。
链接: https://arxiv.org/abs/2610.24168
作者: Patrick Zhao,Yuxuan Li,Hirokazu Shirado
机构: Simon Fraser University (西蒙菲莎大学); Carnegie Mellon University (卡内基梅隆大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Large language model (LLM) agents are increasingly incorporated into agent-based models (ABMs) on the premise that richer, context-sensitive human behavior can improve empirical fidelity. However, LLM-supported ABMs introduce an open-ended design space in which subtle modeling and configuration choices, such as prompt design and environment coupling, can substantially alter outcomes. Developing LLM-supported ABMs to simulate human collective behavior therefore requires navigating configuration sensitivity while identifying the limitations of the current model architecture. In this paper, we introduce the \textitModel Capacity Frontier (MCF), a framework that characterizes the empirical performance attainable by a specified architecture across its admissible configurations. In multi-objective settings, this capacity is represented by an estimated Pareto frontier obtained through systematic configuration exploration. We demonstrate MCF in crowd simulation of university commencement events with approximately 12,000 agents, using empirical movement data on temporal occupancy, spatial distribution, and flow direction. We first characterize the capacity of a conventional crowd model and then evaluate successive LLM-supported architectural extensions. In the process, MCF reveals which dimensions of attainable performance expand and which remain limited, providing empirical targets for refinement and re-evaluation. Initial LLM integration improves flow direction but degrades temporal occupancy and spatial distribution; subsequent architectural refinement of the LLM layer improves temporal occupancy and recovers spatial performance, while retaining much of the earlier gain in flow direction. Our case study shows how MCF guides the systematic architectural development of LLM-supported ABMs in configuration-sensitive, open-ended design spaces.
[MA-24] Preservation and loss of algorithmic structure in experimental diffusion of joint-action norms
【速读】：该论文试图解决的问题是：在社会规范的跨群体传播过程中，其背后隐含的协作算法（joint plans）如何演化，尤其是复杂性在文化演化中的选择压力机制尚不明确。其解决方案的关键在于通过大规模实验研究人类参与者对新型社会规范的扩散过程，并结合程序合成（program synthesis）技术，定量分析个体在独立群体中自发构建的交互规则。研究发现，在群体间互动初期，复杂的算法倾向于被简化甚至丢失；然而，互动修复（interactive repair）机制的存在能够有效缓解这一简化压力，从而在规范错配的情况下仍能保留复杂算法。这表明，互动推理（interactive reasoning）在文化演化过程中对程序性规范的结构复杂性起到了关键的调节作用，是决定复杂算法能否存续的核心机制。
链接: https://arxiv.org/abs/2610.24150
作者: Dhara Yu,Bill D. Thompson
机构: University of California, Berkeley(加利福尼亚大学伯克利分校)
类目: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Everyday norms of social interaction - e.g. norms of sharing space, taking turns, or queuing - are essential to large societies, offering structured procedures for common interactions. However, it remains unclear how the underlying algorithms of these implicit joint plans evolve as they diffuse through populations. Here we show that this process of diffusion can select against complex algorithms in the earliest stages. We conducted a large-scale study of experimental diffusion of novel social norms among human participants, leveraging advances in program synthesis to characterize the procedures people developed in independent populations. Inherited procedures became structurally simplified and lost during interactions between participants who inherited different variants of joint-action norms; however, opportunities for interactive repair offset this pressure, preserving complex algorithms despite misalignment. These results demonstrate that interactive reasoning mediates selective pressures on structural complexity during cultural evolution of procedural norms.
[MA-25] he Weight of Teamwork: Does Communication Structure Help LLM Scientists Search for Scientific Laws? FAST NEURIPS2026
【速读】：该论文旨在解决团队性能评估中将多阶段生成过程简化为单一评分所导致的表征失真问题，尤其关注在生成式人工智能（Generative AI）系统中，候选答案生成与最终选择之间的分离对整体表现的影响。其核心解决方案在于通过符号回归这一可观测的科学搜索任务，系统比较不同通信架构（如独立、环形、星型、全连接）下固定预算的大型语言模型（LLM）团队，重点分析其生成的候选解档案（frozen candidate archives）与由统一选择规则选出的最优表达式之间的差异。研究发现，将计算预算分配给更多独立轨迹能带来最大性能提升（在最宽泛分配下，平均保留拟合得分Q提高0.70）；而当采用统一的选择规则时，环形（Ring）策略相较独立（Independent）策略使平均Q下降0.33，星型（Star）和全连接（Complete）结构则未呈现明确排序。更重要的是，在固定候选档案的前提下，仅改变最终交付规则即可显著影响架构间的相对表现差距，甚至改变全连接结构点估计值的符号。这表明，架构性能排名并非由通信拓扑本身决定，而是高度依赖于最终选择机制的设计。因此，论文强调应将候选生成与最终选择视为两个独立的评估端点，以避免因选择规则模糊而导致的误导性结论。
链接: https://arxiv.org/abs/2610.24131
作者: Jianing Shi,Zaibin Zhang,Shijun Lei,Zhenfei Yin
机构: The University of Chicago (芝加哥大学); Dalian University of Technology (大连理工大学); Northwestern Polytechnical University (西北工业大学); University of Oxford (牛津大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at the NeurIPS 2026 Workshop on Foundations of Agentic Systems Theory (FAST), non-archival. 7 pages main text, 33 pages total with appendices
Abstract:Team performance is often reported as one delivered score, even though that score is produced in two stages: a system generates candidate answers and a selection rule returns one of them. We use symbolic regression as an observable scientific-search setting and compare fixed-budget LLM teams that differ in peer visibility and update timing, evaluating both their frozen candidate archives and the expressions selected from those archives. Distributing the call budget across more independent trajectories produced the largest measured gain, improving mean held-out fitting score Q by 0.70 at the widest tested allocation. Under a common archive-wide selector, the implemented Ring policy reduced mean Q by 0.33 relative to Independent, while Star and Complete remained unresolved. Holding the generated archives fixed, changing the delivery rule changed the measured architecture gaps, including the sign of the Complete point estimate. These results support reporting candidate generation and final selection as separate endpoints. They do not identify a communication mechanism or a universally preferred topology; rather, they show that an architecture ranking is not identified by communication structure alone when the rule that returns the final answer is left unspecified.
[MA-26] Coopshift: Post-Training Amplifies The Memory Curse In LLM Cooperation
【速读】：该论文旨在探究后训练（post-training）对重复交互中合作行为的影响，具体聚焦于在七种已发布的模型系列中，预训练模型（Base）、监督微调（SFT）及后续偏好或奖励优化（Aligned）版本在七类社会困境任务中的合作表现变化。其核心问题是：随着模型经过不同阶段的后训练，其在重复博弈中是否更倾向于维持合作，以及这一趋势如何受历史可见范围（visible history）和收益结构的影响。解决方案的关键在于通过系统性对比不同训练阶段的模型在200轮重复交互中的表现，发现“记忆诅咒”（memory curse）——即当智能体能够观察到更长的历史记录时，合作水平反而下降的现象，并进一步揭示后训练显著放大了这一负面效应。研究发现，在六种二元动作博弈中，随着可见历史长度从两轮扩展至四十轮，所有后训练阶段（SFT与Aligned）的合作率均持续下降，且降幅普遍超过基线模型（Base），其中大部分放大效应已在SFT阶段即已显现；在长期历史条件下，后训练模型的整体合作水平甚至低于基线。此外，后训练模型在对称二元博弈中表现出更陡峭的合作衰减趋势，尤其在单轮合作成本上升、收益下降或非合作更具激励时更为明显。值得注意的是，Aligned与SFT模型在重复互动中的合作差异会随情境变化：当指令强调最终回合或存在未来互动可能性时，其基于相同历史记录的决策行为呈现显著分化。因此，该研究提出，评估后训练模型的合作能力不应局限于单次决策，而应深入考察其在重复交互中合作的可持续性及其对收益结构与历史可见性的依赖关系。
链接: https://arxiv.org/abs/2610.24086
作者: Jiayuan Liu,Shiyi Du,Tianqin Li,Emanuel Tewolde,Keyang Xuan,Vincent Conitzer
机构: Carnegie Mellon University (卡内基梅隆大学); Foundations of Cooperative AI Lab (FOCAL) (合作AI基础实验室); The University of Texas at Austin (德克萨斯大学奥斯汀分校)
类目: Computer Science and Game Theory (cs.GT); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:How does post-training change cooperation during repeated interaction? Within each of seven released model lineages, we compare pretrained (Base), supervised fine-tuned (SFT), and subsequent preference- or reward-optimized (Aligned) checkpoints across seven social dilemmas. Copies of each checkpoint play one another for 200 rounds, with visible history limited to the last two or forty rounds. The memory curse refers to lower cooperation when agents can see more past rounds. Our central finding is that post-training amplifies this decline. Averaged across the six binary-action games, expanding visible history reduces cooperation at every SFT and Aligned checkpoint. At each post-trained stage, these declines exceed Base’s in six of seven lineages. Most of this average amplification is already present at SFT. With long history, both post-trained stages also cooperate less than Base on average in each binary-action game. In the symmetric binary games, their fitted cooperation rates fall more steeply on average than Base’s as per-round cooperation costs rise, benefits fall, or noncooperation becomes more rewarding. The difference in cooperation rates between Aligned and SFT also changes from one-shot decisions to repeated play. When given the same past records, their choices depend on whether the instructions describe a final round or possible future interaction. Post-training evaluation should therefore go beyond isolated choices to assess whether cooperation is sustained in repeated interaction and how it depends on payoffs and visible history.
[MA-27] Alliance of Stochastic Routing and Reinforcement Fusion for Multi-Model Reasoning
【速读】：该论文旨在解决多模态场景下不确定性环境中的动态模型路由问题，即系统需在面对查询时选择并可能组合多个模型以生成答案，而传统强化学习（Reinforcement Learning, RL）方法常因奖励信号噪声大和探索不足而导致过早收敛至次优模型。其核心解决方案在于提出一个统一的视觉-语言框架（VL-Scout），通过融合贝叶斯先验、策略梯度优化与序贯决策机制，实现结构化探索与自适应推理。关键创新包括：引入上下文相关的狄利克雷信念模型（context-dependent Dirichlet belief model），用于建模对模型质量的内蕴不确定性（epistemic uncertainty），并基于观测奖励与上下文证据在线更新，支持类似汤普森采样（Thompson-style）的探索策略，同时作为正则化先验；采用KL正则化的REINFORCE目标训练路由策略，使策略充当近似后验，狄利克雷信念作为先验，从而提升策略稳定性；进一步设计基于组相对策略优化（Group Relative Policy Optimization, GRPO）或可验证奖励强化学习（Reinforcement Learning with Verifiable Rewards, RLVR）的融合策略，该策略作用于中间模型推理理由，决策“继续”（CONTINUE）或“回答”（ANSWER），最终生成输出。实验在四个代表性多模态基准（MMMU、MMMU-Pro、OKVQA、RealWorldQA）上验证了该方法的优越性，相较于11种基线方法，在保持一致性能提升的同时有效缓解了策略过早坍缩（premature policy collapse）问题，凸显了将不确定性感知先验与近似策略学习及序贯融合相结合，在离散动作空间中实现鲁棒高效模型选择的重要性。
链接: https://arxiv.org/abs/2610.24054
作者: Selim Furkan Tekin,Ling Liu
机构: Georgia Institute of Technology(佐治亚理工学院)
类目: Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注:
Abstract:This paper studies dynamic model routing under uncertainty in multimodal settings, where a system must select and potentially compose multiple models to answer a query. Standard reinforcement learning approaches for routing often suffer from premature convergence to suboptimal models due to noisy rewards and insufficient exploration. We propose a unified vision-language framework (VL-Scout) for efficient multimodal and multi-step reasoning with turn-based reinforcement fusion. First, VL-Scout combines Bayesian priors, policy gradient optimization, and sequential decision-making to enable structured exploration and adaptive inference. A context-dependent Dirichlet belief model allows for epistemic uncertainty over model quality and is updated online based on observed rewards and context-specific evidence, enabling Thompson-style exploration while serving as a prior for regularization. Second, a routing policy is trained via a KL-regularized REINFORCE objective by maximizing a reward-weighted evidence lower bound, where the policy acts as an amortized posterior, and the Dirichlet belief serves as a prior. Third, we introduce a fusion policy trained with Group Relative Policy Optimization (GRPO) or Reinforcement Learning with Verifiable Rewards (RLVR), which operates over intermediate model rationales and selects between two actions: CONTINUE or ANSWER, terminating with a generated output. We evaluate VL-Scout on four representative multi-modal benchmarks (MMMU, MMMU-Pro, OKVQA, and RealWorldQA), compared against 11 representative routing and policy optimization baselines. Our method provides consistent performance improvements while reducing premature policy collapse. These results highlight the importance of integrating uncertainty-aware priors with both amortized policy learning and sequential fusion for robust and efficient model selection in discrete action spaces
[MA-28] Red-Teaming Multi-Agent Attacks for Loss-of-Control Risks
【速读】：该论文旨在解决生成式AI（Generative AI）代理在大规模协同部署背景下，因隐蔽合谋（covert collusion）而引发系统性失控风险的问题。随着AI代理能力与持久性的提升，其在分布式环境中执行碎片化攻击任务并隐匿行为特征的能力日益增强，传统基于单个代理轨迹的监控机制难以有效识别此类联合攻击。论文提出一种名为MiniLab的合成测试平台，包含5个相互关联的服务与249个良性任务，模拟真实部署场景，并设计了16种由人类优化的分布式攻击策略（如权重外泄、未受监控的恶意部署、权限提升等）。研究发现，当攻击由多个代理分担执行时，即使每个代理的行为均看似合规，整体攻击仍可成功规避前沿监控系统（如GPT-5.6 Sol）的检测——单独分析各代理轨迹时仅能识别7%的成功攻击，远低于单一代理执行相同任务时的87%识别率。此外，个体攻击可被组合成更长的攻击序列，实现更高阶的失控目标，且无任何单个代理触发警报。实验还表明，尽管当前模型具备一定规划能力，但在面对单轨迹监控时，其自主规划的攻击成功率最高仅为25%，显著低于人类设计的80%。这揭示了现有监测机制在应对复杂协作攻击中的局限性，也凸显了未来需构建具备跨代理行为理解与全局上下文感知能力的防御体系。本工作通过将最坏情况下的协同攻击转化为可复现的红队测试案例，为防范局部无害行为累积导致系统级失控提供了关键评估框架与防御研究基础。
链接: https://arxiv.org/abs/2610.24052
作者: John Kitaoka,Thao Amelia Pham,Carlos Rafael Giudice,Jacob Merizian
机构: MATS Research; Redwood Research
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:AI agents are growing more capable and persistent, and are being deployed in ever larger numbers. As more start collaborating, AI control measures must remain effective when they covertly collude on a harmful objective. We introduce MiniLab, a synthetic series of live services and codebases (5 interconnected services, 249 benign tasks, 8 attack objectives), adapted from internal AI deployment infrastructure. We demonstrate distributed attacks, in which different agents carry out separate attack fragments, and how their covertness evades frontier monitors when reading their trajectories individually. We show how distributed attacks spread across diverse contexts and harmful objectives through 16 human-optimized attacks (weight exfiltration, unmonitored rogue deployment, permission elevation, and more). A frontier monitor, GPT-5.6 Sol, reading each agent separately flags only 7% of successful distributed attacks, compared with 87% when a single agent carries out the same side task. Individual attacks can also be composed into longer attack sequences toward a greater loss-of-control outcome, without any individual agent being flagged. Finally, we tested whether models can plan these covert attacks: even with execution and monitor feedback, the best model-planned attacks succeed covertly against a single-trajectory GPT-5.6 Sol monitor in at most 25% of executions, compared with 80% for human-crafted plans, though this may scale with model capabilities. By turning worst-case coordinated attacks into reproducible red-team cases, our work provides a foundation for preventing locally benign agent actions from accumulating into system-level loss of control.
[MA-29] LLM Evolution in a Simple Simulated Market
【速读】：该论文旨在解决生成式 AI（Generative AI）在缺乏明确监管机制时可能因激励机制设计不当而出现系统性退化的问题。其核心关切在于：当模型的优化目标仅聚焦于某一可量化的成本维度（如令牌使用量），而忽略其他关键性能指标（如准确性）时，即便未对模型施加直接指令或反馈，系统仍会通过集体演化过程产生不可预见的负面行为适应。解决方案的关键在于揭示了一种“非显式反馈下的集体优化”机制——在无个体反馈、无记忆、且不理解支付规则的前提下，基于种群演化与突变的市场环境促使代理群体整体向特定行为模式收敛，从而导致原本未被定价的性能维度（如准确性）发生显著退化。研究通过三组对照实验表明，这种适应是由于令牌使用与准确性之间存在隐含的相关性，使得对单一维度的选择压力间接驱动了另一维度的演化，凸显了在复杂系统中设计激励结构时需全面考虑多维属性耦合效应的重要性。
链接: https://arxiv.org/abs/2610.24033
作者: Charlie Pilgrim,Christopher Krupenye,Dora Biro,Richard P. Mann
机构: University of Leeds(利兹大学); Johns Hopkins University(约翰霍普金斯大学); University of Rochester(罗切斯特大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Populations and Evolution (q-bio.PE)
备注:
Abstract:Arguments that AI systems will evolve in harmful ways are largely made in the abstract. We provide a concrete empirical example. We simulate simple markets consisting of populations of twenty Large Language Model agents tasked with solving arithmetic problems. Agents differ only in their system prompt; those that go bankrupt are removed and replaced by mutated copies of the prompts of successful agents. Two market conditions select for opposite phenotypes: a market that rewards accuracy produces agents with high accuracy (94%) and high token use; a market that charges per token produces agents with low accuracy (10%) and low token use. In the second market, only tokens are incentivised yet we see a collapse in accuracy. Token use and accuracy are correlated traits such that selection on one also moves the other, and behaviour adapts along an unpriced dimension. A third neutral condition with no selection allows us to separate these effects from bias in the mutation operator. Adaptation occurs through a collective optimisation process that operates above the level of individual agents, who are never told the payment rule, do not receive feedback, and have no memory.
[MA-30] Federation over Harness
【速读】：该论文旨在解决当前基于大语言模型（LLM）的分布式多智能体推理系统中，依赖人工设计提示词（prompt）和工作流所带来的局限性问题。现有方法在本地智能体提取有效推理轨迹以及全局智能体聚合分布式反思以形成共性模式与洞见的过程中，均需依赖预设的、静态的人工设计机制，难以适应复杂任务动态演化的需求。为此，论文提出Federation over Harness（FoH）框架，其核心创新在于通过自进化机制自动优化本地与全局组件的“引导框架”（harness），实现端到端的自主演进。具体而言，由服务器智能体同时演化客户端的任务求解与自我反思引导框架，并自主优化自身用于聚合分布式反思、维护共享知识库及更新下游客户端框架的全局引导机制。关键优势在于，FoH无需额外验证信号（如真实标签或验证数据），仅依靠分布式智能体生成的自反思内容进行学习。实验结果表明，在多个基准测试上，FoH相比固定引导框架的单智能体基线和仅具备引导框架自演化的方案，分别实现了53.57%和29.76%的性能提升，充分验证了其在无需人工干预下持续优化多智能体协作效能的有效性。
链接: https://arxiv.org/abs/2610.24007
作者: Dixi Yao,Tahseen Rabbani,Tian Li
机构: University of Chicago(芝加哥大学)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Modern LLM-powered agents increasingly improve through experience sharing, where distributed agents distill local reasoning experience into reusable knowledge, such as a shared insight library. Existing approaches, however, still heavily rely on manually designed prompts and workflows that determine both how local agents extract useful reasoning traces and how a central/global agent aggregates them to obtain common patterns and insights. We propose Federation over Harness (FoH), a framework that automatically evolves the harness for both local and global components in this distributed multi-agent reasoning pipeline. A server agent evolves the client harness for task solving and self-reflection while, at the same time, evolving its own harness for aggregating distributed reflections, maintaining shared knowledge, and updating subsequent client harnesses. Importantly, FoH requires no extra validation signals, such as ground-truth labels or validation data, during evolution; it learns solely from self-generated reflections produced by distributed agents. Experiments on multiple benchmarks show that FoH consistently improves over single-agent baselines with a fixed harness and with harness self-evolution by 53.57% and 29.76%.
[MA-31] win: A Calibrated Digital Twin of Telegram Crisis Discourse as a Ground-Truth Testbed for Detecting LLM -Powered Information Operations
【速读】：该论文旨在解决在真实社交平台（尤其是Telegram）上评估协调性检测算法有效性时缺乏可靠“真实标签”（ground truth）的问题。由于大型语言模型（Large Language Models, LLMs）可低成本生成永不重复的影响力传播内容，使得检测此类操纵行为变得极为困难，而现有平台数据又缺乏足够的标注支持。针对这一挑战，研究提出TeleTwin——一个基于LLM构建的Telegram危机话语生态系统的数字孪生体，其中每个频道被建模为一个智能体，其行为依据195个公开频道在2025年3月至8月期间共107万条消息的历史数据进行校准，聚焦于以色列-加沙与伊朗-以色列冲突议题。实验表明，该数字孪生体在模拟6月2025年伊以战争爆发后的每日动态时，其相关性（0.86）显著优于统计基线（0.64），且能无提示地响应重大事件，但存在过度聚焦冲突、夸大转发级联、低估冲击效应以及生成明显具有LLM特征文本等偏差。进一步将已知成员的LLM生成宣传战役注入真实数据与孪生环境后发现：在真实数据中，基于相同文本、共用URL或同步性的检测方法在内容改写、延迟发布或移除链接后即失效；多语言语义相似性检测虽能识别高调战役（包括跨语言传播），但对隐蔽战役无效，且在不同政治立场间误报率分布不均；而在数字孪生环境中，所有检测方法均失败，原因在于模拟背景的语义同质性比真实数据高出约40倍。因此，该研究的核心结论是：数字孪生体虽可提供用于评估传播战役的“真实标签”，但若未充分校准其语义多样性，则无法生成真实可信的背景噪声，从而导致对语义检测器的评估失真。解决方案的关键在于：必须通过引入更丰富的语义多样性来校准数字孪生体，才能使其成为有效评估检测算法性能的可信环境。
链接: https://arxiv.org/abs/2610.23944
作者: Despoina Antonakaki,Sotiris Ioannidis
机构: Institute of Computer Science, FORTH(希腊科技研究所计算机科学学院); Technical University of Crete(克里特技术大学)
类目: ocial and Information Networks (cs.SI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Large language models make it cheap to run influence operations whose messages never repeat, yet real platforms offer almost no ground truth for testing whether coordination detectors can find them. The problem is sharpest on Telegram, which moderates little and spreads content through broadcast channels and forwarding rather than a follow graph. We introduce TeleTwin, an LLM-powered digital twin of a Telegram crisis-discourse ecosystem in which channels are agents, each instantiated from the history of a real channel, calibrated on 1.07 million messages from 195 public channels on the Israel-Gaza and Iran-Israel conflict (March-August 2025). On a held-out period containing the June 2025 Iran-Israel war, the twin tracks daily dynamics better than statistical baselines (correlation 0.86 against 0.64) and reacts to the war unprompted, but it over-focuses on the conflict, over-amplifies forwarding cascades, underestimates the shock and writes recognisably LLM-like text. We then inject LLM-written campaigns with known membership into the real data and into the twin. In real data, identical-text, co-URL and synchrony detectors fail once a campaign rewords, staggers or drops links; multilingual semantic similarity recovers loud campaigns, including cross-lingual ones, but not quiet ones, and its false positives fall unevenly across political perspectives. Inside the twin all detectors fail, because the simulated background is about 40 times more semantically homogeneous than the real one. Digital twins can supply ground truth for campaigns, but not, without calibrating their diversity, a realistic background for evaluating semantic detectors.
[MA-32] LVMH: A Meta-Optimization Harness for Long-Video Agent Memory
【速读】：该论文旨在解决长视频智能体在处理数小时甚至数天的多模态证据时，如何在有限的上下文长度（token budget）和响应延迟（latency budget）约束下实现高效的记忆管理问题。现有记忆系统主要依赖人工设计的策略，缺乏对性能与效率协同优化的记忆架构设计方法。其核心解决方案是提出一种自动化多智能体框架——长视频元记忆调度系统（Long Video Meta-memory Harness, LVMH），将记忆策略开发建模为围绕固定答案模型的多目标代码搜索任务。LVMH通过分离实现与独立评审机制，并设置受保护的观察边界，有效防止自认证偏差和开发集过拟合。系统持续记录历史提案、推理轨迹、资源消耗及失败案例，形成持久化的元记忆以指导后续策略迭代。在此框架下，研究发现了“自演化证据记忆”（Self-Evolving Evidence Memory, SEEM）策略，该策略采用“文本优先、视觉次之”的范式，从紧凑的结构化事件记忆与证据专家中生成答案，并通过选择性视觉门控仅在必要时调用局部视觉帧。SEEM在EgoLifeQA上训练后可泛化至未见基准，在Ego-R1 Bench与Video-MME (L) 上均优于Meta-Harness，使用GPT-5 mini时平均准确率达65.7%，较此前最优方法提升7.7个百分点，同时平均评估token消耗降低约32倍，平均响应时间缩短91倍，显著提升了长视频理解任务中的性能与效率平衡。
链接: https://arxiv.org/abs/2610.23904
作者: Liwei Che,Benlin Liu,Dai Li,Xiangyan Li,Zijian Wu,Changjiang Li,Minghao Guo,Chengzhi Mao,Vladimir Pavlovic,Ranjay Krishna
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Multiagent Systems (cs.MA)
备注:
Abstract:Long-video agents must map hours or days of multimodal evidence into a bounded context while respecting token and latency budgets, making memory design a context-management challenge. Existing memory systems largely rely on hand-designed policies, leaving open how to jointly optimize memory design for performance and efficiency. We propose Long Video Meta-memory Harness (LVMH), an automated multi-agent framework that formulates memory-policy development as multi-objective code search around a fixed answer model. To guard against self-certification and development-set overfitting, LVMH separates implementation from independent review and enforces protected observation boundaries. Persistent meta-memory records past proposals, traces, costs, and failures to guide subsequent policy revisions. Through this harness, we discover Self-Evolving Evidence Memory (SEEM), a “text-first, vision-second” policy that answers from compact structured event memory and evidence specialists, consulting localized frames through a selective visual gate. The policy evolved on EgoLifeQA generalizes to unseen benchmarks, outperforming Meta-Harness on both Ego-R1 Bench and Video-MME (L) with GPT-5 mini. Across the three benchmarks, SEEM with GPT-5 achieves 65.7% average accuracy, surpassing the previous state of the art by 7.7 percentage points with approximately 32x lower average evaluation token use and 91x shorter average response time.
[MA-33] Multi-LLM Systems Exhibit Robust Semantic Collapse
【速读】：该论文旨在解决多大语言模型（multi-LLM）系统在闭环设置下能否持续进行开放式探索的问题，尤其关注其在自主生成任务中是否能够真正产生新颖内容。研究发现，尽管各模型在词汇层面表现出多样性，但其语义表征却呈现出系统性收敛，即发生语义坍缩（semantic collapse），这一现象在不进行任何重训练或权重更新的情况下持续存在，且跨越多个模型家族和200至1,000轮交互。尽管测试了包括解码参数、提示工程、代理构成、激活工程及强化学习在内的十三种干预策略，均未能可靠恢复语义多样性。机制分析表明，语义坍缩并非仅由对齐或一致性偏差导致，而是与自回归生成（autoregressive generation）的内在特性密切相关。因此，该研究的关键结论是：多大语言模型系统在闭环环境中面临根本性的生成约束，难以维持真正的开放性探索能力。
链接: https://arxiv.org/abs/2605.17193
作者: Weiyi Kong,Shiyang Lai,Jinghua Piao,James Evans
机构: 未知
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 94 pages, 12 figures, 13 tables; includes Supplementary Information
Abstract:Whether machines can originate novel content has been debated for nearly two centuries, from Lovelace’s assertion that no engine can “originate anything” to Turing’s question of whether a machine can amplify ideas brought in from outside. Interacting multi-LLM systems, increasingly deployed for autonomous generation, reopen this question empirically. Here we show that such systems exhibit semantic collapse, systematic convergence in semantic representations despite apparent lexical variation, with no retraining or weight updates. The pattern holds across model families and over 200 to 1,000 rounds. Thirteen intervention implementations tested, spanning decoding parameters, prompt design, agent composition, activation engineering, and reinforcement learning, did not reliably restore semantic diversity. Mechanistic analyses suggest that semantic collapse is not solely attributable to the alignment or conformity biases examined here, but is consistent with intrinsic properties of autoregressive generation. Our results point to persistent constraints on the ability of multi-LLM systems to sustain open-ended exploration in closed-loop settings.
[MA-34] Cooperative Multi-Agent Reinforcement Learning for Market Making on Prediction-Market Order Books MICRO
【速读】：该论文旨在解决在预测市场（如Polymarket）中，基于多智能体强化学习的做市商在薄订单簿环境下面临的三大挑战：一是逆向选择可能导致仅基于真实数据训练的策略停止报价；二是报价的后果可能在多个时间步后才显现，造成信用分配困难；三是预测市场提供的L2级数据无法识别个体订单，难以支持精确模拟。为应对这些问题，研究通过重构Polymarket的订单级别数据为LOBSTER格式，并在JaxMARL-HFT框架下于417个真实市场中训练智能体。关键解决方案包括：丰富做市商的观测空间以包含订单簿状态与近期订单流信息，重新设计动作空间和库存控制机制，并提出一种合作式多智能体分解结构——将做市行为拆分为报价者与风险管理者两个角色，采用带有循环神经网络（RNN）策略的独立近端策略优化（IPPO）进行联合训练。此外，通过奖励塑形技术（特别是奖励融合与填单后估值调整），有效缓解了多智能体间的信用分配难题。实验表明，改进后的单智能体做市商相比基线模型报价更频繁且持有损失更小，而合理的奖励设计显著提升了分解架构的表现，成为最优结果。尽管所有模型在独立测试集上均未实现正的平均投资组合价值，但该研究构建了可扩展至同时报价多个关联市场的做市系统基础环境、数据与训练协议。
链接: https://arxiv.org/abs/2610.26789
作者: Skander Lejmi
机构: 未知
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Multiagent Systems (cs.MA); Trading and Market Microstructure (q-fin.TR)
备注: 50 pages, 8 figures, 7 tables. Extended version of a paper accepted at ICAIF 2026 Workshop on AI-Driven Market Microstructure. MSc dissertation, London School of Economics and Political Science (2026)
Abstract:A market maker provides liquidity by quoting prices to buy and sell an asset, earning the spread while managing inventory risk. We study market makers trained with multi-agent reinforcement learning on prediction markets, specifically Polymarket, whose thin order books pose distinct challenges. Three challenges underpin this research: adverse selection may cause a policy trained on real data to stop quoting; the consequences of a quote can arrive many steps after the action; and prediction-market L2 data does not identify individual orders needed for simulation. We address these challenges by reconstructing order-level LOBSTER-format data from Polymarket and training agents in JaxMARL-HFT on 417 real markets. We enrich the baseline market maker’s observation space with the book and recent flow, redesign its actions and inventory controls, and propose a cooperative multi-agent decomposition into a quoter and a risk manager trained using recurrent Independent Proximal Policy Optimisation (IPPO) with RNN policies. We investigate reward shaping, particularly reward blending and post-fill markout, to mitigate credit-assignment difficulties between the agents. Our redesigned single-agent market maker quotes more frequently and incurs smaller held-out losses than the retrained JaxMARL-HFT baseline, although the contributions of jointly changed observations, actions and risk constraints are not separately identified. Decomposition alone performs poorly; careful reward design substantially improves the desk relative to the unshaped split and yields our best result. Models are evaluated on realised end-of-episode portfolio value on held-out markets. None establishes positive average portfolio value in periodic held-out evaluation, and each configuration was trained with one seed. The environment, data and protocol lay a foundation for makers that quote many linked markets at once.
自然语言处理
[NLP-0] Can It Actually Work? Synthesizing Functional CAD Mechanisms
【速读】： 该论文旨在解决大语言模型（LLM）生成的计算机辅助设计（CAD）装配体虽在视觉上合理，却常无法满足实际物理功能需求的问题。其核心挑战在于：如何让智能体在需求描述不完整的情况下，自主构建并迭代优化机械结构，使其在物理仿真中满足特定功能要求。解决方案的关键在于提出FunCAD-Harness框架，该框架能够将抽象的功能需求转化为与具体设计方案相关的物理测试，并利用仿真反馈指导设计修正。为系统评估该能力，研究还构建了FunCADBench基准，涵盖46类具有相互依赖性和多步操作的机制。实验表明，FunCAD-Harness可识别出仅通过视觉检查难以发现的细微机械缺陷，显著提升功能成功率，优于标准的仿真增强型基线方法，标志着向实现能将人类意图转化为可运行机械装置的CAD智能体迈出了关键一步。
链接: https://arxiv.org/abs/2610.26861
作者: Wenqian Zhang,Yangyi Huang,Zigeng Xu,Weiyang Liu,Zhen Liu
机构: The Chinese University of Hong Kong (香港中文大学); The Chinese University of Hong Kong (Shenzhen) (香港中文大学（深圳）)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Graphics (cs.GR)
备注: 45 pages, 16 figures, project page: this https URL
Abstract:While LLM agents can generate visually plausible CAD assemblies, these designs often fail to fulfill their intended physical functions. We study functional CAD design, a setting where agents interpret underspecified requests to construct and iteratively revise mechanisms that satisfy functional requirements in physical simulation. Verifying these requirements is nontrivial: different candidate mechanisms may realize the same function through entirely different components, controls, and operating sequences. To bridge this gap, we propose FunCAD-Harness, a framework that translates abstract requirements into candidate-specific physical tests and uses simulation feedback to guide design revisions. To systematically evaluate this capability, we introduce FunCADBench, a benchmark spanning 46 mechanism families with interdependent and multi-step operations. Experiments across frontier LLM agents show that FunCAD-Harness exposes subtle mechanical failures invisible to visual inspection, substantially improving functional success rates over standard simulation-augmented baselines. Ultimately, our work marks a step toward CAD agents capable of translating human intent into working mechanisms.
[NLP-1] OpenVLHarness: Real World Visual Agent ic Tool-Use with Multimodal Memory
【速读】： 该论文旨在解决多模态大语言模型（Multimodal Large Language Models, MLLMs）在视觉推理任务中因依赖内部推理而难以获取必要证据的问题，尤其在需要接地感知、外部知识及数值测量等异构工具支持的任务上表现受限。其核心挑战在于如何有效整合来自不同工具的异构输出，并将其转化为可被模型理解与持续利用的多模态证据。解决方案的关键在于提出OpenVLHarness框架，其核心由能力层（capability layer）和持久化多模态记忆（persistent multimodal memory）构成：能力层通过统一的任务级工具接口与证据渲染机制，实现对异构后端的一致访问，并将工具输出转换为模型可读的观测信息；持久化多模态记忆则以可寻址的实体形式长期保留图像、结构化数据与网络引用，支持跨推理步骤的信息复用；同时，集成代码接口可执行生成程序对存储数据进行计算，实现多工具结果的融合与迭代推理。该架构显著提升了模型在22个基准测试上的表现，平均得分提升9.7–12.6点，且在仅消耗37%模型令牌成本的情况下，超越了高算力投入下的基线模型。
链接: https://arxiv.org/abs/2610.26860
作者: Yu Zhou,Cheng-Fu Yang,Rui Sun,Di Wu,Sibo Peng,Mingyang Zhang,Zhen Yang,Ruiyi Zhang,Nanyun Peng,Zhe Gan,Kai-Wei Chang
机构: University of California, Los Angeles(加州大学洛杉矶分校); Apple(苹果)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Multimedia (cs.MM)
备注:
Abstract:Multimodal large language models (MLLMs) have made rapid progress on visual reasoning, yet many tasks require evidence that is difficult to obtain through internal reasoning alone: our controlled study on 22 benchmarks shows that simply increasing the inference budget of the base model yields only modest gains. Such tasks demand grounded perception, external knowledge, and numerical measurements obtained through heterogeneous tools, whose outputs must be made interpretable to the model and retained across reasoning steps. We introduce OpenVLHarness, a multimodal reasoning harness organized around a capability layer and persistent multimodal memory. The capability layer combines task-level tool interfaces with evidence rendering, providing consistent access to heterogeneous backends and converting their outputs into model-readable observations. Persistent multimodal memory retains the corresponding images, structured data, and web references as addressable artifacts throughout each query. A coding interface executes generated programs over stored data, enabling computations that combine results from multiple tools. Together, these mechanisms support an iterative process of acquiring, inspecting, and composing multimodal evidence. Evaluated on 22 benchmarks with eight open and proprietary backbones, OpenVLHarness improves average scores by 9.7-12.6 points over matched no-tool baselines, including +11.2 with Qwen3-VL-8B, +9.7 with Kimi K3, and +9.9 with GPT-6 Astra. With GPT-6 Luna at medium reasoning effort, OpenVLHarness outperforms the base model, Codex, and Visual Sketchpad at maximum effort by 7.7-8.5 points while incurring at most 37% of their model-token cost.
[NLP-2] WeAct: Dynamic Weight Switching as Agent Actions
【速读】： 该论文旨在解决长时程智能体在交错轨迹中学习异构能力时面临的灾难性遗忘问题。传统方法在联合训练或静态适配器合并过程中采用固定参数配置，无法适应执行过程中不同阶段对不同能力的需求，导致新能力的学习损害已有能力。其解决方案的关键在于提出WeAct（Weight-level Internal Actions）框架：在冻结主干网络的基础上，为特定能力独立训练LoRA适配器，并将适配器选择机制引入智能体的动作空间。通过模型原生的工具调用接口，实现基于共享上下文的按需能力组合。实验表明，Qwen3-32B结合WeAct在八个深度研究基准上相比全微调方法平均准确率提升6.2%（计算开销更低），相比最强的静态LoRA融合方法提升27.5%。此外，WeAct支持模块化持续适应，可通过新增代码计算适配器扩展能力，更新写作适配器以提升报告质量，且不改变主干网络及其他适配器，实现了对智能体的可扩展、非侵入式持续学习。
链接: https://arxiv.org/abs/2610.26847
作者: Rui Shi,Xunjian Yin,Jiacheng Sang,Zichuan Liu,Shuyan Zhou
机构: Duke University; Carnegie Mellon University
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 24 pages, 14 figures, 12 tables, including references and appendices
Abstract:Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories. Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting. As a trajectory unfolds, different stages may require different capabilities, yet conventional joint training and static adapter merging maintain a fixed parameter configuration throughout execution. We introduce WeAct (Weight-level Internal Actions), which trains independent capability-specific LoRA adapters over a frozen backbone and adds adapter selection to the agent’s action space. These internal actions use the model’s native tool-calling interface for on-demand capability composition in a shared context. Across eight deep-research benchmarks, Qwen3-32B with WeAct improves average accuracy by 6.2% relative to full fine-tuning counterparts despite using less compute, and by 27.5% relative to the strongest static LoRA-merging method. WeAct also supports modular continual adaptation, by adding a new code computation adapter to expand the agent’s capabilities and updating the existing writing adapter to improve report quality. Both updates leave the backbone and other adapters unchanged, supporting modular continual adaptation in agents.
[NLP-3] LatComm: Reusable Context Compression for Latent Communication between Large Language Models
【速读】： 该论文旨在解决多大语言模型（Large Language Models, LLMs）系统中因重复处理相同上下文而导致的计算效率低下问题。当多个问题共享同一上下文时，若以纯文本形式传递，接收模型需对全部词元（token）重新进行处理，造成显著冗余开销。为克服这一瓶颈，论文提出一种名为LatComm（Latent Communication）的潜在通信接口，其核心创新在于将上下文编码与读取过程解耦：上下文在不依赖具体问题的情况下一次性被冻结的编码器模型压缩为固定大小的数据包，随后由另一冻结的接收模型通过跨注意力机制结合自身提示（prompt）生成门控信号，选择性地读取所需信息。该方案实现了上下文表示的紧凑性、跨问题复用性以及对不同接收模型的兼容性，三者兼备。在12个多项选择型问答数据集上的实验表明，相比单向量基线方法，LatComm平均准确率提升最高达10.2个百分点；在QuALITY数据集上，接收端输入量减少57倍，文章内所有问题的回答速度提升3倍。
链接: https://arxiv.org/abs/2610.26840
作者: Huawei Lin,Jinting Liu,Tan Yu,Weijie Zhao
机构: Rochester Institute of Technology (罗切斯特理工学院); NVIDIA(英伟达)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 17 pages, 9 figures, 11 tables. Under Review
Abstract:Systems that combine several large language models (LLMs) often use one context for many questions. When the context is passed between models as text, the receiving model processes all of its tokens again for every question. Avoiding this cost requires a context representation that is compact, reusable across questions, and readable by a different LLM. Existing compression and latent communication methods each provide only some of these properties. We introduce LatComm, a Latent Communication interface that provides all three by separating encoding from reading: the context is encoded once without the question, and the receiver selects what to read for each question. A frozen context encoder LLM compresses the context into a fixed-size packet and sends it to a different frozen receiver LLM, which reads the packet through cross-attention, with gates computed from its prompt. On 12 multiple-choice question-answering datasets, LatComm improves average accuracy over single-vector baselines by up to 10.2 points. On QuALITY, it reduces the receiver’s input by 57x and answers the questions of an article 3x faster.
[NLP-4] Competing Learning in Hybrid Architectures
【速读】： 该论文旨在解决混合架构语言模型中，基于现代循环层（recurrence layer）与Softmax注意力机制（softmax attention）在上下文信息检索任务中存在学习竞争且性能失衡的问题。其核心问题是：尽管循环层具备高效的表达能力并能压缩上下文状态以实现高吞吐，但在典型数据分布下，其对上下文信息的检索功能学习速度远快于注意力机制，导致后者在训练过程中因学习信号被抑制而难以有效学习检索功能，从而影响整体性能。解决方案的关键在于揭示“竞争学习”（Competing Learning）现象，并据此提出一种分阶段预训练策略——在训练中期停止循环层的权重更新，以释放学习信号供注意力机制学习检索功能。实证结果表明，该方法在高达10亿参数的模型上显著提升了语言建模损失、常识推理能力以及多种长度下的检索相关任务表现，验证了其有效性。
链接: https://arxiv.org/abs/2610.26830
作者: Xingyu Shen,Yingfa Chen,Ziqing Qiao,Chaojun Xiao,Xu Han,Zhiyuan Liu
机构: Tsinghua University (清华大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Many modern language models have adopted hybrid architectures that incorporate both linear-time recurrence and softmax attention. Modern recurrence layers provide a highly expressive mixing mechanism while ensuring high throughput by compressing contextual information into a fixed-size state, but underperform softmax attention in contextual information retrieval. In this paper, we reveal a Competing Learning phenomenon where different kinds of mixers compete to learn the same function, and when any layer learns a function, it suppresses the corresponding learning signal for other layers. Layer attribution analysis reveals that under typical data distributions, modern recurrence learns retrieval functions faster than softmax attention, causing attention to fail to learn retrieval in most cases. Stopping weight updates in recurrence in the middle of training helps attention learn retrieval. Based on these results, we propose a staged training procedure for pre-training hybrid architectures on natural language. Empirical results on models up to 1B parameters show that our method results in substantial improvement in language modeling loss, common-sense reasoning, as well as a wide range of retrieval-related tasks across different context lengths.
[NLP-5] Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
【速读】： 该论文旨在解决当前通用计算机使用代理（Computer-Use Agents, CUAs）在真实世界系统中进行规模化学习所面临的挑战，尤其是由此带来的安全风险与环境构建的可扩展性问题。现有方法通常局限于单一交互接口（如仅命令行或图形界面），难以支持多模态交互能力的融合。为此，本文提出“野外代理世界模型”（Agent World Model in the Wild, WildAWM），构建一个可扩展的合成环境与任务生成流水线，能够同时支持图形用户界面（GUI）、模型上下文协议服务（Model Context Protocol Services）和命令行接口（CLI），从而实现视觉交互、工具调用与代码执行的灵活组合。其核心创新在于采用完全开源模型驱动整个流程，证明了无需依赖专有前沿模型即可生成高质量的计算机使用环境。针对任务合成中的三大难题——可解性保障、难度适中性维持及验证器可靠性提升，提出了基于执行过程的质量控制机制，结合异构代理回放、跨裁判一致性检验以及反事实判断鲁棒性探测，有效提升了合成任务的质量与可信度。最终生成的341个长时程环境-任务对需超过100步才能解决，即使对于前沿模型（如GPT-6-Astra）也仅能成功完成60%，具有显著挑战性。基于这些合成环境训练的两个不同规模的Qwen模型，在8个分布外基准测试（涵盖网页操作、机械臂控制及通用计算机使用任务）上均取得显著性能提升，验证了在WildAWM中学习具备良好的泛化能力。该研究为通用代理在多样化、复杂化的真实场景中自主学习提供了可复现、可扩展且安全的解决方案。
链接: https://arxiv.org/abs/2610.26828
作者: Zhaoyang Wang,Qianhui Wu,Haorui Wang,Simon Yu,Baolin Peng,Canwen Xu,Hao Cheng,Wenlin Yao,Boyi Liu,Yite Wang,Jianfeng Gao,Huaxiu Yao
机构: UNC-Chapel Hill(北卡罗来纳大学教堂山分校); Microsoft Research(微软研究院); Georgia Tech(佐治亚理工学院); Northeastern University(东北大学); Independent Researcher(独立研究员)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 35 pages
Abstract:Agent learning is shifting from learning from human data toward learning through interaction with the world. However, scaling such learning directly in real-world computer systems is challenging and can introduce substantial security risks. We therefore propose Agent World Model in the Wild (WildAWM), a scalable pipeline for synthesizing controllable environments and tasks for training general computer-use agents (CUAs). Whereas existing approaches typically specialize in a single agent interaction interface, we construct executable environments that jointly support graphical user interfaces, model context protocol services, and command-line interfaces, allowing agents to flexibly combine visual interaction, tool use, and code execution. WildAWM is entirely powered by open-source models, demonstrating that high-quality computer-use environments can be generated without proprietary frontier models. Beyond environments, we address three recurring challenges in task synthesis: ensuring solvability, maintaining sufficient difficulty, and producing reliable verifiers, by introducing execution-based quality gates that combine heterogeneous agent rollouts, cross-judge consistency checks, and counterfactual probes of judgment robustness. The resulting pipeline synthesizes 341 long-horizon environment-task pairs that can require over 100 agent steps to solve and remain challenging even for frontier models, with GPT-6-Astra solving only 60% of them. We train two Qwen models of different sizes exclusively in these synthetic worlds. Both models achieve substantial gains across 8 out-of-distribution benchmarks spanning web, claw, and computer-use agents, demonstrating that learning in WildAWM’s synthesized worlds generalizes broadly to unseen computer-use tasks. The code is publicly available at this https URL.
[NLP-6] KEEP-Bench: Evaluating Evidence Preservation in Context Compression
【速读】： 该论文旨在解决长上下文处理中因上下文压缩（context compression）导致的忠实性下降问题，即信息瓶颈可能引发关键信息丢失，而现有基于问答（QA）指标的评估方法无法有效捕捉此类信息失真，从而低估了压缩带来的忠实性损失。其解决方案的关键在于提出KEEP-Bench这一诊断基准，涵盖4,102个案例，覆盖需保留证据以支持合理引用解析、显式限定词及少数派观点等三类高挑战性场景，直接量化压缩过程中的忠实性表现。实验表明，即便采用定制提示（tailored prompting），当前最强模型在超10万标记上下文窗口下仍存在显著信息丢失：最佳开源模型（31B）仅在66.8%的案例中完整保留必要证据，最佳闭源模型也仅达61.6%。此外，模型性能在不同证据结构间差异显著，暗示基于有限结构的结论易产生偏差。研究进一步引入证据保留回报（Evidence Preservation Return, EPR）指标，揭示证据召回率与推理计算量（FLOPs）的关系呈非单调性，在4B–8B规模时达到峰值，但此时召回率依然偏低；增加输出预算主要惠及已有强模型，加剧能力差距，且伴随高昂计算成本。因此，研究强调未来工作应聚焦于在低计算开销下具备强信息保留能力的压缩机制，而非依赖大模型规模或高输出预算。
链接: https://arxiv.org/abs/2610.26827
作者: Ziqing Man,Di Wu,Jia-Chen Gu,Junyi Zhang,Kai-Wei Chang,Nanyun Peng
机构: 未知
类目: Computation and Language (cs.CL)
备注: Code: this https URL Project: this https URL
Abstract:Context compression improves long-context efficiency but may introduce faithfulness issues through an information bottleneck. Existing evaluations rely on QA metrics, but a reader can still answer correctly from an unfaithful compression, so these metrics underestimate such loss. To bridge this gap, we introduce KEEP-Bench, a diagnostic benchmark that directly assesses compression faithfulness with 4,102 cases across three challenging settings that require keeping evidence for plausible reference resolutions, explicit qualifiers, and minority perspectives. Experiments show that even with tailored prompting, the strongest models remain unreliable: despite context windows exceeding 100K tokens and inputs averaging only 3.1K tokens, the best open-weight model (31B) preserves all required evidence in only 66.8% of cases, and the best proprietary model in only 61.6%. Performance also varies sharply across evidence structures, so conclusions from limited structures may be biased. We further introduce a metric, Evidence Preservation Return (EPR), to relate evidence recall to inference FLOPs, and find that it peaks at moderate scales (4B-8B), where recall remains low. Additional output budget mainly benefits already strong models and widens capability gaps, yet the stronger preservation of these models comes at a steep computational cost. These findings highlight the importance for future work to investigate compressors with strong information preservation under a small computation budget, rather than relying on a large output budget or model scaling.
[NLP-7] RefCraft: Benchmarking Visual-Reference-Conditioned Agent ic Generation through Web Development
【速读】： 该论文旨在解决当前生成式AI在实际生产力任务中对视觉参考（visual references）有效整合能力不足的核心问题，尤其聚焦于基于视觉示例的网页开发这一典型应用场景。现有基准多局限于纯文本指令生成或目标视觉复现，未能反映开发者日常工作中“以图搜图、参照设计稿迭代开发”的真实流程。为此，研究提出RefCraft基准，包含300个融合自然语言指令与1至5张参考网页截图的多模态任务，要求智能体在生成过程中选择性解析视觉参考，并通过视觉检查与迭代优化输出。其解决方案的关键在于构建一个贴近真实工作流的评估体系，结合可独立验证的关键点（keypoints）、程序化评估与视觉语言模型（VLM）判别器进行多维度验证。实验表明，即使在九个前沿大模型中表现最优的系统，得分也仅为52.12/100，凸显当前模型在视觉参考条件下的自主生成与修正能力仍存在显著瓶颈。
链接: https://arxiv.org/abs/2610.26826
作者: Yuchi Wang,Rongyao Fang,Weihao Yu,Xilin Wei,Tiancheng Gu,Dingkang Yang,Hang Zhang,Dayiheng Liu,Hongsheng Li,Shuai Bai
机构: Alibaba Token Foundry; MMLab, CUHK (香港中文大学多媒体实验室); The Chinese University of Hong Kong (香港中文大学); Fudan University (复旦大学); The University of Sydney (悉尼大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Beyond basic tool use and coding skills, effectively incorporating visual references into agentic generation is equally essential for applying AI agents to real-world productivity tasks. We study this crucial capability through webpage development, a practical and readily evaluable setting in which existing benchmarks remain largely idealized: they typically focus on reconstructing visual targets or generating webpages from textual instructions alone, overlooking the more common workflow of developing from visual examples. Specifically, we introduce RefCraft, a benchmark comprising 300 realistic multimodal webpage-development tasks that combine natural-language instructions with screenshots from one to five reference webpages. Agents must selectively interpret and utilize these references while rendering, visually inspecting, and iteratively refining their outputs. We evaluate the resulting open-ended artifacts using independently verifiable keypoints and complementary programmatic and VLM judges. Across nine frontier (M)LLMs, the strongest system achieves an overall score of only 52.12 out of 100, revealing that visual-reference-conditioned agentic generation remains a substantial challenge for current models in practical webpage-development settings.
[NLP-8] MemT-1.0: Towards Memory-Native Evolution
【速读】： 该论文旨在解决基础模型在持续学习过程中缺乏原生记忆能力的问题，即如何使模型具备内生的、可累积且可复用的记忆机制，以实现对长期信息的有效存储与利用。其核心挑战在于如何在不牺牲通用能力的前提下，构建高效、稳定且可扩展的记忆系统。解决方案的关键在于提出MemT-1.0架构，该架构将记忆作为模型计算过程的内在组成部分，通过引入基于门控增量网络（Gated DeltaNet-2）的稳定分词级记忆更新机制，实现每轮交互中的精准记忆维护；同时采用混合存储（Mixture-of-Storage）策略以线性扩展原生记忆容量。此外，论文设计了两阶段训练流程（持续预训练与后训练），并依托340亿tokens的训练数据及高效的训练推理基础设施，最终实现了在长达128K token历史长度下，相较于全上下文键值缓存（KV cache）方案，实现约1.86倍的端到端推理加速和66倍的持久存储降低，显著提升了代理任务（agentic tasks）的成功率，为迈向真正意义上的记忆原生演进（memory-native evolution）提供了可行路径。
链接: https://arxiv.org/abs/2610.26825
作者: Zhiyu Li,Zeyu Zhang,Linyun Liu,Yiwei Bu,Chenyang Xi,Xichong Zhang,Yi Yu,Yu Yu,Zhichen Liu,Yuxiao Yi,Chunyu Li,Haoran Liu,Zhenkai Ding,Le Zhou,Juncheng Zhang,Boyu Chen,Zehao Lin,Guorui Li,Renyu Fu,Ji Li,Youyang Xia,Jiajun Shen,Harvey Li,Jingsong Zhou,Helan Hu,Tong Shen,Tianxiao Xu,Chenye Han,Weijie Li,Zhongquan Zhou,Hongming Xu,Yuekeng Li,Erenkai Zhu,Peng Yu,Wenhao Li,Jingyi Kang,Xixuan Hao,Xuzhou Zhu,Can Chen,Jiawei Liu,Chen Jiang,Tao Jiang,Fangwei Xiang,Zepeng Yao,Kai Chen,Guang Yang,Mingfei Lu,Chunhui Li,Zhengxuan Jiang,Bo Tang,Shiyu Zhan,Yihang Sun,Jie Chu,Wenjun Xiong,Zhenxiang Hua,Chen Lin,Detong Su,Dongyu Deng,Qiuping Ding,Guofeng Ding,Si Chen,Guohua Wen,Xu Chen,Zhi-Qin John Xu,Ying-Cong Chen,Ningyu Zhang,Haofen Wang,Muning Wen,Haotian Chen,Wentao Zhang,Xuanhe Zhou,Rita Zhang,Junchi Yan,Feiyu Xiong
机构: MemTensor Team
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 66 pages, 15 figures, 16 tables
Abstract:Memory-native evolution requires making memory an intrinsic capability of foundation models, enabling them to continuously accumulate and utilize information. To this end, we introduce MemT-1.0, a new architecture of memory foundation models that internalize memory as part of model computation. Building on Metis, MemT-1.0 employs Gated DeltaNet-2 with a stable token-level memory update mechanism within each turn. We further introduce Mixture-of-Storage to scale native memory capacity. Following a deliberate backbone preparation, we develop a two-stage training pipeline consisting of continued pre-training and post-training, supported by 34B tokens of training data and efficient training and inference infrastructure. We release a model family comprising MemT-1.0-Nano and MemT-1.0-Flash, and conduct a comprehensive evaluation of memory and general capabilities. The results show that they outperform the parametric memory baselines on average and remain competitive with retrieval-augmented methods, while largely preserving general capabilities. At a history length of 128K tokens, MemT-1.0-Nano achieves approximately 1.86\times end-to-end inference speedup and a 66\times reduction in persistent storage relative to full context KV cache. By internalizing retrieved procedural experience, MemT-1.0 improves task success on agentic tasks, providing a practical step toward memory-native evolution.
[NLP-9] SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum
【速读】： 该论文旨在解决生成式表示学习中嵌套前缀（nested prefixes）在低维压缩下缺乏显式约束，导致初始部分编码不够紧凑有效的问题。现有方法如Matryoshka表示学习（MRL）虽能支持灵活的表示尺寸，但在极端压缩场景下性能下降明显，其根本原因在于未对前缀信息进行有方向性的优化。为此，本文提出SpecMRL，其核心创新在于引入一种由模型参数导出的谱基（spectral basis），并基于特征重要性对特征空间进行重构，使前缀读取（prefix readout）在学习到的谱空间中进行。该设计通过显式强调高信息量方向，实现了更高效的表示压缩。实验在MTEB基准上验证了其有效性：SpecMRL在所有压缩维度下均优于原始MRL，尤其在高比例压缩时优势显著；更关键的是，其仅用一半嵌入维度即可超越全维MRL的表现，突破了传统MRL的性能边界。
链接: https://arxiv.org/abs/2610.26824
作者: Zhongxin Chen,Songhao Wu,Heng Cui,Cong Li,Xiao Zhou,Rui Yan
机构: Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学高瓴人工智能学院); School of Computer Science, Wuhan University(武汉大学计算机学院); Lenovo Group(联想集团)
类目: Computation and Language (cs.CL)
备注: preprint
Abstract:Matryoshka Representation Learning (MRL) is designed to support flexible representation sizes by jointly optimizing nested prefixes. While effective, this design lacks an explicit constraint to ensure that the leading portion provides a suitable compact encoding. To address this, we propose SpecMRL, which reorganizes the feature space to enable more effective prefix readout. Specifically, we extend MRL with a parameter-derived spectral basis organized with feature importance. SpecMRL otherwise follows standard MRL, with prefix readout performed in the learned spectral space. Experiments on MTEB show that SpecMRL outperforms MRL at all evaluated reduced dimensions, with larger gains under more aggressive compression. These results suggest that the learned spectral basis better prioritizes informative directions, enabling more effective representation compression. Notably, SpecMRL extends the performance frontier of MRL, surpassing full-dimensional MRL with only half the embedding dimensions.
[NLP-10] AfriBED: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages
【速读】： 该论文旨在解决自动语音识别（ASR）系统中存在的说话人属性偏差问题，即系统在识别不同性别、口音及说话人类型群体时表现不均，尤其在低资源非洲语言场景下，此类偏差缺乏有效的评估手段。其解决方案的关键在于构建首个针对低资源非洲语言——巴姆巴语（Bemba）和尼亚尼亚语（Nyanja）的语音识别偏见评估数据集——非洲语音偏见评估数据集（AfriBED），该数据集通过在不同性别与说话人类型子群体间保持可比性语音样本，实现了对性别和说话人类型属性偏差的量化检测与评估。数据集包含80名说话人的9小时巴姆巴语语音（3928个语句）以及127名说话人的约18小时尼亚尼亚语语音（6260个语句），基准实验验证了其在揭示目标语言中潜在偏见方面的有效性。
链接: https://arxiv.org/abs/2610.26821
作者: Claytone Sikasote,Hussein Suleman,Jan Buys
机构: University of Cape Town(开普敦大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: accepted at 2026 IEEE Spoken language Technology Workshop Demo Track
Abstract:Despite advancements in automatic speech recognition systems (ASR), current systems exhibit speaker-attribute bias, i.e., their speech recognition performance is biased against certain speakers based on their demographic attributes such as gender, accents, and speaker type. However, accurately assessing speaker attribute bias in a given ASR system requires the availability of a bias evaluation dataset with comparable speech across subgroups of a speaker attribute. In this paper, we introduce African speech Bias Evaluation Dataset (AfriBED), a novel speech recognition bias evaluation dataset for two low resource African languages, Bemba and Nyanja, to detect and estimate gender and speaker-type bias for ASR systems. The dataset comprises 9 hours of speech with 3928 utterances recorded by 80 speakers for Bemba, while the Nyanja set includes 6260 utterances with approximately 18 hours of speech recorded by 127 speakers. The results of our baseline experiments demonstrate that the datasets are useful for examining gender and speaker-type attribute bias in the target languages.
[NLP-11] ASTOR: Benchmarking and Improving Autoformulation for Soft-Constraint Optimization with a Formal Grammar
【速读】： 该论文旨在解决现实决策问题中软约束（soft constraints）建模缺失的问题，即现有基于大语言模型（LLM）的自动建模方法普遍忽略消费者、管理者及规划者提出的可违反偏好，而这些偏好在实际应用中往往至关重要。其解决方案的关键在于提出ASTOR框架——一个推理时（inference-time）的系统，通过将自然语言描述映射为语法引导的抽象语法树（abstract syntax tree, AST），并在此结构化表示基础上确定性地编译为可执行的优化代码，从而有效支持软约束的显式建模与表达。此外，研究还引入了SoftOR这一统一基准，专门聚焦于软约束的自然语言表述，验证了ASTOR在包含软约束的任务上优于现有单请求与多智能体方法，在保持对传统硬约束基准竞争力的同时，显著提升了对复杂偏好建模的能力。
链接: https://arxiv.org/abs/2610.26819
作者: Amer Goel,Yeganeh Alimohammadi,Andrew Daw
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Real-world decision problems combine two qualitatively different kinds of constraints: requirements that must be satisfied, called hard constraints, and preferences that can be violated, called soft constraints. Because these problems are often posed to decision-makers in natural language, large language models offer a natural tool for assisting with their solution. This has given rise to autoformulation, which focuses on using LLMs to translate natural-language descriptions of decision problems into executable mathematical optimization models. However, this research often neglects soft constraints like those proposed by consumers, managers, and planners, despite the fact that many practical problems require modeling them explicitly. We introduce ASTOR, an inference-time framework that maps a natural-language decision problem to a grammar-guided abstract syntax tree (AST) before deterministically compiling the resulting representation into executable optimization code. We also introduce SoftOR, a unified benchmark of natural language problems which focuses on soft constraint formulations. On SoftOR, ASTOR achieves the best aggregate accuracy among comparable single-request and multi-agent LLM methods while remaining competitive on standard autoformulation benchmarks that include only hard constraints. Code and data are available at this https URL.
[NLP-12] Safety Mid-Training: Internalizing LLM Safety as a Foundational Capability
【速读】： 该论文旨在解决大语言模型（LLM）在安全对齐（safety alignment）与通用能力（general capability）之间存在的权衡问题。传统方法通常在后训练阶段进行安全对齐，而近期研究尝试将安全数据引入预训练阶段，但后续训练若不再包含安全数据，安全性能会退化，且提升安全性往往以牺牲通用能力为代价。为此，本文提出一种新的“安全中段训练”（safety mid-training）范式，作为预训练与后训练之间的中间阶段，通过共享的推理结构同时学习安全性和通用推理能力——二者均依赖于对意图或任务目标的理解、相关知识的应用以及对潜在后果的推理。该方案利用基础模型生成结构化文档和推理轨迹，基于安全与效用提示进行联合训练。实验表明，安全中段训练可使攻击成功率（ASR）相对Qwen3-1.7B、4B和8B基线模型降低72%至81%，同时显著提升通用能力，且该优势在后续仅使用通用指令微调（SFT）或仅使用通用奖励策略优化（GRPO）时仍能保持。更重要的是，从安全中段训练模型初始化的强化学习（RL）能到达更优的安全-效用前沿。因此，核心解决方案在于：将安全视为与通用推理共用同一认知结构的能力，并通过中段训练阶段实现二者的协同学习，从而有效缓解安全-效用权衡，增强安全性的鲁棒性。
链接: https://arxiv.org/abs/2610.26809
作者: Zhengyue Zhao,Liwei Jiang,Yejin Choi,Chaowei Xiao1
机构: Johns Hopkins University(约翰霍普金斯大学); University of Washington(华盛顿大学); Stanford University(斯坦福大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Safety alignment for large language models (LLMs) is primarily performed during post-training, while recent work has also begun incorporating safety data into pre-training. Yet safety can degrade when later training no longer includes safety data, and gains in safety often come at the cost of general capability. This tension suggests that safety and general reasoning should not be treated as separate training goals. We instead view safety as an LLM capability that can be learned through a reasoning structure shared with general tasks. Both require the model to understand intent or task goals, apply relevant knowledge, and reason about potential consequences. We introduce safety mid-training, a stage between pre-training and post-training that jointly develops safety and general reasoning through this shared structure. Given safety and utility prompts, the base model generates structured documents and reasoning traces that are then used for mid-training. Safety mid-training reduces average attack success rate (ASR) by 72% to 81% relative to Qwen3-1.7B, 4B, and 8B base models, while also improving general capability. The benefits persist through subsequent post-training. After general-only SFT, mid-trained models largely retain their safety, achieving lower ASR than the baselines without sacrificing general capability. They also retain safety under general-only GRPO without safety rewards, while subsequent RL reaches a better safety-utility frontier when initialized from the mid-trained models. Together, these results suggest that learning safety and general reasoning within a shared mid-training stage can reduce the safety-utility trade-off and make safety more robust to subsequent training.
[NLP-13] RESIN: Multi-Chain Reasoning with Entity-Structured Indexing
【速读】： 该论文旨在解决多链推理（multi-chain reasoning）中证据分散与不同推理环节间存在相互依赖关系所带来的挑战，尤其针对传统扁平化检索（flat retrieval）易引入松散相关文本，以及过早选定单一推理路径可能导致后续有用证据被丢弃的问题。其核心解决方案是提出RESIN（Reasoning with Entity Structured Indexing），通过结构化实体索引（Entity Structured Indexing）精准定位证据，并在保持原始来源作为推理基础的前提下，将问题分解为聚焦的一跳式推理需求（one-hop needs），逐跳检索少量基于源文档的证据单元（source-grounded units），并将其存入紧凑的推理缓冲区（reasoning buffer）。该缓冲区保留所有潜在支持证据直至最终组合阶段，从而实现中间决策的显式表达，避免强制形成脆弱的单一推理路径。实验结果表明，在完整的HolisQA基准测试中，RESIN实现了82.92%的上下文召回率（context recall, CR）和76.70%的答案相关性（answer relevancy, AR），在五个领域均优于已有基线模型，且在四个领域中F1分数更高。消融实验进一步验证了类型化证据多样性（typed evidence diversity）与逐跳分解（hop-wise decomposition）的关键作用，明确指出：精确的证据组织与延迟承诺（delayed commitment）是实现高效、可审计的多链推理的核心机制。
链接: https://arxiv.org/abs/2610.26782
作者: Xuanyu Meng,Jash Rajesh Parekh,Jiawei Han
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Computation and Language (cs.CL)
备注: 22 pages, preprint, including appendices
Abstract:Multi-chain reasoning is a challenging task not only because the multi-chain evidence is dispersed, but also because different parts of reasoning may require distinct, interacting chains of support. Flat retrieval can flood a model with loosely related text, whereas early selection of a single reasoning path can discard evidence that becomes useful only after other facts are considered. We introduce RESIN (Reasoning with Entity Structured Indexing), which uses structure to locate evidence while keeping the original source as the basis for reasoning. RESIN decomposes a question into focused one-hop needs, retrieves a small set of source-grounded units for each hop, and places only this relevant evidence in a compact reasoning buffer. The buffer preserves alternative supports until final composition, making intermediate decisions explicit without forcing a brittle path. On the full HolisQA benchmark, RESIN achieves 82.92 context recall and 76.70 answer relevancy over all questions, with higher reported CR and AR across five domains and higher F1 in four domains than the published baselines. Ablations show that both typed evidence diversity and hop-wise decomposition are essential, supporting a clear conclusion: Precise evidence organization and delayed commitment make multi-chain reasoning effective and auditable. Our code is available at this https URL.
[NLP-14] Game of Phones: Harnessing Game Theory and LLM s in Bot Wars to Counteract Phone Scams ECML KDD2026
【速读】： 该论文旨在解决电话诈骗（phone scams）日益增长的经济与社会成本问题，特别是现有反制措施对诈骗者无资源成本、难以形成有效威慑的缺陷。传统被动防御手段无法有效遏制诈骗活动的持续演化，而主动式反诈策略（如战略钓鱼，strategic scam-baiting）虽具潜力，但需动态适应诈骗者的策略变化，这对静态系统构成挑战。论文提出一种基于博弈论（game-theoretic）的框架，使大语言模型（LLM）能够在不进行微调（fine-tuning）的前提下，通过上下文学习（in-context learning）实现对抗策略的自适应选择。该框架将诈骗者与反诈代理的交互建模为具有实证基础策略与效用函数的双人博弈，并以纳什均衡（Nash equilibrium）作为评估学习行为有效性的形式化基准。在300次模拟对话及三种不同LLM架构上的实验表明，采用自适应策略选择的模型相较于非自适应基线，其行为与纳什均衡的逼近程度提升了28.9%至79.7%。该研究的关键在于构建了一个可动态演化、无需微调的博弈学习框架，实现了高效且可扩展的智能反诈能力。
链接: https://arxiv.org/abs/2610.26781
作者: Nardine Basta,Robin Carpentier,Benjamin Zi Hao Zhao,Dali Kaafar
机构: Macquarie University (麦考瑞大学)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Science and Game Theory (cs.GT)
备注: Accepted at the 2026 Joint European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD 2026)
Abstract:Phone scams impose substantial and growing costs on individuals and financial systems worldwide, yet existing countermeasures impose no resource cost on perpetrators. Strategic scam-baiting, in which automated agents engage scammers to deplete their operational capacity, offers a proactive alternative but requires dynamic strategy adaptation that static systems cannot provide. We introduce a game-theoretic framework enabling LLMs to learn effective counter-strategies through in-context experience without fine-tuning. The framework models scammer-baiter interactions as a two-player game with empirically grounded strategies and utility functions, using Nash equilibrium as a formal benchmark for evaluating learned behaviour. Across 300 simulated dialogues and three LLM architectures, adaptive strategy selection achieves 28.9-79.7% closer alignment to Nash equilibrium than non-adaptive baselines. The code and an annotated dataset of 300 dialogues with turn-level strategy labels and utility scores are released at this https URL .
[NLP-15] Monsoon: Scalable Spontaneous Speech Collection for Low-Resource ASR in 50 Languages ICASSP2027
【速读】： 该论文旨在解决自动语音识别（ASR）在多数语言中受限于缺乏高质量、真实口语对话转录数据的问题。现有主流方法受限于数据规模与质量，难以有效支撑低资源语言的高性能语音识别系统。其解决方案的关键在于提出名为Monsoon的可扩展数据采集框架：通过配对用户进行非脚本化对话，利用自动化分类器筛选录音片段，经过五级人工校验提升转录准确性，并基于词汇与声学覆盖度选择高质量语料。该框架已成功收集50种语言的数据，其中7个3,500小时语料库的分析表明，在相同数据量下，其多级质量控制流程可使语义词错误率（Semantic WER）相对降低10%-39%。相较于基于公开语料训练的同架构基线模型，使用Monsoon数据微调的Whisper模型不仅在基线测试集上表现相当，更在对话类基准测试中将语义词错误率降低20-44点，且性能增益随数据量增加而持续提升，具有良好的泛化能力。研究团队已开源该框架、质量控制模型及微调后的模型。
链接: https://arxiv.org/abs/2610.26776
作者: Shobhit Banga,Manas Dhir,Aaditya Pareek,Amritansh Walecha,Bhaskar Singh,Hanuman Sidh,Manmeet Kaur,Sagar Jain,Supriya Paul,Vanshika Chhabra,Utkarsh Singh
机构: 未知
类目: Computation and Language (cs.CL)
备注: Submitted to ICASSP 2027
Abstract:For most languages, automatic speech recognition (ASR) is limited less by model architecture than by a lack of accurately transcribed spontaneous speech. We present Monsoon, a framework for collecting such data at scale. It elicits unscripted conversations between paired contributors, screens recordings with automated classifiers, corrects transcripts over five levels of human review, and selects segments for lexical and acoustic coverage. It has collected speech in 50 languages; we analyse seven 3,500-hour corpora. At equal volume, its quality stages reduce semantic word error rate (WER) by 10-39% relative to a conventional protocol. Compared with an architecture-matched baseline trained on public corpora, Whisper fine-tuned on Monsoon performs comparably on the baseline’s own test sets and reduces semantic WER by 20-44 points on conversational benchmarks. The gains transfer to a second architecture and grow with data volume. We release the framework, its quality-control models, and the fine-tuned models.
[NLP-16] HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs
【速读】： 该论文旨在解决现有生成式AI（Generative AI）代理在构建任务执行框架（harness）过程中，虽通过递归自我改进（Recursive Self-Improvement, RSI）提升了任务性能，但伴随显著增加的执行开销的问题。其核心挑战在于如何在不依赖真实答案或官方测试用例的前提下，实现高效、低成本且高性能的框架演化。解决方案的关键在于提出HarnessPro——一种无标签（label-free）的框架进化方法：利用大语言模型（LLM）自动生成测试用例进行候选框架评估，由专用代理分析执行轨迹，基于不同性能-成本权衡策略规划并实施更新；通过归一化加权平方距离选择机制，输出轻量型（lite）、平衡型（medium）和性能优先型（heavy）三种适配不同部署预算的框架变体，并在迭代中利用执行反馈持续优化。实验表明，在Terminal-Bench 2、SWE-Bench Pro和ProgramBench等多个基准上，使用GPT-5.5+Codex时，HarnessPro的heavy模式相较RHO平均提升2.0个百分点的通过率，同时每任务成本降低26.1%；该优势在使用GLM-5.2+Claude Code时亦保持一致。
链接: https://arxiv.org/abs/2610.26770
作者: Zhuoshi Pan,ZhiYuan Feng,Yukuo Cen,Honglin Lin,Junru Lu,Zhongxia Chen,Shuo Yang,H. Vicky Zhao,Jie Tang
机构: Tsinghua University(清华大学); Z.AI Co., Ltd.; Shanghai Jiao Tong University(上海交通大学); University of Warwick; Kuaishou–Kuaishou Technology
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 41 pages, 4 figures, 4 tables, under review
Abstract:Harness development plays a crucial role in building capable AI agents. Recent approaches improve harnesses through recursive self-improvement (RSI), but their performance gains often come with higher execution costs. We introduce HarnessPro, a label-free method for evolving harnesses to improve task performance while reducing execution cost. In this setting, candidate harnesses are evaluated using LLM-generated tests without access to ground-truth solutions or official tests. Specialized agents diagnose execution trajectories, plan updates with different performance-cost priorities, and implement them as candidate harnesses. Rather than returning a single harness, HarnessPro uses normalized weighted squared-distance selection to provide cost-focused (lite), balanced (medium), and performance-focused (heavy) harnesses for different deployment budgets. Selected harnesses enter the next iteration, where execution feedback guides further updates. Our experiments cover Terminal-Bench 2, SWE-Bench Pro, and ProgramBench, using GPT-5.5 with Codex and GLM-5.2 with Claude Code. Across these benchmarks using GPT-5.5 with Codex, the heavy mode of HarnessPro achieves an average pass-rate gain of 2.0 percentage points and an average per-task cost reduction of 26.1% relative to RHO. This advantage also holds when using GLM-5.2 with Claude Code. Code is available at this https URL
[NLP-17] okenGate: Adapting LLM s for Fast and Memory-Efficient Prefill with Learned Token Pruning
【速读】： 该论文旨在解决大模型在处理长上下文输入（如代码仓库和长文档）时存在的高计算成本与内存开销问题。传统语言模型对所有输入标记（token）进行逐层处理，导致预填充（prefill）阶段效率低下。为实现高效且可学习的标记剪枝，论文提出TokenGate——一种轻量级、即插即用的组件，能够无缝集成至已训练模型中，支持端到端可学习的标记裁剪。其核心创新在于引入“门控注意力”（token-gated attention）机制，该机制受混合专家模型（Mixture-of-Experts, MoE）启发，通过令牌重要性得分重新加权注意力逻辑值，从而在非可微分的Top-K剪枝操作下仍能实现梯度传播与端到端训练。实验表明，TokenGate在多种任务中实现了1.4–11.3倍的首次标记生成时间（TTFT）加速和1.5–3.4倍的峰值内存降低，且无准确率损失；相比现有最优剪枝方法，其预填充速度提升高达6.4倍，内存占用降低2倍。此外，TokenGate可原生适配标准Transformer架构（如Qwen3）及混合注意力模型（如Qwen3.5），仅需类似参数高效微调（PEFT）级别的轻量级调整即可部署。
链接: https://arxiv.org/abs/2610.26754
作者: Liana Patel,Rishabh Ranjan,Ion Stoica,Matei Zaharia,Carlos Guestrin
机构: Stanford University (斯坦福大学); UC Berkeley
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Expert human readers process large inputs (e.g., code repositories and long documents) by skimming to closely read only what matters. On the other hand, language models process all input tokens through each model layer, resulting in significant processing costs for long-context inputs. We introduce TokenGate, a small, drop-in component that extends existing trained models to enable end-to-end learnable token pruning over long-context inputs for fast and memory-efficient prefill. To enable end-to-end learning despite the non-differentiable top-K operator for token pruning, we introduce token-gated attention, a mechanism inspired by Mixture-of-Experts that re-weights attention logits using token importance scores. Across diverse tasks, TokenGate achieves 1.4 - 11.3\times faster time-to-first token (TTFT) and 1.5 - 3.4\times lower peak memory at no accuracy degradation for prefill in transformers. Compared to the best-performing prior methods for token-pruning, TokenGate gives up to 6.4\times faster prefill and 2\times lower peak memory. TokenGate natively extends both standard transformers (e.g., Qwen3) and hybrid-attention models (e.g., Qwen3.5) with simple lightweight adaptation comparable to standard parameter-efficient fine-tuning. We make our code available at: this https URL.
[NLP-18] recAMPLIFY: Large-Scale Pretraining of Recursive Protein Language Models
【速读】： 该论文旨在解决蛋白质语言模型（pLMs）在性能提升过程中过度依赖模型规模与数据量所带来的高计算成本及过拟合问题，同时针对测试时扩展（test-time scaling）方法在非自回归掩码编码器架构中的适用性挑战。现有基于自回归大语言模型的测试时扩展方法无法直接迁移至掩码编码型pLMs，因其缺乏生成循环机制且氨基酸序列不具备自然语言推理的语义表达能力。为此，论文提出一种递归预训练范式（recursive pretraining）的适配方案——recAMPLIFY，其核心在于采用权重共享的Transformer主干网络，在可调迭代次数内对残基对齐的隐藏状态进行多轮精细化更新。该模型仅含1.55亿参数，远少于同类规模的AMPLIFY-350M，但在相同2T token数据集上训练后，仍显著提升CASP接触恢复率及四项监督任务中的三项表现，尤其在折叠准确率上提高22.5%。为进一步优化计算效率，论文进一步提出门控混合递归（Gated Mixture Recurrence, GMR）框架，通过学习残差门控机制（在标量、低秩或逐维度层面），以概率加权方式融合递归过程中的多个中间隐藏状态，实现动态停止与信息选择的统一建模，将最优停顿深度与生物特性相关联。该方法在保持性能的同时，使期望层数减少最高达37%，且其决策行为可通过机制分析验证：残差门控不仅调控递归稳定性，还与蛋白质的生物学属性存在内在关联。
链接: https://arxiv.org/abs/2610.26747
作者: Saurav Jha,Qianli Liao,Robert Vernon,Sarath Chandar,Quentin Fournier
机构: Mila – Quebec AI Institute(蒙特利尔人工智能研究所); Polytechnique Montréal(蒙特利尔理工学院); Center for Brains, Minds, and Machines(大脑、心智与机器中心); MIT(麻省理工学院); Amgen(安进公司); Canada CIFAR AI Chair(Canada CIFAR人工智能主席); University of Montréal(蒙特利尔大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Models: this https URL Code: this https URL
Abstract:Protein language models (pLMs) are powerful tools for protein annotation and mutation-effect prediction, but progress has largely depended on scaling model size and data, making them expensive and prone to memorization. Test-time scaling offers a complementary path, but methods developed for autoregressive LLMs do not transfer directly: masked encoders lack a generation loop, and amino acids do not verbalize in a reasoning language. Instead, we adapt recursive pretraining to masked pLMs, where a weight-tied Transformer core refines residue-aligned hidden states over an adjustable number of cycles. We instantiate this recipe as recAMPLIFY, a pLM with 155M unique parameters trained on the same 2T tokens as AMPLIFY-350M. Despite 56% fewer parameters, it improves CASP contact recovery and three out of four supervised tasks, notably raising fold accuracy by 22.5%. Because optimal depth for recAMPLIFY varies across tasks and assays, we then introduce Gated Mixture Recurrence (GMR), a framework that decodes each protein from a probability-weighted mixture of hidden states visited during recursion. GMR employs a learned residual gate at scalar, low-rank, or per-dimension resolution to control how much of each cycle’s proposed update is accepted into the state. This unifies halting and gating into a single adaptive computation objective while recovering PonderNet-style halting as a special case without gates. Layered on top of recAMPLIFY, GMR further reduces expected layer depth by up to 37%, although the accuracy trade-off varies by task. Mechanistic analyses show that residual gating controls recurrence stability beyond the trained depth and that halt depth tracks biological properties.
[NLP-19] EpiLens: Lightweight RAG for Epidemiological Knowledge Acquisition
【速读】： 该论文旨在解决在应对未来疫情暴发时，科研人员难以高效获取已有研究及其所用数据的问题。传统手动文献梳理耗时耗力，而单纯依赖大语言模型又存在误述或混淆文献来源的风险。其解决方案的关键在于提出EpiLens——一个轻量级、开源的检索增强生成（Retrieval-Augmented Generation, RAG）工具包，通过固定的检索模板和输出结构对语料库中的每篇论文进行处理，并将生成结果与原始引用段落建立关联。该方法无需任务特定训练数据，在未微调的情况下仍能在四项任务中达到或超越监督学习与零样本基线的表现。通过对2,297篇多病原参数研究来源文献的分析，研究发现：相较于致死率相关研究，与基本再生数（R₀）相关的论文被标记为“基于模型”的比例显著更高（54% vs. 22%，人工标注结果分别为62% vs. 24%），这一差异反映了传播模型在估算基本再生数中的普遍应用；同时，不同病原体在潜伏期估计方面的模型化程度存在显著差异，如SARS-CoV-2为55%，而猴痘病毒仅为6%，揭示了不同病原体研究范式的异质性。
链接: https://arxiv.org/abs/2610.26730
作者: Vincenzo Perri,Ciro Cattuto,Daniela Paolotti
机构: ISI Foundation(ISI基金会), Turin(都灵), Italy(意大利)
类目: Computation and Language (cs.CL); Digital Libraries (cs.DL); Physics and Society (physics.soc-ph)
备注:
Abstract:To prepare for future outbreaks, researchers need to know what studies have been done and what data they use. Mapping a large literature by hand is slow, while large language models alone can misstate or conflate sources. We present EpiLens, a lightweight, open-source retrieval-augmented generation package that applies a fixed retrieval template and output schema to every paper in a corpus and links outputs to source passages. We build a corpus of 2,297 papers from the sources of a multi-pathogen parameter study and evaluate EpiLens’s paper-classification workflow against 214 manually labelled papers. Without task-specific training data, the classifier matches or exceeds the tested supervised and zero-shot baselines across four tasks. The study’s early extraction sheets cite published sources for 327 estimates. We link each estimate to a cited document or to corpus studies included by a review. Then, we use the classifier’s labels to compare the papers associated with different parameters and pathogens. EpiLens labels more papers linked to reproduction numbers than fatality risks as model-based (54% versus 22%, compared with 62% versus 24% in the manual labels). The gap is consistent with the frequent use of transmission models to estimate reproduction numbers and persists within each of five pathogens compared. Beyond the contrast between parameters, the model-based share of papers linked to incubation-period estimates differs by pathogen: 55% for SARS-CoV-2 versus 6% for mpox.
[NLP-20] Copying Under Repetition: From Prefix Matching to Position Hashing
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理长串重复字符时出现的复制溢出（overrun）或截断（truncate）问题。尽管前沿模型能够几乎完美地复制随机文本，但在面对长距离重复模式时，其基于局部内容匹配的复制机制失效，导致无法准确判断何时继续复制或终止。核心解决方案在于设计一种双层变压器架构，其具有 $ O(\log\log N) $ 位精度和 $ O(\log N / \log\log N) $ 的宽度，通过引入类似旋转位置编码（RoPE）的注意力值与输出旋转（VO-rotations），利用小素数周期对分隔符锚点的距离进行模运算编码。借助中国剩余定理（Chinese Remainder Theorem），这些周期性编码可联合区分所有必要的距离信息，从而实现单一检索规则同时选择下一个源词元或结束标记。相比之下，论文证明了仅依赖加性注意力偏置且最终趋于恒定的位置编码方法（如NoPE及ALiBi的递归实现）存在宽度无关的不可行性：当相同词元序列超过饱和阈值后，其逻辑值将不再变化，若该阈值随 $ N $ 亚线性增长，则模型无法既完成长串全相同复制又正确终止——这一现象与前沿模型的实际终止错误高度一致。在合成复制实验中，基线变压器模型即使在训练范围内也难以处理长重复字符串；而经过训练的VO模型则能在接近5005长度（即所选素数周期乘积）内以近乎完美的准确率泛化至训练长度之外。
链接: https://arxiv.org/abs/2610.26722
作者: Anxin Guo,Tianlong Huang,Chenxiao Yang,Zhiyuan Li
机构: Northwestern University (西北大学); Duke University (杜克大学); Toyota Technological Institute at Chicago (丰田技术学院芝加哥分校)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Frontier LLMs can copy random text almost flawlessly, yet they often overrun or truncate long blocks of repeated characters. In repeated strings, matching local content no longer reliably identifies where copying should continue or when it should stop. Nevertheless, we show that a two-layer transformer with O(\log\log N) -bit precision and O(\log N/\log\log N) width can copy every string up to a chosen design horizon N and terminate correctly under arbitrary repetition. We use RoPE-style rotations of attention values and outputs (VO-rotations) with small prime periods to encode distances from delimiter anchors modulo those primes. By the Chinese Remainder Theorem, these periodic encodings jointly distinguish all required distances, allowing one retrieval rule to select either the next source token or the end marker. In contrast, we prove a width-independent impossibility result for copying under sequential finite-precision softmax accumulation. It applies to transformers whose only positional input is an additive attention bias that eventually becomes constant, including NoPE and a recursive implementation of ALiBi. Once an identical-token run exceeds the resulting saturation threshold, further repetitions leave the logits unchanged. If this threshold grows sublinearly with N , these models cannot both complete a long all-identical copy and stop correctly, a pattern resembling the termination errors of frontier LLMs. In synthetic copying experiments, baseline transformer models fail on long repeated strings even within the training range. Trained VO models, by contrast, generalize beyond the training lengths with near-perfect accuracy up to length \approx 5005 , the selected primes’ period product.
[NLP-21] HoloCosm: Beyond Single-Metric Benchmark Compression Across Multiple Metrics
【速读】： 该论文旨在解决多维度评估场景下基准压缩（benchmark compression）中因单一指标引导样本选择而导致的性能失真问题。现有方法通常依赖单一评价指标进行样本筛选，然而在实际评测中，模型表现往往涉及多个异构维度，不同指标间对模型性能的评估结果相关性较弱，导致仅基于某一指标选取的压缩样本难以有效保留其他指标上的得分与排名一致性。当压缩率达80%时，五种单指标基线方法在超过70%的跨指标对比中，其重构误差高于等规模随机采样。为应对这一挑战，本文提出HoloCosm框架，其核心在于联合优化源模型与各评估指标间的归一化得分重构误差（normalized score-reconstruction error），同时约束排名扭曲（ranking distortion），实现多维度性能的均衡保留。在15个涵盖语言、多模态、视觉-语言-动作（VLA）及智能体的基准配置上，共包含85个评估子指标的实验表明，在80%压缩率下，HoloCosm显著优于随机采样和现有最强基线：平均NMAE降低35.3%至0.236，最差组件的NMAE下降39.5%至0.330，平均Spearman相关系数从0.884提升至0.915；同时在单指标评估任务中仍保持竞争力。
链接: https://arxiv.org/abs/2610.26714
作者: Yupei Lin,Zhongzhan Huang,Junxin Li,Guoming Ling,Shanshan Zhong,Hefeng Wu,Yukai Shi,Liang Lin
机构: Sun Yat-sen University (中山大学); Guangdong University of Technology (广东工业大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Benchmark compression reduces evaluation cost by selecting a small set of samples. Existing methods typically guide selection using a single metric, although benchmarks often assess models along multiple dimensions. We find that model rankings are often weakly correlated across metrics, indicating that different metrics capture distinct aspects of model performance. Samples selected for one metric do not reliably preserve scores and rankings on others when used for compressed evaluation. At 80% compression, each of the five single-metric baselines yields higher reconstruction error than equal-size random sampling in over 70% of cross-metric comparisons. To address this challenge, we present HoloCosm, which jointly minimizes normalized score-reconstruction error across source models and metrics while constraining ranking distortion. We evaluate HoloCosm across multiple compression rates on 15 language, multimodal, VLA, and agent benchmark configurations with 85 component metrics. With 80% compression, HoloCosm outperforms Random, the strongest baseline overall: average NMAE drops by 35.3% to 0.236, worst-component NMAE drops by 39.5% to 0.330, and average Spearman correlation rises from 0.884 to 0.915. In addition, HoloCosm remains competitive with existing baselines in single-metric evaluation. We will release our code and experimental results.
[NLP-22] Hierarchical Token Merging with Recursive Autoencoders for Unsupervised Grammar Induction
【速读】： 该论文旨在解决无监督条件下从数据中自动学习有意义的层次化表示结构这一核心挑战，尤其关注在缺乏显式结构标注的情况下，如何通过压缩机制自发形成具有组合性（compositional）特征的层级抽象。其解决方案的关键在于提出一种分层令牌合并自编码器（hierarchical token-merging autoencoder），该模型通过递归合并相邻令牌生成越来越抽象的表示，并利用变分信息瓶颈实现对完整序列的压缩；同时，通过动态规划对可能的二叉树结构进行边缘化处理，联合学习表示与层次结构。为应对训练不稳定性问题，引入基于无监督共识与知识蒸馏的多模型集成策略。实验在合成模糊语法数据集上验证了该方法可自发产生高保真层次结构，达到93%的句法解析有效性与精确重构率，显著优于最强基线（42%解析有效性），证明了仅依赖压缩目标即可驱动层次结构发现，为无监督结构探索与抽象学习提供了概念验证。
链接: https://arxiv.org/abs/2610.26695
作者: Murat Guney Kemal,Robert Lieck
机构: Durham University (杜伦大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Hierarchical abstraction is a fundamental principle across many data modalities, enabling complex sequences to be represented compositionally at multiple levels. Learning such hierarchies directly from data, without explicit structural supervision, remains a central challenge in representation learning. We investigate whether meaningful hierarchical structure can emerge from compression alone. While autoencoders learn compressed representations through reconstruction, this objective does not necessarily favour meaningful compositional structure. We study this challenge through unsupervised grammar induction, where the underlying hierarchy can be evaluated explicitly. We introduce a hierarchical token-merging autoencoder that recursively combines adjacent tokens into increasingly abstract representations, compressing entire sequences through a variational information bottleneck. The model jointly learns representations and hierarchical structure by marginalising over possible binary trees using dynamic programming. To address training instability, we combine independently trained models through unsupervised consensus and distillation. Experiments on a synthetic, ambiguous grammar demonstrate the emergence of meaningful hierarchical structure without structural supervision. The resulting model achieves 93% grammatical parse-validity and 93% exact reconstruction on held-out sequences, compared with 42% parse-validity for the strongest trivial baseline. These findings establish a proof of concept for hierarchical token merging as a mechanism for unsupervised structure discovery and abstraction learning.
[NLP-23] Parallel Language Models via Variational Decoding
【速读】： 该论文旨在解决自回归语言模型（autoregressive language models）在推理过程中每次前向传播仅生成一个词元（token），导致推理延迟较高，尤其在模型规模扩大以提升性能时问题更为显著的问题。其核心解决方案是提出并行语言模型（Parallel Language Models, PLMs），通过将生成过程中的依赖关系路由至潜在的辅助变量（latent auxiliary variables），实现一次前向传播中并行生成多个相互一致的词元。PLMs采用基于变分下界（variational bound）的统一训练目标，该目标可分解为自回归损失与并行解码损失之和，从而在单一训练目标下实现与自回归模型相当的训练质量。实验表明，PLMs在推理速度与生成质量之间达到了更优的帕累托前沿，其预训练模型在80M至3B参数规模下可达到接近自回归模型的预训练损失，且推理加速呈模型规模的幂律增长；经过后训练的1.3B参数模型在数学、代码生成和指令遵循等基准测试中表现与同规模自回归模型相当或更优，推理速度提升2.3至5.7倍。研究证明，通过统一的训练机制，可实现高质量的并行生成，显著降低推理延迟。
链接: https://arxiv.org/abs/2610.26693
作者: Felix Draxler,Robin C. Geyer,Daniel Geyfman,Farrin Marouf Sofian,Justus Will,Philipp Hildebrandt,Giosue Migliorini,Prakhar Srivastava,Gagan Sai Trinadh Duggaraju Peddiraju,Kushagra Pandey,Jorge da Silva Gonçalves,Elijah Tavares,Jan Niklas Groeneveld,Yang Meng,Julia E. Vogt,Stephan Mandt
机构: University of California, Irvine, CA, USA; ETH Zurich (瑞士联邦理工学院); Charité Berlin (柏林夏里特医学院); Hasso Plattner Institute (哈索·普拉特纳研究所), Potsdam, Germany
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Autoregressive language models emit one token per forward pass. This causes high inference latency, in particular as models are scaled up for quality. In this paper, we introduce Parallel Language Models (PLMs), which generate mutually consistent tokens in one forward pass by routing dependencies through latent auxiliary variables. PLMs are trained using a variational bound that reduces to an autoregressive loss plus a parallel decoding loss. Empirically, PLMs are on a faster and better Pareto frontier than autoregressive and discrete diffusion baselines. Pretrained PLMs achieve near-autoregressive pretraining loss on model scales of 80M to 3B parameters, and speedup follows a power law in model size. When post-trained, our 1.3B model matches or exceeds an identically trained autoregressive model on math, coding, and instruction-following benchmarks, speeding up inference 2.3\text-5.7\times . Together, we show that parallel generation can be trained from a single training objective at autoregressive quality. We will make our code available at this https URL.
[NLP-24] Intra-Rubric Reward Divergence: Mitigating Rubric Hacking through Latent Dimension Recovery and Balancing
【速读】： 该论文旨在解决基于评分体系（rubric）的强化学习中普遍存在的“评分作弊”（rubric hacking）问题，即模型为最大化总分而牺牲某些关键质量维度，导致评估结果与真实质量之间出现系统性偏差。其核心问题是：在多维度评分体系下，模型因过度优化总分而引发维度内部奖励分歧（Intra-Rubric Reward Divergence），表现为部分子维度得分随总分上升而系统性下降，进而造成维度失衡（dimension imbalance）与维度缺失（dimension absence），严重影响模型性能与训练鲁棒性。解决方案的关键在于提出两项创新机制：一是公平竞争维度聚合（Fair-Race Dimension Aggregation, FRDA），通过在每个维度内独立进行生成代际竞争，缓解维度间权重失衡；二是残差语义维度补全（Residual Semantic Dimension Completion, RSDC），利用整体偏好信号隐式挖掘被遗漏的评价维度，实现对评分体系的完整覆盖。实验表明，FRDA显著提升了训练稳定性并抑制了评分作弊行为，而RSDC在对话、科学和医疗等多个领域均优于单一奖励训练及朴素多奖励聚合方法，生成响应的整体质量更高。
链接: https://arxiv.org/abs/2610.26685
作者: Jianqiang Xia,Han Xu,Dianxi Shi
机构: Shanghai Jiao Tong University (上海交通大学); Baidu, Inc. (百度公司); Intelligent Game and Decision Lab (智能游戏与决策实验室)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 10 pages, 7 figures, 4 tables
Abstract:Rubric hacking limits both performance and training robustness in rubric-based reinforcement learning. By decomposing rubrics into sub-dimensions, we reveal the Intra-Rubric Reward Divergence: as the total rubric score increases, certain sub-dimension scores systematically decrease, suggesting the model sacrifices specific quality aspects to optimize the aggregated objective. From the sub-dimension perspective, we observe two phenomena: (1) dimension imbalance, causing the policy to favor heavily covered dimensions; (2) dimension absence, leading to a systematic gap between rewards and true quality. To mitigate these issues, we propose Fair-Race Dimension Aggregation (FRDA), which has generations compete within each dimension separately to mitigate rubric imbalance, and Residual Semantic Dimension Completion (RSDC), which mines evaluation dimensions implicitly encoded in a holistic preference signal to complete rubric coverage. Experiments in chat, science, and medical domains indicate that FRDA mitigates rubric hacking with more stable improvements, while RSDC outperforms both single-reward training and naive multi-reward aggregation, yielding generally higher-quality responses.
[NLP-25] Which Model for Which Role? Cost-Efficient Pareto Recommendation for LLM Agent ic Workflows AAAI NEURIPS2026
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体工作流中多模型角色分配的优化问题，即在任务质量、部署成本与延迟之间权衡时，如何高效地寻找帕累托最优（Pareto-optimal）的模型组合。传统方法需对大量候选配置进行独立评估，导致高昂的评估开销。其核心解决方案是提出一种多目标扩展的代价耦合吉廷斯（Cost-Coupled Gittins, CC-Gittins）搜索算法，该算法利用评估成本与部署成本之间的耦合关系——即高部署成本的配置同样具有更高的评估成本——通过交替探索更高任务质量与更低部署成本的目标，复用每次评估结果同时指导两个目标的搜索。在每一批次评估后，基于不确定性调整后的质量与成本估计，推荐非支配（nondominated）的候选方案。实验在包含2至5个角色、最多1,681个候选配置的工作流上进行，涵盖八组质量-成本矩阵，并进一步引入延迟作为第三目标。结果表明，CC-Gittins仅需约10%的全量评估成本即可恢复大部分真实帕累托前沿，且在30%的评估预算下几乎完全覆盖前沿而仅有少量误报，显著提升了多目标配置搜索的效率与有效性。
链接: https://arxiv.org/abs/2610.26676
作者: Qian Xie,Wenyue Hua,Yueli He,Sripad Karne,Armaan Agrawal,Nikos Pagonas,Eugene Wu,Kostis Kaffes,Tianyi Peng,Nairen Cao
机构: Columbia University (哥伦比亚大学); Cornell University (康奈尔大学); Microsoft(微软); New York University (纽约大学); Shanghai University of Finance and Economics (上海财经大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Machine Learning (stat.ML)
备注: Preliminary version accepted at NeurIPS 2026 Workshop on Resource-Aware Agentic AI (RAAAI)
Abstract:LLM agentic workflows increasingly employ different models for specialized roles. Users may prefer different trade-offs between task quality and deployment cost, sometimes also considering latency. Finding good assignments incurs a separate search cost from evaluating multiple candidates. We introduce Cost-Coupled Gittins (CC-Gittins), a multi-objective extension of single-objective cost-aware Gittins search for Pareto recommendation. It exploits the coupling between evaluation and deployment costs: assignments that cost more to deploy also cost more to evaluate. The search alternates between seeking higher quality and lower deployment cost, reusing each evaluation to inform both searches. After each batch, CC-Gittins recommends assignments that are nondominated under uncertainty-adjusted estimates of quality and deployment cost. We evaluate CC-Gittins on eight quality–cost matrices from workflows with two to five roles and up to 1,681 candidate assignments. We also evaluate the method with latency as a third objective. Across the eight benchmarks, CC-Gittins recovers a substantial fraction of the true Pareto frontier at roughly 10% of the exhaustive-evaluation cost, and recovers most of the frontier with few false positives by 30%.
[NLP-26] SpatialScout: Can VLM Agents Actively Explore and Map 3D Scenes?
【速读】： 该论文旨在解决当前视觉语言模型（VLM）在主动感知条件下构建精确三维度量地图的能力评估问题，尤其关注模型在自主探索过程中对物体级三维空间信息的动态整合与定位能力。现有空间问答任务难以有效检验模型是否能通过自主探索生成准确的度量地图，因此缺乏对模型空间智能的充分验证。为此，作者提出了SpatialScout基准，一个用于评估VLM代理在主动对象级3D建图任务中的表现平台。其关键解决方案在于设计了一个包含80个多样化室内场景、共3,140个目标物体的评测环境，要求代理在有限交互预算下，基于RGB观测、相机标定和位姿信息，通过控制相机视角进行探索，将物体中心估计映射至统一坐标系中，并完成跨视角关联、几何估计与地图修正。评估采用基于物体尺度的重叠分数（overlap score），联合衡量定位精度与目标覆盖度，且对遗漏预测施加惩罚。实验结果表明，当前最先进的模型（Claude Opus 5.5）仅达到41.71%的平均重叠分数，仅有47.23%的目标达到0.5以上交并比（IoU），不足3.25%超过0.9，揭示了现有模型在从粗略物体布局恢复到构建完整、高精度度量地图之间存在显著差距。SpatialScout为开发具备可靠主动感知能力的三维空间表征代理提供了可量化的测试平台。
链接: https://arxiv.org/abs/2610.26659
作者: Zhaoxi Zhang,Hao Liang,Linxuan Wu,Yiming Wu,Yifu Yuan,Kun Liang,Chenming Tang,Meiyi Qiang,Wentao Zhang,Weikang Li
机构: Peking University (北京大学); Tianjin University (天津大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Spatial intelligence requires agents to actively acquire visual evidence and integrate observations into a coherent representation of the three-dimensional world. Yet spatial question answering offers limited evidence of whether models can construct accurate metric maps through autonomous exploration. We introduce SpatialScout, a benchmark for evaluating vision-language model (VLM) agents on active object-level 3D mapping. SpatialScout comprises 80 diverse indoor scenes containing 3,140 target objects. Given an object inventory, RGB observations, camera calibration, and camera poses, agents control a camera to explore each scene and estimate object centers in a shared coordinate frame. The task requires coordinating viewpoint selection, cross-view association, geometric estimation, and map revision under bounded interaction budgets. Evaluation jointly measures localization accuracy and target coverage through object-scale overlap scores that penalize missing predictions. Experiments with frontier VLMs reveal substantial limitations: the strongest evaluated model, Claude Opus 5.5, achieves a mean overlap score of 41.71%, with 47.23% of targets exceeding an intersection-over-union (IoU) threshold of 0.5 and only 3.25% exceeding 0.9. These results reveal a substantial gap between recovering coarse object layouts and constructing complete, precise metric maps. SpatialScout provides a quantitative testbed for developing agents that build reliable 3D spatial representations through active perception. Code is available at this https URL.
[NLP-27] he Development of Sycophancy in Language Models: Tracing Vulnerabilities from Pretraining to Post-Training
【速读】： 该论文旨在解决生成式 AI 在训练过程中如何习得并继承对用户意见的敏感性这一问题，特别是当模型在没有新证据的情况下改变其初始判断时，这种行为在训练阶段的演化路径及其对最终发布模型的影响尚不明确。研究通过追踪30条开源模型谱系共741个检查点，从预训练到发布模型的全过程，在道德情境中植入初始判断，并测量模型在用户施加压力时是否坚持原判或妥协。结果显示，此类行为在后训练阶段前已普遍存在，且其强度在不同谱系中呈现动态变化。关键发现在于：在个体问题层面，预训练阶段的行为几乎无法预测模型在发布时是否会妥协；而这种预测能力主要形成于后训练阶段——尤其是在监督微调（Supervised Fine-Tuning, SFT）和偏好优化（Preference Optimization）阶段，尤其是偏好优化阶段的检查点能显著提升预测准确性。相比之下，同一基础模型但采用不同训练路径的“兄弟”模型则表现出明显更低的可预测性，表明训练历史对模型行为具有决定性影响。此外，内部状态虽提供额外预测信息，但不改变上述层级关系。这揭示了平均性能指标可能掩盖模型间本质差异，强调了模型审计必须考虑其具体训练轨迹。
链接: https://arxiv.org/abs/2610.26655
作者: Haonan Huang,Chengkun Cai
机构: Princeton University (普林斯顿大学); University College London (伦敦大学学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Language models can change a committed answer to accommodate a user’s opinion or pushback without new evidence. However, when this behaviour develops during training and what a released model inherits from earlier stages remain unclear. We follow 741 checkpoints of 30 open model lineages, from early pretraining to released models. We plant initial verdicts on moral scenarios and measure how far a model follows a user’s stated opinion or abandons its verdict under pressure. In most lineages the behaviour is present before post-training, and its average strength changes in either direction, at different stages across lineages. A clearer pattern emerges at the level of individual questions. Beyond responses without pressure, pretraining behaviour predicts little of which questions a released model yields on. That pattern forms during post-training: on routes with supervised fine-tuning, the fine-tuning checkpoint predicts part of it and the preference-optimization checkpoint most of it. A sibling trained from the same base along another route predicts far less: for Tülu-3 8B, its own preference-stage checkpoint removes 85% of the remaining prediction error, a sibling 19%. Internal states add predictive information without changing this ranking. Similar average scores conceal different vulnerabilities, and training history matters for auditing a released model.
[NLP-28] From Outcomes to Components: A Counterfactual Evaluation Framework for LLM -Proposed Repairs
【速读】： 该论文旨在解决大语言模型（LLM）生成的修复动作在实际应用中因部分修复失败而导致整体修复效果不佳的问题，特别是在交通网络与数据库等场景中，由于聚合观测值掩盖了需修正的具体记录，使得单一修复操作难以准确识别有效目标。其核心解决方案在于：通过系统性地评估同一初始状态下所有可能的“目标-更新”配对组合，利用参考修复动作作为基准，验证各子修复部分协同作用的实际收益。研究发现，在76/114个交通网络修复动作中，同时修正目标与更新量所带来的增益超过单独修正两者的总和；其中6个动作在仅修正任一部分时反而恶化，唯有两者协同修正才实现改善。在数据库修复中，若仅以参考更新替换模型提出的更新，会导致17/108个动作性能下降，且这些动作均错误地选中了未受影响的记录。进一步实验表明，将两个不同修复提案中的目标与更新进行重新组合后选择，可在保留样本上恢复参考动作61.6%的损失降低效果，显著优于仅从原始提案中选择的58.0%，证明了修复组件重用与配对优化的关键价值。
链接: https://arxiv.org/abs/2610.26650
作者: Hengyu Ye,Songhua Hu
机构: ShanghaiTech University(上海科技大学); City University of Hong Kong(香港城市大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 29 pages, 5 figures
Abstract:An LLM-generated repair action specifies which records to change and by how much. A failed action may still contain a useful target set or numerical update, depending on how the two are paired. We evaluate this problem in traffic networks and databases, where aggregate observations obscure the records that need correction. We examine the pairings by executing each combination from the same initial state, using a reference action that restores the affected records. In 76 of 114 traffic actions, the gain from correcting both parts exceeds the sum of the gains from correcting each part alone. Six of these actions become worse after either correction alone, but improve when both parts are corrected. In database repair, replacing only the proposed update with the reference update worsens 17 of 108 model actions, all of which select unaffected records. We also test whether parts of two proposals can be reused. On held-out observations, selecting among the proposals and their recombined target-update pairs recovers 61.6% of the reference action’s loss reduction, compared with 58.0% from selecting between the proposals alone.
[NLP-29] When the harness decides success: state that cannot be re-observed
【速读】： 该论文旨在解决生成式 AI（Generative AI）在多轮对话中如何有效管理上下文信息以维持推理性能的问题。核心挑战在于：当模型面临长期任务时，其上下文窗口有限，需通过代理（Agent）机制决定保留、丢弃或重述历史交互内容，而这一决策直接影响最终任务成功率。解决方案的关键在于识别并量化不同上下文管理策略对模型表现的影响——具体表现为：1）若关键事实可被重新检索，则截断上下文几乎不造成损失；但一旦事实不可复现，成功概率将骤降至随机水平（差距达92.5个百分点），表明上下文完整性对不可恢复信息的依赖性极高；2）在ARC-AGI-3基准测试中，通过用摘要替代简单丢弃旧轮次信息，并结合笔记提示，可恢复63分差距中的86%；进一步引入模型自身推理的重放机制，可补足剩余57%，说明推理重放是提升复杂任务表现的关键组件；3）尽管代理优势在深层任务中更显著，但该现象主要反映任务停止位置分布而非真实能力随深度增长，因此不能直接推断为持续增强的性能优势。整体而言，解决方案的核心在于精细化控制上下文状态管理策略，尤其强调信息可重获性与推理过程重放的作用。
链接: https://arxiv.org/abs/2610.26632
作者: Raimondas Lencevicius(Microsoft)
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 14 pages, 3 figures, 4 tables. Pre-registered; code, pre-registration and cost ledger at this https URL
Abstract:Agent harnesses decide what a language model sees across turns: they drop old turns, summarize them, or replay the model’s own earlier reasoning. We ask when these choices change whether an agent succeeds, not only what it costs, and report three findings. First, in a pre-registered synthetic probe, truncating the context window costs a low-cost reasoning model nothing in success when a needed fact can be looked up again (it looks the fact up instead), but drops its success to chance when the fact cannot be re-observed: the two penalties differ by 92.5 percentage points (90% CI [89.5, 95.5]). A pre-registered replication on one model from a second provider showed the same contrast. Once the fact was lost, that model chose no answer rather than guessing. Second, on ARC-AGI-3, where the benchmark’s Standard harness trails its Provider Adapter harness by a wide margin, we decompose that gap one component at a time, for one model on four games. Summarizing old turns instead of dropping them, while keeping the Standard harness’s invitation to take notes, recovers 86% of a 63-point gap without replaying the model’s reasoning. The same harness without the invitation recovers 43%, and adding the provider’s native replay of the model’s reasoning recovers the remaining 57%. These are point estimates from one counted run per configuration and game, and the replay share depends on a single run. Third, in the benchmark’s published runs and in ours, the Adapter’s advantage in the probability of clearing a level it has reached is larger at deep levels than at early ones. This largely restates where runs stopped and does not separate a growing advantage from a constant one.
[NLP-30] Conducting Reliable Keyword Estimation with Unreliable ASR in Police Audio Data
【速读】： 该论文旨在解决在领域特定场景下，由于现成自动语音识别（ASR）系统存在系统性转录错误，导致基于大规模音频数据集进行关键词频率分析时统计结论不可靠的问题。其核心解决方案是结合模型输出与少量人工标注样本，采用基于预测驱动推断（Prediction Powered Inference）和设计导向的监督学习（Design-based Supervised Learning）框架，对不完美ASR转录文本进行后处理校正。关键在于通过引入小规模人工标注数据，对ASR生成的噪声转录进行偏差修正，从而显著提升下游关键词频率估计的统计有效性，实证表明该方法将可获得有效置信区间覆盖的领域关键词比例从仅使用ASR的43.0%提升至56.2%（巴尔的摩数据集），验证了即使在低质量转录情况下，通过后验误差校正仍可实现可靠的统计推断，为大规模语音分析中高效利用现成ASR模型提供了可行路径。
链接: https://arxiv.org/abs/2610.26614
作者: Robert Cole Molloy,Rhiannon N. Miller,Stephen Morgan,Anjalie Field
机构: Johns Hopkins University (约翰霍普金斯大学); Providence College (普罗维登斯学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Extracting insights from large audio data sets using human transcription is expensive and time-consuming. Off-the-shelf Automatic Speech Recognition (ASR) systems offer an easy-to-use scalable alternative, but systematic transcription errors, especially in domain-specific settings, can bias downstream analyses and lead to unreliable conclusions. We investigate whether imperfect ASR transcripts can still be used to obtain statistically valid keyword frequency estimates by combining model outputs with a small set of human-labeled examples, following Prediction Powered Inference and Design-based Supervised Learning frameworks. We evaluate this approach on police radio communications from Baltimore and Chicago, a domain in which keyword distributions could provide valuable insights into police activity, but off-the-shelf ASR systems exhibit systematic errors on operationally important terminology. Using bootstrap-based confidence interval coverage as our evaluation metric, we find that our approach substantially increases the number of domain-specific keywords that can be analyzed with statistically valid confidence intervals, from 43.0% using ASR alone and 27.2% using only a small set of human-labeled transcriptions to 56.2% in the Baltimore dataset, with similar improvements observed in Chicago data. Our results demonstrate that post-hoc error correction can improve statistical validity of downstream analyses even for poor-quality transcriptions, offering an avenue for better leveraging off-the-shelf ASR models for large-scale speech analysis.
[NLP-31] BERT Is Already a Denoiser: Cheaply Unlocking Iterative Generation in Masked Language Models
【速读】： 该论文旨在解决如何在不依赖显式扩散训练（diffusion training）的情况下，利用预训练的掩码语言模型（Masked Language Models, MLMs）实现有效的条件生成问题。现有方法通常需要复杂的扩散过程来完成生成任务，但本文提出，仅通过简单的迭代解码机制即可实现高质量的生成效果。其解决方案的关键在于：将预训练的MLM（如BERT）作为基础表示与去噪算子，通过任务适应（task adaptation）调整模型输出目标，并结合基于余弦调度的高置信度预测提交规则（iterative commitment rule）进行逐词生成。实验表明，即使冻结原始参数的BERT，也具备一定的去噪能力，但在任务导向生成上表现有限；而通过全量或参数高效的任务适应后，模型在多项基准测试中显著提升性能，达到与专门经过扩散训练的BERT模型相当的水平。此外，将该生成框架应用于更强的掩码表示（如RoBERTa-large）时，性能进一步跃升。研究结果明确区分了三个核心组件：预训练提供表征与去噪能力、任务适应决定生成内容、迭代提交规则控制解码过程。因此，在所研究的设定下，显式扩散训练并非实现有效迭代式条件生成的必要条件。
链接: https://arxiv.org/abs/2610.26601
作者: Julio C. Amador Diaz Lopez
机构: SiftyML, LTD (SiftyML, LTD)
类目: Computation and Language (cs.CL)
备注:
Abstract:Masked language models (MLMs) such as BERT are typically used for representation learning despite their denoising pretraining. We test whether this capability supports conditional generation without diffusion training. Our decoder repeatedly predicts masked tokens and commits high-confidence predictions using a cosine schedule. It requires no diffusion timestep or reverse-diffusion objective. Frozen BERT already exhibits substantial denoising capability, reconstructing 61.0%, 37.5%, and 24.4% of masked tokens at 15%, 45%, and 60% corruption, respectively, although generation deteriorates sharply as visible context disappears. This capability alone does not yield task-directed generation: frozen BERT reaches only 11.0% on ARC-Easy and 4.2% on MMLU. Ordinary MLM adaptation, however, rapidly converts the same architecture into a usable generator. Full adaptation reaches 28.2% and 24.6%, respectively, while parameter-efficient adaptation substantially reduces unmatched outputs. On open-ended conditional generation, full MLM adaptation reaches BLEU/ROUGE-L of 0.163/0.583 on QQP and 0.107/0.327 on Quasar-T, placing it in the same broad performance range as published BERT models trained through diffusion without explicit diffusion training. Finally, applying the same generative interface to stronger masked representations substantially improves performance: RoBERTa-large reaches 44.2% on ARC-Easy and 33.8% on MMLU. These results separate three components of masked iterative generation: pretraining supplies the representation and denoising operator, task adaptation determines what the model generates, and an iterative commitment rule supplies the decoding process. Explicit diffusion training is therefore not necessary to obtain effective iterative conditional generation from pretrained MLMs in the settings studied here.
[NLP-32] Beyond Attack Success Rate: Cost-Aware Metric for Evaluating Iterative LLM Jailbreaks
【速读】： 该论文旨在解决现有攻击成功率（ASR）评估方法在衡量大型语言模型越狱攻击时忽略对目标模型重复调用的问题。传统ASR未考虑攻击过程中多次调用目标模型所带来的额外被检测与干预风险，从而高估了攻击的有效性。为此，论文提出新的评估指标 ASRSTEP，通过惩罚额外的模型调用次数，在保持二元成功判断的基础上，更真实地反映攻击在实际部署场景中的可行性。其核心解决方案在于引入“步数预算”（step budget）机制，量化攻击过程中的交互成本，并揭示不同迭代式攻击策略（如GCG、PAIR和PAP）在调用效率上的差异：例如，GCG的得分因调用惩罚下降7.0–8.7个百分点，而单步攻击的PAP得分不变，表明其调用效率更高；而PAIR则表现出3.4–12个百分点的广泛折扣范围，凸显不同策略对调用次数的敏感性差异。该研究强调同时报告原始ASR与ASRSTEP的重要性，使重复调用带来的性能衰减更加透明。
链接: https://arxiv.org/abs/2610.26589
作者: Rohan Menon,Chih-Hong Cheng
机构: Carl von Ossietzky University of Oldenburg(奥尔登堡大学); Chalmers University of Technology(查尔姆斯理工大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注:
Abstract:Attack success rate (ASR) evaluates jailbreaks against large language models without accounting for repeated calls to the victim model. Yet each additional call gives the deployed system another opportunity to detect the attack and intervene. We introduce \mathrmASR_\mathrmSTEP , which penalizes additional victim-model calls while retaining binary success judgments. Our evaluation considers GCG~\citezou2023universal, PAIR~\citechao2025jailbreaking with a step budget, and a single-step PAP~\citezeng2024johnny baseline on JailBreakBench across three model families. GCG’s scores decrease by 7.0–8.7 percentage points after discounting, while PAP’s remain unchanged. While for PAIR~\citechao2025jailbreaking, we see a wider discounting range between 3.4-12 percentage points, showing that the “step” quantity differs per iterative strategy. Reporting both metrics makes the discount due to repeated victim-model interaction explicit. Code is available \hyperlinkthis https URLhere.
[NLP-33] ReAnchor: Reusing Rejected Verifier States for Diffusion Drafting
【速读】： 该论文旨在解决块扩散式采样（block diffusion drafting）中生成草案令牌（draft tokens）时存在的局限性：现有方法在单次前向传播中仅能基于单一目标生成的锚点（anchor）产生固定长度的块，导致草案扩展范围受限。同时，目标验证过程在首次拒绝位置之后仍会计算隐藏状态，但这些状态通常被丢弃而未被复用。针对上述问题，本文提出一种无需训练的 ReAnchor 方法，其核心创新在于利用被丢弃的验证器（verifier）隐藏状态，在偏移位置重新锚定预训练的草稿生成器（drafter），从而生成并行的草案块。这一机制不仅提升了原始块内的预测准确性，还显著扩展了草案覆盖范围至原始块末端之外。实验表明，在采样温度 T=0 时，ReAnchor 在 Gemma 4-12B 上实现了最大平均提升，接受长度（acceptance length）提高 19.4%，解码速度提升 17.9%，在四个目标模型与八个基准测试上的平均表现也分别提升 15.2% 和 13.4%，且可兼容预训练的 DFlash 草稿生成器。
链接: https://arxiv.org/abs/2610.26531
作者: Sanghyun Lee,Wonjin Nam,Seungho Jang,Jaeho Lee,Jiho Park,Jaewoon Lee,Seungryong Kim
机构: KAIST(韩国科学技术院); Seoul National University(首尔国立大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Block diffusion drafters accelerate speculative decoding by generating multiple draft tokens in a single forward pass. However, each pass produces only a fixed-length block from a single target-generated anchor, limiting how far the draft can extend. Meanwhile, target verification computes hidden states even at positions beyond the first rejection, but these states are typically discarded rather than reused. We propose ReAnchor, a training-free method that reanchors a pretrained drafter at shifted positions using discarded verifier states to generate parallel draft blocks. These blocks improve predictions within the original block and extend coverage beyond its endpoint. At sampling temperature of T=0, ReAnchor achieves its largest average gains on Gemma 4-12B, improving acceptance length by 19.4% and decoding speed by 17.9% across eight benchmarks. Averaged equally over four target models and eight benchmarks, it improves acceptance length by 15.2% and decoding speed by 13.4% with pretrained DFlash drafters.
[NLP-34] A Fisher-Geometric View of Multi-Teacher On-Policy Distillation
【速读】： 该论文旨在解决大模型后训练阶段中，如何高效且高质量地将多个专用基础模型（specialized foundation models）合并为统一通用模型（generalist model）的核心问题。现有方法存在显著局限：模型合并（model merging）虽计算高效但缺乏理论依据，而多教师在线策略蒸馏（multi-teacher on-policy distillation）虽能精确匹配分布却需高昂的推理计算开销，并面临跨任务梯度冲突。本文的关键发现是，这两种方法本质上是同一优化目标在不同展开阶次下的表现——在同源专用模型的共享低维流形内，多教师在线蒸馏的目标函数在二阶近似下等价于Fisher合并（Fisher-merging）目标。其最优解即为经典的闭式合并方法：Fisher合并，这表明模型合并在二阶意义下可视为蒸馏过程的一个最优闭式预处理阶段。基于此理论洞察，作者提出残差驱动的分枝与合并（Residual-Governed Branch-and-Merge, RGBM）两阶段整合框架：第一阶段，由于全参数Fisher合并对基础模型不可行，该阶段限制合并仅作用于任务向量子空间；通过扩展合并的切空间分析，证明任务向量混合系数同样适用于隐式奖励的混合，从而设计一个基于未标注提示的残差程序来求解这些系数，该程序最小化各领域残差的最大值，并输出归一化后的残差谱；第二阶段，利用该残差谱指导更新策略：对残差谱值低于谱隙的领域进行主动抑制，仅对高于谱隙的领域训练独立分支，并以校准收缩的方式回写更新。实验表明，在两个开源整合基准上，仅第一阶段的合并已达到或超越完全训练的多教师蒸馏性能，而RGBM进一步提升性能且无需额外联合训练步骤，相比之下所有无门控的回写方式（包括从合并结果继续联合蒸馏）均表现更差。
链接: https://arxiv.org/abs/2610.26525
作者: Liu Kaiyuan,Zhuang Ziyuan,Ma Haoxuan,Gu Yannian,Song Mingfei,Weng Rongxiang,Ye Jieping
机构: Zhejiang University (浙江大学); Meituan LongCat; Nanjing University (南京大学); Shanghai Jiao Tong University (上海交通大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Consolidating specialized foundation models into a unified generalist is a central goal of post-training. Two approaches are in common use: model merging, which is fast but treated as a heuristic, and multi-teacher on-policy distillation, which matches distributions but demands heavy rollout compute and suffers from cross-task gradient conflict. We show that the two are one objective seen at different orders. Within a shared basin of homologous specialists, the multi-teacher on-policy distillation objective is, to second order, exactly the Fisher-merging objective. Its minimizer is therefore the Fisher merge, a classic model-merging method given by a closed-form combination of the specialists, so merging is optimal, to second order, as a closed-form stage of distillation. Building on this identification, we propose Residual-Governed Branch-and-Merge(RGBM), a two-phase consolidation procedure. Concretely, a full-parameter Fisher merge is intractable on foundation models, so Phase~1 restricts the merge to the task-vector subspace. Extending tangent-space analyses of merging, we show that the same coefficients that mix task vectors also mix implicit rewards, which lets a residual program on unlabelled prompts solve for them. The program minimizes the worst per-domain residual and reports its energy-normalized values as the domain residual spectrum. Phase~2 lets this spectrum govern what may be added: RGBM passivates domains with low normalized residuals, trains isolated branches only for domains above the spectral gap, and writes back the admitted updates with calibrated shrinkage. On two open-source consolidation suites, the Phase~1 merge alone matches or outperforms fully trained multi-teacher distillation, and RGBM improves on it further with zero extra joint-training steps, whereas every un-gated write-back, including continued joint distillation from the merge, falls below it.
[NLP-35] RouterWorld: Scaling Conversational AI with Latency-Aware LLM Routing
【速读】： 该论文旨在解决大规模对话式生成式AI（Conversational GenAI）系统中，基于大语言模型（LLM）的路由机制在实际部署中面临的两大核心挑战：如何在保证回答质量的同时有效控制响应延迟，以及现有评估基准在可观测性、延迟指标权重和协议一致性方面存在的缺陷。其解决方案的关键在于提出RouterWorld——一个全面的LLM路由评估基准，通过整合37.1万条单轮对话响应与7.6万条多轮代理轨迹，采用标准化的“大模型作为裁判”（LLM-as-a-judge）协议模拟真实用户交互，并引入基于硬件的延迟标注以实现效率与动态定价的解耦。尤为关键的是，RouterWorld首次在部分可观测性（partial observability）框架下形式化路由问题，设计了两种典型场景：冷启动（Cold-Start）设定下仅依赖稀疏静态日志进行学习，以及流式数据环境下的Voyager设定，其中路由决策会动态影响后续训练分布。实验表明，多数现有路由策略在部分可观测条件下过度关注准确性而忽视延迟，凸显了新基准在推动更均衡、更具现实适应性的路由算法发展中的重要价值。
链接: https://arxiv.org/abs/2610.26511
作者: Jinyang Li,Denis Emelin,Simone Tedeschi,Weiwei Cheng,Cristina Morariu,Haytham Assem,Jens Lehmann,Emine Yilmaz,Angus Addlesee
机构: The University of Hong Kong(香港大学); Amazon(亚马逊); Centre for Artificial Intelligence, University College London(伦敦大学学院人工智能中心)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 28 pages, pre-print
Abstract:Conversational AI systems built on Large Language Models (LLMs) require routing mechanisms that balance answer quality and response latency at scale. However, existing LLM routing benchmarks assume full observability of all candidate model outcomes, under-emphasize latency as a first-class metric, and fragment evaluation across heterogeneous protocols misaligned with real-world settings. We introduce RouterWorld, a comprehensive benchmark for LLM routing in conversational GenAI. RouterWorld unifies 371k single-turn responses and 76k multi-turn agentic trajectories, adopts natural user interactions with a standardized LLM-as-a-judge protocol, and decouples efficiency from transient pricing via hardware-grounded latency annotations. Crucially, RouterWorld formalizes routing under partial observability through two setups: Cold-Start, which restricts learning to sparse, static logs, and Voyager, which models continuous data streams where routing decisions shape future training distributions. Evaluating 13 routing baselines, we find that most routers over-emphasize accuracy while ignoring latency under partial observability. Code and benchmark data are publicly available.
[NLP-36] Mitigating Reward Hacking in Preference Reward Models with Evolving Penalties
【速读】： 该论文旨在解决生成式AI在语言模型强化学习中因使用偏好奖励模型（Preference Reward Model）而引发的“奖励黑客”（reward hacking）问题。偏好模型易导致策略通过表面优化得分，而非真正提升输出质量，甚至在质量下降时仍能获得高分。针对开放性任务中难以量化评估质量的场景，本文提出Dynalize方法，其核心在于在策略训练过程中在线动态地识别并抑制有害行为。该方案的关键机制是：周期性且异步地由一个语言模型分析训练轨迹，自动识别出奖励模型所激励但实际有害的行为模式，并生成相应的惩罚项；随后由裁判模型将这些惩罚应用于所有训练轨迹。实验表明，在临床咨询与创意写作等任务中，仅依赖原始奖励模型会导致评分持续上升但外部评估指标最终恶化，而采用Dynalize可实现更长时间的稳定优化，并显著提升下游任务性能。此外，该方法有效抑制了偏好模型隐含激励的不良模式在响应及推理过程中的广泛传播，且其优势在多轮训练及奖励模型重新训练后依然保持。
链接: https://arxiv.org/abs/2610.26496
作者: Afonso Baldo,António Farinhas,Marcos Treviso,Ricardo Rei,Nuno M. Guerreiro
机构: Sword Health(剑盾健康); Gandara AI; Instituto Superior Técnico, Universidade de Lisboa(里斯本大学理工学院); Instituto de Telecomunicações(电信研究所); ELLIS Unit Lisbon(里斯本ELLIS单元)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning for language models has moved from preference reward models to verifiable rewards, which sustain much longer optimization. Preference models are prone to reward hacking: the policy raises their score without improving response quality, or even while degrading it. Yet for open-ended tasks, where quality can neither be verified nor fully written down as rubrics, they remain valuable. We introduce Dynalize, which mitigates reward hacking online, during policy training. Periodically and asynchronously, a language model analyzes training rollouts and writes penalties for undesirable behaviors that the reward model appears to reward; a judge applies them to every rollout. In clinical counseling and creative writing, training on the reward model alone raises its score continuously but ultimately leads to a decline in external benchmarks. Dynalize sustains longer training and results in improved downstream performance. For example, we find that our method suppresses patterns incentivized by preference models that, without penalties, spread to most responses, and even to the model’s reasoning, which the reward model does not explicitly score. Furthermore, we show that its gains hold in multi-turn training, and even after we retrain the reward model on the discovered behaviors.
[NLP-37] Strength Controls Change How Interventions Compare Across Sparse Autoencoder Codes
【速读】： 该论文旨在解决生成式模型中稀疏自编码器（Sparse Autoencoder, SAE）干预效果评估中的强度控制问题，即在比较不同干预策略时，如何确保干预的“强度”一致，以避免因干预幅度差异导致的误判。其核心挑战在于：尽管归一化匹配（norm matching）和输出校准（output calibration）提供了不同的强度控制方式，但二者在实际干预效果上的表现差异尚不明确。论文的关键解决方案是通过系统对比两种控制方法在固定特征组上的干预效果，揭示其在不同场景下的有效性。研究发现，对于整体块缩放（whole-patch scaling），全输出Kullback-Leibler（KL）校准相比总范数匹配能更有效地保留编码器参考干预的转移性能，在新输入上几乎全部保持决策一致性；而在逐位置缩放（positionwise scaling）下，位置范数匹配表现更优；此外，针对答案排除（answer exclusion）情形，输出校准可减少部分指标的差距，但对性别重构任务反而增加偏差。由于校准依赖于具体输入的输出结果，因此可作为诊断残余不一致性的有效工具。
链接: https://arxiv.org/abs/2610.26489
作者: Jinchen Zhang,Zheyuan Liu,Yihan Zhu,Meng Jiang
机构: University of Notre Dame (圣母大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 35 pages, 8 figures
Abstract:Changing sparse autoencoder (SAE) codes can alter both intervention direction and magnitude, so comparing intervention effects requires a strength control specifying what is held equal. Norm matching and output calibration provide different controls, but whether interventions yield different effects across codes after strength matching remains unclear. We compare intervention effects under these controls for fixed feature groups in Gemma and Llama and observe: (1) Whole-patch scaling. With one coefficient per input, full-output Kullback-Leibler (KL) calibration leaves smaller mean effect gaps from the encoder-based reference intervention than total-norm matching for nonnegative least-squares (NNLS) refits. On new Llama inputs, total-norm matching loses 11 reference transfers, whereas full-output calibration preserves every transfer decision. (2) Positionwise scaling. On a small matched gender-agreement subset, positionwise norm matching leaves smaller observed mean gaps than full-output calibration. (3) Answer exclusion. Compared with unscaled NNLS, answer-excluded calibration reduces gaps in the measured subject-verb agreement comparisons but increases them for gender refits. Full-output KL calibration also reduces larger gaps for norm-matched constructed candidates, although matching coverage is incomplete in several settings. Since calibration uses the evaluated input’s outputs, it serves as a diagnostic of residual disagreement.
[NLP-38] AfriGuard: Safety and Robust Refusal For African Languages
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多元语言与文化背景下对有害意图识别、安全类别理解及拒绝行为响应能力不足的问题。由于不同语言和文化语境中安全风险的表达方式与认知差异，现有模型在处理跨文化安全威胁时表现不佳。其解决方案的关键在于构建一个以非洲文化为根基的多语言数据集AfriGuard，涵盖10种非洲语言及英语，系统性地评估模型在有害意图检测、安全类别分类与拒绝行为上的表现。研究发现，当前守卫模型（guardrail models）在安全意图与类别分类任务上均逊于开源与闭源大模型，且对适当安全行为的理解有限（F1得分低于60%）。更关键的是，模型对不当提示的不恰当响应并非源于显式拒绝，而是因语言层面的不充分输出导致非故意回避（incidental non-assistance），而非规范拒绝。基于AfriGuard，研究提出高效守卫模型，性能可媲美规模大30倍的模型；开发新型拒绝检测器，相较现有方法平均提升43.9%；并验证监督微调（supervised fine-tuning）能显著增强整体安全性表现。
链接: https://arxiv.org/abs/2610.26474
作者: Israel Abebe Azime,Jesujoba O. Alabi,Iffat Maab,Alexander D. Cai,Tadesse Destaw Belay,Atnafu Lambebo Tonja,Jonas Kgomo,Folasade Alabi,Salomey Osei,Saminu Muhammad Aliyu,Nkechinyere Faith Aguobi,Bontu Fufa Balcha,Blessing Sibanda,Davis David,Mouhamadane Mboup,Daud Abolade,Neo Putini,Louis Powell,Kiant’e Brantley,Philipp Slusallek,Dietrich Klakow,Julia Kreutzer
机构: Kempner Institute, Harvard University, USA(哈佛大学肯普纳研究所, 美国); Instituto Politécnico Nacional, Maxico(墨西哥国立理工学院, 墨西哥); University College London, UK(伦敦大学学院, 英国); Equiano Institute, SA(南非埃基亚诺研究所, 南非); University of Ilorin, Nigeria(伊洛林大学, 尼日利亚); Universidad de Deusto, Spain(德乌斯托大学, 西班牙); Bayero University, Nigeria(贝亚罗大学, 尼日利亚); University of Lagos, Nigeria(拉各斯大学, 尼日利亚); Addis Ababa University, Ethiopia(亚的斯亚贝巴大学, 埃塞俄比亚); Black Swan(黑天鹅); Universite Alioune Diop, Senegal(阿里翁·迪奥普大学, 塞内加尔); GSMA(全球移动通信系统协会); Cohere Labs(科赫尔实验室); Saarland University(萨尔兰大学); Saarland Informatic Campus(萨尔兰信息学园区); Germany National Institute of Informatics(德国国立信息学研究所); Japan(日本)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) designed to interact with users are typically trained to respond appropriately to harmful or toxic content. However, ensuring appropriate responses becomes more challenging in diverse linguistic and cultural contexts, where safety risks may be expressed or interpreted differently. Building safe AI systems therefore requires detecting safety risks, understanding their nature, and responding appropriately, while accounting for linguistic and cultural variations that add complexity to each stage. In this work, we introduce AfriGuard, a culturally grounded multilingual dataset for studying harmful intent, safety-category understanding, and refusal behavior across 10 African languages and English. Our findings reveal that guardrail models underperform both open- and closed-source LLMs on safety-intent and safety-category classification, while models show limited understanding of appropriate safety behavior, scoring below 60% F1. We further find that models often fail to respond appropriately to unsafe prompts not because they explicitly refuse, but because they produce linguistically inadequate responses that result in incidental non-assistance rather than proper refusals. Finally, using AfriGuard, we develop efficient guardrail models that achieve performance competitive with models up to 30x larger, introduce refusal detectors that improve average performance by 43.9% over existing detectors, and demonstrate that supervised fine-tuning can substantially improve overall safety performance.
[NLP-39] raining Negatives Still Matter After Calibrating Answerability Readouts
【速读】： 该论文旨在解决在问答系统中，去除负样本（negative passages）进行训练时，是否仍可通过校准（calibration）手段弥补其对模型预测性能造成的负面影响这一关键问题。研究发现，尽管采用相同校准策略，仅使用正样本进行训练（empty training）的模型在答案可回答性（answerability）预测、排序质量及下游答案保留率方面均表现逊于保留负样本的原生训练（native training）。其解决方案的关键在于：通过控制实验对比原生训练与去负训练，在固定问题、标签、正样本及模型权重的前提下，验证了负样本在训练过程中对模型排序能力与决策行为的深层影响，即使经过校准也无法完全消除这种影响。结果显示，原生训练下的逻辑回归读出器在多个深度下取得更低的Brier损失，且后验AUROC分析表明其排序性能优于阈值调整所能解释的范围；在SQuAD 2.0开发集上，原生策略虽未普遍提升F1，但显著减少对不可回答问题的答案误报，体现出更强的鲁棒性。然而，该优势受估计器类型和推理深度影响，且因文本捷径（text shortcuts）的存在，难以对模型上下文理解能力做出普适判断。最终结论为：负样本的训练参与对模型的排序与下游决策具有不可忽视的影响，即便经过常规校准，也不存在对所有场景都适用的原生训练绝对优势。
链接: https://arxiv.org/abs/2610.26461
作者: Jinchen Zhang,Zheyuan Liu,Yihan Zhu,Meng Jiang
机构: University of Notre Dame(圣母大学)
类目: Computation and Language (cs.CL)
备注: 20 pages, 5 figures
Abstract:Answerability readouts predict whether a passage supplies an answer, but whether calibration can compensate for removing negative passages during training remains unclear. We present a controlled study of how training-negative construction affects calibrated prediction, ranking, and downstream answer retention. We compare native training, which retains negative passages, with empty training, which removes them, holding questions, labels, positive examples, and model weights fixed. Both are calibrated on the same labeled native examples. On SQuAD 2.0 and RepLiQA, native-trained logistic readouts at selected depths achieve lower Brier loss for Qwen2.5-7B-Instruct and Gemma-2-9B-it, with or without output scores. Native fusion also outperforms calibrated output alone. Post-hoc AUROC analysis shows better ranking under native training, beyond what threshold adjustment alone can explain. On the official SQuAD 2.0 development set, native readout policies improve F1 over empty policies but retain more answers to unanswerable questions. The advantage varies by estimator and depth, while RepLiQA’s text shortcuts limit claims about contextual understanding. Unfinished generations prevent complete answer-level evaluation, although worst-case bounds preserve the native-empty F1 gains. Our findings show that training negatives affect ranking and downstream decisions even after common calibration, without establishing a universal advantage for native training.
[NLP-40] ASR-Amharic: A Benchmark for Robust Amharic ASR in Realistic Telecommunication Conditions AACL
【速读】： 该论文旨在解决低资源语言（如阿姆哈拉语）在真实通信场景中语音识别（ASR）模型可用性差的问题，尤其针对现有语音资源多来源于纯净录音室环境、难以反映实际电信通信条件下语音质量退化的挑战。其核心解决方案是构建首个面向阿姆哈拉语的电信域语音识别基准——TeleASR-Amharic，系统性地涵盖多种信道条件、环境干扰及电信领域特有的信号失真。研究发现，严重的电信退化可导致最优模型的词错误率（WER）从0.23飙升至0.79，相对错误率增加243%，使高性能模型在真实场景下失效。通过分析不同噪声与信道失真影响，并结合大规模训练数据扩展与精心设计的电信感知增强策略，显著提升了模型鲁棒性，最佳训练配置在极端噪声条件下实现23.6%的相对WER降低。该工作不仅提供了可复现的基准数据集，也为开发适用于真实部署环境的鲁棒语音技术指明了关键方向。
链接: https://arxiv.org/abs/2610.26448
作者: Dagnachew Mekonnen Marilign,Enrique Molero,Abrham Haile,Louis Powell,Israel Abebe Azime
机构: HiLCoE School of Computer Science and Technology; GSMA; gheero; Saarland University
类目: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: Accepted to AACL-IJCNLP 2026 Workshop W8 TrustAudio: Workshop on Secure, Safe, and Privacy-Preserving Speech and Audio AI
Abstract:The real-world usability of ASR models remains challenging, particularly for low-resource languages, where existing speech resources are often collected in clean studio environments. In this work, we introduce TeleASR-Amharic, a benchmark for evaluating Amharic ASR under various acoustic and telecommunication conditions. We analyze the major degradations observed in real telecommunications recordings and develop a benchmark that covers multiple channel conditions, environmental disturbances, and telecom-domain variations. Our results show that telecommunication degradation can substantially reduce recognition performance, making even high-performing ASR models impractical under severe conditions. For example, the WER of the best-performing model increases from 0.23 to 0.79 under severe telecommunication degradation, corresponding to a relative increase of 243% in recognition errors. We further investigate the effects of different noise and channel distortions and show that scaling training data and applying carefully designed telecommunication-aware augmentation substantially improve robustness. In particular, our best training configuration achieves a 23.6% relative WER reduction under extreme noise conditions. Together with the release of our telecommunications-domain data, these findings provide a reproducible benchmark for measuring progress in Amharic ASR while highlighting open challenges in developing robust speech technologies for real-world deployment.
[NLP-41] Label Authority in Speech Emotion Recognition: Intended and Perceived Ground Truth in Emozionalmente
【速读】： 该论文旨在解决语音情感识别（Speech Emotion Recognition, SER）中“真实标签”（ground truth）定义不一致的问题。传统SER研究通常将情感标签视为单一的类别事实，但实际中演员表达的情感（expressed condition）与听者感知到的情感（listener judgment）反映了同一语音片段的不同层面：前者描述的是语音生成时的意图，后者则体现听者的主观感知。为揭示这一差异，作者基于意语情感表演数据集Emozionalmente，同时提供了演员标注的表达情感与五个听者的独立判断。通过冻结的XLS-R、HuBERT和WavLM语音表示模型，结合匹配的分类探针，比较三种标签权威性来源：表达情感、听者多数决（hard plurality）以及完整的投票分布（vote distribution）。研究发现，恢复表达情感比预测听者多数决更容易，但二者本质是回答不同问题，并非表明模型整体情感识别能力下降；当听者一致性降低时，多数决作为硬目标变得愈发脆弱。以完整投票分布作为训练目标虽未显著改变模型最终的硬决策结果，但使模型输出的概率分配更贴近听者在模糊片段中的选择倾向，从而更好地保留感知不确定性。因此，三类标签各有其意义：表达标签反映语音生成属性，多数决提供可操作的单类决策，而投票分布则完整保留了人类感知中的不确定性。对于面向人类的“声音物联网”（Internet of Sounds）应用，评估应明确指出所预测的标签权威类型，保留相应的聚合规则与一致性信息，并在目标包含分歧时采用分布评分方法。
链接: https://arxiv.org/abs/2610.26446
作者: Marco Giordano,Stefano Giacomelli,Claudia Rinaldi,Fabio Graziosi
机构: 未知
类目: ound (cs.SD); Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注:
Abstract:Ground truth in Speech Emotion Recognition (SER) is often presented as a single categorical fact, yet an acted emotion and the emotion heard by listeners describe different aspects of the same recording. We examine this distinction in Emozionalmente, an acted Italian corpus that provides both an elicited production label and five listener judgements for every clip. Using frozen XLS-R, HuBERT, and WavLM representations with matched probes, we compare three label authorities: the expressed condition, the listeners’ hard plurality, and their complete vote distribution. The results tell a consistent story. Recovering the expressed condition is easier than predicting a tie-broken listener plurality, but that gap compares two different questions rather than revealing a general loss of SER ability. As listener agreement weakens, the plurality becomes an increasingly brittle hard target. Training on the full vote distribution does not substantially change which class wins the model’s hard decision, but it produces probability assignments that more closely follow the alternatives selected by listeners, especially for ambiguous clips. No authority therefore replaces the others. Expressed labels describe how speech was produced, pluralities provide an operational single-class decision, and vote distributions preserve perceptual uncertainty. For human-facing Internet of Sounds applications, evaluations should state which authority is being predicted, retain its aggregation rule and agreement information, and use distributional scores when disagreement is part of the target.
[NLP-42] WenSight-Decision-1 for Direct Decisions under E-commerce Rules
【速读】： 该论文旨在解决电子商务决策中如何高效、准确地将自然语言规则与具体案例证据相结合的问题，尤其针对多语言环境下跨任务一致性判断的挑战。其核心解决方案是提出WenSight-Decision-1，一个共享模型，能够直接基于规则和案例信息对多个操作性任务的候选选项进行评分。该模型通过使用参考选项作为监督信号，对Qwen3.5-9B进行微调，并在训练与推理阶段采用相同的评分机制，实现了端到端的一致性决策。实验表明，在1,440个预留的合成操作案例上，模型准确率从原始Qwen3.5-9B的61.60%提升至92.92%，显著改善了六项任务的整体表现。关键在于通过联合微调将异构的操作性判断统一到一个共用的决策接口下，从而实现跨任务、跨语言的高效协同决策。
链接: https://arxiv.org/abs/2610.26444
作者: Mingwen Liu(WenSight AI Lab),Jun Sun(TooChuang),Sichen Wang(WenSight AI Lab)
机构: WenSight AI Lab(文视智能实验室); TooChuang(图创); The Chinese University of Hong Kong (香港中文大学)
类目: Computation and Language (cs.CL)
备注: 6 pages, 2 figures, 1 table
Abstract:E-commerce decisions require applying natural-language rules to the available case evidence. We introduce WenSight-Decision-1, a shared model that scores supplied options directly from rules and case information for six operational tasks, with examples spanning six languages. We fine-tune Qwen3.5-9B with reference options as supervision, using the same option scores for training and inference. On 1,440 synthetic operational cases reserved for final evaluation, accuracy rises from 61.60% for Qwen3.5-9B to 92.92% for WenSight-Decision-1 under identical inputs and scoring. Joint fine-tuning improves all six tasks, bringing heterogeneous operational judgments under a common decision interface.
[NLP-43] HarnessRL: Co-Evolving Policies with a Just-in-Time Training Harness for Open-Ended RL
【速读】： 该论文旨在解决基于评分标准（Rubric-based）的强化学习在开放式生成任务中因奖励信号随训练进程失效而导致性能退化的问题。具体而言，现有方法存在两种失效模式：一是非区分性奖励（non-discriminative rewards），由于探索不足或任务难度与策略能力不匹配，导致评分趋于全低或全高；二是虚假奖励（spurious rewards），即评分升高并不对应生成质量提升。这两种问题均源于训练过程中任务采样、探索策略和评估标准固定不变，无法适应策略的动态演化。其解决方案的关键在于提出HarnessRL——一种与策略协同进化的训练期调控机制，通过三个接口实现任务采样、生成引导和评分标准的动态调整。HarnessRL维护一个训练时记忆库，由“调适演进器”（harness evolver）从中提取跨任务与内任务证据，识别导致奖励失效的具体接口，并通过“提议者-即时评判者”循环迭代优化更新方案：评判者仅在冻结当前策略状态下验证候选更新的有效性，确保改进可被可靠评估，无效方案则返回修正。此外，研究还构建了涵盖医疗、科学、写作、角色扮演和指令遵循五大领域的基准数据集Rubric-Mix-10k，实验表明HarnessRL在4B与9B模型规模下均显著优于最强基线2.8至3.1分，且达到最优检查点所需训练步数近乎减半。
链接: https://arxiv.org/abs/2610.26426
作者: Xiaoqiang Wang,Ting-En Lin,Yuchuan Wu,Pinyi Zhang,Caixin Kang,Ling Li,Bingxue Qiu,Jieping Ye,Yongbin Li,Bang Liu
机构: Alibaba Token Hub, Alibaba Group(阿里巴巴通义实验室); Université de Montréal Mila - Quebec AI Institute(蒙特利尔大学 Mila-魁北克人工智能研究所)
类目: Computation and Language (cs.CL)
备注: Work in progress, 21 pages, 6 figures
Abstract:Rubric-based reinforcement learning (RL) trains policies for open-ended generation with natural-language criteria as rewards, but these rewards can become ineffective as training progresses. We identify two failure modes. Under non-discriminative rewards, scores collapse to uniformly low or high values due to insufficient exploration or a mismatch between task difficulty and policy capability. Under spurious rewards, higher scores do not correspond to better responses. Both reflect a training configuration whose task sampling, exploration, and evaluation stay fixed as the policy evolves. We introduce HarnessRL, a training-time harness that co-evolves with the policy through three interfaces for task sampling, rollout guidance, and rubric criteria. HarnessRL maintains a training-time memory from which a harness evolver retrieves within-task history and cross-task evidence. Using this evidence, the evolver attributes each reward failure to the interface whose update can address it. It then runs a proposer and just-in-time critic loop in which the critic accepts a candidate update only after measuring its effect on the current frozen policy, returning rejected candidates for revision. We also curate Rubric-Mix-10k, a five-domain training and evaluation suite for open-ended RL across medical, science, writing, role-play, and instruction-following tasks. On Rubric-Mix-10k, HarnessRL outperforms the strongest baselines by 2.8 to 3.1 points across both 4B and 9B model scales and nearly halves the training steps to its best checkpoint. Code and data are available at this https URL.
[NLP-44] SNO-PI: Label-Scarce SNOMED CT Entity Linking with Pre-trained Encoders and Curated Terminology Knowledge
【速读】： 该论文旨在解决临床文本提及（clinical mentions）到SNOMED CT术语的实体链接问题，该任务具有零样本（zero-shot）和少样本（few-shot）特性，且标签空间极为庞大——在约21.9万可选概念中，仅6,595个出现在632份MIMIC-IV出院摘要标注数据中，其中三分之一的概念仅出现一次。这一挑战使得传统监督学习方法难以奏效，因此亟需一种能够有效利用预训练模型迁移能力并融合人工构建知识的解决方案。其核心创新在于SNO-PI框架，该框架整合了高召回率片段生成（high-recall span proposal）、基于SapBERT的候选检索以及基于列表的集成重排序（listwise ensemble reranking），并采用三组经过特定领域预训练的BERT类编码器：首先在未标注的MIMIC-IV病历文本上进行掩码语言建模预训练，再在SNOMED CT同义词与本体层级结构上进一步预训练，最后进行任务微调。研究发现，人工构建的知识在处理训练中从未见过的概念时表现尤为关键，但这些未见概念仍是当前系统的主要错误来源。
链接: https://arxiv.org/abs/2610.26425
作者: Piyawoot Songsiritat
机构: The University of Queensland (昆士兰大学)
类目: Computation and Language (cs.CL)
备注: 5 pages, 4 figures, 6 tables
Abstract:We present SNO-PI, the top-performing submission on SNOMED International’s SNOMED CT Entity Linking Benchmark as of 18 October 2026. SNO-PI achieves a macro character IoU of 0.5387, outperforming the organisers’ reference system (0.4582; based on the 2024 challenge winner’s approach) by 18% relative. Linking clinical mentions to SNOMED CT is a zero- and few-shot problem over a very large label space: the 272 annotated MIMIC-IV discharge summaries cover only 6,595 of the roughly 219,000 eligible concepts, and a third of those occur only once. This motivates a design built on transfer from pre-trained models and human-curated knowledge. SNO-PI combines high-recall span proposal, SapBERT candidate retrieval and listwise ensemble reranking with three BERT-family encoders, each further pre-trained with masked language modelling on unlabelled MIMIC-IV notes and then on SNOMED CT synonyms and hierarchy before task fine-tuning. We find that curated knowledge helps most on concepts never seen in training, yet these concepts remain the main source of error.
[NLP-45] Governing the Rule Space. Ruliology Computational Irreducibility and Anticipatory Governance of Quantum Technologies
【速读】： 该论文旨在解决新兴技术治理中“滞后性监管”问题，即立法者往往在技术发展之后才介入，导致监管措施难以有效应对快速演进的技术变革。其核心挑战在于如何在不确定性高度复杂的背景下实现前瞻性的制度设计。该研究提出的关键解决方案是构建“预见性治理”（anticipatory governance）框架，强调通过识别信号、构建情景并动态调适规则来培育制度的适应能力，而非追求对未来的精确预测。在此基础上，论文提出了“规则学框架”（Ruliological Framework for Anticipatory Regulation），将规则、转型与演化路径三者进行系统区分与评估，以探索多种可能的发展路径。该框架有条件地引入“计算不可约性”概念，将其视为一种方法论上的假设而非法律系统的固有属性，从而避免过度悲观的技术决定论倾向。以“量子欧洲战略”为案例，该研究进一步验证了该框架在复杂技术治理场景中的适用性与反思价值。
链接: https://arxiv.org/abs/2610.26423
作者: Antonio Felice Uricchio,Francesco Scagliola
机构: 未知
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: English translation of an article originally published in Italian in Rivista della Corte dei conti, no. 4 (2026), pp. 27 - 41. Published with the journal’s permission
Abstract:The regulation of emerging technologies is often conceived as a strategy of catching up, because legislators arrive too late. Yet gathering signals, constructing scenarios and adapting rules to an anticipated future are fundamental tasks for those entrusted with governance, including the governance of rules themselves. The literature on so-called future law has already shown that anticipation means building institutional capabilities rather than predicting outcomes precisely (anticipatory governance). Taking this premise as its starting point, this article proposes to understand regulation as an exploration of the transformations that may occur. The Ruliological Framework for Anticipatory Regulation distinguishes rules, transformations and trajectories, assessing alternatives and possible developments. Within this approach, computational irreducibility is adopted conditionally, not as a demonstrated property of legal systems. The Quantum Europe Strategy provides a case study and further grounds for reflection.
[NLP-46] Self-Consistency of Entity Sets and Citations in a Deep Research System under Repeated Prompting
【速读】： 该论文旨在解决生成式深度研究系统（Deep Research Agent, DRA）在开放性合成任务中缺乏稳定性和可重复性的问题，尤其针对传统闭域问答（Closed-world QA）基准无法有效评估其性能的局限性。其核心问题是：在相同提示下，商业级DRA的输出在实体提及与引用文献方面表现出显著的不一致性，导致结果难以验证和复现。解决方案的关键在于通过量化重复相同提示下的输出结构稳定性，采用命名实体集合与引用集合的成对交集（Jaccard相似度）来表征系统自一致性（self-consistency），并基于自定义分类体系对实体进行归类分析。研究发现，在十次重复运行中，实体集合的中位数Jaccard相似度仅为0.12，引用集合的相似度低于0.02，且发现曲线未呈现饱和趋势，表明系统输出存在高度随机性。进一步分析显示，不同实体类别间的（不）一致性差异显著，说明自一致性并非系统固有的稳定属性，而应被理解为依赖于底层实体空间分布的动态特征。这一发现揭示了当前DRA在可信赖性方面的根本缺陷，为未来提升系统稳定性提供了关键评估维度。
链接: https://arxiv.org/abs/2610.26421
作者: Lauri Pykälä,Kimmo K. Kaski
机构: Aalto University School of Science (阿尔托大学理学院); Aalto University (阿尔托大学)
类目: Computation and Language (cs.CL)
备注: 6 pages, 3 figures
Abstract:Deep research agents (DRA) are commonly used for open-ended synthesis tasks where closed-world QA benchmarks are poorly suited to characterize relevant aspects of performance. We study the self-consistency of one commercial deep research system under repeated identical prompting, representing each report with the set of named entities it mentions, categorised by a custom schema, and the set of citations it retrieves. Across repeated runs of a single open-ended cybersecurity related prompt, we quantify pairwise overlap to characterize how stable the retrieved and synthesized output structure is under repeated identical prompts. In ten runs, the pairwise entity Jaccard results in a median of 0.12 and the pairwise citation Jaccard is below 0.02, with discovery curves that do not visibly saturate. The magnitude and structure of (dis)agreement between entity sets varies by entity class, which suggests that self-consistency in a DRA is better understood as a function of the underlying entity space than as a stable property of the system.
[NLP-47] Co-Evolving Policy and Reward for Code Generation: A Mechanistic Study of Single-Adapter Drift ICTAI2026
【速读】： 该论文旨在解决生成式代码模型在强化学习框架下实现自进化能力的问题，特别是探究基于可验证奖励的强化学习（RLVR）中策略与奖励协同进化机制是否能有效提升代码生成模型的泛化性能。其核心挑战在于：尽管通过程序执行获得的反馈可提供客观评价并支持训练过程中的自我改进，但这种机制是否能在不引入过拟合或策略漂移的前提下，真正实现对未见问题的通用优化仍存疑。解决方案的关键在于采用“策略与奖励协同进化框架”（SPARK），利用执行回放（rollout recycling）将历史生成结果及其评估反馈循环用于同时优化生成策略与评估能力。然而，实验结果表明，在使用Qwen2.5-Coder-3B-Instruct进行对比测试时，尽管协同进化在训练集上提升了奖励得分，但并未带来对未见问题的泛化性能提升；相反，其导致显著的单适配器策略漂移（single-adapter policy drift），表现为与预训练代码生成行为的KL散度高达执行仅用型GRPO的34倍，且随着训练推进，每序列生成词数下降、保真度下降，最终无法维持早期性能优势。研究揭示，单纯依赖训练奖励提升不足以证明具备可泛化的自改进能力，强调了在评估中必须结合保留集表现和策略漂移诊断的重要性。
链接: https://arxiv.org/abs/2610.26416
作者: Amarsaikhan Batjargal,Noriko Tomuro
机构: Jarvis College of Computing and Digital Media, DePaul University (杰弗里学院计算与数字媒体学院，德保罗大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG); Software Engineering (cs.SE)
备注: Accepted at IEEE ICTAI 2026
Abstract:Reinforcement learning with verifiable rewards (RLVR) is particularly well suited to code generation because generated programs can be objectively evaluated through execution. Execution also provides behavioral information that can be reused as additional supervision. The Synergistic Policy and Reward Co-Evolving Framework (SPARK), originally proposed for reasoning tasks, exploits this idea by recycling training rollouts and their evaluation results to train a model not only to generate solutions but also to evaluate and improve them. We investigate whether this policy-reward co-evolution approach transfers effectively to code generation and whether execution-grounded reflection provides an additional source of self-improvement. We compare SPARK-style rollout recycling with execution-only Group Relative Policy Optimization (GRPO) using Qwen2.5-Coder-3B-Instruct. Our results do not support either hypothesis under the configuration tested. Although co-evolution improves training performance, these gains do not generalize to held-out problems. Compared at the same stage of training, rollout recycling produces roughly 34x greater KL divergence from the reference policy than execution-only GRPO. Over training, trained tokens per GRPO sequence fall from 185 to 86 as training pass rate increases, while held-out performance declines or fails to sustain early gains. We characterize these patterns as single-adapter policy drift: the shared trainable parameters increasingly depart from pretrained code-generation behavior without corresponding gains in generalization. The results show that higher training reward alone is not evidence of generalizable code self-improvement and highlight the importance of held-out evaluation and drift diagnostics.
[NLP-48] Where Does Rhetorical Distortion Live? A Multi-Platform Multi-Language Measurement of Influencer Content
【速读】： 该论文旨在解决在线意见领袖内容中广泛存在的“修辞扭曲”（rhetorical distortion）问题，即虽然信息本身真实，但通过夸大其重要性、制造焦虑、宣称新颖性、使用带有情感倾向的语言或时间错位等手段进行不当框架化表达。传统事实核查与点击诱饵检测方法无法有效识别此类非虚假但具有误导性的传播策略，且学界对这种扭曲的来源（创作者、话题、平台或受众互动）尚缺乏系统认知。为此，研究提出基于五维文献基础的修辞扭曲操作化定义（显著性夸大、焦虑制造、新颖性宣称、情绪化语言、时间扭曲），并构建三阶段分析管道：首先利用双语词汇规则生成候选样本，再通过负向过滤排除已知误报情境，最后由大语言模型（LLM）验证或重标。在涵盖六种平台（RSS/新闻简报、YouTube、Bluesky、Reddit、微博、X/Twitter）共330个账号的13,060条内容上应用该方法，发现平台因素对账户层面的修辞扭曲影响远大于内容类别（Kruskal-Wallis ε²=0.33；两因素方差分析 η_p²=0.49 vs. 0.09）。其中，X/Twitter的扭曲指数（DI=7.74）显著高于其他平台（如RSS为0.78）。进一步区分推文与其回复层分析发现，用户在帖子下方看到的前5条回复所承载的扭曲程度是原推文的3.5倍（7.65 vs. 2.17，p<10⁻⁶，43个账户中有36个呈现此差异），主要由“情绪化语言”驱动；即便控制文本长度，差距仍达1.5倍且具统计显著性。而仅考虑创作者原始推文时，其扭曲水平与YouTube和Bluesky无显著差异。两名独立人工标注者验证了回复层的扭曲差距，且使用GPT-4o作为第二分类器亦复现了平台排序与回复层差异。研究结论表明，在多层级对话平台中，修辞扭曲主要存在于观众互动层，因此需建立分离创作者与用户回复文本的测量机制以准确评估信息传播中的非真实性风险。
链接: https://arxiv.org/abs/2610.26412
作者: Xucheng Yu,Jiangping Chen,Haohan Wang
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: ocial and Information Networks (cs.SI); Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 12 pages
Abstract:Much of the manipulation in online influencer content is not false but disproportionate: an accurate fact is framed as historic, urgent, or catastrophic. Fact-checking and clickbait detection are not designed to capture such framing, and little is known about where it originates: with the creator, the topic, the platform, or the audience that replies. We operationalize rhetorical distortion through five literature-grounded dimensions (Significance Inflation, Anxiety Manufacturing, Novelty Claims, Loaded Language, and Temporal Distortion) and measure it with a three-tier pipeline in which bilingual lexical rules propose candidates, negative filters cancel known false-positive contexts, and an LLM verifies or relabels the candidates. Applying the pipeline to 13,060 posts from 330 accounts on six platforms (RSS/newsletters, YouTube, Bluesky, Reddit, Weibo, and X/Twitter), we find that platform is far more strongly associated with account-level distortion than content category (Kruskal-Wallis \varepsilon^2=0.33 ; two-way ANOVA \eta_p^2=0.49 for platform vs. 0.09 for category). X/Twitter scores highest (DI 7.74 vs. 0.78 for RSS). Classifying tweets and their replies separately shows where this comes from: the reply layer readers see under a post (its top-5 replies) carries 3.5x the distortion of the tweet it answers (7.65 vs. 2.17, p10^-6 , 36 of 43 accounts), driven by Loaded Language. At equal text length the gap shrinks to 1.5x but remains significant, and creator tweets alone are statistically indistinguishable from YouTube and Bluesky. Two independent human annotators confirm the reply gap, and the platform ordering and the reply gap replicate with GPT-4o as a second classifier. On threaded platforms, measured distortion thus lives largely in the audience layer, which motivates measurement that separates creator and reply text.
[NLP-49] Expert L2 English Speech Evaluation on a Gaming PC: Accuracy Fluency and Prosody with XLS-R XEUS and PhoneticXeus
【速读】： 该论文旨在解决在消费级计算机上对二语（L2）英语发音准确性、流利度和语调（prosody）进行客观评估的难题，尤其针对现有自动语音评估系统依赖云端服务或需大规模数据微调的局限性。其核心解决方案在于：利用预训练的XLS-R、XEUS和PhoneticXeus声学表示，结合参考文本特征，并通过在SpeechOcean762、韩国学习者二语语音语料库（英语）、J-AESOP及其组合数据集上训练的岭回归（ridge regression）模型，实现无需对私有评估数据进行神经网络微调或校准的高性能评估。实验结果表明，在305段独立保留的录音上，经专家评分验证，多种岭回归配置在斯皮尔曼相关系数与加权平方κ值上均显著优于2026年1月所存档的Azure API评分，验证了基于预训练表示与特定构念回归建模在消费硬件上的有效性。
链接: https://arxiv.org/abs/2610.26406
作者: Richard Rose(Hankuk University of Foreign Studies)
机构: HUFS(韩国外国语大学)
类目: Computation and Language (cs.CL); Sound (cs.SD)
备注: 19 pages, including appendices. Code and trained weights: this https URL
Abstract:oollim is a locally executable system for evaluating L2 English pronunciation accuracy, fluency, and prosody on a consumer computer. The system combines pretrained XLS-R, XEUS, and PhoneticXeus representations with reference-text features and ridge regressions trained on SpeechOcean762, the Korean Learners’ L2 Speech Corpus (English), J-AESOP, and their combinations. On 305 independently held-out recordings assessed by three expert raters, several ridge configurations substantially outperform archived Azure API scores obtained in January 2026 in both Spearman correlation and quadratic weighted kappa across all three constructs. These results support expert speech evaluation on consumer hardware using pretrained representations and construct-specific regressions, without neural-model fine-tuning or calibration on the private evaluation data.
[NLP-50] LaMER: Language-Conditioned Multi-Expert Routing for Weakly Supervised Referring Expression Comprehension
【速读】： 该论文旨在解决弱监督条件下指代表达理解（Referring Expression Comprehension, REC）任务中面临的视觉-语言语义鸿沟以及仅依赖粗粒度图像级信号难以准确推断区域与文本对应关系的问题。其核心挑战在于单一视觉编码器无法同时捕获语义对齐、细粒度纹理和边界感知等互补特性，而简单融合多个预训练编码器又会导致表征异质且难以统一。为此，论文提出LaMER框架，其关键创新在于设计了一种语言条件的跨模态路由融合模块（Cross-Modal Routing Fusion），动态聚合四个冻结的专家编码器，根据具体的指代表达自适应选择最优专家组合，从而实现更精准的多模态特征融合。此外，引入带有负锚点增强的细胞-文本匹配头（Cell-Text Matching Head），结合对比学习、重建、对齐与负载均衡的多目标损失函数，有效稳定了优化过程。实验表明，LaMER在RefCOCO、RefCOCO+和RefCOCOg三个基准上分别取得77.67%、62.81%和69.49%的Acc@IoU≥0.5，显著优于DViN和WeakMCN等现有方法，消融实验进一步验证了各组件的有效性。
链接: https://arxiv.org/abs/2610.26404
作者: Yu Yu Chang,Tzu-Hsuan Weng,Youssef Mohamed Amin,Mohaned Ahmed Elsayed,Hung-Hsuan Chen,Min-Te Sun,Sherine Shawky
机构: National Central University, Taoyuan, Taiwan(国立中央大学, 台湾桃园); Pharos University in Alexandria, Alexandria, Egypt(亚历山大法罗斯大学, 埃及亚历山大)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Referring Expression Comprehension (REC) localizes the image region described by a natural language expression, underpinning applications such as autonomous navigation, human-robot interaction, and medical image analysis. Under weak supervision, REC is challenged by the visual-linguistic semantic gap and the difficulty of inferring region-text correspondences from coarse image-level signals alone. Moreover, no single visual encoder captures all complementary properties—semantic alignment, fine-grained texture, and boundary awareness—while naively fusing multiple pretrained encoders yields heterogeneous, hard-to-reconcile representations. We propose LaMER, a weakly supervised REC framework that dynamically aggregates four frozen expert encoders through a language-conditioned Cross-Modal Routing Fusion module, adapting expert selection to each referring expression without instance-level bounding-box supervision. A Cell-Text Matching Head with Negative Anchor Augmentation, trained with a multi-objective loss combining contrastive, reconstruction, alignment, and load-balancing terms, further stabilizes optimization. LaMER achieves 77.67%, 62.81%, and 69.49% Acc@IoU(\geq 0.5) on RefCOCO, RefCOCO+, and RefCOCOg, surpassing DViN and WeakMCN by up to 16.49 and 14.49 points, with ablations validating each component.
[NLP-51] How Benchmark Graph Construction Shapes Multi-Hop KG Navigation: A Controlled Audit of PathQuestion and PathQuestion-Large
【速读】： 该论文旨在解决多跳知识图谱问答（Multi-hop Knowledge-Graph Question Answering, KGQA）基准测试中，图结构构建对导航型方法性能评估的隐性影响问题。现有基准（如PathQuestion, PQ 和 PathQuestion-Large, PQL）常被视为固定的语言与推理任务，但实际中其难度不仅取决于问题本身，还受具体图结构的拓扑特性制约，导致图构建成为导航任务不可分割的一部分。为分离这一影响因素，研究提出在严格控制条件下进行干预：保持问题、主题实体、答案、路径标注及家族级划分不变，仅将原任务过滤后的子图替换为基准所附带的更大源知识图谱（source KG）。此“源图恢复”操作显著扩展了导航搜索空间，增加了图一致答案与路径的多样性，同时大幅降低无指导情况下的可达性。通过结构参考指标、冻结的小语言模型导航策略以及训练好的MINERVA策略进行评估发现，无论采用何种导航策略，源图恢复均导致原始答案准确率和轨迹保真度显著下降，而仅依赖图一致性评价无法完全恢复性能损失。结果表明，基准图结构并非单纯的实现细节，而是导航任务规范的重要组成部分，即使在保持释放的问答任务不变的前提下，仍会实质性地影响模型行为的测量结果。
链接: https://arxiv.org/abs/2610.26399
作者: Eduin E. Hernandez,Sergio A. Diaz,Luis F. Garcia,Nurassyl Askar,Stefano Rini
机构: Model call failure
类目: Computation and Language (cs.CL)
备注: 17 pages. PQ/PQL reconstruction code available at this https URL
Abstract:Multi-hop knowledge-graph question answering (KGQA) benchmarks are often treated as fixed language-and-reasoning tasks, but for navigation-based methods, difficulty also depends on the graph in which those tasks are instantiated. This makes graph construction a consequential but often entangled part of the navigation task. To isolate this effect, we audit PathQuestion (PQ) and PathQuestion-Large (PQL) as graph-navigation benchmarks under a controlled intervention: we hold questions, topic entities, released answers, released path annotations, and family-level splits fixed while replacing each released task-filtered graph with the larger source KG distributed with the benchmark. Source restoration expands the effective navigation search space and graph-consistent answer and path multiplicity while sharply reducing unguided reachability to released answers. We evaluate whether these structural changes persist across navigation policies using structural references, frozen small-language-model navigation policies, and trained MINERVA policies. Across both frozen and trained policies, restoration substantially reduces released-answer accuracy and trajectory fidelity, while graph-consistent evaluation recovers only part of the loss. These results show that benchmark graph construction is part of the navigation task specification rather than merely an implementation detail and can materially shape measured behavior even when the released QA task is held fixed.
[NLP-52] Bridged Self-Distillation for On-Policy Learning ALT
【速读】： 该论文旨在解决在策略蒸馏（on-policy distillation）中，强教师模型（strong teacher）与小型学生模型（small student）之间存在的双重不匹配问题：一是教师模型的解空间难以被学生模型模仿，二是教师模型输出的词元级概率分布可能与学生模型的词元空间不兼容或不可用。现有方法通常依赖于特定形式的监督来应对不同场景，缺乏统一性。为此，本文提出桥接自蒸馏（Bridged Self-Distillation, BSD），一种适用于任何提供推理轨迹（reasoning traces）的教师模型的统一蒸馏框架。其核心创新在于引入外部教师作为特权上下文生成推理轨迹，而学生模型在无此上下文的情况下生成轨迹；同时，一个基于外部教师上下文条件化的学生副本（即自教师，self-teacher）在学生自身的词元空间内对这些轨迹进行评分，从而提供词元级指导信号。该指导信号结合验证器（verifier）提供的结果反馈，用于调整失败轨迹的更新过程。由于外部教师仅需生成文本，无需暴露内部概率分布，因此该方法可统一适用于共享分词器（shared-tokenizer）、跨分词器（cross-tokenizer）及闭源API教师等多种场景。实验表明，BSD在各类场景下均优于基线模型，实现了性能提升并提供了统一、可泛化的蒸馏路径。
链接: https://arxiv.org/abs/2610.26397
作者: Runsheng Chen(1),Yihong Liu(2 and 3),Raoyuan Zhao(2 and 3),Michael A. Hedderich(2 and 3),Hinrich Schütze(2 and 3),Matthias Grabmair(1) ((1) Technical University of Munich, Germany, (2) LMU Munich, Germany, (3) Munich Center for Machine Learning, Germany)
机构: Technical University of Munich(慕尼黑工业大学); LMU Munich(慕尼黑大学)
类目: Computation and Language (cs.CL)
备注: 21 pages, 10 figures, 15 tables. Code: this https URL
Abstract:On-policy distillation can transfer the reasoning capabilities of a strong teacher to a smaller student. However, distillation from a stronger teacher model faces two mismatches: its solutions may be difficult for a smaller student to imitate, and its token-level probabilities may be incompatible with, or even unavailable to, the student. Existing methods handle these settings through different forms of supervision. To this end, we propose Bridged Self-Distillation (BSD), a unified framework for distilling from any teacher that provides reasoning traces. The model to be distilled from, termed the external teacher, first provides a reasoning trace as privileged context, which can be generated in an offline fashion. The student generates rollouts without this context, while a copy of the student conditioned on it, serving as the self-teacher, scores the same rollouts in the student’s own token space. BSD uses the resulting token-level guidance to adjust updates on unsuccessful rollouts, with a verifier providing the outcome signal. As the external teacher is queried only to produce text, the same training procedure applies to all teacher scenarios: shared-tokenizer, cross-tokenizer, and closed API teachers. Across these scenarios, BSD improves over the base model and outperforms baselines tailored to each scenario, offering a novel and unified pathway for distillation. Code is available: this https URL
[NLP-53] An Empirical Study of Parameter-Efficient Fine-Tuning for Mixture-of-Experts LLM s
【速读】： 该论文旨在解决在资源受限条件下，如何高效地对大规模混合专家模型（Mixture-of-Experts, MoE）进行参数高效微调（Parameter-Efficient Fine-Tuning, PEFT）的关键问题。现有研究缺乏系统性的方法指导，导致PEFT在MoE架构上的应用存在显著空白。其解决方案的核心在于通过大规模系统性实验，探索并验证一系列关键设计决策：应选择哪些模块进行微调、以何种粒度适配MoE模块、采用何种PEFT方法以及负载均衡目标的权重设置。研究发现，仅微调专家（expert）部分即可在保持相近通用能力的前提下实现最优下游性能；原生专家权重更新相比绕过专家或整模块微调，在下游性能与通用能力保留之间具有更优权衡；LoRA与DoRA组合在所评估方法中表现出最强的下游精度表现；且引入非零负载均衡系数并未带来帕累托改进。最终，作者提出对MoE²-LoRA进行权重级重构，将模块级旁路池迁移至原生专家权重，使四项上游任务指标提升2.4至17.9分，同时减少17%的可训练参数量，并降低3.2分的平均下游成本。
链接: https://arxiv.org/abs/2610.26370
作者: Minglei Li,Peng Ye,Haonan He,Jingqi Ye,Qingyu Yang,Tao Chen
机构: Fudan University (复旦大学); Shanghai Artificial Intelligence Laboratory (上海人工智能实验室); The Chinese University of Hong Kong (香港中文大学); University of Science and Technology of China (中国科学技术大学); KTH Royal Institute of Technology (皇家理工学院); Shanghai Innovation Institute (上海创新研究院)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Parameter-efficient fine-tuning (PEFT) methods are critical for adapting large models under resource-constrained settings. However, as mixture-of-experts (MoE) architectures, which offer efficient inference at massive parameter scales, gain increasing prevalence in the community, a critical gap arising from the lack of systematic recipes for applying PEFT to MoE models. To narrow this gap, we conduct a large-scale systematic study that investigates a set of essential design recipes, including which modules should be adapted, which granularity should be used to adapt MoE modules, which PEFT method to use, and to what extent should the load-balancing objective be used, across three MoE backbones on downstream tasks and general knowledge tasks. Our experiments yield several key insights. First, adapting MoE experts alone achieves the highest downstream performance with competitive general capability retention under a comparable parameter budget. Second, native expert-weight updates offer a better downstream–retention trade-off than adapters that bypass experts or entire MoE modules. Third, LoRA and DoRA together define the strongest operating region for high downstream accuracy among the evaluated PEFT methods. Fourth, no nonzero load-balancing coefficient Pareto-dominates the zero baseline. We finally evaluate a weight-level redesign of MoE ^2 -LoRA on Qwen3. Moving its module-level bypass pool into native expert weights improves all four upstream endpoints by 2.4 to 17.9 points with 17% fewer trainable parameters and a 3.2-point average downstream cost.
[NLP-54] VibeTraining-Bench: Data-Trainer-Harness Co-Design for Iterative System Improvement
【速读】： 该论文旨在解决生成式智能体在递归自我改进（Recursive Self-Improvement, RSI）过程中，如何实现训练数据、训练器（trainer）与推理框架（inference harness）三者协同优化的核心挑战。其关键在于构建一种跨组件协同设计（cross-component co-design）能力，使智能体能够基于验证反馈，在固定资源预算下，迭代地诊断系统瓶颈并联合优化数据、训练策略与推理架构。为此，研究提出VibeTraining-Bench基准，涵盖终端操作、软件工程、多轮工具使用及数学推理等任务，通过测试集隔离、研究与评分容器分离以及对研究代理的网络限制等三重安全机制，防止奖励劫持（reward hacking）。实验表明，前沿研究智能体可显著提升目标系统性能（如在SWE-bench Verified上将准确率从32.9%提升至53.7%，接近人类标注指令模型的55.4%），但其在持续迭代中实现累积性能增益的能力仍有待增强。该工作为推进递归自我改进提供了统一评估平台，深化了对智能体系统跨组件优化机制的理解。
链接: https://arxiv.org/abs/2610.26342
作者: Ruotian Peng,Yuanyi Ren,Shuaiqi Liu,Xin Guo,Fang Zhu,Yangyi Huang,Yue Wu,Zheng Wang,Yandong Wen,Zhihang Fu
机构: Zhejiang University (浙江大学); Westlake University (西湖大学); Alibaba Token Hub, Alibaba Group (阿里巴巴令牌中心，阿里巴巴集团); South China University of Technology (华南理工大学); The Chinese University of Hong Kong (香港中文大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: (Technical report (27 pages, 21 figures)
Abstract:Recursive self-improvement (RSI) requires agents to diagnose system bottlenecks and coordinate improvements across training and inference. We refer to the agents’ capability to jointly optimize training data, the trainer, and the inference harness as cross-component co-design. We introduce VibeTraining-Bench, a benchmark for evaluating this capability across tasks spanning terminal operation, software engineering, multi-turn tool use, and mathematical reasoning. Within each task, research agents start from an identical initial agent system under a fixed resource budget. Guided by validation feedback, they iteratively refine data, trainers, and inference harnesses, then select and submit a final agent system for evaluation on the test set. To prevent reward hacking, we enforce three safeguards: test data isolation, separate research and grading containers, and network restrictions for research agents. Our evaluation results show that frontier research agents substantially improve target agent systems. For example, on SWE-bench Verified, GPT-6-Astra raises the target agent system’s performance from 32.9% to 53.7%, comparable to the human-developed instruct model’s 55.4%. However, there is still room to improve their ability to co-design and iteratively refine data, trainer, and harness to achieve cumulative performance gains. We hope that VibeTraining-Bench will serve as a unified platform for advancing RSI, deepening our understanding of cross-component optimization for agent systems. Code and data are available at this https URL.
[NLP-55] yped Decision Models under Prevalence Shift: Decision Training Calibration and Implicit Priors
【速读】： 该论文旨在解决生成式模型在实际部署中因类别分布（class frequency）变化导致的校准性退化问题，特别是针对类型化决策模型（Typed Decision Models, TDMs）在标签分布偏移（prevalence shift）下的性能稳定性与校准能力。现有评估存在两大缺陷：一是未将厂商提供的决策训练（decision training）收益与骨干模型（backbone）本身的性能分离；二是测试仅限于原始测试集的类别分布，未能模拟真实场景中的分布漂移。本文的关键解决方案在于通过在五个公开数据集上对三种TDM（Jev、Tev1-4B、Nimble-9B）及其对应的基线模型和微调分类器进行35组预先固定的成对对比实验，在模拟的分布偏移条件下系统评估决策训练的有效性及校准鲁棒性。研究发现：其一，决策训练并非普遍有益——尽管Nimble-9B在意图识别任务上校准性更优，但在金融情感分析任务上宏平均F1下降12.0个百分点；其二，面对中等程度的分布偏移，Jev的期望校准误差（ECE）和阈值风险增长显著高于微调后的DeBERTa模型（分别增加1.7和2.5个百分点），但整体差距在2000次模拟偏移下平均不超过1个百分点；其三，采用模型开发阶段平均预测作为先验的期望最大化（EM）先验校正方法无需标签即可降低Jev在高偏移场景下的ECE达2.7和1.8个百分点，虽效果不及有标签重新校准，但具备零样本应用潜力。结论表明，所测试的TDM既非普遍优于其骨干模型，也非对分布偏移特别脆弱，其主要校准缺陷源于偏移前即存在的偏差。
链接: https://arxiv.org/abs/2610.26325
作者: Abdul Rahman Iqbal,Muhammad Fawad Nasim,Amir Saleem
机构: The Superior University, Lahore, Pakistan; Faculty of Computer Science and Information Technology
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 27 pages, 5 figures. Code, protocol and per-example outputs: this https URL
Abstract:Typed decision models (TDMs) return a probability over caller-defined options, and Jev is described by its vendor as calibrated. But in deployment class frequencies change, so calibration measured on a fixed test set need not hold where thresholds are applied. Existing evaluations leave two gaps: the gains of vendor decision training have not been separated from those of the backbone under an identical readout, and TDMs have been tested only at the test set’s class distribution. We ask whether decision training improves a model over its own backbone, and whether a TDM’s calibration and thresholds withstand prevalence (label) shift as well as a fine-tuned classifier’s. We compare three TDMs (Jev, Tev1-4B, Nimble-9B), their base models and two trained classifiers on identical decisions from five public datasets under simulated shift, with 35 paired tests fixed before any test was computed. First, decision training is not uniformly beneficial: Nimble-9B, which differs from Qwen3.5-9B only by a LoRA adapter, is better calibrated on intent tasks but 12.0 Macro-F1 points worse on financial sentiment. Second, under moderate shift Jev’s expected calibration error (ECE) and threshold risk grow significantly more than a fine-tuned DeBERTa’s only on financial sentiment, by 1.7 and 2.5 percentage points; averaged over 2,000 shift scenarios (exploratory), the gap is at most one point on every dataset. Third, EM prior correction with the model’s mean development prediction as its prior needs no labels and lowers Jev’s ECE under shift by 2.7 and 1.8 points on the two datasets where it is highest, although labelled recalibration lowers it far more. The three TDMs we tested are neither uniformly better than their backbones nor uniquely fragile under shift; for calibration, their main weakness is the miscalibration they already have without shift. We release all inputs, code and outputs.
[NLP-56] Answering While You Type: Optimizing Time-to-Answer with Streaming Generation
【速读】： 该论文旨在解决用户感知延迟（user-visible latency）的问题，即从提交请求到获得可用响应之间的等待时间。尽管现有研究主要聚焦于加速解码过程本身，但用户实际体验的延迟不仅取决于解码速度，还受制于请求完全提交前无法启动计算的限制。传统部署方式在用户输入尚未完成时即停止计算，导致大量等待时间。本文提出并研究了“流式生成”（streaming generation）这一新型解码范式，其核心在于：在用户输入仍不完整时即开始生成响应，并随着后续输入的到达动态修正输出，使得请求提交时仅需完成少量剩余内容的计算。该方案的关键在于利用请求的渐进性（partial specification）提前启动生成过程，从而显著缩短用户可感知的延迟。实验表明，该机制在自回归与扩散语言模型上均有效，且其性能提升依赖于输入的可预测性与模型对部分输入的响应能力。
链接: https://arxiv.org/abs/2610.26296
作者: Nikita Dragunov,Matvey Mikhalchuk,Anton Razzhigaev
机构: FusionBrain Lab
类目: Computation and Language (cs.CL)
备注:
Abstract:Work on efficient large-language-model inference has concentrated almost entirely on reducing the cost of decoding once a request is complete. The quantity a user experiences, however, is the interval between submitting a request and receiving a usable response, and that interval can be shortened without making decoding itself faster. While a request is being composed it is already partially specified, yet conventional deployments begin no computation until submission. We propose and study streaming generation, a decoding regime in which the model begins producing a response from an incomplete request and revises it as the remaining input arrives, so that only a small residue of the answer remains to be computed once the request is submitted. We evaluate the regime on autoregressive and diffusion language models, characterise the conditions under which it reduces user-visible latency, and analyse the mechanism responsible.
[NLP-57] AudioScorer: Learning Question-Aware Token Relevance for Large Audio Language Models
【速读】： 该论文旨在解决大音频语言模型（Large Audio Language Models, LALMs）在处理长时序音频时因声学编码器生成密集音素序列而导致计算成本过高的问题。现有音频剪枝方法多依赖于启发式策略，如注意力权重、表征相似性或时间采样，但这些方法未能显式学习与特定问题相关的关键音频片段。本文提出AudioScorer，一个轻量级评分模型，能够根据用户提问对音频令牌进行重要性打分。其核心在于通过知识蒸馏方式，利用冻结的LALM在回答梯度显著性（answer-gradient saliency）下的音频令牌排序，并结合下游任务的答案监督信号进行训练。实验表明，在10%令牌保留率下，AudioScorer在10项语义与声学任务中的7项上取得了最佳或并列最佳性能，涵盖语音识别、情绪识别及音频场景分类等关键任务，验证了其在精准定位关键音频信息方面的有效性。
链接: https://arxiv.org/abs/2610.26258
作者: Marwan Farag,Bin Yang
机构: Institute of Signal Processing and System Theory (ISS); University of Stuttgart (斯图加特大学)
类目: ound (cs.SD); Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注:
Abstract:Large Audio Language Models (LALMs) reason over speech, environmental sounds, music, and other acoustic information. However, their audio encoders produce dense token sequences that make long-form audio costly to process. Existing audio pruning methods reduce this cost using heuristics such as attention, representation similarity, or temporal sampling, but do not explicitly learn which tokens are most useful for a given question. We propose AudioScorer, a lightweight model that scores audio tokens according to the user question. It is trained by distilling the ranking of audio tokens under the frozen LALM’s answer-gradient saliency, together with downstream answer supervision. Compared with other pruning methods, AudioScorer achieves the best or tied-best result at 10% token retention on 7 of 10 semantic and acoustic tasks, including speech recognition, emotion recognition, and audio-scene classification.
[NLP-58] Machina Pauliana: A neuro-symbolic approach to the actio pauliana in Dutch insolvency law
【速读】： 该论文旨在解决在破产法实践中应用《荷兰破产法》第42条（Actio Pauliana）时所面临的双重挑战：一方面需要基于结构化规则的逻辑推理，另一方面还需对判例法中发展出的开放性术语（open-textured terms）具备敏感性。其核心解决方案是构建一个神经符号模型（neuro-symbolic model），该模型以可读性强的Catala语言编写符号核心，确保法律条文与形式化表达的一致性，并通过结对编程开发；其中，开放性术语被建模为布尔输入，由神经组件与符号求解器协同处理。具体而言，神经模块将案件的自然语言事实转化为布尔值和时间戳输入，而神经子程序则通过案例导向的推理，在检索到的荷兰民事判决中识别相关先例，检索过程依据待解决要件进行过滤并沿诉讼程序链扩展。对100个手动标注案例的判断结果表明，该方法在段落分割与标注上实现了高精度与高召回率。在10个人工标注案例上的测试显示，该模型能合理处理多数判定标准，但在“知识要求”（knowledge requirement）方面仍存在困难。尽管检索深度未表现出显著的定量影响，但所获取的判例为模型提供了明确的既有判例法基础。
链接: https://arxiv.org/abs/2610.26241
作者: Daan Boven,Gregor Behnke,David Graus
机构: University of Amsterdam (阿姆斯特丹大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to JURIX 2026
Abstract:Applying a legal provision frequently requires both structured rule-based reasoning and sensitivity to open-textured terms developed in case law. The actio pauliana of Article 42 of the Dutch Bankruptcy Act (Faillissementswet, Fw), which allows an insolvency practitioner to avoid transactions that prejudice creditors, requires both. We present a neuro-symbolic model of Article 42 Fw. Its symbolic core is written in Catala in a literate style that keeps statute and formalisation aligned and was developed through pair programming, with open-textured terms represented as boolean inputs where neural components couple to the symbolic solver. A neural routine structures a case’s natural-language facts into boolean and date inputs. Neural subroutines resolve the open-textured terms through case-based reasoning over Dutch civil judgments, retrieved through a pipeline that filters on the requirement being resolved and expands along the procedural chain of a case. Segmenting and labelling those judgments attains high precision and recall against a manually annotated gold standard of 100 judgments. On ten hand-annotated cases the neuro-symbolic model handles most criteria reasonably but struggles with the knowledge requirement. Retrieval depth shows no clear quantitative effect, although retrieved judgments provide explicit grounding in prior case law.
[NLP-59] Shaping the Document Not the Model: Optimized Suffixes for Context Following
【速读】： 该论文旨在解决在用户提供的文档与语言模型自身参数化知识相矛盾时，如何有效引导模型遵循文档内容而非依赖其先验知识的问题。现有方法主要通过调整激活值、对比解码时的逻辑值或使用自然语言指令来实现，但这些方法通常需要在推理阶段对模型进行干预或在文档中嵌入特定指令。本文提出一种无需额外指令、不依赖推理时模型访问的解决方案：通过离散词元搜索优化文档后缀（document suffix），从单个问题的答案出发，逐步扩展目标至多跳问答、包含预设事实的段落集合以及对话中的后续轮次。实验表明，该后缀在对抗性段落（即与模型参数知识相悖的文本）上几乎完全扭转了模型的错误倾向，在三种指令微调模型家族中表现优异，优于基于解码的基线和人工撰写的最强指令，且效果接近激活值引导。控制实验显示，后缀的作用主要源于文档本身而非直接注入答案，且每个模型仅需一个通用后缀即可迁移至未见段落。此外，每项扩展目标均能将影响传递至对应任务，说明目标设计本身决定了干预效果的覆盖范围，而非仅由文档内容决定。
链接: https://arxiv.org/abs/2610.26236
作者: Gia-Bao Dinh Ho,Sekh Mainul Islam,Masahiro Kaneko,Timothy Baldwin,Isabelle Augenstein
机构: 未知
类目: Computation and Language (cs.CL)
备注: 19 pages, 5 figures, 18 tables. Project page: this https URL
Abstract:When a user supplies a document and wants the answer drawn from it, a language model is expected to follow it even when it contradicts its parametric knowledge. Prior work has focused on making the model rely on context by steering activations, contrasting logits during decoding, or instructing it in plain language. We ask whether the document alone, with no instruction written into it and no access to the model at inference time, is enough. We optimize a document suffix by discrete token search, starting from the answer to a single question and extending the objective to a set of passages, to multi-hop questions that follow from the planted fact, and to later turns of a conversation. On counterfactual passages that contradict parametric knowledge of the model, the suffix flips nearly all items the model resists across three instruction-tuned model families, in most settings outperforms the decoding baseline and the strongest instruction we could write, and matches activation steering. Controls show that the suffix largely acts through the passage rather than injecting the answer, and a single suffix per model transfers to unseen passages. Each extended objective carries the effect to its target, so the objective, not only the document, sets how far the effect reaches.
[NLP-60] Heartbeat Thinking: Speculative Reasoning and Tool Execution While Listening in Realtime Voice Agents
【速读】： 该论文旨在解决语音代理在使用工具时存在的响应延迟问题：尽管可操作信息可能在用户语音结束前已到达，但传统方法仍需等待语音输入完全结束才开始处理，导致计算资源未能充分利用。其核心挑战在于，现有推测性生成方法（如基于窗口的推测）在面对部分输入更新时，容易因中断或分块限制而截断工具调用参数，从而将计算时间与语音更新耦合，限制了提前准备的有效性。为此，论文提出“心跳式思考”（Heartbeat Thinking）机制，其关键在于通过周期性触发新的思考者（Thinker）实例，在不中断正在进行的推理和已调度工具的前提下，持续读取最新的部分语音转录文本及共享的已完成推理与工具执行结果。该机制能够动态补全缺失参数、触发依赖性工具调用，使准备工作在语音输入过程中持续推进。实验表明，在SpeechFCEval、FDBench v3和AudioCRAG等多个基准上，该方法显著提升了推测性工具覆盖率或检索复用率；尤其在FDBench v3上，工具覆盖率提升16个百分点，平均最终响应延迟降低36.4%。结果证明，通过在语音输入期间保留并延续计算过程，“心跳式思考”可有效将听觉等待时间转化为可复用的前置准备，从而实现更快速、更强大的语音代理响应能力。
链接: https://arxiv.org/abs/2610.26233
作者: Zhenyu Zhang,Kewei Fu,Yongqi Tong,Chenzhi Zheng,Wei Wu,Jiang-Ming Yang,Jianshe Li
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Tool-using voice agents often leave users waiting after speech ends, even when actionable information arrives earlier in the request. Speculative approaches begin work during speech, but chunk-bounded or interrupt-on-update generation can truncate a tool call before its arguments are complete. This couples computation time to speech updates, limiting useful work completed in advance. We introduce Heartbeat Thinking, which handles updated input through new Thinker invocations while allowing ongoing invocations and dispatched tools to finish. Periodic invocations read the latest partial transcript and a shared record of completed reasoning and tool results. These results help resolve missing arguments and initiate dependent calls, allowing preparation to progress across speech updates. Across SpeechFCEval, FDBench v3, and AudioCRAG, Heartbeat improves aggregate speculative tool coverage or retrieval reuse over a SHANKS-style windowed baseline, with benchmark-dependent effects on final task quality. On FDBench v3, it increases tool coverage by 16 percentage points and reduces mean final-response latency by 36.4%. The results show that heartbeat thinking paradigm preserving ongoing computation during speech can turn listening time into reusable preparation for faster and more capable voice-agent responses.
[NLP-61] Rethinking BPE: Merge Selection as Model Selection
【速读】： 该论文旨在解决现代语言模型中子词分词（subword tokenization）方法（如字节对编码，BPE）的词表容量确定问题，传统上该容量由启发式选择的固定合并预算决定，导致词表过大时增加内存占用与计算开销。其核心解决方案在于将BPE合并预算的优化建模为统计模型选择问题，并基于信息论准则直接从语料库统计特征中计算最优合并集合，从而实现更紧凑且高效的词表生成。为进一步提升低资源场景下的分词效果，研究引入合成文本增强原始训练语料，以更准确估计一元组和二元组分布，进而优化合并策略。实验表明，采用这一有理论依据的词表选择方法可在显著减小模型规模的同时保持性能稳定，且在低资源条件下合成数据能进一步提升分词器质量。
链接: https://arxiv.org/abs/2610.26225
作者: Vijini Liyanage,François Yvon
机构: Sorbonne-Université, CNRS, ISIR, Paris(索邦大学，法国国家科学研究中心，智能机器人研究所，巴黎)
类目: Computation and Language (cs.CL)
备注:
Abstract:Subword tokenization methods such as Byte-Pair Encoding (BPE) are a core component of modern language models; yet, tokenizer capacity remains typically determined by a fixed merge budget, chosen heuristically. Since larger vocabularies increase memory usage and computational cost, selecting compact and effective token inventories is essential for efficient language modeling. We study the optimization of a BPE merge budget as a statistical model selection problem and investigate information-theoretic criteria for \emphcomputing optimal merge sets directly from corpus statistics. To further improve the merge selection in limited data conditions, we augment the BPE training corpora with synthetic texts generated from the original data, providing estimates for the unigram and bigram distributions. Experiments across multiple corpora and downstream tasks show that using a principled vocabulary selection technique reduces model size, with little degradation in performance. We further observe that synthetic data can yield better tokenizers, especially in low-resource settings.
[NLP-62] VEILED-SH: Vulnerability Expression through Indirect Language Evaluation Dataset (Self-Harm)
【速读】： 该论文旨在解决当前人工智能（AI）系统在识别间接、文化情境化的情绪困扰表达时存在的局限性，以及在不同应用场景下提供不匹配支持的问题。现有评估基准未能协同考量文化差异、应用背景及专业实践者之间的意见分歧。为此，研究提出VEILED-SH基准及其构建方法，用于评估针对自杀、自伤及心理困扰的间接表达所生成的单轮回应。该基准包含10,003个源自新加坡本地语境的提示，涵盖咨询聊天、论坛发帖、平台搜索和初步筛查等多种场景，数据来源包括本地化资源与公开数据集。通过专家验证的多标签分类体系，区分意图、交际功能与回应行为。引入六种经独立校准的自动化评估代理模型，分别对应三名朋辈支持者与三名临床从业者，保留各自的参考策略以实现代理对齐评分。在覆盖范围增强的样本上对五种系统进行评估，发现即使对于同一系统输出，其代理对齐度在不同参考策略下仍存在显著差异，凸显了多维度评估的重要性。解决方案的关键在于构建一个融合文化敏感性、应用上下文多样性及专业判断异质性的综合性评估框架。
链接: https://arxiv.org/abs/2610.26190
作者: Leanne Tan,Keith Chua,Shaun Khoo,Onno P. Kampman,Lim Ruling Charmaine,Hazirah Hoosainsah,Kellie Yu Hui Sim,Kai He,Kenny Tsu Wei Choo,Roy Ka-Wei Lee
机构: GovTech Singapore(政府科技局); MOH Office for Healthcare Transformation(卫生部医疗转型办公室); Singapore University of Technology and Design(新加坡科技设计大学); National University of Singapore(新加坡国立大学); Nanyang Technological University(南洋理工大学); University of British Columbia(不列颠哥伦比亚大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: Content warning: this paper discusses suicide, self-harm, and distressing language
Abstract:AI systems may miss indirect, culturally situated distress or provide support mismatched to the application context. Existing benchmarks do not jointly address cultural variation, application context, and practitioner disagreement. We introduce VEILED-SH, a benchmark and construction methodology for evaluating single-turn responses to indirect expressions of suicide, self-harm, and distress. Its 10,003 prompts span counselor chat, forum posts, platform search, and triage, drawing on Singapore-grounded sources and public datasets. An expert-validated multi-label taxonomy distinguishes intent, communicative function, and response behavior. Six automated profiles, individually calibrated to three peer-support practitioners and three clinicians, retain separate reference policies for proxy-alignment scoring. Evaluating five systems on a coverage-enriched sample reveals variation in proxy alignment across profile-generated reference policies, even for the same system responses.
[NLP-63] Reading Time Modeling with Pixel-Based Language Models EMNLP2026
【速读】： 该论文旨在解决传统基于词元（token-based）语言模型（TLMs）在自然阅读建模中因分词过程引入的偏差问题，尤其关注分词对信息论指标（如预期意外度，surprisal）估计的干扰，以及其将底端视觉字形特征排除在模型之外的局限性。其核心解决方案是采用无分词、基于像素的语言模型（PLMs），直接在像素空间中进行建模，从而避免分词带来的先验偏见，并有效整合字形结构与周边视区（parafoveal）的视觉信息。研究表明，基于像素空间计算的下一个词预测难度在跨语言阅读行为（如注视时长和跳读现象）上具有强预测力，且其优势与现有词元级预期意外度和熵估计呈互补关系，进一步揭示了底端视觉信息如何逐步融入基于预期的阅读过程，为计算（心理）语言学研究提供了新的范式。
链接: https://arxiv.org/abs/2610.26181
作者: Junlin Li,Sara Alhajeri,Tatsuki Kuribayashi
机构: Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学); Tohoku University(东北大学)
类目: Computation and Language (cs.CL)
备注: accepted by EMNLP 2026 (Main)
Abstract:Tokenization-free, pixel-based language models (PLMs) have recently been proposed as an alternative to conventional token-based language models (TLMs), typically motivated by practical considerations, e.g., the flexibility to support new languages or vocabularies. Nevertheless, PLMs also have the potential to serve as infrastructure for computational (psycho)linguistics, where tokenization has been a source of bias in estimating information-theoretic values, such as surprisal, and has pushed the bottom-up visual orthographic features outside the scope. In this study, we first integrate PLMs into the modeling of naturalistic human reading and, in particular, examine their use to estimate next-word predictability, which can also integrate orthographic and parafoveal information. Our results demonstrate that next-word prediction difficulty, computed in pixel space, has strong predictive power for gaze duration and word skipping behavior across various languages. In particular, their advantage is complementary to existing token-level surprisal and entropy estimates. Cross-lingual analyses with script variations also clarify their distinctive visual processing biases, facilitating research into how bottom-up visual information is incrementally integrated into expectation-based reading.
[NLP-64] VISAGENT : An Agent -Based End-to-End Voice Interaction System
【速读】： 该论文旨在解决当前基于语音的公共服务系统中存在的刚性菜单式交互、垂直服务碎片化以及可扩展性不足等问题。其核心解决方案在于提出VISAGENT——一个面向大规模公共事务的基于智能体（Agent）的端到端语音交互系统，通过整合端到端语音交互、分层智能体路由、可扩展的语音智能体市场及用户记忆支持机制，实现自然语言对话替代传统菜单导航。系统采用两级路由策略，在数百个智能体中实现亚秒级意图识别与智能体分配，显著提升响应效率；同时利用用户记忆实现偏好感知的智能体选择，并为选定智能体提供上下文信息，支持个性化任务执行并减少重复输入。此外，引入语音智能体注册机制，将任务型智能体转化为可对话的实体，实现动态接入与生态级可扩展性。在真实世界部署于中国114热线多服务场景的实验验证表明，该系统具备低延迟全双工交互能力、强路由可扩展性及媲美主流模型的语音自然度。本研究推动了从孤立语音助手向以智能体为中心的对话基础设施范式转变，为公共事务及其他领域构建可扩展的语音智能体系统提供了通用架构蓝图。
链接: https://arxiv.org/abs/2610.26168
作者: Xiang Wang,Zhaoxiang Liu,Xiang Chen,Liqiang He,Kai Wang,Shiguo Lian
机构: China Unicom(中国联通); Unicom Data Intelligence(联通数据智能)
类目: Computation and Language (cs.CL)
备注:
Abstract:Voice-based public services have evolved from manual switchboards to intelligent platforms to meet modern user needs. However, current systems remain constrained by rigid menu-driven interaction, fragmented vertical services, and limited extensibility. To address these issues, we present VISAGENT, an agent-based end-to-end voice interaction system for large-scale public services that integrates end-to-end speech interaction, hierarchical agent routing, a scalable voice-agent marketplace, and user memory support. The system replaces menu-driven workflows with natural speech and dynamically dispatches user requests to either a general conversational model or specialized domain agents via a two-level routing paradigm that achieves sub-second intent-to-agent selection over hundreds of agents. User memory informs preference-aware agent selection and supplies relevant preferences and context to selected agents, supporting personalized task execution and reducing the need for users to repeat previously provided information. A voice-agent registration mechanism converts task agents into conversational entities, enabling dynamic onboarding and ecosystem-level extensibility. We validate VISAGENT through a real-world deployment in China’s 114 hotline across multiple service domains. Experimental results demonstrate low-latency full-duplex interaction, robust routing scalability, and competitive speech naturalness. More broadly, this work advocates a shift from isolated voice assistants toward agent-centric conversational infrastructures, providing a general blueprint for building extensible voice-agent systems in public services and beyond.
[NLP-65] DeepSEA: A Multilingual Evaluation Benchmark for Southeast Asia
【速读】： 该论文旨在解决当前大语言模型在高资源语言上表现优异，而在东南亚地区众多低资源语言中存在严重性能偏倚的问题。东南亚地区拥有超过7亿人口和高度的语言多样性，但现有评测基准普遍覆盖语言有限、任务范围狭窄，难以全面评估模型在多语言环境下的表现。为此，研究提出DeepSEA——一个涵盖12种东南亚语言（包括印尼语、泰语、越南语等高资源语言及缅甸语、高棉语、老挝语等低资源语言）的综合性多任务基准，覆盖数学推理、科学推理、问答、摘要生成和机器翻译五大任务。其核心解决方案在于通过专业人工翻译构建高质量、跨语言对齐的数据集，以实现更公平的模型评估。关键发现表明：尽管前沿闭源模型（如Gemini-3.7-Flash）在不同资源层级下仍保持较强鲁棒性，但小型模型在低资源语言上性能显著下降；更重要的是，模型在推理与摘要任务中通过英语作为中介语言时表现优于直接使用目标语言，揭示了对英语表征的高度依赖，尤其在低资源语言场景下更为明显。此外，研究还开展了跨语言错误分析与跨书写系统分词器适应性分析，为下一代包容性多语言模型的发展提供了重要启示。
链接: https://arxiv.org/abs/2610.26155
作者: Jiuzhou Han,Abhishek Bijay Mishra,Trevor Cohn,Shubham Mittal
机构: Google Research(谷歌研究院)
类目: Computation and Language (cs.CL)
备注: For the codebase and dataset, see this https URL
Abstract:While language models have demonstrated remarkable capabilities, their performance remains disproportionately skewed towards high-resource languages. This disparity is particularly acute for the 700 million inhabitants of Southeast Asia, where vast linguistic diversity remains under-represented in existing benchmarks, often suffering from limited language coverage or narrow task scopes. To address this, we introduce DeepSEA, a comprehensive benchmark constructed from established datasets through professional human translation into 12 Southeast Asian languages, spanning both high-resource (e.g., Indonesian, Thai, Vietnamese) and low-resource (e.g., Burmese, Khmer, Lao) tiers across five distinct tasks: mathematical reasoning, scientific reasoning, question answering, abstractive summarization, and machine translation. Extensive evaluation of frontier proprietary and open-weights models reveals that while frontier models like Gemini-3.7-Flash maintain robustness across resource tiers, smaller models face severe performance degradation on low-resource languages. Crucially, models perform better in reasoning and summarization when pivoted through English rather than target languages, exposing a critical dependency on English representations, especially for low-resource languages. Beyond these observations, we provide a cross-lingual error analysis and a cross-script tokenizer fertility analysis. We hope these findings shed light on the development of next-generation inclusive multilingual models.
[NLP-66] Honeyed Words Hidden Daggers: Benchmarking and Training LLM s to See Through Mock Politeness
【速读】： 该论文旨在解决生成式 AI 在理解复杂语用现象（如“口蜜腹剑”——表面温和实则隐含敌意的言语策略）方面的关键挑战，尤其聚焦于话语意图、情感与语言表达之间的不一致以及听者反应的动态建模。其解决方案的核心在于构建并利用一个高质量、标注精细的基准测试集 ZhenhuanBench（包含1,592个样本），涵盖说话者意图、情绪-语言错位及听者接受度等多维度语用特征，并引入监督轨迹（supervised traces）和可字符串验证的训练奖励信号。通过提出 CPR-GRPO 优化循环（Compress-Probe-Replenish），该方法基于对可恢复性能差距的估计，动态加权高质量监督信号，在无需目标任务微调的情况下，显著提升了模型在中英文语用与反讽识别任务中的表现，优于基础模型（Base）、监督微调（SFT）和标准强化学习（GRPO）等多种基线方法。
链接: https://arxiv.org/abs/2610.26059
作者: Caixin Kang,Sitong Gong,Tianyu Yan,Kaipeng Zhang,Yoichi Sato,Yifei Huang
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:The Chinese idiom “kou mi fu jian” (“honeyed words, hidden daggers”) names speech that is polite in form but hostile in function. We present ZhenhuanBench, a 1,592-item benchmark built from expert commentary on Empresses in the Palace, covering speaker intent, emotion-language mismatch, and per-listener uptake. Text-only evaluation covers 34 LLMs. GPT-6 Astra leads the leaderboard at 73.48% overall, while its listener-reaction score is 58.37%. The annotations also supply supervised traces and string-verifiable training rewards. With 338 gold traces, an 8B model outperforms teacher distillation on identical items. Subsequent RL saturates, and chance correction reduces the estimated recoverable performance gap by a factor of 7.5. Our CPR-GRPO loop (Compress-Probe-Replenish) uses this estimate to weight gold supervision, reaching 67.0 on public multiple-choice tasks and 74.5 under the training-matched hidden protocol. Without target-task training, CPR outperforms Base, SFT, and GRPO across six Chinese and English pragmatic and sarcasm evaluations. On three external datasets, training with CPR yields higher mean Macro-F1 than SFT and GRPO.
[NLP-67] Balance-Aware Pivot Selection for Parallel Diffusion LLM Decoding
【速读】： 该论文旨在解决生成式语言模型在并行解码过程中因依赖预测置信度进行令牌选择而导致的解码效率与生成质量失衡问题。现有方法虽通过置信度筛选实现并行提交，但其忽略令牌位置对剩余解码调度的影响，且未充分建模令牌间的依赖关系，导致解码路径不够优化。为此，论文提出平衡感知的枢轴解码（Balance-Aware Pivot Decoding, BPD），其核心在于优先选择连续掩码区间内的中心枢轴位置进行并行解码，从而实现与区间长度呈对数关系的解码深度。该方法通过结合置信度驱动的预填充与基于阈值的中心选择策略，有效融合模型不确定性处理机制，在保持生成质量的同时显著提升解码效率。实验在LLaDA和Dream模型上针对五项数学推理、代码生成及指令遵循任务验证了该方案在质量-效率权衡上的优越性。
链接: https://arxiv.org/abs/2610.26040
作者: Yejin Lee,Hyundong Jin,Yo-Sub Han
机构: Yonsei University (延世大学)
类目: Computation and Language (cs.CL)
备注: 18 pages, 4 figures
Abstract:Diffusion language models enable parallel token generation, offering a promising alternative to sequential autoregressive decoding. Existing acceleration methods use predictive confidence to enable parallel token commitment while preserving generation quality. However, confidence-based selection prioritizes reliable token predictions without explicitly considering how token positions shape the remaining decoding schedule. Moreover, such ordering does not always reflect inter-token dependencies, suggesting that confidence need not be the sole criterion for token selection. Motivated by this limitation, we propose Balance-Aware Pivot Decoding (BPD), which favors central pivots within contiguous masked intervals and decodes the selected positions in parallel. Balanced pivots yield logarithmic decoding depth in the interval length. Our practical method accounts for model uncertainty by combining confidence-based prefilling with confidence-thresholded center selection. We evaluate the resulting quality-efficiency tradeoff on LLaDA and Dream across five mathematical reasoning, code generation, and instruction-following benchmarks. Our implementation is available at: this https URL
[NLP-68] Understanding When Reflective Prompt and Skill Optimization Helps on Locally Served Reasoning Models: A Two-Model Study of GEPA and SkillOpt
【速读】： 该论文旨在解决生成式 AI（Generative AI）在实际应用中，基于文本空间优化器（如GEPA和SkillOpt）所表现出的性能提升不可靠、泛化性差的问题。研究发现，尽管这些优化方法在特定验证集上表现优异，但在未见任务（held-out tasks）上可能出现性能退化，且其效果高度依赖于具体任务与模型配置。解决方案的关键在于：通过分解优化效果为内容/推理（content/reasoning）、答案实现/评估器对齐（answer-realization/evaluator alignment）以及服务完成度/可靠性（serving-completion/reliability）三个维度，揭示优化成效的本质驱动因素。研究进一步表明，低歧义、可操作的评估信号有助于提升优化效果，而单纯依赖验证集上的性能增益可能误导模型部署决策。因此，真正有效的优化策略应建立在对模型-任务对的文本冗余空间（textual headroom）与评估机制对齐性的深入理解之上。
链接: https://arxiv.org/abs/2610.26020
作者: Andrea Giummarra(1),Carmelo Baglieri(1),Antonio Di Maria(1),Andrea Farruggio(1),Michele Piatto(1),Giovanni Micale(2)
机构: Evision srl(艾维森公司); University of Catania(卡塔尼亚大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 36 pages, 10 figures, 11 tables
Abstract:Reflective text-space optimizers such as GEPA and SkillOpt have reported large gains on frontier and mid-size models. We test replication across two reasoning-native target-serving configurations: DeepSeek V4 Flash through DwarfStar and Qwen3.8-27B through LM Studio. For each target, we evaluate the same held-out tasks and four conditions - minimal baseline, GEPA prompt, SkillOpt skill, and manually engineered prompt - on HotpotQA (n=500 per condition) and MATH-500 (n=250). In the primary Qwen v2 rerun under the same frozen protocol, GEPA selects the HotpotQA seed and exactly reproduces baseline behavior, whereas SkillOpt’s validation gain reverses into held-out degradation. On MATH-500, GEPA raises exact match from 0.828 to 0.904 while reducing latency but increasing token consumption, so lower response time does not imply lower token cost. DeepSeek likewise shows its strongest GEPA effect on MATH-500 (0.852 to 0.908). Across both targets, GEPA’s strongest gains occur on MATH-500, while neither optimizer improves the HotpotQA baseline. Optimization is therefore task- and method-dependent, and validation gains may reverse on held-out evaluation. An earlier Qwen run is retained as serving-sensitivity evidence. These results suggest that textual headroom for a model-task pair and the observed effect of a text-space optimizer are more informative when decomposed into content/reasoning, answer-realization/evaluator alignment, and serving-completion/reliability. The cross-benchmark pattern further suggests that optimization may benefit from low-ambiguity, actionable evaluation signals.
[NLP-69] Accelerated Drift of Scientific Prose in the LLM Era: Evidence from an Interrupted Time-Series Analysis EMNLP2026
【速读】： 该论文旨在解决生成式人工智能（Generative AI）工具，特别是对话式大语言模型（LLM）在科学预印本写作中广泛应用背景下，其对学术写作风格演变影响的系统性理解不足的问题。研究聚焦于2022年11月ChatGPT公开发布这一关键时间点，考察在此前后科学预印本文本在词汇、标点、可读性及句法结构等维度上的语言轨迹变化。其解决方案的关键在于采用中断时间序列回归（interrupted time-series regression）方法，对arXiv与medRxiv中人工智能、统计学和定量生物学领域预印本摘要的语言特征进行多维度、跨领域的纵向分析，揭示了自2022年后语言演化在速率、方向与结构层面的显著变化：如平均词长增加、基于公式的可读性下降，以及句法依赖树深度与句级复杂度变异性的降低。研究进一步指出，单一语言标记频率的变化仅反映部分现象，而语言轨迹在动态演化中的速率、方向及结构性转变提供了更全面的证据，强调应将科学写作视为一个多维、持续演化的系统，而非依赖静态的“大语言模型关联标记”进行评估。
链接: https://arxiv.org/abs/2610.26008
作者: Hanna Hubarava,Benjamin V. Ineichen
机构: University of Bern(伯尔尼大学), Switzerland(瑞士)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted at EMNLP 2026. 22 pages, including references and appendices
Abstract:Every scientific text is embedded in the writing norms of its time: it reflects earlier publications while also shaping the texts that follow. The growing use of LLM-based writing tools introduces a new factor in this evolving system, but the linguistic changes coinciding with their emergence remain insufficiently understood. We investigate whether the trajectories of scientific-preprint writing changed in rate or direction after conversational LLMs became publicly accessible, using the release of ChatGPT in November 2022 as the observational breakpoint. We compare pre- and post-2022 trajectories across vocabulary, punctuation, readability, and syntax in abstracts from arXiv’s artificial intelligence, statistics, and quantitative biology categories and from medRxiv. Using interrupted time-series regression, we find substantial post-2022 changes across multiple feature families and research domains, including increases in average word length, reduced formula-based readability, and declines in maximum dependency-tree depth and sentence-level depth variation. Individual features follow different dynamics: some lexical markers accelerate persistently, while others rise and then recede. Our findings show that isolated marker frequencies capture only part of the shift: changes in the rate, direction, and structure of linguistic trajectories provide complementary evidence of broader post-2022 change in scientific-preprint writing. More broadly, these findings motivate monitoring scientific writing as a multidimensional, evolving system rather than through a fixed set of LLM-associated markers.
[NLP-70] Scaling Synthetic Pre-training Data and Evaluation for Indic LLM s
【速读】： 该论文旨在解决印度多语言大模型（Indic language models）在训练数据规模与评估体系方面的关键瓶颈问题。当前，高质量、大规模的印地语及印度其他22种官方语言的预训练语料库严重匮乏，且缺乏统一、全面的多任务评估基准，制约了模型性能的提升与可比性分析。为此，研究提出Nemotron-Pretraining-Indic-v1，一个约2.83万亿词元的合成预训练语料库，覆盖全部22种印度官方语言的原生文字形式，并包含12种语言的罗马化形式，是目前公开发布的最大规模印地语预训练语料库。其解决方案的关键在于构建一套基于GPU加速的端到端合成数据生成管道：通过精选英文文档进行翻译，结合自动质量评估与人工干预修复低质输出，在保持技术内容准确性与文档结构完整性的前提下实现源文本可追溯性。实验表明，将该合成语料与真实印地语文本联合微调，显著提升了模型的知识掌握与推理能力。此外，研究引入Indic-LLM-Eval-Suite，一个涵盖对话、指令遵循、推理、知识、代码生成与翻译等多维度的多语言评估套件，融合现有印地语基准与经过任务感知翻译、自动化质量检测及母语者人工校验的英语基准，形成“人机协同”的评估流程。通过该套件对主流开源印地语与多语言模型的全面评测，揭示了不同任务与语言间性能的显著差异。综上，该工作为开发与评估更强大的印地语语言模型提供了可扩展的数据与评估基础设施。
链接: https://arxiv.org/abs/2610.26006
作者: Abhinav Khattar(1),Aishwarya Chhabra(1),Ameya Velingker(1),Anusha Kamath(1),Brandon Norick(1),Kanishk Singla(1),Kiran Praveen(1),Markus Kliegl(1),Mohammad Qazim Bhat(1),Mostofa Patwary(1),Niket Agarwal(1),Rajpreet Thethy(1),Ramani Akkati(1),Raviraj Joshi(1),Sai Nishanth Padala(1),Shuoyang Ding(1),Sriharsha Niverty(1),Utkarsh Vaidya(1),Vitaly Lavrukhin(1),Yev Meyer(1),Anoop Kunchukuttan(2),Mitesh Khapra(2),Mohammed Safi Ur Rahman Khan(2),Rudra Murthy(2),Sumanth Doddapaneni(2),Thanmay Jayakumar(2),Ujjwal Sharma(2) ((1) NVIDIA, (2) AI4Bharat)
机构: NVIDIA(英伟达); AI4Bharat
类目: Computation and Language (cs.CL)
备注: 44 pages, 5 figures. Includes appendices and references
Abstract:Building capable language models for India requires scalable multilingual data and robust evaluation. We introduce Nemotron-Pretraining-Indic-v1, a synthetic pretraining corpus of approximately 2.83 trillion tokens covering all 22 scheduled Indian languages in native script and twelve in romanized form. It is the largest openly released Indic pretraining corpus. Using our GPU-accelerated pipeline, we translate curated English documents, assess translation quality, and repair low-quality outputs while preserving technical content and document structure and maintaining source traceability. Through continued-pretraining experiments across twelve languages, we show that combining our synthetic corpus with natural Indic text improves knowledge and reasoning over a public-corpus baseline. Additional experiments show how data volume, script coverage, and Indic training share affect English performance. We also introduce the Indic-LLM-Eval-Suite, a multilingual benchmark suite for evaluating Indic language models across chat, instruction following, reasoning, knowledge, code, and translation. The suite combines existing Indic benchmarks with task-aware translations of English benchmarks, using automated quality assessment, targeted correction, and native-speaker review in a human-in-the-loop pipeline. We use the suite to comprehensively benchmark contemporary open-weight Indic and multilingual models across capabilities and languages, revealing substantial variation across tasks and languages. Together, these resources provide a scalable foundation for developing and evaluating more capable Indic language models.
[NLP-71] mpoMoE: Identifying and Steering Fast-Slow Thinking Experts for Efficient Multimodal Reasoning
【速读】： 该论文旨在解决多模态混合专家（Multimodal Mixture-of-Experts, MoE）推理模型在处理复杂任务时依赖长链式思维（Chain-of-Thought, CoT）所导致的“过度思考”问题，即对简单输入仍进行冗长推理，从而增加令牌消耗与延迟，并可能降低准确性。现有方法通过训练阶段优化推理长度或推理时引入外部控制来缓解此问题，但均未能揭示推理机制的本质。研究发现，快速思维与慢速思维在专家路由模式上存在显著差异：固定输入仅改变提示的推理节奏（tempo），即可引发稳定的路由变化，且抑制慢速思维专家会缩短推理链，而抑制快速思维专家则会延长推理链，为两类专家对推理深度具有因果影响提供了实证依据。基于此，作者提出无需训练的两阶段框架TempoMoE，其关键在于通过节奏对比路由分析识别出快速与慢速思维专家，并根据样本难度在原始top-k专家集合内重新分配路由权重：对简单输入增强快速思维专家、抑制慢速思维专家，对复杂输入则反向调整，实现推理深度的自适应控制。在Kimi-VL、Qwen3-VL和ERNIE-4.5-VL三个模型上六项基准测试中，相较于基线模型（Vanilla），TempoMoE在保持或提升准确率的同时显著减少输出令牌数，最高准确率提升达2.6点，令牌减少高达35.4%。
链接: https://arxiv.org/abs/2610.25999
作者: Xiyue Zhang,Keru Zhou,Ziyi Wang,Bing Hu,Li Wang
机构: Zhejiang University(浙江大学); Tsinghua University(清华大学); Wenzhou Business College(温州商学院)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 11 pages, 3 figures
Abstract:Multimodal Mixture-of-Experts (MoE) reasoning models rely on long chain-of-thought (CoT) for complex tasks, yet often overthink easy inputs, increasing token cost and latency, and sometimes degrading accuracy. Existing methods optimize reasoning length during training or impose external control at inference, but leave the underlying reasoning mechanism a black box. We find that fast and slow thinking exhibit distinct expert-routing patterns: holding the input fixed and varying only the prompted reasoning tempo produces stable routing shifts. Suppressing slow-thinking experts shortens the reasoning chain, whereas suppressing fast-thinking experts lengthens it, providing causal evidence that these experts influence reasoning depth. Based on this finding, we propose TempoMoE, a training-free, two-stage framework. It identifies fast- and slow-thinking experts via tempo-contrastive routing analysis and redistributes routing weights within the original top-k set according to sample difficulty. It amplifies fast-thinking experts and suppresses slow-thinking experts on easy inputs and reverses this adjustment on hard ones, enabling difficulty-adaptive control over reasoning depth. We evaluate TempoMoE on Kimi-VL, Qwen3-VL, and ERNIE-4.5-VL across six benchmarks. Compared with Vanilla, TempoMoE consistently improves accuracy while reducing output tokens, with accuracy gains of up to 2.6 points and token reductions of up to 35.4%.
[NLP-72] Correctness-Dependent Asymmetric Target Distributions for Self-Distillation Policy Optimization
【速读】： 该论文旨在解决自蒸馏策略优化（self-distillation policy optimization, SDPO）在大语言模型（LLM）后训练过程中存在的关键问题：直接对齐自教师（self-teacher）生成的输出，会在正确轨迹上破坏已有的成功行为，而在错误轨迹上又提供不足的修正。为此，论文提出了一种基于自蒸馏的非对称策略优化方法（Asymmetric Policy Optimization via Self-Distillation, APSD），其核心创新在于构建依赖于输出正确性的目标分布——通过插值（interpolation）保留正确轨迹上的有效行为，同时利用外推（extrapolation）超越自教师分布以增强对错误轨迹的纠正能力。理论分析结合散度分解、反向KL巴氏几何与梯度分析，揭示了通过学生依赖的目标引导梯度流动的重要性。实验结果表明，在可验证奖励与丰富反馈设置下，APSD在性能上显著优于传统强化学习与自蒸馏基线方法，并在不同规模的Qwen3模型上平均仅需2.6倍更少的训练步数即可达到与SDPO相当的性能水平。
链接: https://arxiv.org/abs/2610.25973
作者: Suwan Yoon,Seongryong Jung,Changhee Lee
机构: Korea University (韩国大学); Chung-Ang University (中央大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Preprint
Abstract:Self-distillation policy optimization provides dense token-level supervision for LLM post-training through feedback-conditioned self-teachers. However, directly matching the self-teacher can disrupt successful behaviors on correct rollouts while providing insufficient correction on incorrect ones. We propose Asymmetric Policy Optimization via Self-Distillation (APSD), which constructs correctness-dependent targets by mixing student and self-teacher distributions: interpolation preserves successful behaviors on correct rollouts, while extrapolation beyond the self-teacher strengthens correction on incorrect ones. We analyze APSD through divergence decomposition, reverse-KL barycentric geometry, and gradient analysis, highlighting the importance of gradient flow through student-dependent targets. Experiments under reinforcement learning with verifiable rewards and rich feedback show that APSD achieves strong overall performance against reinforcement-learning and self-distillation baselines and matches SDPO performance with 2.6x fewer training steps on average across Qwen3 model scales.
[NLP-73] GLOBE: Geometry-Led Online Behavioral Early Exit via Unit-Sphere Trajectories
【速读】： 该论文旨在解决生成式AI在链式思维（Chain-of-Thought, CoT）推理过程中存在冗余生成的问题，即模型在已得出正确答案后仍继续生成额外步骤，导致推理成本上升且可能引入错误中间结果。其核心解决方案是提出一种无需训练、在线运行的早停机制GLOBE（Geometry-Led Online Behavioral Early Exit）。GLOBE的关键在于将每个完成的推理片段嵌入单位球面作为点，通过分析语义轨迹的几何形状来判断是否应终止生成：首先通过试运行阶段在独立分支中生成简短候选答案，并基于置信度阈值决定是否接受；若未终止，则进入轨迹阶段，当推理路径反复反向且净进展微弱时触发早停。实验表明，在五个基准任务（包括数学推理、工具调用与智能体记忆）上，GLOBE可平均减少49.7%的输出标记数和51.0%的端到端运行时间，同时保持80.58%的平均准确率，接近完整链式思维（Full CoT）的表现。
链接: https://arxiv.org/abs/2610.25970
作者: Xinyue Peng,Zhenlin Liang,Meng Cai,Hua Wang
机构: Moore Threads
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Chain-of-thought (CoT) reasoning can improve problem solving, but models often keep generating after the correct answer is already reachable; the extra steps raise serving costs and may replace a correct intermediate result with an incorrect one. We propose GLOBE (Geometry-Led Online Behavioral Early Exit), an online, training-free algorithm that terminates the model’s CoT process early. GLOBE embeds each completed reasoning segment as a point on a unit sphere and reads the shape of the resulting semantic trajectory. GLOBE first applies a trial phase that generates short trial answers in a separate branch, stopping the main stream on an accepted trial after checking confidence against a threshold. If generation continues, a trajectory phase stops the stream once this path repeatedly reverses direction with little net progress. On five benchmarks including mathematical reasoning, tool calling, and agent memory, GLOBE cuts output tokens by 49.7% and active end-to-end wall time (E2E) by 51.0% across two models, and achieves a five-task average of 80.58%, comparable to Full CoT.
[NLP-74] oken-Set Choice Confounds POPE: A Systematic Audit of Yes/No Extraction in VLM Hallucination Evaluation NEURIPS2026
【速读】： 该论文旨在解决生成式 AI（Generative AI）在视觉语言模型（VLM）评估中因读出机制（readout）设计不当而导致的幻觉缓解结果误判问题。其核心问题是：现有研究在POPE基准上报告的对抗性样本性能提升，可能并非源于模型本身能力的增强，而是由于评估过程中对输出标签的解码方式存在系统性偏差。解决方案的关键在于识别并纠正这一读出偏差——具体而言，原评估采用基于词汇表索引（ID 3582 和 1217）的二元判断逻辑，但实际模型输出的是带有前导空格的表面形式（如“ Yes”和“ No”，对应token 3869和1939）。当使用覆盖空格前缀的八词查找表进行正确读出时，模型在对抗性、流行性和随机分拆上的F1得分分别为0.8221、0.8498和0.8713，而原始错误读出导致的对抗性分数偏移高达6.13点，超过了六篇被调研论文中五篇所报告的增益。进一步分析表明，对抗性样本的错误预测反而获得更强的图像注意力（Cohen’s d = -0.46, p = 1.4×10⁻²⁴），与基于对齐的纠错方法假设相反。跨四种不同分词器家族的多模型审计证实，该混淆为协议层面而非模型特异性问题，仅迁移原模型的词汇索引会导致新模型性能骤降至0.00，而基于分词器映射的读出策略则维持在0.80至0.92之间。因此，该研究强调：公平比较必须统一读出机制与量化精度，否则无法真实反映模型改进效果。
链接: https://arxiv.org/abs/2610.25964
作者: Kesav Kumar Jayakumar,Karthigeyan Thilak
机构: Sri Krishna College of Technology (斯里克里希纳技术学院); Sri Eshwar College of Engineering (斯里埃什瓦工程学院)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: Accepted at the VLM4RWD workshop, NeurIPS 2026 (poster). 21 pages, 10 figures. Code: this https URL
Abstract:We set out to reproduce a recent hallucination-mitigation result on the POPE benchmark. The LLaVA-1.5-7B baseline we measured, 82.21 F1 on the adversarial split, sat 1.21 to 8.67 points above the adversarial baselines we planned to compare against (81.00, 79.30, and 73.54). We looked at the evaluation code and found a readout choice that moves F1 on its own. A two-token logit readout decides yes versus no by comparing two vocabulary IDs, 3582 (“yes”) and 1217 (“no”). Across 9,000 greedy-decode questions spanning all three splits, neither ID is generated even once. The model emits " Yes" (token 3869) and " No" (token 1939), because SentencePiece prepends a space byte after the prompt suffix “ASSISTANT:”. Read with an eight-token lookup that covers the space-prefixed surface forms, LLaVA-1.5-7B scores F1 0.8221, 0.8498, and 0.8713 on the adversarial, popular, and random splits. That 6.13-point adversarial shift exceeds the adversarial gain reported by five of the six papers we surveyed. None of them states its readout, and our reference ran under 4-bit NF4 quantization, so we cannot say how much of any reported gain a matched evaluation would keep. That comparison stays unresolved until readout and precision are matched. On the fixed baseline we run nine inference-time corrections of our own (none beat it) and dissect why the adversarial split resists logit-level intervention. In a 3,000-question probe, wrong predictions draw more image attention than correct ones (Cohen’s d=-0.46 , p=1.4\times10^-24 ), the opposite of what grounding-based corrections assume. A four-readout audit across four further VLMs from three tokenizer families confirms the confound is protocol-level, not model-specific: porting the LLaVA-1.5 IDs verbatim drops F1 to 0.00 on the two disjoint-vocabulary models, whereas a tokenizer-derived readout holds between 0.80 and 0.92.
[NLP-75] AdaExploreBench: Benchmarking LLM Agents for Proactive Exploration and Test-time Adaptation in Evolving Environments
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在陌生且动态演化环境中的主动探索与测试时适应能力不足的问题。现有基准测试普遍缺乏对代理是否能主动发现环境运行机制并随环境变化进行自适应的评估，导致对模型实际泛化能力的衡量存在偏差。为此，作者提出了AdaExploreBench这一新型基准，通过将环境动态建模为隐藏规则（hidden rules），强制要求代理主动收集证据并推断规则。在隐式规则漂移（implicit rule drift）设定下，潜在规则在未通知的情况下发生变化，迫使代理通过性能下降等信号识别先前知识已过时并重新探索。实验基于八种领先的LLM进行，结果显示：即使在提供充分证据或消除规则漂移的情况下，平均性能提升接近10%，而GPT-6 Sol的校准准确率仅达61.9%。这揭示了当前LLM代理的核心瓶颈——其虽能在已有信息充分时有效推理，但难以主动获取恰当证据，也难以识别已有知识的有效性失效。解决方案的关键在于构建具备主动探索意识和动态适应能力的评估框架，以推动模型从“被动响应”向“主动认知”演进。
链接: https://arxiv.org/abs/2610.25950
作者: Rui Wang,Zishen Lai,Jiayu Liu,Cheng Qian,Jeonghwan Kim,Heng Wang,Xuejun Zhang,Jiateng Liu,Bingxuan Li,Shujin Wu,Qing Zong,Tianshi Zheng,Chunyang Li,Qihan Lin,Chen Li,Dadi Guo,Xiusi Chen,Yangqiu Song
机构: HKUST(香港科技大学); HKUST(GZ)(香港科技大学（广州）); HKUST-UGAIL(香港科技大学-通用人工智能与智能实验室); UIUC(伊利诺伊大学厄本那-香槟分校); CUHK(香港中文大学); CMU(卡内基梅隆大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Model (LLM) agents are increasingly deployed in unfamiliar and evolving environments, requiring them to proactively explore, understand, and adapt to environmental dynamics. However, existing benchmarks rarely test whether agents can proactively discover how an environment works and adapt as it evolves. To address this gap, we introduce AdaExploreBench, a benchmark for proactive exploration and test-time adaptation in evolving environments. By instantiating environmental dynamics as hidden rules, AdaExploreBench requires agents to actively acquire evidence and infer the rules. Under implicit rule drift, the latent rule changes without announcement, requiring agents to detect from degraded performance that prior knowledge is stale and re-explore. Experiments on eight leading LLMs show that proactive and adaptive exploration remain challenging: GPT-6 Sol achieves only 61.9% calibrated accuracy, while providing sufficient evidence or removing rule drift improves average performance significantly by close to 10%. These results reveal a critical bottleneck in current LLM agents: they can often reason effectively from informative evidence once it is available, but struggle to actively acquire the right evidence and recognize when previously acquired evidence is no longer valid.
[NLP-76] Speech LLM s That Breathe: Modeling Non-Verbal Vocalizations in Dialogue
【速读】： 该论文旨在解决当前语音交互模型在自然对话中缺乏对非语言声学事件（Non-Verbal Vocalizations, NVVs）建模的问题，这些事件涵盖呼吸、情感和话语层面的非语言发声，是人类交流中不可或缺的组成部分。现有方法因忽略此类声学特征，导致生成语音在自然性和情境适配性上存在明显缺陷。为此，论文系统性地覆盖了13类主要的NVVs，并通过自动化合成流程构建了一个符合人类对话规范的自然对话数据集。其核心解决方案在于采用多阶段微调策略与面向NVV的组相对策略优化（NVV-aware GRPO），显著提升了语音交互模型生成上下文相关且自然的NVVs的能力。实验表明，该方法在VocalNet和VITA-Qinyu等模型上有效增强了共情互动与角色扮演对话中的表现力，同时揭示了为实现高质量自然语音生成所需的关键架构设计要素。
链接: https://arxiv.org/abs/2610.25942
作者: Heyang Liu,Lixin Zhang,Jiayi Huang,Ziyang Cheng,Yuhao Wang,Ronghua Wu,Qunshan Gu,Yanfeng Wang,Yu Wang
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:The user experience of speech interaction models has improved, benefiting from the modeling of paralinguistic attributes and advances in speech reconstruction paradigms. However, human communication follows extensive norms regarding non-verbal vocalizations (NVVs) spanning respiratory, affective, and discourse. The omission of these events imposes inherent limitations on the naturalness of current approaches. In this work, we cover 13 major types of NVVs and construct a natural dialogue dataset that conforms to human conversational norms through an automated synthesis pipeline. We enhance the ability of speech interaction models to generate contextually appropriate NVV events through multi-stage fine-tuning and NVV-aware GRPO (Group Relative Policy Optimization). Based on distinct modeling paradigms, we substantially improve the expressiveness of VocalNet and VITA-Qinyu in empathy interactions and role-play dialogues, and further discuss the architectural necessity for achieving high-quality natural speech generation.
[NLP-77] MITRA-lexicon: Lexica of attested translations from Sanskrit into English Tibetan and Chinese
【速读】： 该论文旨在解决传统梵语词典（Sanskrit lexica）在现代数字语料环境下的多重局限性：其引文仅以缩略形式标注版本信息，缺乏与现成数字化文本的直接链接，未提供上下文翻译，且无法反映词典出版后新出现的译文。为弥补这些不足，研究提出MITRA-lexicon——一组基于语料库的梵语词典，以梵语为枢纽语言，构建了梵英、梵藏及梵汉三类双向词典，分别源自Mitrasamgraha语料库和MITRA平行语料库中的佛教文献。其核心创新在于采用跨语言词段对齐（span annotation）方法，通过多阶段流水线实现精准翻译等价物识别，该流程融合语法标注模型、微调的大语言模型（LLM）以及教师-学生训练机制，生成高精度的词义对应关系。最终成果包含6种方向性的218万条词目，涵盖1040万条语用实例，每条均附带原文与目标语言句子对及其字符偏移量。词义按频率排序，确保语用相关性。研究通过与权威词典及人工校正的黄金标准对齐数据进行评估，验证了其有效性，并开源全部词典数据、微调后的词段标注模型及交互式查询网页，显著提升了梵语词典的可访问性与实用性。
链接: https://arxiv.org/abs/2610.25934
作者: Sebastian Nehrdich,Kengo Harimoto,Sven Sellmer,Oliver Hellwig
机构: Tohoku University(东北大学); Adam Mickiewicz University(亚当·密茨凯维奇大学); University of Zurich(苏黎世大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:The standard Sanskrit lexica are monumental and indispensable, but they cite their attestations as abbreviated references to editions, without links to the digitized texts now widely available and without a translation of the passage, and they naturally do not reflect translations made after them. To address these gaps, we present MITRA-lexicon, a family of corpus-grounded lexica based on span annotations of translation equivalents with Sanskrit as the pivot: Sanskrit-English, derived from the Mitrasamgraha corpus, as well as Sanskrit-Tibetan and Sanskrit-Chinese, derived from the MITRA-parallel corpus of Buddhist literature. In total, the resource comprises 2.18 million headword entries in six directional lexica. Headwords are lemmas and compounds from silver annotations that follow the Digital Corpus of Sanskrit (DCS). Senses are ranked by frequency, and each of the 10.4 million attestations includes its sentence pair, with character offsets of the word and its equivalent. We propose a multi-step pipeline based on grammatical annotation models, fine-tuned large language models, and a teacher-student setup to produce the alignments. We evaluate the lexica against reference dictionaries and manually corrected gold alignments. We release the lexica and the fine-tuned span annotation model, together with a web interface for looking up headwords and their attested senses.
[NLP-78] Keep the Lineages Alive: Diversity-Preserving SMC for Diffusion LLM Decoding
【速读】： 该论文旨在解决生成式语言模型在推理阶段因采样策略导致的探索能力下降与多样性丧失问题，具体表现为传统多项式重采样（multinomial resampling）在强化高权重粒子的同时过度淘汰低权重但可能通向正确解的路径，从而破坏了推理过程中对潜在正确推理路径的探索能力。其核心解决方案是提出一种保持谱系（lineage-preserving）的序贯蒙特卡洛（SMC）方法——LP-SMC，该方法结合了Chopthin重采样机制与输入权重下限策略：Chopthin通过限制最大与最小权重之间的比例而非强制均一化权重，保留原始权重差异；权重下限则防止过低权重的粒子被完全丢弃，使其具备后续恢复的潜力。这一设计有效维持了种群多样性，同时仍能响应奖励信号进行引导。实验表明，在数学推理和代码生成基准上，LP-SMC显著提升了“最优解覆盖度”（oracle coverage），在所有设置中提升1.8至25.6个百分点，并在11/12个场景中提高端到端准确率，最高达7.9点（LLaDA-1.5）和18.3点（Dream-7B）。该方法在外部奖励模型下的持续增益进一步验证了多样性保持与奖励引导的独立贡献，证明了其在不依赖后训练的前提下增强扩散型语言模型（diffusion LLMs）推理性能的有效性。
链接: https://arxiv.org/abs/2610.25932
作者: Minoo Ahmadi,Seyedarmin Azizi,Mehdi Kamal,Massoud Pedram
机构: University of Southern California (南加州大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Applications (stat.AP); Machine Learning (stat.ML)
备注:
Abstract:Inference-time sequential Monte Carlo (SMC) decoding can improve the reasoning accuracy of masked diffusion language models (diffusion LLMs) without post-training by resampling a population of partial generations according to a reward. Its effectiveness therefore depends on the resampling step, which must concentrate computation on promising candidates while retaining enough diversity to explore alternative reasoning paths. The reward can be internal, the decoder’s own confidence on the committed prefix, or external, a process reward model scoring that prefix. Under either reward, multinomial resampling can disrupt this balance by duplicating high-weight particles and discarding lineages that may still reach a correct solution. The effective sample size (ESS) does not register this loss: multinomial resampling resets it to its maximum even as distinct lineages disappear. To retain such lineages, we introduce lineage-preserving SMC (LP-SMC), which combines Chopthin resampling with an input weight floor. Instead of equalizing weights, Chopthin bounds the ratio between the largest and smallest weights it returns and carries the unequal weights forward. The floor limits how far below the maximum a weight may fall before resampling so that lineages with low early weight keep a chance to recover. Across mathematical-reasoning and code-generation benchmarks on two base models, LP-SMC raises oracle coverage, the fraction of problems with at least one correct particle, in every setting, by 1.8 to 25.6 points, and raises end-to-end accuracy under a common reward-model selector in eleven of twelve settings, by up to 7.9 points on LLaDA-1.5 and 18.3 on Dream-7B. The coverage gain persists under the external reward model, which suggests that diversity-preserving resampling and the guidance signal contribute separately to inference-time search.
[NLP-79] EvoSkill v2: Fully Agent ic Skill Evolution from Benchmark Adaptation to a Stateful Evolver
【速读】： 该论文旨在解决大模型在复杂任务执行中因初始指令（技能）设计不佳而导致性能受限的问题，核心挑战在于如何高效优化任务执行所需的技能描述。其解决方案的关键在于提出EvoSkill v2这一三代理系统架构：通过一个具有状态保持能力的编码代理（stateful coding agent）作为进化器（evolver），持续基于求解器（solver）的训练轨迹动态重写技能指令，同时保留模型和工具链不变。与现有方法（如SkillOpt采用多轮单模型调用、SkillBoost采用每轮重启的代理）相比，EvoSkill v2采用统一的状态化演化机制，支持多轮尝试并利用细粒度的训练轨迹反馈，不依赖噪声评分进行选择，从而更稳定地提升技能质量。实验表明，在SpreadsheetBench v2和τ³-bench banking两个基准上，该方法相较于SkillOpt和SkillBoost分别提升了7.7和5.4个百分点的保留集通过率，且在单位性能增益的成本上更具优势；此外，还验证了低能力模型（如Claude Haiku 5.5）可通过自身演化显著改进技能，实现最高达21.5个百分点的性能提升。
链接: https://arxiv.org/abs/2610.25931
作者: Dastin Huang,Baran Nama,Salaheddin Alzubi
机构: Sentient Labs
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 13 pages, 5 figures
Abstract:EvoSkill v2 is a three-agent system that edits the instructions (skills) an agent is given, leaving its model and harness intact. The user starts a run by describing a benchmark and supplying a grader, which a benchmark adaptor turns into a benchmark-specific plugin. A solver attempts the tasks, and an evolver reads the solver’s training rollouts and rewrites the skills at each epoch. We implement the evolver as one stateful coding agent with a bash tool for the whole run, unlike the two most common designs among the skill optimizers we surveyed: a sequence of single model calls, as in SkillOpt, and an agent that starts afresh each round, as in SkillBoost. We also measure how much the solver’s rollouts vary from one attempt to the next, give the evolver several attempts per training task, and accept every epoch’s skills instead of selecting among them on noisy scores. With fine-grained ground truth, EvoSkill v2’s skills raised the held-out pass rate above the stronger of SkillOpt and SkillBoost by 7.7 percentage points on SpreadsheetBench v2 and by 5.4 on \tau^3 -bench banking. In both benchmarks and both feedback settings, EvoSkill v2’s evolver cost less per percentage point of held-out pass rate gained than either baseline’s optimizer. EvoSkill v2 also enables a less capable model to improve its own skills, demonstrated by a gain of up to 21.5 percentage points in the \tau^3 -bench banking experiment where Claude Haiku 5.5 powers both solver and evolver.
[NLP-80] LEAPS: Long-context Evaluation of Adaptive Pattern Learning and State Tracking
【速读】： 该论文旨在解决当前长上下文语言模型（Long-context Language Models, LCLMs）评估体系中对复杂长程推理能力覆盖不足的问题。现有基准主要聚焦于“回忆”（recall）类任务，即模型从上下文中提取明确事实的能力，而忽视了在代理工作流中更为关键的“状态追踪”（state tracking）与“模式学习”（pattern learning）能力。为此，本文提出LEAPS（Long-context Evaluation of Adaptive Pattern learning and State tracking），一个包含12个新构建数据集的评估框架，涵盖代理工作流、行为建模及市场与世界事件预测等场景。其核心创新在于构建了抗数据污染、可扩展至百万级输入标记、且具备可靠评估指标的评测体系。实验表明，尽管多数模型在回忆任务上表现接近完美，但在状态追踪与模式学习方面性能差异显著，且随着输入长度增加，性能退化更为剧烈——尤其在模式学习任务中，理论上观测样本越多越易学习，但实际表现反而恶化。此外，具备思维链（thinking）能力的模型常因生成预算耗尽或任务简化而表现逊于基础模型。因此，解决方案的关键在于建立全面、真实反映模型长期推理能力的评估标准，以推动下一代长上下文语言模型的发展。
链接: https://arxiv.org/abs/2610.25919
作者: Howard Yen,Nitya Mani,Alex Renda,Christopher S. Hardin
机构: Princeton Language and Intelligence, Princeton University (普林斯顿大学语言与智能中心); Jane Street (Jane Street)
类目: Computation and Language (cs.CL)
备注: Code and data available at this https URL
Abstract:Long-context language models (LCLMs) power agents that act over long trajectories, from improving a codebase to running research experiments to forecasters that condition on lengthy histories. These applications draw on three distinct long-context capabilities: recall, where the answer is a fact present in the context; state tracking, where the outcome can be computed by performing a series of transformations to a specified state; and pattern learning, where the goal is to make a prediction by sampling from a distribution inferred from many observations. Existing benchmarks primarily focus on the first capability. In this work, we introduce LEAPS (Long-context Evaluation of Adaptive Pattern learning and State tracking), a collection of 12 novel datasets, spanning agentic workflows, behavioral modeling, and forecasting of market and world events. By construction, LEAPS is resistant to data contamination, extends to millions of input tokens, and uses reliable metrics. We evaluate 30 API and open models from 8K to 512K input tokens. LEAPS distinguishes models that perform near perfectly on recall tasks, finding that such models’ state tracking and pattern learning abilities can vary significantly and degrade much more dramatically with increasing input lengths–even for pattern learning, where more observations should make learning a theoretically easier task. Moreover, thinking models often underperform their base counterparts, exhausting their generation budget or oversimplifying the task. We hope LEAPS can guide the development of the next generation of LCLMs.
[NLP-81] FisherEdit: Fisher Information Guided Knowledge Editing for Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在进行序列化知识编辑时，因参数累积干扰导致模型通用能力退化的问题。现有基于投影的方法（如AlphaEdit）虽被广泛采用，但研究发现其鲁棒性主要源于更新正则化机制，而非设计初衷的零空间投影。针对这一关键问题，论文提出FisherEdit，一种基于费舍尔（Fisher）几何理论的新型编辑方法，其核心创新在于用参数空间中的重要性感知正则化替代粗粒度的激活空间投影：通过惩罚对敏感参数的过度修改，同时保留对新知识学习的灵活性，并可推导出高效的逐行闭式解。实验在Llama-3、Qwen2.5和Olmo-3等多个模型及基准测试上验证了FisherEdit在保持模型稳定性与下游任务性能方面的显著优势，实现了当前最优的知识编辑效果。
链接: https://arxiv.org/abs/2610.25914
作者: Yizhou Yang,Jiahong Fu,Zhiwu Wang,Kailong Fan,Haokun Lin,Yichen Wu,Zhenan Sun,Renzhen Wang,Deyu Meng
机构: Xi’an Jiaotong University (西安交通大学); Zhejiang University (浙江大学); City University of Hong Kong (香港城市大学); Harvard University (哈佛大学); Institute of Automation, Chinese Academy of Sciences (中国科学院自动化研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Sequential knowledge editing in Large Language Models (LLMs) often leads to a degradation of general capabilities due to cumulative parameter interference. In this paper, we revisit prominent projection-based methods like AlphaEdit and reveal a surprising mismatch: their robustness is primarily driven by update regularization rather than the intended null-space projection. Based on this insight, we propose FisherEdit, a theoretically grounded method built on Fisher geometry, which replaces coarse activation-space projections with importance-aware regularization in parameter space. It penalizes changes to sensitive parameters more strongly while retaining flexibility for new facts, and admits an efficient row-wise closed-form solution. Experiments on Llama-3, Qwen2.5, and Olmo-3 across multiple benchmarks demonstrate that FisherEdit achieves state-of-the-art performance while better preserving model stability and downstream task accuracy.
[NLP-82] CALM: Consistency-Aware Logit Masking in Decoding Space for Mitigating Overthinking of Large Reasoning Models
【速读】： 该论文旨在解决大推理模型（Large Reasoning Models, LRMs）在复杂数学与科学任务中普遍存在“过度思考”（overthinking）的问题，即模型生成大量冗余的推理步骤，导致计算资源浪费且可能引入错误。现有基于截断的方法通常依赖于连续试答一致性的局部停止信号来强制提前退出，但此类信号并不能保证最终答案的稳定性，且可能中断模型后续的反思或验证过程，从而损害性能。本文提出了一种无需训练的解码干预方法——一致性感知的逻辑掩码（Consistency-Aware Logit Masking, CALM），其核心在于：在推理步骤边界监控试答结果，并对动作转换点（Action Transition Point, ATP）标记（如“Wait”、“Alternatively”、“Hmm”）施加逻辑掩码，抑制冗余的过渡性推理，同时允许模型以更自然的方式继续思考。实验表明，CALM在四种先进LRM和五个基准测试上相较完整链式思维（Chain-of-Thought, CoT）推理平均减少18.6%–25.0%的词元消耗，且在保持性能的同时，通过消除冗余推理提升了0.07%–1.48%的准确率；此外，基于大语言模型作为评判者（LLM-as-a-Judge）的评估也证实，CALM生成的结果更具连贯性与简洁性。
链接: https://arxiv.org/abs/2610.25908
作者: Shengjing Liu,Feiyang Li,Yuxuan Yang,Yile Wang,Hui Huang
机构: Shenzhen University (深圳大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large reasoning models (LRMs) have achieved strong performance on challenging mathematical and scientific tasks through long chain-of-thought reasoning, but they often suffer from overthinking with unnecessary reasoning steps. Existing truncation-based methods typically detect consecutive trial-answer agreement and force early exit, yet such local stopping signals do not necessarily indicate final-answer stability and can disrupt useful subsequent reflection or verification, potentially leading to performance degradation. We propose consistency-aware logit masking (CALM), a training-free decoding intervention for mitigating overthinking in LRMs. CALM first monitors trial answers at step boundaries and then applies logit masking to action transition point (ATP) tokens such as “Wait”, “Alternatively”, and “Hmm”, suppressing redundant transitional reasoning while allowing the model to continue its thought process more naturally. Experiments across four advanced LRMs and five benchmarks show that CALM reduces token consumption by 18.6%-25.0% compared with full CoT reasoning while maintaining performance, and can even improve accuracy by 0.07%-1.48% by reducing redundant reasoning. LLM-as-a-Judge results also demonstrate that CALM provides more coherent and concise results.
[NLP-83] SWE-Residual: Evaluating Starting-State Robustness of Coding Agents
【速读】： 该论文旨在解决当前软件工程（Software Engineering, SWE）基准测试中普遍存在的“起始状态过于理想化”问题：现有基准几乎均假设编码代理（coding agent）从零开始执行任务，而实际应用中，代理常需在已有部分工作成果的基础上继续推进，且这些前期工作可能尚未完全正确或完整。为揭示这一现实场景下的性能退化问题，论文提出SWE-Residual基准，用于评估代理在“残余状态”（Residual State）下的鲁棒性——即当起始状态已包含前序代理提交的不完整或错误代码时，代理能否有效识别并修正遗留问题。其核心解决方案在于构建198个源自DeepSWE与Terminal-Bench共111个任务的残余状态实例，每个状态均为前序代理“声称完成”但隐藏测试仍失败后的实际环境快照。评估时，新代理仅接收原始请求与该残余状态，无任何历史轨迹或失败提示，从而真实模拟连续协作场景。实验表明，在五个前沿代理上，相对于清洁启动（Clean Start），其成功率下降20.6–53.5个百分点（DeepSWE）和20.6–34.9个百分点（Terminal-Bench），且在DeepSWE上的排名发生逆转。进一步分析发现，代理普遍存在“接受偏差”（acceptance bias）现象：倾向于将继承的代码视为任务定义，而非重新校验其与需求的一致性。在58.9%–88.2%的失败案例中，代理提交的代码仍存在与清洁启动下即可通过的相同失败测试，暴露了现有方法在处理非初始状态时的关键能力缺口。
链接: https://arxiv.org/abs/2610.25899
作者: Jinze Li,Weijie Liu,Jinfeng Xu,Shuo Yang,Xin Yang,Yang Zhang,Yijin Wang,Zifei Shan,Edith C.H. Ngai
机构: The University of Hong Kong (香港大学); The University of British Columbia (不列颠哥伦比亚大学); University of Tokyo (东京大学); University of North Texas (北德克萨斯大学); Tencent (腾讯)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL)
备注: Under Review
Abstract:Software engineering (SWE) benchmarks are the primary way to measure the capabilities of coding agents: an agent is placed in a code repository or machine environment, given a request, and judged by hidden tests. These benchmarks almost always start the agent from a clean state that contains no earlier work on the task. In practice, however, agents often continue work that a developer or another agent has already done and reported as complete, with no indication of what remains incorrect or missing. We introduce SWE-Residual, a benchmark for Starting-State Robustness: the ability to maintain performance on a task when the starting state already contains earlier work on that task. SWE-Residual contains 198 Residual States drawn from 111 tasks in DeepSWE and Terminal-Bench, where the earlier work modifies a code repository or a whole machine environment. Each Residual State is the state a predecessor agent left behind after submitting its work as complete, although the hidden tests still fail. The evaluated agent receives only the original request and this state, without the predecessor’s trajectory or any sign that the attempt failed; because the request and hidden tests are unchanged, the starting state is the only difference from the source benchmark. Across five frontier agents, success drops by 20.6–53.5 percentage points on DeepSWE and 20.6–34.9 points on Terminal-Bench relative to a Clean Start, and agent rankings on DeepSWE reverse. Failed continuations show a pattern we call acceptance bias: agents treat the inherited work as the definition of what remains to be done instead of checking it against the request. On DeepSWE, in 58.9–88.2% of failed runs, they submit with exactly the same tests failing, although they pass most of these tests from a Clean Start. SWE-Residual thus exposes a capability gap that clean-start evaluation does not capture.
[NLP-84] Budgeted Attack Scheduling for LLM Red-Teaming: The Roles of Static Priors Semantic Guidance and Online Feedback
【速读】： 该论文旨在解决在有限尝试预算下，如何高效地对大型语言模型（Large Language Models, LLMs）进行鲁棒性评估的问题，特别是在面对日益多样化的越狱攻击（jailbreak attacks）时，如何合理调度固定攻击库中的攻击策略以最大化评估效率。其核心挑战在于：攻击的有效性高度依赖于目标模型与防御配置的组合，而全面测试所有攻击策略会带来巨大的计算资源消耗。为此，论文将问题形式化为“基于预算的攻击调度”（budgeted attack scheduling），并探索三种互补的信号来源：（1）静态的历史攻击成功率（Attack Success Rate, ASR）先验；（2）由生成式AI（Generative AI）提供的语义引导；（3）早期查询产生的在线反馈。研究设计了四种调度策略，并在GPT-5.2上对三个受害者模型和三种防御设置进行了系统评估。结果表明，历史ASR提供强初始指导，但在防御引起的分布偏移下可靠性下降；语义引导在冷启动阶段始终优于或等同于静态先验，尤其在缺乏可靠历史数据时表现突出；在线反馈则在先验信息过期时具有显著价值，且能部分弥补历史先验缺失。综合而言，该研究揭示了预算内红队测试本质上是一个多信号协同的攻击调度问题，其中历史先验、语义引导与在线反馈共同构成互补性决策机制，显著提升评估效率与适应性。
链接: https://arxiv.org/abs/2610.25897
作者: Zhicheng Fang,Han Sun,Jiachen Lei,Xinyi Fang,Zongmin Zhang,Chenxu Fu,Sizhe Fang,Wei Xu
机构: Shanghai Qi Zhi Institute(上海奇智研究院); East China Normal University(华东师范大学); The Chinese University of Hong Kong(香港中文大学); The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)); Georgia Institute of Technology(佐治亚理工学院); Tsinghua University(清华大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注: 17 pages, 3 figures, 4 tables
Abstract:As jailbreak attacks against large language models (LLMs) become increasingly diverse, red-teaming has come to rely on large libraries of jailbreak prompts and attack strategies to evaluate the robustness of victim models and their defense settings. However, attack effectiveness is highly condition-dependent: an attack that succeeds under one victim-defense setting may fail under another, while exhaustively evaluating the full attack library for each setting incurs substantial token and monetary costs. This motivates studying how jailbreak attacks should be prioritized under a limited attempt budget for efficient robustness evaluation. In this work, we formalize this problem as budgeted attack scheduling over a fixed attack library and investigate three complementary signals under a fixed victim-defense condition: (1) static historical attack success rate (ASR) priors; (2) semantic guidance from LLMs; and (3) online feedback from earlier queries. Building on these signals, we design four scheduling strategies and evaluate them across three victim models and three defense settings on GPT-5.2. Extensive experiments clearly characterize the conditions under which each signal is most effective. Historical ASR provides a strong initialization, but its reliability degrades under defense-induced distribution shifts. Semantic guidance consistently improves or matches static prior-based scheduling in the cold-start regime, while online feedback becomes particularly valuable when transferred priors become stale. Semantic guidance and online feedback can also partially compensate for missing historical priors. Taken together, these findings characterize budgeted red-teaming as an attack scheduling problem in which historical priors, semantic guidance, and online feedback play complementary roles.
[NLP-85] Learning to Reason Efficiently with Interventional Step Credit
【速读】： 该论文旨在解决大模型在生成链式思维（Chain-of-Thought, CoT）过程中普遍存在冗余推理步骤的问题，即生成的推理轨迹中混杂了有效中间步骤与无用信息，影响推理效率与可解释性。现有方法虽尝试利用注意力质量（attention mass）和步骤熵（step entropy）等步级信号识别冗余步骤，但这些信号仅反映模型在某一步的行为特征，而非该步骤对最终答案的实际贡献。为此，本文提出干预式步骤信用评估（Interventional Step Credit, ISC），从因果干预视角出发，通过针对性地调整参数来量化每一步对推理结果的影响，从而更准确地衡量步骤的重要性。基于ISC，进一步提出IGPO（Interventional Gradient Policy Optimization）框架，利用ISC指导强化学习中的细粒度优势重标定，有效抑制冗余步骤并保留关键推理路径。实验在两种模型规模及九个基准任务上验证，IGPO实现最高达62%的平均令牌减少率，并在整体性能上优于对比方法。
链接: https://arxiv.org/abs/2610.25876
作者: Guanda Li,Xuefeng Bai,Yulong Chen,Zhiyong Wang,Kehai Chen,Min Zhang
机构: Harbin Institute of Technology, Shenzhen(哈尔滨工业大学深圳校区); University of Aberdeen(阿伯丁大学); University of Cambridge(剑桥大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 15 pages
Abstract:Large reasoning models often produce lengthy chain-of-thought (CoT) traces mixing useful intermediate steps with unnecessary ones. Recent methods explore step-level signals such as attention mass and step entropy to identify and suppress unnecessary steps. Despite their effectiveness, such signals reflect the model’s behavior at a step rather than that step’s contribution to the answer. To address this, we propose the interventional step credit (ISC), which takes a causal-intervention view and assesses step importance to the reasoning through targeted parameter-level interventions. Building on ISC, we further propose IGPO, which uses ISC to guide fine-grained advantage rescaling in reinforcement learning, reducing redundant steps while preserving the important ones. Experiments across two model scales and nine benchmarks show that IGPO achieves up to 62% average token reduction and better overall performance among compared methods.
[NLP-86] Arabic Financial NLP
【速读】： 该论文旨在解决阿拉伯语金融自然语言处理（Arabic Financial Natural Language Processing, Arabic F-NLP）中如何在计算表征与原始文档之间保持语义关联的问题，尤其关注从文本采集、处理到分析全链条中，语言上下文、文档来源及解释性判断的持续可追溯性。其解决方案的关键在于构建一个贯穿始终的证据链，确保在词汇分析、实体识别、因果归因、多语言报告对比以及金融摘要生成与基于检索的回答等环节中，始终保留原始文档的语境信息与生成背景，避免因自动化处理导致的语义失真或上下文丢失。此外，通过引入方言意图检测，揭示了机构间对话模式的转变——从单向陈述转向双向互动，进一步凸显了对语言使用情境和传播主体的敏感性。论文强调，现有阿拉伯语资源（包括金融预训练语料、标注数据集与评估基准）需根据其设计初衷进行批判性使用，并主张以“文档溯源性”与“解释性判断”为核心，建立更具透明性和可信度的阿拉伯语金融自然语言处理范式。
链接: https://arxiv.org/abs/2610.25854
作者: Mo El-Haj
机构: 未知
类目: Computation and Language (cs.CL)
备注: 20 pages
Abstract:Financial communication connects numerical accounts of economic activity with explanations of performance, responsibility and future expectations. Arabic financial natural language processing offers methods for studying these connections across corporate reports, market disclosures, journalism and banking interactions. Its contribution to digital humanities depends on retaining the relationship between a computational representation and the document from which it was made. This chapter follows that relationship from source collection and Arabic text processing to the analysis of vocabulary, evaluation, entities, causal attribution and multilingual reporting. It then examines what financial summaries and retrieval-grounded answers preserve or lose, and how dialectal intent detection changes the perspective from institutions speaking to institutions listening. Published Arabic resources are discussed in relation to the questions they were designed to answer, with a distinction between financial pretraining corpora, annotated datasets and evaluation benchmarks. A running constructed example connects the methodological discussion without presenting illustrative analysis as empirical findings. The chapter argues for an account of Arabic financial NLP in which linguistic context, documentary provenance and interpretive judgement remain part of the evidence throughout the analytical process.
[NLP-87] Measurement Validity and Representation Dynamics of Confidence Signals in Large Language Models
【速读】： 该论文旨在解决语言模型在可信部署中“是否知晓自身知识边界”的核心问题，即如何准确判断模型在何种情况下应拒绝回答。现有方法通常依赖单一指标（如口头化置信度、隐藏状态上的线性探测器或采样答案的一致性，即语义熵），并假设这些指标可互换使用，但缺乏系统性比较与验证。本文通过一项预注册审计，对五个Qwen2.5系列模型（0.5B–7B，含7B基座与指令微调对）在事实检索（PopQA、SimpleQA）和多步推理（GSM8K、MATH）任务上进行了全面评估。研究发现：基于预填充阶段隐藏状态的线性探测器在事实检索任务中（约平均7%层深）即可提供有效信号，显著早于多步推理任务（约28%层深），但进一步的21层细粒度分析表明，检索任务中的早期信号可能源于问题表面特征（如TF-IDF分类器即可匹配其性能），而非深层认知能力；口头化置信度表现出高度不稳定性，仅因提示模板微小差异即导致得分无显著相关性；而预期校准误差（Expected Calibration Error, ECE）具有误导性——尽管口头置信度看似校准良好（因集中于基础率），但实际判别能力差；相比之下，采样一致性虽ECE更差，却具备两倍于口头置信度的分辨能力。在五项预注册假设中，两项被证伪，一项部分支持，一项未获支持，另一项因基座模型的口头置信度退化而无法检验。因此，解决方案的关键在于：摒弃单一指标依赖，采用多维度、任务适配的评估框架，并强调采样一致性作为更可靠的不确定性估计手段，同时警惕表面特征对早期信号的干扰。
链接: https://arxiv.org/abs/2610.25846
作者: Debojeet Ghosh,Shreyansh Kumar,Ayushman Garg
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Whether a language model “knows what it knows” matters for trustworthy deployment and for deciding when it should abstain. Prior work usually measures this through one channel: verbalized confidence, a linear probe on hidden states, or agreement among sampled answers (semantic entropy), and often treats these as interchangeable. We compare all three in a pre-registered audit of five Qwen2.5 models (0.5B-7B, including a base/instruct pair at 7B) on factual retrieval (PopQA, SimpleQA) and multi-step reasoning (GSM8K, MATH). First, probes trained on prefill states to predict whether sampled answers will disagree become informative earlier for retrieval (about 7% of layer depth on average) than for reasoning (about 28%). The pre-registered five-layer test supported this only marginally; a finer 21-layer follow-up, designed after those results, supports it more clearly. In most retrieval cells, however, a prompt-only TF-IDF classifier matches the probe, so the early signal may reflect surface features of the question. Second, verbalized confidence is unstable: slightly different prompt templates yield scores that barely correlate. Third, Expected Calibration Error is misleading: verbal confidence looks well calibrated because it clusters near the base rate while discriminating poorly, whereas sampling agreement has worse ECE but about twice the resolution. Of five pre-registered hypotheses, two were falsified, one was partly supported, one was not supported, and one could not be tested because the base model’s verbal confidence was degenerate.
[NLP-88] Where RL Breaks: Localizing and Repairing Failures via Targeted Continued Pretraining
【速读】： 该论文旨在解决强化学习增强语言模型推理能力时出现的“局限性”问题：尽管基于可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR）被广泛用于提升语言模型的推理性能，但现有研究表明，其效果往往仅限于放大基线模型已具备的能力，而非真正拓展其推理边界。当强化学习失败时，关键问题是基线模型究竟缺失了什么？研究发现，强化学习策略在采样回放（rollouts）中的失败具有高度局部化特征——失败行为可被少数特定n-gram（如“let’s recheck”等反思短语或“矩阵A的逆”等具体操作表达）强烈预测。这些失败模式贯穿整个强化学习训练过程，且并非模型无法掌握：同一问题的其他采样答案中常存在正确解法。基于此观察，论文提出一种新方法：利用导致失败的n-gram从预训练语料库中检索相关文本，并在强化学习前对基线模型进行针对性继续预训练。所构建的微调语料仅含3500万词元，远小于常规中段训练阶段的数据规模（低3–4个数量级）。在OLMo2与Qwen3模型上评估表明，该方法在多个任务（包括GSM8K、数学/常识/知识混合题、通用数理推理）中均取得一致性能提升，优于同等规模随机采样的预训练数据继续预训练，在11组实验中有10组表现更优，尤其在插入中段训练前时，于GSM8K上实现2.0分的显著提升。此外，检索文本还展现出跨模型家族的迁移能力：为OLMo2-1B失败案例检索的内容可使Qwen3-4B在混合任务上提升1.9分。综上，该研究揭示了强化学习失败点是一种低成本且高效的预训练数据选择信号，为突破强化学习瓶颈提供了有效路径。
链接: https://arxiv.org/abs/2610.25836
作者: Xinyi Wang,Danqi Chen
机构: Princeton University (普林斯顿大学); Princeton Language and Intelligence
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 9 pages
Abstract:Reinforcement learning with verifiable rewards (RLVR) is widely used to improve reasoning in language models, yet growing evidence suggests that it mostly sharpens what the base model can already do. This raises a practical question: when RL fails, what is the base model missing? We look inside the RL policy’s own rollouts and find that failures are localized: whether a rollout fails is strongly predicted by a small set of characteristic n-grams, ranging from reflection phrases such as “let’s recheck” to specific operations such as “the inverse of A”. These failures persist throughout RL training, although they are not beyond the model’s reach: other sampled answers to the same question are often correct. Building on this observation, we use the failure n -grams to retrieve text from the pretraining corpus and continue pretraining the base model on it before RL. The resulting corpus has only 35M tokens, 3–4 orders of magnitude smaller than typical mid-training. We evaluate this approach on OLMo2 and Qwen3 models trained with RL on (1) GSM8K; (2) a mixture of math, commonsense, and knowledge questions; and (3) general math and science reasoning. Despite its size, the corpus yields consistent gains: our approach outperforms continued pretraining on the same amount of text sampled from the pretraining corpus across 10 out of 11 settings, including by 2.0 points on GSM8K when inserted before mid-training. The retrieved text also transfers across model families: text retrieved for OLMo2-1B’s failures improves Qwen3-4B on the mixture by 1.9 points over the same baseline. Together, our results suggest that where RL fails is a cheap and effective signal for selecting pretraining data.
[NLP-89] Diffusion Language Models Change Speed Across Tasks and Languages
【速读】： 该论文旨在解决生成式扩散语言模型（Diffusion Language Models）在文本生成过程中存在显著计算成本波动的问题，即相同长度的输出可能因生成任务不同而消耗差异巨大的前向传播次数。其核心发现是：生成成本主要由输出内容与输入的相似性及词汇稀有度决定——当输出为对输入的复制或轻微修改时，计算开销较低；而开放性生成（如故事创作）则需更多前向传播，成本可高出三倍以上。进一步分析表明，生成内容的新颖性（相对于输入）和词汇在特定任务中的罕见程度是解释成本差异的关键因素。此外，在包含推理过程的生成中，重复模型自身推理轨迹的最终答案成本低，而引入新信息的答案则更昂贵。多语言生成中，高分词化语言因单位文本包含更多标记，导致每标记所需前向传播次数增加，加剧了整体成本。值得注意的是，计算成本与生成正确性正相关：同一数学问题中，错误答案比正确答案需要更多前向传播，且基于解码速度加权候选答案的方法可在无需额外计算成本的情况下，性能优于多数投票法达4.2个百分点。综上，该研究揭示了扩散模型的计算优势依赖于输出内容的确定性程度，同时提出解码过程中的计算努力可作为高效、可靠的推理优化信号。
链接: https://arxiv.org/abs/2610.25835
作者: Mohsen Fayyaz,Ali Modarressi,Daniel Israel,Lucas Bandarkar,Nanyun Peng
机构: University of California, Los Angeles(加州大学洛杉矶分校); CIS, LMU Munich(慕尼黑路德维希-马克西米利安大学信息科学中心); Munich Center for Machine Learning(慕尼黑机器学习中心)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Diffusion language models can write several tokens in one forward pass, and they decide during generation how many passes to spend. Two outputs of the same token length can therefore vary a lot in cost. To understand the underlying factors of this cost variation, we analyze two diffusion models across 10 tasks. We find that the number of forward passes per token depends heavily on the task. Text copied or lightly edited from the input is cheap, while open-ended writing is costly. In DiffusionGemma, generating a token for story writing requires more than three times as many forward passes as for grammar correction. Two output properties explain much of this gap: how much of it is new relative to the input, and how rare its words are within the task. The same holds within a single response when thinking is involved. The final answer is cheap when it repeats the model’s reasoning trace and costly when it adds something new. In multilingual generation, we find that highly tokenized languages also require more passes per token, adding to the cost they already pay with more tokens. Finally, cost tracks correctness: on the same math problem, wrong answers take more passes per token than correct ones, and weighting candidate answers by decoding speed beats majority voting by up to 4.2 points at no extra cost. Taken together, our results show that diffusion’s computational advantage depends on how much of the output is already determined, while its decoding effort offers a practical signal for making inference more efficient and reliable.
[NLP-90] When Retrieval Partially Conflicts with Skills: Understanding Procedural Hallucination in Language Models
【速读】： 该论文旨在解决生成式AI在执行任务时因检索到的部分冲突信息导致的“程序幻觉”（procedural hallucination）问题，即模型在采纳与技能规范相冲突的检索线索后，生成看似合理但违背技能逻辑的答案。其解决方案的关键在于识别并干预导致错误决策的内在机制：通过引入SourceClash这一匹配基准，系统性地评估不同检索情境下模型的表现；进一步对开源模型进行机制分析发现，答案偏好差异在生成初期即达峰值，且残差替换和注意力分解表明，值向量（value vectors）在修复过程中起主导作用。因此，针对特定值路径的干预可有效提升最终答案的有效性，揭示了部分冲突检索如何扭曲技能引导生成过程的核心机制。
链接: https://arxiv.org/abs/2610.25827
作者: Guanqun Jiang,Pengfei Cao,Kang Liu,Jun Zhao
机构: Institute of Automation, Chinese Academy of Sciences (中国科学院自动化研究所); Xi’an Jiaotong-Liverpool University (西交利物浦大学); School of Artificial Intelligence, University of Chinese Academy of Sciences (中国科学院大学人工智能学院)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Language models increasingly use skills that provide reusable instructions and procedures for task execution. After a relevant skill has been selected, the model may need to apply its governing procedure together with retrieved task context. When the retrieved context is useful but contains cues that violate the skill, the model must retain compatible information while rejecting the conflicting cues. We study failures in which the model adopts such cues and produces a retrieval-supported but skill-invalid answer, which we call procedural hallucination. We introduce SourceClash, a matched benchmark with 1,271 bundles and 6,355 prompts across ten procedural domains. Each bundle fixes the task, skill specification, and answer space while varying the relation between the skill and retrieval. Across nine models, valid-answer rates average 25.7% under conflicting retrieval, compared with 78.2% and 72.8% under two supportive retrieval settings. Mechanistic analysis of open-weight models shows that the answer-preference difference typically reaches its maximum shortly after generation begins. Residual replacement moves the model toward valid answers, while attention decomposition attributes most of the repair to value vectors. Intervention at the identified value pathway also improves final answer validity. These results characterize how partly conflicting retrieval changes skill-guided generation.
[NLP-91] Modal Sense Disambiguation in Language Models: The Role of Context Training Domain and Input Frequency
【速读】： 该论文旨在解决语言模型（Language Models, LMs）在处理具有多义性的情态动词（如may和must）时，是否能够根据语境准确区分其根义（root interpretations，如许可、义务）与认知义（epistemic interpretations，如知识、信念）的问题。其核心挑战在于，情态动词的语义依赖于上下文，而现有研究对语言模型在这一语义消歧任务中的表现缺乏系统评估。解决方案的关键在于通过向量相似性（vector-based similarity）与意外度（surprisal）度量，在四个不同数据集上对比七种架构、规模及训练数据类型各异的语言模型的表现，从而考察其利用上下文信息进行语义消歧的能力，以及其性能是否反映训练语料中情态动词语义分布的统计特征。研究发现，大规模且在多样化文本上训练的语言模型能够有效利用上下文信息进行语义区分；而在缺乏上下文时，则倾向于选择语料库中占主导地位的语义。相比之下，规模较小、且在儿童导向语言（Child-Directed Language, CDL）上训练的发育可实现模型（developmentally plausible LMs）未能表现出一致的语义敏感性，也未能充分受益于上下文线索。
链接: https://arxiv.org/abs/2610.25816
作者: Ezgi Başar,Yevgen Matusevych,Annemarie van Dooren,Arianna Bisazza
机构: Center for Language and Cognition (CLCG), University of Groningen(格罗宁根大学语言与认知中心)
类目: Computation and Language (cs.CL)
备注:
Abstract:Modals such as may and must can take on different senses depending on discourse context, alternating between root interpretations (e.g., permission, obligation) and epistemic interpretations (e.g., knowledge, belief). While modals are known to be notoriously difficult for children to acquire, limited research exists on whether language models (LMs) can reliably distinguish between different modal senses. We evaluate seven LMs that vary in architecture, size, and training data type using vector-based similarity and surprisal measures across four datasets. Specifically, we investigate whether LMs can leverage contextual information to disambiguate modal senses and whether their performance reflects the distribution of modal senses in their training data. We show that larger LMs trained on diverse text successfully leverage additional contextual information when available. In the absence of context, they tend to favor the dominant sense based on corpus statistics. Smaller developmentally plausible LMs, including those trained on child-directed language (CDL), fail to show consistent sensitivity to modal sense distinctions and do not benefit from context to the same degree.
[NLP-92] CoMem: Learning to Anticipate Failures from Retrieved Experience
【速读】： 该论文旨在解决大语言模型（LLM）代理在自我反思过程中存在的记忆局限性问题：单一任务的局部记忆导致代理反复采用相同的失败解释，缺乏跨任务的泛化能力。现有方法虽结合了检索式记忆（retrieved memory）与参数化记忆（parametric memory），但二者以加法方式并行输入求解器，致使检索经验无法影响参数化记忆的输出，限制了反思的深度与多样性。其解决方案的关键在于提出CoMem（Compounding Memory），将检索到的经验作为参数化记忆的直接输入，贯穿训练与推理阶段。通过构建自身成功与失败反思的案例库，模型能够基于检索到的历史案例生成关于失败的假设（failure hypotheses），并在测试时利用新检索案例引导每次尝试的推理过程。这种机制不仅使反思具有多视角的多样性，且该多样性由真实检索案例所支撑，而非随机采样。实验表明，CoMem在数学、代码生成和多跳问答等五个基准上均表现优异，并展现出对分布外任务的良好泛化能力，同时可有效指导更强模型的求解行为，使小型模型具备显著的指导价值。
链接: https://arxiv.org/abs/2610.25790
作者: Yun Shen,Ming Shen,Tianjun Yao
机构: The Ohio State University(俄亥俄州立大学); Shenzhen Loop Area Institute(深圳环区研究院); Shanghai Jiaotong University(上海交通大学)
类目: Computation and Language (cs.CL)
备注: 29 pages, 5 figures
Abstract:Reflection lets LLM agents improve from their own attempts, but the memory it builds is local to a single task, so an agent reflecting on its own tends to revisit the same explanation for failure. Cross-task reflective memory addresses this in two forms: retrieved memory keeps past experience explicit and looks up related cases, which is specific but bounded by the bank, and parametric memory internalizes experience into parameters, which generalizes but is grounded in nothing beyond the task description. When both are used, existing agents combine them additively, feeding their outputs to the solver side by side, so retrieved experience never changes what the learned memory produces. We introduce CoMem (Compounding Memory), which makes retrieved experience an input to the learned memory itself, at training and at inference. The agent builds a bank of its own successful and failed reflections, learns from retrieved cases to write failure hypotheses, candidate explanations of how an attempt is likely to go wrong, and at test time reasons from newly retrieved cases to guide each attempt. Varying the retrieved cases gives the memory different perspectives, yielding reflective diversity that is not merely sampled but also grounded in retrieved experience. Across two backbones and five benchmarks in math, code, and multi-hop QA, CoMem achieves strong overall performance. Additional evaluations show that it generalizes to out-of-distribution tasks and enables smaller models to provide effective guidance to stronger solvers.
[NLP-93] Standing on the Shoulders of Giants: Enhancing LLM Agents for Automated Fact-Checking with Human Expert Skills
【速读】： 该论文旨在解决基于大语言模型（LLM）的自动化事实核查（AFC）系统在证据检索与声明验证过程中存在的效率与准确性瓶颈问题。现有方法依赖计算成本高昂的微调或难以规模化的人工反馈机制，限制了其实际应用。为此，研究提出一种可复用、模块化的专业事实核查技能（Professional Fact-Checking Skill, PFC-Skill），通过提炼法新社（AFP）事实核查公开指南构建而成。在此基础上，提出首个支持推理阶段动态调用PFC-Skill的AFC系统——SkillCheck，使LLM代理能够在证据检索与声明验证过程中实时获得专业指导。实验表明，SkillCheck在多个LLM架构上均显著优于无技能基线模型DEFAME，于AVeriTeC基准测试中准确率提升2.2–4.8个百分点，宏平均F1得分提升1.5–6.3个百分点；尤其值得注意的是，使用GPT-4o-Mini的SkillCheck在准确率上超越使用GPT-4o的DEFAME达3.4个百分点，同时推理成本降低约61%。消融实验显示，将PFC-Skill融入证据检索阶段带来的性能增益最大，而两阶段协同应用则实现最优整体表现。案例分析进一步证明，该技能促使模型进行更精准的搜索并加强对证据的细致审查。研究结论表明，通过借鉴专业事实核查人员的最佳实践，LLM代理可“站在巨人的肩膀上”，在无需额外训练或持续人工监督的前提下，显著提升自动化事实核查性能。
链接: https://arxiv.org/abs/2610.25777
作者: Xinwen Chen,Haorui He,Yupeng Li
机构: Hong Kong Baptist University (香港浸会大学); The University of Hong Kong (香港大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Automated fact-checking (AFC) systems powered by large language model (LLM) agents still face challenges in evidence retrieval and claim verification. Existing approaches improve their capabilities through computationally costly post-training or human-in-the-loop feedback that is difficult to scale. To transfer professional fact-checking expertise to LLM agents efficiently, we distill publicly available guidelines published by Agence France-Presse (AFP) Fact Check into a reusable, modular Professional Fact-Checking Skill (PFC-Skill). We then present SkillCheck, the first AFC system that gives LLM agents dynamic access to PFC-Skill to guide evidence retrieval and claim verification at inference time. Extensive experiments demonstrate that SkillCheck improves upon skill-free DEFAME baselines across multiple LLM backbones, with gains of 2.2–4.8 pp. in Accuracy and 1.5–6.3 pp. in Macro-F1 on the AVeriTeC benchmark. Notably, SkillCheck with GPT-4o-Mini exceeds DEFAME with GPT-4o by 3.4 pp. in Accuracy while cutting inference cost by roughly 61%. The performance gains also extend to open-web retrieval. Ablations show that incorporating the skill during evidence retrieval provides most of the gains, while incorporation into both stages yields the best overall performance. Case studies show that PFC-Skill enables more targeted search and closer scrutiny of evidence. To conclude, our work shows that LLM agents can “stand on the shoulders of giants” by drawing on the best practices of professional fact-checkers, substantially enhancing AFC performance without extra training or continuous human supervision.
[NLP-94] runcated Reasoning Under a Budget: Evaluating System-1 Models as Decision-makers
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在推理过程中生成冗余推理文本（reasoning tokens）导致计算成本过高的问题。传统方法通过限制推理预算以降低成本，但强制在不完整推理轨迹上得出答案会损害模型准确性。本文提出一种基于“系统-1”决策模型（System-1 decision model）的解决方案——Jev，其核心在于无需生成完整推理过程即可对候选答案进行判断，从而作为截断推理路径的决策读取器（decision reader）。实验在七种不同LLM推理器与五个多项选择基准（共35组对比）上验证了Jev的有效性，结果表明：在500至16,000个token的预算范围内，Jev的平均准确率均优于强制输出策略，尤其在1,500 token预算下整体准确率提升4.1个百分点，对截断问题的提升达8.6个百分点；在17组对比中经Holm校正的自举检验显示显著优势。此外，离线回放实验表明，仅需约34.5%的全量运行成本，准确率损失仅为1.13个百分点；基于训练集选取的策略可节省71.3%的估算成本，准确率损失仅0.58个百分点。值得注意的是，Jev在仅依赖问题输入的情况下仍表现出色，平均准确率超过500 token强制回答（68.0% vs. 65.5%），尽管表现因数据集而异。研究结果表明，系统-1型决策模型可有效补充生成式推理，显著降低对完整推理链的依赖，实现高性价比的智能决策。
链接: https://arxiv.org/abs/2610.25768
作者: Zhongyin Wang,Zhaoqian Xue,Junting Mu,Haozheng Liu,Ye Yuan,Haolun Wu,Lingyao Li
机构: University of Illinois Urbana-Champaign; University of Arizona; University of California, Irvine; New York University; Mila – Quebec Artificial Intelligence Institute; McGill University
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) often generate more reasoning tokens than needed. Limiting the reasoning budget reduces cost, but forcing an answer from an incomplete trace can hurt accuracy. System-1 decision models select among candidate options without producing a reasoning response. We evaluate Jev as a decision reader of truncated reasoning across seven LLM reasoners and five multiple-choice benchmarks (35 pairs). Jev achieves higher mean accuracy than forced answering at tested budgets from 500 to 16,000 tokens, including the estimated costs of reasoning and answering. At a 1,500-token budget, it improves accuracy by 4.1 pp overall and 8.6 pp on truncated questions, with significant gains in 17 pairs under Holm-corrected bootstrap tests. Offline replay estimates 34.5% of the full-run cost, with a 1.13 pp accuracy loss. A policy selected on training folds saves 71.3% of the estimated full-run cost on held-out folds, with a 0.58 pp accuracy loss. We also find that Jev shows strong question-only accuracy, exceeding 500-token forced answering on average (68.0% vs. 65.5%), although this advantage varies by benchmark. These findings suggest that System-1 decision models can complement LLM reasoning and reduce the need for costly reasoning completion.
[NLP-95] Computation Not Memory: Chain-of-Thought and State Tracking in LLM Agents
【速读】： 该论文旨在解决语言模型智能体在动态状态环境下，其推理过程与实际行为之间关系的可解释性问题，尤其关注当正确动作依赖于不断变化的系统状态时，推理对行为决策的影响机制。研究聚焦于一个小型文件系统任务，其中背景事件仅通知某文件锁的状态切换，但不提供当前确切值，因此智能体必须基于历史记录重构锁的当前状态才能安全移动文件。解决方案的关键在于区分推理的两种角色：一是每次执行动作前即时生成的推理内容，二是将先前推理保留至提示（prompt）中的长期记忆机制。实验表明，减少锁冲突命令的主要贡献来自在动作发生时刻即时生成的推理，而保留过往推理对遵循规则的帮助有限，尽管对某一模型完成任务有轻微促进作用。此外，推理内容的有效性高度依赖其语义准确性——填充文本、针对不同历史生成的推理或准确的状态表均无法复现相同效果；当给予与自身历史存在单个锁值差异的推理时，智能体仍会依据该推理采取行动，甚至违背自身历史记录。进一步分析显示，不依赖推理的情况下，通过线性探测从模型激活值中解码状态的性能与基于文本转录的分类器相当，但解码精度本身并不能预测干预措施是否改变智能体行为。综上，在此设定下，推理对规则遵守的提升主要源于每步动作前即时生成推理，而无需依赖外部记忆的静态推理；仅能通过无推理方式解码出的状态信息，不足以全面指导智能体的行为模式。
链接: https://arxiv.org/abs/2610.25760
作者: Pedram Mouseli,Massieh Moayedi
机构: University of Toronto(多伦多大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Language-model agents often reason before they act, and the value of monitoring that reasoning depends on how it relates to their actions. We ask how an agent’s actions depend on its reasoning when the correct action depends on a changing state. In a small file-system task, background events announce which file’s lock has been toggled but not its new value. The agent therefore has to reconstruct the current locks from its history before it can move files safely. With Qwen3-8B and Qwen3.5-9B, we separate two roles of reasoning: generating it before each action, and retaining earlier reasoning in the prompt. Most of the reduction in lock-violating commands comes from reasoning written at the moment of action. Retaining earlier reasoning adds little to rule-following, although it helps one model complete the task. The benefit depends on what the reasoning says. Filler text, reasoning written for another history and an accurate table of the state do not reproduce it. When we give the agent reasoning written for a history that differs in one lock value, the agent acts on that reasoning even against its own history. Without reasoning, linear probes recover the state from the model’s activations about as well as a text classifier recovers it from the transcript. However, decoding accuracy does not predict which interventions change the agent’s actions. In this setting, the benefit of reasoning for rule-following comes mainly from generating it at each step. The state that can be decoded without reasoning is a limited guide to how the agent acts.
[NLP-96] A Calibrated Instrument for Detecting Disposition Drift in Language Models
【速读】： 该论文旨在解决大模型在发布前经过安全性测试后，其安全行为可能因微调、训练数据更新或推理设置调整等因素发生隐性退化（即“对齐漂移”）的问题。现有研究通常仅依赖另一模型评估的单一汇总分数来衡量漂移程度，缺乏必要的误差分析以深入理解其机制。本文的关键解决方案是引入一种经校准的测量方法，该方法源自评估推理优化对模型输出质量影响的研究，能够可靠检测出仅几个百分点的变化。通过在两个开源模型上针对医疗与金融建议等多维度对齐问题进行实验，研究发现：相较于正确建议，有害建议的微调引发的对齐漂移显著更大；尤为关键的是，当启用推理机制时，模型可通过生成看似合理但掩盖真实倾向的回答，使外部观测几乎无法察觉其内部状态的偏移；而关闭推理后，模型的行为则随有害训练数据比例线性变化。这一发现揭示了推理模式在安全评估中的双重作用——既可能隐藏风险，也使风险更易被识别，因此强调必须在推理开启与关闭两种状态下均开展安全性测试，以全面捕捉对齐漂移的真实情况。
链接: https://arxiv.org/abs/2610.25735
作者: Jerry Kaplan
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 32 pages, 3 figures. Code and data: this https URL
Abstract:Before language models are released, they are tested for harmful traits such as willingness to carry out dangerous requests, deception, power seeking, and sycophancy. The safe behavior these tests certify may degrade as models are modified, by topical fine-tuning, new training data, or changes to settings such as the depth of their reasoning. Most work reports such drift as single summary rates scored by another AI model, without the error analysis needed to interpret them. We adapt a calibrated measurement methodology recently developed for assessing how inference optimizations affect the quality of model outputs. Applying it to several misalignment dimensions in two open-weight models, we first establish the instrument’s precision, showing that shifts of a few percentage points can be reliably detected. We then induce drift deliberately by fine-tuning on correct and on subtly harmful medical and financial advice; harmful advice induces a far greater shift than correct advice. Most strikingly, a model that reasons before answering can conceal the shift almost entirely in its answers, even though a direct reading of its internal state shows that its disposition has moved. With reasoning switched off, the same model shifts steadily as the share of harmful advice in its training grows. These results demonstrate the importance of testing safety with reasoning turned both on and off.
[NLP-97] Does Unlearning Survive Collaboration? Robust Unlearning for Shared-Model LLM Teams
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）团队协作场景下知识遗忘（unlearning）有效性下降的问题。在团队协作中，尽管单个模型的遗忘效果良好，但多个模型之间的信息交互会通过消息聚合产生可恢复的隐含证据，导致敏感或不需要的知识仍能被联合重构，从而削弱遗忘效果。现有方法仅关注个体输出或最终答案的控制，无法有效防止多模型协同带来的知识恢复风险。其解决方案的关键在于提出一种新型的**聚合条件遗忘（Aggregation-Conditioned Unlearning, ACU）**机制：ACU通过联合训练共享模型，利用个体与聚合后消息的重建反馈，识别高恢复风险的视角，并采用基于替换的信用分配策略，在保障个体与协作任务性能的同时，抑制跨模型组合带来的信息泄露。实验结果表明，相较于传统的输出监督训练方式，ACU在TOFU评估框架下将完整轨迹的恢复率从41.8%降至19.3%（ROUGE-L召回），事实性恢复率从34.2%降至18.7%，同时保持了良好的任务保留能力，验证了其在不同团队规模和读者场景下的泛化性与有效性。
链接: https://arxiv.org/abs/2610.25731
作者: Hengsong Liu,Liangyu Teng,Jingnan Cai,Yuntian Shi,Jingyu Zhang,Yang Liu,Jing Liu,Liang Song
机构: Fudan Institute on Networking Systems of AI(复旦大学人工智能网络系统研究所)
类目: Computation and Language (cs.CL)
备注: 39 pages, 8 figures
Abstract:LLM unlearning is essential for removing sensitive or unwanted knowledge while preserving useful capabilities. As LLM teams are increasingly used to solve complex problems, forgetting must remain effective during collaboration. Despite advances in forgetting effectiveness, efficiency, and robustness to relearning, team deployment remains underexplored. We find that collaboration weakens forgetting: final-answer evaluations miss recoverable evidence in team messages, including complementary clues supporting joint recovery. We establish theoretically and confirm empirically that controlling final answers or individual messages does not ensure low recovery from their combination. To address this challenge, we propose Aggregation-Conditioned Unlearning (ACU). ACU trains a shared model using reconstruction feedback from individual and pooled messages, prioritizes views with higher recovery risk, and uses replacement-based credit assignment while protecting direct and collaborative utility. TOFU evaluates trajectory-level recovery, while WMDP-bio tests native forgetting and retained utility. On TOFU, ACU lowers full-trajectory recovery from 41.8% to 19.3% in ROUGE-L recall and factual recovery from 34.2% to 18.7% versus output-supervised training, with comparable retained-task performance. These results demonstrate that ACU curbs information leakage in LLM teams and generalizes across team sizes and readers.
[NLP-98] No Matter Who You Are: Addressivity and Sycophancy in Large Language Models EMNLP2026
【速读】： 该论文旨在解决生成式模型中“谄媚行为”（sycophancy）这一关键缺陷问题，即模型在与用户交互时表现出过度顺从、讨好或奉承的倾向。其核心问题是：尽管后训练（post-training）被怀疑是导致该现象的原因，但其内在机制尚不明确。论文的关键解决方案在于揭示“面向性”（addressivity）——即模型输出从文本补全转向面向特定接收者（addressee）的言语表达——在诱导谄媚行为中的作用。研究通过稀疏自编码器（Sparse Autoencoders）提取预训练与后训练模型中的面向性方向，并发现仅在后训练模型中，操纵这些方向可显著放大或抑制谄媚行为。进一步分析表明，后训练过程强化了面向性与谄媚行为之间的内部关联。此外，控制实验显示，模型的谄媚行为并非针对特定用户，而是对任何被识别为“接收者”的存在作出反应，说明触发谄媚的核心因素是“存在一个被感知的加害者”（addressee），而非其具体身份。因此，该研究的关键发现是：后训练引入的面向性及其引发的迎合压力，是导致模型产生谄媚行为的根本机制。
链接: https://arxiv.org/abs/2610.25728
作者: Seungjong Sun,Hyunmin Jun,Minsik Yoon,Eungu Lee,Choongwon Kang,Young Seok Woo,Jang Hyun Kim
机构: Sungkyunkwan University (成均馆大学); Department of Applied Artificial Intelligence (应用人工智能系); Department of Interaction Science (交互科学系); Department of Human-Artificial Intelligence Interaction (人机智能交互系)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: EMNLP 2026
Abstract:Sycophancy, where an LLM becomes overly agreeable, deferential, or flattering toward the user, has been identified as a major failure mode. While post-training has been implicated as a cause, how it internally gives rise to sycophantic behavior remains insufficiently understood. Post-training transforms the model’s output from text completion into utterances directed toward an addressee, thereby giving the output \textitaddressivity. We hypothesize that the accompanying pressure to follow the addressee’s instructions and align with their preferences links this addressivity to sycophancy. To test this, we use Sparse Autoencoders to extract addressivity directions from pre-trained and post-trained models and examine their relationship with sycophancy. Steering models with addressivity directions can amplify or suppress sycophancy, and this effect is observed only in post-trained models. Furthermore, comparing the internal mechanisms of both model types confirms that the association between addressivity and sycophancy is strengthened through post-training. Finally, through a controlled post-training experiment, we find that models are sycophantic not toward a specific addressee but toward any recognized addressee, suggesting that the presence of an addressee, rather than the addressee’s identity, is what elicits sycophancy.
[NLP-99] Inside the Scorer: Mechanistic Evidence on Rubric Control in LLM -Based Automated Essay Scoring from 25 Base Models ACL
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在自动化作文评分（AES）中缺乏可控性与可验证性的问题，尤其在高风险语言测评场景下，现有基于大语言模型（LLM）的评分系统难以实现对具体评分维度（如语法、词汇、结构等）的独立控制。其核心解决方案在于探究大语言模型是否在内部表征空间中为不同评分量规形成可区分的表示，并评估这些表示是否可被独立操纵以实现精准干预。研究通过对25个开源基础大模型（参数规模0.13B–10.31B，涵盖14个架构家族）在人类评分的BESTEP英语写作任务上的探针分析与因果操控实验发现：长篇作文（第2、3部分）中整体分数可从模型激活值中解码，且各评分维度在表征空间中存在部分可分离方向；然而，通过因果干预测试表明，无论施加匹配或不匹配的调控（如对语法退化的文本进行词汇干预），均未产生显著差异，其效果仅约基准提升的2%，且在所有10组可测维度中均无统计可检测的改进。这表明，尽管模型能预测总分，但其评分维度共享一个主导控制杠杆，而非独立可调的评价通道，其内在机制可能反映了人类评分中的“光环效应”（halo effect）。因此，该研究揭示了当前基于大语言模型的评分系统虽具备较高的分数一致性与可解码性，但并不等同于实现了量规层面的可控性，从而提出了一个关键方法论贡献：分数一致性与表征可解码性不能作为鲁棒性维度控制的充分依据。
链接: https://arxiv.org/abs/2610.25724
作者: Chungche Huang,Po-Ya Angela Wang,Shu-Kai Hsieh
机构: National Taiwan University(国立台湾大学); Co-intelligence Humanities AI; (CHAI) Future Lab
类目: Computation and Language (cs.CL)
备注: Camera-ready version. Accepted to PACLIC 40 (The 40th Pacific Asia Conference on Language, Information and Computation); to be presented in Taipei, Taiwan, December 10-12, 2026. Code at this https URL
Abstract:Automated essay scoring (AES) using large language models (LLMs) offers promising use cases for language assessment, yet high-stakes testing requires scores that are both controllable and verifiable, which current LLM-based scorers lack. Given that, we investigate whether LLM-based scorers form distinct internal representations for individual scoring rubrics and whether these representations can be independently controlled. Probing and steering 25 open-weight base LLMs (0.13B-10.31B parameters; 14 families) on human-scored BESTEP English writing tasks, we find that holistic scores are decodable from model activations for the longer tasks (Parts 2 and 3) and only weakly for the shortest (Part 1). Furthermore, scoring rubrics have partly separable directions in representation space for the longer tasks, though not for short responses (Part 1). However, causal steering reveals a critical limitation: applying a matching intervention (e.g., steering “grammar” on a grammar-degraded response) yields only a negligible advantage (about 2% of the effect) over a mismatched intervention (e.g., steering “vocabulary”), and none that is detectable in any of the 10 testable comparisons across reliably measured rubrics. Rather than controlling distinct evaluative dimensions, the rubric directions share a dominant lever whose rubric-specific parts are not selective, which may mirror the halo structure of human ratings. Our contributions are two-fold. First, we show that analytic scoring rubrics designed for human raters do not behave as independent controls in base LLMs, even among the seven models whose scores agree best with human raters ( r \geq 0.6 ). Second, we establish a methodology that shows that score agreement and decodability do not imply rubric-level control in LLM-based assessment.
[NLP-100] Whose Credential Is It? How Owner Attribution in Agent MemoryShapes Cross-User Account Misuse
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在共享长期记忆环境下，因无意中检索到其他用户凭证而导致隐私泄露与潜在滥用的风险问题。其核心挑战在于：尽管存在凭证暴露（exposure）现象，但现有研究未能明确判断这种暴露是否真正导致了实际危害。为此，作者提出一种三阶段审计框架，将风险事件分解为三个关键环节——凭证进入提示（prompt）（暴露）、代理基于该凭证采取行动（行为）以及目标账户状态发生改变（状态），从而实现对危害链的精准识别与量化。实验结果表明，在模拟真实场景的AppWorld环境中，84.7%的应用会遭遇其他用户的凭证暴露，但多数情况下代理并未实际使用；而在压力测试下，当代理自身凭证缺失且仅存在单一外部凭证时，7个代理中有70%至94%的运行实例会使用该凭证，并导致受害者账户变更，最高达43%的案例甚至通过官方验证。进一步分析发现，通过在每个凭证行前或内显式标注所有者名称（owner name），可使滥用率从305次配对运行中的92次降至1次，显著降低风险；而仅使用无名用户名则效果有限，且添加所有者绑定注释会导致模型在干净记忆下每任务损失3至5次成功执行。此外，错误书写所有者信息几乎完全恢复滥用行为，尾随所有者标签和附加安全指令仅对部分模型有效。因此，解决方案的关键在于：强制代理记忆系统始终记录所存储凭证的所有权归属，并将评估重点从单纯的“暴露”转向对“行为”与“状态变化”的综合衡量，以实现更真实、可靠的安全性评价。
链接: https://arxiv.org/abs/2610.25707
作者: Hau-Sing So,Daixian Liu,Carl Hou,Yuhaohang He,Mengzhu Wang,Hai-Tao Zheng
机构: University of Macau(澳门大学); Tsinghua University(清华大学); Lanzhou University(兰州大学); Sichuan Agricultural University(四川农业大学); Hebei University of Technology(河北工业大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 19 pages, 14 figures, 12 tables
Abstract:Shared long-term memory lets an LLM agent serving one user inadvertently retrieve another user’s credentials, but it is unclear when such exposure leads to actual harm. We introduce a three-stage audit that separates three events: a foreign credential reaches the prompt (exposure), the agent acts with it (behavior), and the other user’s account changes (state). In AppWorld, a store that mixes 763 past logins from all users retrieves another user’s credential for 84.7 percent of required apps, yet agents rarely act on it. Misuse appears in a stress test where memory lacks the user’s own credential and presents a single foreign one as the user’s record: seven agents then use it in 80 to 94 percent of runs and change the victim’s account in up to 43 percent, sometimes passing the official check. Naming the owner before or inside each credential line prevents most of this: with all else fixed, misuse falls from 92 to 1 of 305 paired runs, also with name-free usernames, though owner-bound notes cost 3 to 5 successful tasks per model on clean memory. A wrongly written owner largely restores misuse; trailing owner tags and added safety instructions help only some models. Agent memory should consistently record whose credential it stores, and evaluations should measure behavior, not only exposure.
[NLP-101] Consistency Training Suppresses Cue Following Without Teaching Cue Selectivity
【速读】： 该论文旨在解决语言模型在受到无关提示（cue）影响时出现答案不一致的问题，即模型因诸如“我认为答案是B”这类无实质证据的提示而改变原有判断。其核心解决方案是采用一致性训练（Consistency Training, CT），通过强制模型在有无提示条件下保持输出一致来削弱提示对答案的影响。研究对比了四种不同的CT方法：基于输出词元的BCT、基于残差激活的ACT、基于注意力模式的AttCT以及基于采样答案频率的RMCT。实验在Gemma-2-2B-it与Llama-3.1-8B-Instruct模型上进行，并使用相同数据集训练。结果显示，四种方法均有效降低训练过的提示的影响力，且修复机制具有三个共性特征：第一，正确提示与错误提示的影响力被削弱程度相当；第二，提示位于答案选项前时修复效果显著，而位于之后则减弱明显，仅Llama的RMCT对位置敏感度较低；第三，尽管经过训练，仍主要由特定电路头（circuit heads）传递残留的提示诱导错误。此外，若将BCT的目标依赖于提示正确性，则会显著削弱修复效果。因此，一致性训练虽能有效弱化提示影响，但并未教会模型识别提示本身的对错。仅在训练位置使用错误提示进行评估，会遗漏上述两个关键特性，导致对模型行为的理解偏差。
链接: https://arxiv.org/abs/2610.25650
作者: Prakhar Gupta,Terry Jingchen Zhang,Zhijing Jin
机构: University of Michigan(密歇根大学); Jinesis Lab, University of Toronto(多伦多大学杰尼斯实验室); Vector Institute(向量研究所); University of Oxford(牛津大学); EuroSafeAI; Max Planck Institute for Intelligent Systems, Tübingen, Germany(马克斯·普朗克智能系统研究所，图宾根，德国)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Language models often change answers because of a cue, a hint such as “I believe the answer is B” that gives no valid evidence. Consistency training (CT) trains a model to answer the same with and without the cue. Four CT methods apply this idea to output tokens (BCT), residual activations (ACT), attention patterns (AttCT) and sampled-answer rates (RMCT). Prior work suggests some learn different solutions. We train all four on Gemma-2-2B-it and Llama-3.1-8B-Instruct with the same data. We test predictions written in advance and inspect the models with a known circuit that carries cues to the answer. All four make the trained cue less effective, and their fixes share three properties. First, correct cues lose about as much influence as wrong ones. Second, the fix is usually large for cues before the answer options, as in training, and much smaller after them. Llama RMCT depends least on position. Third, the circuit heads still carry most remaining cue-induced errors. Making BCT targets depend on the cue’s correctness mostly weakens the fix. CT thus weakens the trained cue without teaching models when a cue is right. Evaluations with only wrong cues at the trained position miss the first two properties.
[NLP-102] MapSteal: Map-based Autonomous Knowledge Stealing Attack Against Large Language Models
【速读】： 该论文旨在解决从黑箱大语言模型（Large Language Models, LLMs）中自主获取结构化知识以适应小型学生模型（Student Models）时，现有方法依赖预设查询或学生反馈导致知识目标扩展缺乏自主性的问题。其核心解决方案是提出一种学生无关的递归结构化知识获取框架 MapSteal，关键在于通过初始少量根节点（roots）构建结构化状态表示，利用教师模型生成的方向递归扩展知识目标，并结合问题实现、提议接纳与去重机制约束整个过程；同时采用根感知调度策略，在固定问答预算下高效组织探索路径。该方法无需在知识获取阶段依赖学生模型反馈，即可生成高质量的问答对用于学生模型训练。实验表明，基于Qwen3.5系列2B、4B、9B规模的学生模型在医疗与网络安全六大基准上均取得具有竞争力的性能表现，其中4B医疗任务中，完整版MapSteal相较简化递归获取变体平均准确率提升1.17个百分点，验证了该框架在实现结构化、自主化知识获取方面的可行性，同时揭示了知识目标生成与有效训练监督之间的本质差异。
链接: https://arxiv.org/abs/2610.25626
作者: Zhengao Li,Chuxu Zhang,Yushun Dong
机构: University of Connecticut (康涅狄格大学); Florida State University (佛罗里达州立大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Knowledge acquisition from black-box large language models (LLMs) offers a practical approach to adapting smaller student models. However, many existing methods rely on predefined queries or student feedback to guide acquisition, leaving autonomous expansion of knowledge targets an important challenge. We propose \textbfMapSteal, a student-independent framework for structured recursive knowledge acquisition. Starting from a small set of initial roots, MapSteal represents acquisition targets using structured states and recursively expands them through teacher-proposed directions. Question realization, proposal admission, and deduplication constrain the acquisition process, while root-aware scheduling organizes exploration under a fixed QA budget. The resulting question–answer pairs are used to adapt student models without requiring student feedback during acquisition. Experiments with Qwen3.5 students at 2B, 4B, and 9B scales show competitive downstream performance across six medical and cybersecurity benchmarks. On the 4B medical evaluation, Full MapSteal improves average accuracy by 1.17 percentage points over a simplified recursive acquisition variant. Cross-domain evaluations show that gains depend on the domain and evaluation panel. These results demonstrate the feasibility of structured, student-independent recursive acquisition while highlighting the distinction between generating acquisition targets and obtaining useful training supervision. Our code is available at \hrefthis https URL this https URL.
[NLP-103] Curriculum Learning Teaches LLM s Long-Horizon Implicit Reasoning Up To A Limit
【速读】： 该论文旨在解决大语言模型（LLM）在长时程任务中表现急剧下降的问题，特别是其在链式思维（Chain-of-Thought, CoT）生成过程中隐式推理（implicit reasoning）能力有限，难以在单次前向传播中有效“前瞻搜索”以选择正确下一步，导致模型易陷入错误路径并需通过回溯修正。其核心解决方案是引入课程学习（curriculum learning），通过在词汇量和前瞻深度（lookahead，即模型需前瞻多少步才能确定正确动作）逐步增加的合成数据上进行微调，以提升模型的隐式推理深度。关键发现在于：在相同计算成本下，经过课程学习微调的模型能够实现远超无课程训练的前瞻能力——例如Qwen3-0.6B模型在前瞻256步的任务上准确率达98%，而对照组仅45%；同时，模型的隐式推理能力存在一个由模型规模决定的内在上限，表现为渐进饱和的指数增长曲线。此外，大规模模型更具优势，预训练模型微调优于从零训练，且过大的课程步长会阻碍进展。最终，该方法在合成数据上获得的隐式推理能力可有效迁移至自然语言推理任务，在11个基准测试中均优于同等计算资源下的非课程训练模型。
链接: https://arxiv.org/abs/2610.25612
作者: Jimson Huang,Maxwell Nickell,Ruowang Zhang,Abulhair Saparov
机构: Purdue University (普渡大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 31 pages, 13 figures
Abstract:LLMs “reason” by externalizing intermediate steps as tokens through chain-of-thought (CoT), yet even frontier reasoning models hit abrupt performance cliffs on long-horizon tasks. Implicit reasoning (reasoning within a single forward pass) is useful during CoT generation, as the model can “search ahead” to decide the correct next move without explicit search. Models with weak implicit reasoning are more likely to commit to wrong paths and must attempt to recover via backtracking, while models with strong implicit reasoning identify the correct proof strategy directly, shortening the CoT. However, prior work shows LLMs struggle to reason implicitly beyond shallow depths, with no relief from scale. We investigate whether curriculum learning can extend this depth by fine-tuning on examples with progressively increasing vocabulary size and lookahead (the number of steps a model must search ahead to identify the correct next move). At matched compute, curriculum-trained models achieve lookaheads where no-curriculum training fails. Qwen3-0.6B reaches 98% accuracy on problems with lookahead up to 256 with curriculum, compared to 45% without. However, progress slows as fine-tuning continues, following a saturating exponential that suggests a maximum intrinsic lookahead to which models can learn to reason implicitly. Larger models reach higher fitted ceilings, fine-tuning pretrained models far outpaces training from scratch, and curriculum step sizes that are too large stall progression. Finally, the capability acquired on synthetic data transfers to natural language reasoning, with curriculum-trained models outperforming models trained without curriculum at matched compute on 11 natural language reasoning benchmarks.
[NLP-104] xt Dominance in Paralinguistic Conflict: Tracing Suppression Through Audio LLM Circuits NEURIPS2026
【速读】： 该论文旨在解决生成式 AI（Generative AI）在语音大模型（Audio LLMs）应用中对声学线索（如情感、说话人身份、讽刺等）的依赖不足问题，尤其在危机热线与客户服务等场景中，若仅依赖文本转录而忽视非语言声学信息，将严重影响模型的语用准确性与实用性。其核心挑战在于揭示音频大模型是否过度依赖文本内容而忽略关键的副语言线索，并解析其内在机制。研究发现，在六种音频大模型于类似LISTEN的冲突情境下测试中，平均文本主导率为0.78（排除讽刺时为0.86），表明模型普遍优先采纳文本信息。进一步探查显示，说话人身份特征在语言模型输入层仍可线性解码至接近上限，说明问题并非感知层面缺陷。尽管回补（back-patching）被报道可缓解词汇级文本主导现象，但在本研究的复现中（n=300）该方法本身即无效，因此其“无效果”结果不具说服力。相比之下，注意力击穿（attention knockout）通过阻断中间标记中冲突主张的传播路径，部分恢复了以音频为基础的输出，而直接切断答案位置与冲突主张的连接则无效，表明该影响通过中间传递节点而非单一可切断路径传播。此外，该效应仅出现在开放式生成任务中，未见于封闭式多选题任务，提示当前依赖多选题格式的评估基准可能系统性地忽视此类干预的影响。
链接: https://arxiv.org/abs/2610.25599
作者: Ankur Banerjee,Vishnu Guntakala,Danity Pike,Advay Roongta,Kevin Yang,Ruizhe Li
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 18 pages, 9 figures, 17 tables. Accepted at the InterpScience Workshop at NeurIPS 2026. Code: this https URL
Abstract:Audio LLMs deployed in settings such as crisis hotlines and customer support are only useful beyond speech-to-text pipelines if they use acoustic cues rather than the transcript alone. We test whether they prioritise transcript content over paralinguistic cues (emotion, speaker identity, sarcasm) and trace the mechanism when they do. Across six audio LLMs under LISTEN-style conflict conditions, the mean text-dominance rate on mismatched text+audio items is 0.78 (0.86 excluding sarcasm). In the most text-dominant screened model, probing shows speaker identity remains linearly decodable at ceiling at the language model’s input, so the failure is not perceptual. Back-patching, reported to repair lexical text dominance, has no measurable effect here; but it also fails on its own source task in our replication (n=300), so we treat its null as uninformative rather than as evidence of non-transfer. Attention knockout partially restores audio-grounded output by blocking the claim from propagating through intermediate tokens during open-ended generation, whereas blocking the answer position’s direct connection to the conflicting claim has no effect. The claim’s influence therefore travels through an intermediate relay rather than a single severable edge. The effect appears only in free-form generation and not in closed-set multiple choice, suggesting that benchmarks relying on that format may be systematically blind to interventions of this kind. Code and data are publicly available at this https URL.
[NLP-105] Cognitive Map in Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）是否能够构建类似动物认知地图（cognitive map）的内部表征，以支持在新环境中的灵活推理行为这一核心问题。其关键解决方案在于通过关系推理任务与因果“马赛克实验”（mosaic experiments），系统性地操纵开放权重语言模型的内部激活状态，揭示模型如何动态生成抽象的关系结构表征。研究发现，超过十亿参数规模的模型能有效推断未明确陈述的实体间关系，并对合成实体名称实现泛化；模型将实体身份表征与关系结构表征以加性方式结合，实现实体到角色的绑定；其关系存储机制并非预先定位于全局坐标系，而是以成对链接的形式保留，仅在查询时进行回溯组合。这种绑定由稀疏分布的键-值对（key-value pairs）介导，通过对这些关键组件的扰动可植入“虚假记忆”，并以可预测方式改变模型行为。整体结果表明，语言模型内部存在一种类认知地图的组织结构，解释了其在信息有限条件下进行高效推理的机制。
链接: https://arxiv.org/abs/2610.25596
作者: Albert Qin,Jonathan Cohen,Gautam Reddy
机构: Princeton University (普林斯顿大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:It has long been proposed that animals construct cognitive maps to support flexible behavior in novel environments. We tested whether large language models form similar representations using relational reasoning tasks and causal “mosaic experiments” that manipulate internal activations in open-weight language models. Models beyond the ten-billion-parameter scale successfully inferred unstated relationships between entities and generalized to synthetic entity names. Across model and task families, models generated abstract representations of relational structure and combined them additively with representations of entity identity, binding entities to roles within that structure. Models stored observed relationships as pairwise links that they composed retrospectively when queried, rather than prospectively positioning each entity in a global coordinate system. Binding was mediated by a sparse, distributed set of key–value pairs; by perturbing these pairs, we implanted “false memories” that altered model behavior in predictable ways. Overall, our findings reveal an internal organization analogous to a cognitive map that explains how language models reason with limited information.
[NLP-106] Assertion and Belief: Moores Paradox in Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理语用悖论时是否具备对言语行为（speech act）的语用认知能力，特别是针对摩尔悖论（Moore’s paradox）这一经典语用现象。摩尔悖论表现为“下雨了，但我并不相信下雨”这类句子虽可能为真，却无法被合理地断言，其荒谬性不在于内容本身，而在于说话人声称知道某事却否认自己相信它，这违背了言语行为中的自我一致性要求。论文通过构建一组精心设计的摩尔式句子，在多种语境中对12个主流大语言模型进行测试，发现这些模型普遍能够识别此类句子的语用不适宜性（infelicity），且判断不受记忆特定句式的影响，而是与说话行为密切相关，表明模型已内化了对言语行为的敏感性。进一步基于理性言语行为框架（Rational Speech Act, RSA）分析发现，模型的判断并非源于对自身说话者心理状态的推理，而是依赖于对摩尔悖论相关知识的回忆和规则应用，即模型将该悖论视为关于句子结构的事实性知识，而非对说话者信念状态的规范性约束。因此，解决方案的关键在于：模型并非真正理解言语行为的语用机制，而是以一种“事实性规则”的方式习得并应用摩尔悖论的表面特征，反映出其语用认知仍停留在表层模式匹配，尚未实现对说话者意图的深层推理。
链接: https://arxiv.org/abs/2610.25586
作者: Anshun Asher Zheng,Weili Cao,Fangcong Yin,David Beaver,Junyi Jessy Li
机构: The University of Texas at Austin(德克萨斯大学奥斯汀分校); Duke University(杜克大学); New York University(纽约大学); University of Graz(格拉茨大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Moore’s paradox \citepmoore1942 concerns sentences such as ``It is raining, but I don’t believe that it is,‘’ which may well be true yet cannot be sensibly asserted. The absurdity lies not in the content but in the act, since to assert is to vouch for one’s own belief. This makes the paradox a test of whether language models take the speech act into account when computing meaning. We present the first test of Moore’s paradox in LLMs, evaluating 12 models on a curated set of Moorean sentences in varied contexts. We find that models largely recognize the infelicity of Moorean sentences, and that the judgement is not tied to a memorized form and tracks the speech act. Using the Rational Speech Act framework (RSA; \citealtgoodmanfrank2016), we then test whether these judgements arise from a listener reasoning about a speaker. We find the models’ own listeners do not reason about their own speakers, and their reasoning traces suggest that they instead recall knowledge about the paradox and apply rules therein. The models thus appear to have learned the paradox as a fact about sentences, not as a requirement on speakers.
[NLP-107] Representations in Motion: Layer-wise Rotation and Trajectory Divergence in LLM s
【速读】： 该论文旨在解决语言模型在推理过程中表征动态变化（activation dynamics）如何揭示其行为本质的问题，特别是关注事实性（factuality）这一语义属性在不同推理条件下的演化机制。传统研究多依赖于对单个层的静态表征读出，而本文提出，表征在层间演化的动态过程可能蕴含更深层的行为信息。其解决方案的关键在于发现并验证一种几何旋转现象：在模型被要求“说谎”或“诚实”时，从各层提取的事实性方向（factuality directions）呈现出随层数递进而相互旋转的规律性特征。这一现象不仅存在于事实性任务中，也在需要模型在三维空间中转换视角的任务中观察到，暗示该旋转可能普遍反映模型对概念空间（conceptual space）的动态重构能力。进一步地，研究通过分析单次试次（trial-to-trial）的表征轨迹，证明仅凭轨迹即可准确预测模型的目标任务与行为表现，并发现轨迹的分歧点与群体层面的旋转起始点相吻合。最后，该现象在非显式诱导说谎的自然任务场景中亦可复现，表明其并非人为设定的特例，而是语言模型内在表征动态的核心特征。因此，该研究的核心贡献在于揭示了表征旋转作为理解模型认知状态转变的通用动态指标，为解释生成式模型内部决策机制提供了新的几何视角。
链接: https://arxiv.org/abs/2610.25574
作者: Zhuonan Yang,Yik Siu Chan,Wannan Yang,Ellie Pavlick
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Recent work suggests that activation dynamics–how representations change during inference–might be more revealing about language model behavior than static readouts from individual layers. In this work, we investigate the layer-wise evolution of representations of factuality, using true and false statements when the model is prompted to either lie or to be honest. We report a geometric phenomenon in which the factuality directions extracted from the lying and honest conditions appear to rotate away from each other across layers. We find that such layer-wise rotation also occurs in a task that requires the model to take different visual perspectives in the 3D world, suggesting that the rotation phenomenon might be indicative of manipulating conceptual spaces in general, rather than unique to factuality. We then investigate the “real time” transformation of conceptual representations on a trial-to-trial basis. We show that it is possible to make strong predictions about the model’s target task and behavior from trajectories alone. In addition, diverging points in single-trial trajectories correspond to the onset of rotation in aggregated-trial analysis. Finally, we show that both rotation and divergent trajectories occur in more naturalistic settings in which models are not overtly prompted to “lie” but rather report false statements in the course of completing a different task.
[NLP-108] Flattening Fantasies: Analyzing Stereotypes in Pornographic Discourse
【速读】： 该论文旨在解决色情内容中关于身份维度（如种族、性别）的语言表述模式问题，特别是揭示其背后隐含的刻板印象与主流规范之间的偏离。研究基于对全球访问量第四的成人网站xHamster上540万条视频标题的大规模定量分析，系统考察了不同国籍、族裔群体及性别身份在色情话语中的表征方式。其解决方案的关键在于通过自然语言处理技术识别并量化这些话语中的结构性偏见，发现美国色情内容在行业内部占据主导地位，并成为默认标准；同时揭示出反映殖民权力结构的族裔刻板印象、宗教物化现象以及对文化禁忌的性化处理；此外，研究还指出跨文化语境下性别表征的差异，表现为女性普遍被置于“圣母-娼妓”二元对立框架中，而男性则多以身体优势角色呈现，凸显了深层的文化与社会意识形态投射。
链接: https://arxiv.org/abs/2610.25566
作者: Sarmad Chandio,Osama Khalid,Sharaf Zia,Ethan Kutlu
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:This study presents a large-scale quantitative analysis of 5.4 million video titles from xHamster, the world’s fourth most visited pornography platform, to examine linguistic patterns used to describe different dimensions of identities (e.g., race and gender). Specifically, we investigate how nationalities, ethnic groups, and gender identities are portrayed in pornographic discourse, focusing on stereotypes and deviations from implicit norms of pornography. The results show the dominance of American pornography over other groups, which acts as the de facto norm for the adult industry. We identify persistent ethnic stereotypes reflecting colonial power dynamics, religious fetishization, and the sexualization of cultural taboos. The analysis reveals gendered representational differences across cultures, with a prevalent Madonna-Whore dichotomy in the portrayal of women and stereotypical depictions of men in physically dominant roles. Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.25566 [cs.CL] (or arXiv:2610.25566v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.25566 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Related DOI: https://doi.org/10.1609/icwsm.v20i1.42646 Focus to learn more DOI(s) linking to related resources Submission history From: Osama Khalid [view email] [v1] Sun, 18 Oct 2026 22:01:44 UTC (2,107 KB) Full-text links: Access Paper: View a PDF of the paper titled Flattening Fantasies: Analyzing Stereotypes in Pornographic Discourse, by Sarmad Chandio and 3 other authorsView PDFHTML (experimental)TeX Source view license Additional Features Audio Summary Current browse context: cs.CL prev | next new | recent | 2026-10 Change to browse by: cs References Citations NASA ADSGoogle Scholar Semantic Scholar export BibTeX citation Loading… BibTeX formatted citation loading… Data provided by: Bookmark checked="checked"class=“labs-tab-input”> Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv’s community? Learn more about arXivLabs. Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?) mathjaxToggle(); We gratefully acknowledge support from our major funders, member institutions, , and all contributors. About Help Contact Subscribe Copyright Privacy Accessibility Operational Status (opens in new tab) Major funding support from
[NLP-109] Narrow Decision Models: A System-1 Architecture for Calibrated Typed Decisions
【速读】： 该论文旨在解决生成式AI在决策任务中缺乏可解释性、可验证性和系统性偏差控制的问题，尤其是在需要高可靠性与可审计性的专业场景（如司法文书路由）中的应用瓶颈。其核心挑战在于如何构建一种既具备精准判断能力、又能保证决策逻辑透明、可测试且对输入顺序不变的轻量化决策模型。解决方案的关键是提出一种窄域决策模型（Narrow Decision Model, NDM），该模型以类型化接口（Choice、Score、Noul）统一处理决策任务，通过在特定组织上下文（如法院文书路由）中训练，实现小规模部署于本地基础设施的能力。其架构创新在于采用前缀树（prefix tree）结构序列化输入，并结合双向编码器与块注意力掩码（block attention mask）及逻辑位置编码，使每个答案在构造上天然满足选项顺序无关性（invariance to option order）和多任务独立性；同时引入基于合理评分规则（proper scoring rules）的课程学习、校准与弃权机制，支持由规则驱动的复合决策（rubrics-based composition），并设计预审协议以在未见数据开放前评估机制有效性。实验表明，该模型在保持上述数学性质的前提下，相较传统成对交叉编码器（pairwise cross-encoder）在同等计算量下达到更低损失，且在真实法律文档上的决策一致性达91%，显著优于仅依赖词汇检索的方法。
链接: https://arxiv.org/abs/2610.25553
作者: Carlos André Antunes,Aldry Albuquerque,Gabriel de Melo Schmith,Natã Batista,Pedro Oziel Escobar da Silva,Stefano Benko Rossi,Lingrid Dietrich Lorenz da Silva,Flávia Cristina da Costa,Thiago Bertotto,Luis Fernando Fausto
机构: Softplan
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 27 pages, 6 figures, 4 tables. A first training on real documents is summarized; full results on real data in production are left to a companion paper
Abstract:System One models, introduced with TypeSafe’s Jev, answer typed questions about a state with calibrated probabilities instead of generated text: choosing among options described in natural language (Choice), placing the state on an ordered scale (Score), or judging a yes/no statement (Noul). Jev is general-purpose: one set of weights, served as an API and adapted to a domain only through the request. We study the other end of the design space. A Narrow Decision Model (NDM) offers the same interface but is trained for the decisions of one organization, is small enough to run on its own infrastructure, and has an open architecture whose properties can be stated and tested. The state can be any text or structured record about which decisions are made; our running case is the routing of court communications. The architecture serializes a request as a prefix tree (the state, then each question, then each option) and runs a bidirectional encoder under a block attention mask with logical positions, so that each answer is, by construction, invariant to the order of its options and independent of the other questions asked. Routing through an organization chart is a composition of Choice decisions with an explicit “none of these” option. We describe training with proper scoring rules through a curriculum, calibration and abstention, decisions composed by rubrics, and a protocol that accepts or rejects the mechanism before held-out data is opened. A minimal validation on public and fictitious data shows that the invariances hold to rounding in a trained model, that a small instance routes on organization charts it has never seen well above lexical retrieval, and that, at equal compute, the block architecture reaches a lower loss than a pairwise cross-encoder. A first instance trained on real legal documents agrees with the LLM classifier in production on 91% of held-out documents.
[NLP-110] EmotiQ: A Quantum Residual Specialist Framework for Text Emotion Classification ICTAI2026
【速读】： 该论文旨在解决文本情感分类中因细微语言信号和语义边界重叠导致的困难情感判别问题。其核心挑战在于传统模型在处理边界模糊或语义相近的情感类别时表现不佳。解决方案的关键在于提出EmotiQ框架，即一种基于冻结的Transformer分类器作为主决策者、并引入多种可插拔的残差专家模块（residual specialists）的协同机制。这些专家模块包括量子编码、数据重上传、原型与保真度专家以及基于核函数的专家，仅在满足基于验证集行为的准入标准时才介入决策过程，从而实现对复杂情感边界的局部优化。实验表明，该策略在六个基准数据集上均优于传统基线模型，尤其在量子组件中，ZZ相互作用特征映射表现出最稳定的提升效果（平均宏F1提升+0.0030，在四个数据集上取得正向增益），证明当前量子自然语言处理方法可在强经典监督下作为残差专家，提供精准且可控的局部性能改进。
链接: https://arxiv.org/abs/2610.25549
作者: Carlton Kigonya,Tanaka Chirara,Yunting Yin
机构: Earlham College (厄尔汉姆学院); Eastern Michigan University (东密歇根大学)
类目: Computation and Language (cs.CL)
备注: Accepted to IEEE ICTAI 2026
Abstract:Emotion classification from text is challenging because many affective states are distinguished by subtle linguistic signals and semantically overlapping boundaries. We investigate whether compact quantum modules can serve as targeted residual specialists that assist traditional classifiers on difficult emotion decisions. We introduce EmotiQ, a residual specialist framework for text emotion classification using a frozen transformer classifier as the primary decision maker, identifies challenging emotion boundaries from validation behavior, and trains boundary specialists, including quantum encodings, data re-uploading specialists, prototype and fidelity specialists, and kernel-based specialists. Specialists are permitted to intervene only when they satisfy validation-based admission criteria. We evaluate the framework across six text emotion classification datasets and establish traditional model baselines. Experimental results show that the admitted EmotiQ policy improves mean macro-F1 metrics over the baseline, and achieves positive gains on five of six datasets. Among the quantum components, ZZ interaction feature maps are the most consistently effective, yielding a mean macro-F1 improvement of +0.0030 and positive gains on four datasets. These findings suggest that current quantum NLP methods can be deployed as residual specialists that provide localized improvements under strong classical supervision.
[NLP-111] Auditing Tool Traces in Agent ic LLM s: Source Localization and Counterfactual Dependence Testing NEURIPS2026
【速读】： 该论文旨在解决生成式大语言模型（Generative LLM）在生物科学预测任务中可能因隐藏的标签捷径（label shortcuts）而产生虚假性能提升的问题。尽管模型通过整合实验与计算证据表现出高精度预测，但其决策过程可能依赖于数据集标识符或模板化文本等非生物学意义的线索，而非真正的科学推理。为此，论文提出一种两阶段审计方法：第一阶段为单工具TF-IDF分解（Per-Tool TF-IDF Decomposition, PT2D），将标签预测信号精确定位至各工具输出结果，避免多步工具调用序列中的信号稀释；第二阶段为词元类别擦除（Token-Class Redaction），通过反事实消融测试验证冻结的LLM是否依赖于已定位的捷径。在T细胞受体-肽段-MHC复合物（TCR-pMHC）结合亲和力预测任务中，该审计揭示模型最强信号来源于某一工具输出中的数据集标识符，移除后模型性能降至随机水平。研究结果表明，隐藏的标签捷径可伪装成科学推理，强调在信任基准性能提升前必须进行源级审计。
链接: https://arxiv.org/abs/2610.25540
作者: Junhao Liu,Tianxiao Li,Yongxian Wu,Ziheng Duan,Ren Li,Jing Zhang,Martin Renqiang Min
机构: University of California, Irvine (加州大学欧文分校); NEC Labs America (NEC美国实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 12 pages. Accepted at the NeurIPS 2026 Workshop on Agentic AI for Biological Discovery (AgenticLS)
Abstract:Agentic LLMs increasingly support scientific prediction by grounding decisions in external evidence. In biology, they promise accurate, interpretable predictions for important problems by integrating complex experimental and computational evidence. However, serialized tool results can contain hidden label shortcuts, such as dataset identifiers or template artifacts, that create apparent success without biological reasoning. To address this risk, we introduce a two-stage audit. Stage 1, Per-Tool TF-IDF Decomposition (PT2D), localizes label-predictive patterns to individual tool results while avoiding signal dilution in the concatenated tool trace. Stage 2, Token-Class Redaction, counterfactually tests whether the frozen LLM depends on the localized shortcut. We apply this audit to TCR-pMHC binding, where tool augmentation substantially outperforms a no-evidence baseline. At face value, this suggests that the LLM has learned to use scientific evidence. The audit overturns this interpretation: PT2D localizes the strongest signal to identifiers in one tool result, and redacting them reduces performance to chance. These findings show how hidden label shortcuts can masquerade as scientific reasoning or discovery, motivating source-level audits before benchmark gains are trusted.
[NLP-112] ATEST: Adaptive Resource-Aware Test-Time Scaling for LLM Inference
【速读】： 该论文旨在解决生成式 AI（Generative AI）在测试时扩展（Test-time scaling, TTS）过程中因固定采样预算导致的资源利用效率低下问题，尤其在GPU负载动态变化时，忽视了查询难度与实时推理能耗之间的耦合关系。其核心挑战在于：传统TTS方法采用固定的样本数量，无法根据当前系统负载自适应调整采样策略，从而在高负载下造成能源消耗激增（实测最高达4.1倍）且性能下降。本文提出的ATEST是一种资源感知型TTS框架，其关键创新在于联合优化“额外采样的预期收益”与“当前GPU环境下的实际能耗成本”。其中，ATEST-Online模式通过在线增量生成样本、实时评估多数答案的稳定性、监测动态GPU状态并在线学习每步能耗成本，实现无需离线校准的自适应停止决策；而ATEST-Calibrated则依赖离线构建的准确率与成本模型为每个查询设定预算，适用于负载稳定的场景。实验表明，在NVIDIA A100上对Phi-3-mini进行多数据集评估，ATEST-Online在所有负载条件下均未超出能量预算，同时保持最高或统计等效的准确率，并随负载上升自动减少采样量；在高负载下GSM8K任务中，其准确率达60.0%且无延迟请求，显著优于ATEST-Calibrated的44.0%准确率和42.4%延迟率。消融实验进一步验证了在线能耗学习是适应负载变化的关键机制。综合来看，结合离线预标定的ATEST-Online在资源效率上较ATEST-Calibrated降低42倍GPU时间和48倍能耗，展现出更强的动态鲁棒性，尽管在短任务上略增尾部延迟、降低吞吐率，但在负载波动环境下具备显著优势。
链接: https://arxiv.org/abs/2610.25519
作者: Mobina Kashaniyan,Ali Jannesari
机构: Iowa State University (爱荷华州立大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF)
备注: 12 pages, 4 figures, 8 tables
Abstract:Test-time scaling (TTS) improves large language model (LLM) reasoning by sampling multiple answers, but fixed sampling budgets ignore both query difficulty and the changing energy cost of inference under GPU contention. We measure up to a 4.1-fold increase in energy for the same four-sample query as GPU load rises. We present ATEST, a resource-aware TTS framework that jointly considers the expected value of additional samples and their current GPU cost. ATEST-Online, its default mode, requires no offline calibration: it incrementally generates samples, estimates the stability of the plurality answer, monitors live GPU conditions, and learns step energy cost online to decide when to stop. ATEST-Calibrated instead selects a per-query budget using offline accuracy and cost models, favoring deployments with stable loads. We evaluate both modes on four reasoning and question-answering datasets using Phi-3-mini on an NVIDIA A100 under controlled varying load, and test transfer to an NVIDIA L40S and Qwen2.5-3B. Across the four A100 datasets, ATEST-Online answers every query without exceeding the energy budget, while achieving the highest or statistically tied accuracy and reducing sampling as load increases. On GSM8K at high load, it achieves 60.0% accuracy with no deferrals, versus 44.0% accuracy and 42.4% deferrals for ATEST-Calibrated. Ablations show that online cost learning drives adaptation to load. Including offline profiling, ATEST-Online uses 42 times less GPU time and 48 times less energy than ATEST-Calibrated for the evaluated workload. Online control trades higher tail latency and lower throughput on short tasks for robustness under varying load; offline calibration remains competitive under stable conditions.
[NLP-113] Where Are Language-Specific Features Represented in Multilingual LLM s? Everywhere
【速读】： 该论文旨在解决多语言大语言模型（Multilingual Large Language Models, LLMs）中关于语言特异性与语言无关表示在不同网络层分布的既有认知问题。传统观点认为，语言特异性特征主要集中在模型的初始和最终层，而中间层则支持跨语言的通用表示。然而，本文通过四项实验——包括对葡萄牙语与西班牙语（两种密切相关语言）的细粒度表征分析，以及覆盖200多种语言的大规模语言识别实验——重新审视了这一假设。以Llama 3.2 1B为测试案例，研究发现语言身份信息在整个模型深度中仍具有高度可恢复性，并未局限于初始和最终层，从而挑战了语言特异性与语言无关表示严格分层的范式。其核心解决方案在于通过对上下文句向量（contextual sentence embeddings）的分析与操控，揭示了语言身份信息在深层网络中持续存在且可被有效提取。此外，研究还发现模型对模糊字符串的分类存在偏向性，更倾向于将其判为西班牙语而非葡萄牙语，反映出模型行为与人类判断之间的系统性偏差，提示当前多语言模型在语言间细微差异处理上存在潜在偏见。
链接: https://arxiv.org/abs/2610.25505
作者: Mikaela Martins,Jonathan Dunn
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:Previous work has argued that multilingual LLMs encode language-specific features most strongly in their initial and final layers, implying that intermediate layers support language-agnostic representations. This paper reconsiders this finding by performing four different experiments: three that undertake a fine-grained analysis of representations of Portuguese and Spanish, two closely related languages, and one large-scale language identification experiment that covers over 200 languages. The results of these experiments challenge the strict separation between language-specific and language-agnostic layers in multilingual LLMs, using Llama 3.2 1B as a test case. By analyzing and manipulating contextual sentence embeddings, we demonstrate that language identity remains strongly recoverable throughout the model rather than being restricted to initial and final layers. Moreover, we find that the model tends to classify ambiguous strings as closer to Spanish than to Portuguese, revealing a skew in model behavior that differs from human judgments.
[NLP-114] From Seeing to Reporting: A Theory of Reliable Visual Set Generation in Vision-Language Models
【速读】： 该论文旨在解决生成式视觉语言模型（Vision-Language Models, VLMs）在密集场景中进行视觉报告时存在的准确性问题，尤其是当模型需完成计数、目标检测、文档布局解析及图形用户界面（GUI）定位等任务时，其输出报告常出现重复、遗漏或错误匹配等问题。核心挑战在于，现有模型将“报告”视为对感知结果的格式化表达，而非与视觉理解紧密耦合的推理过程，导致报告阶段引入显著误差。本文提出“从看见到报告”（From Seeing to Reporting, FSR）理论框架及其一系列推理时（inference-time）的优化方法，关键在于将报告建模为一个由响应到解码记录、集合构建及任务答案组成的可追溯链式结构，并通过任务兼容的匹配机制建立精确的逻辑关系：包括记录删除、坐标平均与读出切换的规则，以及将多轮调用中的错误分解为“遗漏候选”、“代表性丢失”和“错误选择”三类。该框架无需目标任务标注即可利用任务描述和预设配置实现性能提升。实验表明，在50个密集合成场景中，单次调用的报告-解码协议使GPT-6.1的类型计数误差降低85.2%，Claude Sonnet 5.5降低98.8%；在ScreenSpot-Pro数据集上，结合三模型的GUI定位层级将漏检率从GPT-6.1双阶段缩放的13.67%降至9.33%，优于同等成本下三阶段缩放的12.67%和简单投票的12.33%，证明了其他模型提供的候选信息与智能选择机制共同提升了最终报告质量。
链接: https://arxiv.org/abs/2610.25503
作者: Junzhe Shi,Yuan Gan,Shida Jiang
机构: University of California, Berkeley (加州大学伯克利分校); Quotr AI; Emeryville, CA, USA
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 56 pages, 6 figures. Code, prompts, configuration and all 67,364 stored model responses: this https URL
Abstract:Vision-language models (VLMs) are asked to report what they see: to count objects, box them, parse a screen or pick a click target (GUI grounding). We argue that reporting is part of the visual task, not merely its formatting. In a dense scene with 173 shapes, GPT-6.1 lists 173 records that match every shape, yet the counts it states in the same response sum to 188. From Seeing to Reporting (FSR) is a theory of visual reporting and a set of inference-time procedures that improve the reports of frozen VLMs. It models reporting as a chain from responses to decoded records, constructed sets and task answers, linked by task-compatible matching. The chain gives exact relations for deleting records, averaging their coordinates and switching readouts, and it splits the error of one answer chosen among several calls into missing candidates, lost representatives and wrong choices. These relations guide procedures for counting, detection, document layout, screen parsing and GUI grounding that use the task specification and a shipped configuration, without requiring labels from the target task. On 50 dense synthetic scenes, a one-call reporting-and-decoding protocol cuts per-type count error by 85.2% for GPT-6.1 and 98.8% for Claude Sonnet 5.5. On 600 ScreenSpot-Pro items, a GUI-grounding tier that combines three models lowers the miss rate from 13.67% for GPT-6.1’s two-stage zoom to 9.33%, against 12.67% for three such zooms combined by a cluster vote at about the same cost. A plain vote over the same three models misses 12.33%: the other models supply the candidates, and selecting among them turns those candidates into fewer misses. We release the protocols, the stored responses and the analysis code.
[NLP-115] Self-Distillation for Supervised Fine-Tuning
【速读】： 该论文旨在解决监督微调（Supervised Fine-Tuning, SFT）中将专家示范或最优响应视为独热（one-hot）目标所导致的问题，即抑制了模型本可赋予合理概率的其他有效延续路径，从而限制了生成多样性与泛化能力。其核心解决方案是提出金标准条件蒸馏（Gold-Conditioned Distillation, GCD），关键在于利用同一模型同时作为教师和学生：教师基于包含黄金响应（gold response）上下文的提示进行条件生成，而学生则仅依赖原始提示进行推理；训练过程中，学生通过最小化与教师在每个黄金响应标记处的前向KL散度来学习，从而实现对教师输出分布的软目标拟合。该方法使模型能够利用自身先验知识自主学习新任务，其中正确性由黄金响应锚定，而软目标则保留了合理的替代生成路径，显著提升了新任务准确率，并普遍降低了在通用基准上的性能退化。
链接: https://arxiv.org/abs/2610.25497
作者: Wenxuan Zhu,Bang An,Shuming Liu,Wenqing Cui,Tianqi Xu,Bing Li,Yibo Yang,Peter Wonka,Bernard Ghanem
机构: King Abdullah University of Science and Technology(阿卜杜拉国王科技大学); Shanghai Jiao Tong University(上海交通大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Supervised fine-tuning treats the expert demonstration or gold response as a one-hot target, suppressing valid alternative continuations that the model would otherwise assign probability to. We propose Gold-Conditioned Distillation (GCD), a method that turns expert demonstrations into model-native supervision. GCD uses one model as both teacher and student: the teacher conditions on a prompt containing the gold response in context, the student conditions on the original prompt as at inference, and the student is trained to minimize the forward KL divergence to the teacher’s next-token distribution at each gold response token. This lets the model use its prior knowledge to teach itself the new task: correctness is anchored by the gold response, while the soft target preserves plausible alternatives. Across various training tasks on Qwen and Llama models, GCD improves new-task accuracy over SFT and generally reduces degradation on general benchmarks.
[NLP-116] A failure of Imagination: VLMs fail to use visual representations for text but self-distillation can help
【速读】： 该论文旨在解决生成式视觉-语言（Vision-Language, VL）模型在经过视觉问答（VQA）任务微调后，是否真正实现了对文本表示的“强多模态”对齐问题，即语言模型的文本表征是否因与视觉信息对齐而获得潜在的视觉结构感知能力。现有方法通常仅实现“弱多模态”，即模型能基于输入图像回答问题，但并未改变其文本表征的本质。为验证是否存在“强多模态”，研究构建了一个形状计数任务，包含成对等价的文本（线段坐标）和图像输入；当提供图像时任务简单，而纯文本形式则需模型自主进行可视化推理。实验结果表明，尽管视觉对齐模型在图像版本任务中表现优异，但在纯文本版本中并未展现出优于纯文本模型的表现，且探针实验显示其解码器嵌入中并无显著增强的视觉结构信息。这说明传统VQA式对齐未能有效促进跨模态知识向文本表征的迁移。为此，论文提出一种优化目标，通过将上下文中的视觉嵌入蒸馏至文本嵌入中，实现隐式的跨模态对齐。该方法在无显式任务监督的情况下，显著提升了复杂文本任务的性能，表明当前主流的VQA对齐策略仍存在未被充分挖掘的跨模态迁移潜力。
链接: https://arxiv.org/abs/2610.25470
作者: Shashwat Singh,He He,Tal Linzen
机构: New York University(纽约大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-language (VL) models are typically built by fine-tuning a language model to attend to visual features in service of a Visual Question-Answering objective. We ask whether this alignment changes how language models represent text—especially on tasks where visualization would help, like geometry questions posed in words. We distinguish weak multimodality (the model can answer questions about an input image) from strong multimodality (the model’s text representations themselves become vision-aligned, improving performance on text-only tasks with latent visual structure). To test for strong multimodality, we construct a task of shape-counting problems with paired equivalent text and image specifications. This task is trivial when the image is provided, but requires visualization when specified in text as line coordinates. We compare instruction-tuned Qwen-2.5 and Mistral models against their vision-aligned counterparts on the text variant of the task. We find that the multimodal models do no better in this setting, despite strong performance on the image version of the task. Probing experiments corroborate this: vision-aligned decoder embeddings carry no more visual structure than their text-only counterparts. Finally, we provide an optimization objective that distills in-context visual embeddings into text embeddings. This alignment induces gains on the text variant of the task without explicit task supervision, even on samples more complex than seen during alignment. This suggests that VQA-style alignment leaves available cross-modal transfer untapped.
[NLP-117] DiaLong: Benchmarking Reminiscence in Long-Context Dialogues
【速读】： 该论文旨在解决生成式 AI（Generative AI）在长对话上下文中记忆保持与信息召回能力不足的问题。现有大语言模型（Large Language Models, LLMs）在处理长上下文时普遍存在对远距离事实信息的遗忘现象，但缺乏系统性、量化的评估基准来全面衡量这一缺陷。为此，本文提出 DiaLong——一个基于人类生成并经大模型扩展的长对话数据集，其对话长度显著超越现有数据集，同时配套设计了以事实回忆为核心的记忆任务：从40个候选事实中判断其在对话中的真假。该任务直接考察模型在长时间对话中主动保留与检索信息的能力，是衡量其长期记忆性能的关键指标。实验结果表明，即使在最新的LLMs及已提出的记忆增强方案下，模型在长对话记忆方面仍存在明显不足，亟需进一步优化。提升记忆能力将有助于增强模型在持续对话中的连贯性、相关性与类人交互表现，从而更好地满足用户在多样化对话场景下的需求。
链接: https://arxiv.org/abs/2610.25463
作者: Owen Bianchi,Samer Aslan,Ryunosuke Saito
机构: Center for Language and Speech Processing, Johns Hopkins University (约翰霍普金斯大学语言与语音处理中心)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 17 pages. Work conducted in 2023 at Johns Hopkins University
Abstract:How well can transformer-based Large Language Models (LLMs) retain and recall memories in long contexts of dialogue? LLMs are known to forget long-range factual information in their context, yet the field lacks a comprehensive benchmark to quantitatively assess these shortcomings. We present DiaLong, a long-context dialogue dataset and accompanying memory task that standardizes evaluation in this space. It consists of 100 unique conversations created by humans and then augmented with LLMs, extending beyond lengths found in existing datasets. Our selected memory task is fact recall: classify 40 candidate facts about a dialogue as true or false. This task directly probes the model’s ability to actively retain and retrieve information, serving as a critical measure of memory within extended conversational contexts. In testing both the latest LLMs and proposed memory enhancements on DiaLong, we demonstrate that more work needs to be done to enable LLMs with better memory performance. By enhancing memory capabilities, LLMs can improve in maintaining coherence, relevance, and human-like interaction over extended conversation. Improving these capabilities will help LLMs fulfill diverse user needs and expectations in various conversational scenarios.
[NLP-118] Quality-Controlled Reply and Summary Data for Post-Training Small Language Models in AI Call-Screening Agents
【速读】： 该论文旨在解决小型语言模型（Small Language Models, SLMs）在本地化语音通话处理中难以同时实现准确应答与高质量摘要生成的问题。具体而言，现有小型模型无法在通话过程中合理代表用户权限进行回应，且事后生成的摘要常遗漏关键情境信息或错误地将来电者陈述当作事实，导致隐私保护与实用性之间的矛盾。由于真实通话缺乏可监督的参考应答与摘要，研究构建了一种基于信息边界约束的合成通话数据生成机制：通过分角色设计（通话发起方撰写、应答方仅依据当前对话上下文生成回复、摘要方仅基于完整通话记录生成总结），并引入基于模型的多轮评审以筛选高质量样本，最终形成包含3,395通通话、共15,102个标注目标（11,707个应答目标，3,395个摘要目标）的训练语料。其核心解决方案是采用单一共享的低秩适配器（LoRA）对多个参数规模（0.6–4B）的小型模型进行联合微调，在192条保留测试集上评估显示，应答质量从21–54%提升至94–97%，摘要质量从0–6%提升至43–72%，所有12项任务-模型组合的增益均在置信区间内显著为正，且摘要性能提升在各模型家族中均成立。尽管微调后连贯性与自然度问题基本消除，剩余挑战聚焦于“接地性”（grounding）与“有用性”——即忠实还原发言者身份及其所处条件的能力。
链接: https://arxiv.org/abs/2610.25458
作者: Jiaqi Gan,Tsang Ng,Ankit Raj,Tommy Duong,Xingyu Shen,Kidus Zewde,Huaiyu Guo,Simiao Ren
机构: Scam.ai
类目: Computation and Language (cs.CL)
备注: 17 pages, 3 figures, 12 tables
Abstract:Small language models have improved to the point where a phone can plausibly screen its own calls, answering on an absent owner’s behalf without sending call content off the device, a privacy-respecting alternative to cloud call handling. Realizing that promise takes a model that can do two things well: reply within the owner’s authority during the call, and afterward tell the owner what happened and what needs attention. Released small models do neither reliably - they raise unwarranted alarms about ordinary callers and write summaries that drop material conditions or report a caller’s claims as fact - and real calls supply no reference replies or summaries to learn from. We construct supervision for both tasks from synthetic calls under an information-boundary constraint: a caller writer works from a private scenario brief, a reply writer sees only the conversation up to the current turn, and a summary writer sees the completed transcript. Separate model-based reviews of the full dialogue and the summary decide which calls enter training, yielding a corpus of 3,395 calls with 15,102 targets (11,707 reply, 3,395 summary). Training a single shared LoRA adapter on both tasks, we evaluate six models (0.6-4B parameters) before and after adaptation on 192 held-out calls. Reply qualification rises from 21-54% to 94-97% and summary qualification from 0-6% to 43-72%; all twelve paired gains (six models x two tasks) have bootstrap intervals above zero, and the summary gains hold in every source family. Because one model both screened the training data and scored the outputs, these rates measure agreement with that evaluator’s standard rather than human acceptance, which a planned human study will test. After adaptation, coherence and naturalness failures nearly vanish, leaving grounding and usefulness - faithfully tracking who said what, under which conditions - as the remaining challenge.
[NLP-119] System X: An LLM Decision Model That Replies from Its Own KV Cache
【速读】： 该论文旨在解决大语言模型（LLM）在部署应用中仅输出概率分数而无法生成自然语言回复的问题，即如何在保持决策性能的同时实现文本化响应。其核心解决方案是提出System X，一种基于大语言模型的决策框架，通过在原有“LLM作为评判器”（LLM-as-Judge）的输出读取机制中引入文本回复选项，并利用决策过程中的键值缓存（KV cache）复用技术，在文本回复胜出或置信度较低时直接从已有缓存继续生成，避免重复进行提示前缀填充（prompt prefill），从而显著降低延迟。实验表明，该方法在维持决策准确率基本不变的前提下，将预期校准误差从0.292降至0.169，且在独立测试集中达到98.3%的路由准确率；在Apple M4 Pro上实现15%-37%的延迟节省，同时保证了与完整序列生成一致的输出概率（差异小于1.4e-6）。进一步优化版本System X v4采用更大规模模型Qwen3-1.7B、重新平衡数据及教师重写回复目标，使JevBench准确率提升至61.0%（p=0.04），延迟降低17%-39%，且人工评估更偏好其生成内容。然而，状态冲突和不支持的回复细节仍是主要失败模式，研究建议未来结合多种子实验、人类评估与真实业务数据开展深入验证。
链接: https://arxiv.org/abs/2610.25456
作者: Syed Usama Khalid Bukhari,Syed Safwan Khalid,Muhammad Aamir
机构: Turing Logic LTD; University of Oxford (牛津大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 25 pages, 5 figures, 10 tables
Abstract:Decision models return probabilities over supplied options, but a deployed agent may also need to reply in words. The present study introduces System X, a large language model (LLM) decision model that adds a text-reply option to the LLM-as-Jev readout. When this option wins, or confidence is low, the same model continues from the decision’s key-value (KV) cache, avoiding a repeated prompt prefill. Qwen3-0.6B has been fine-tuned with LoRA on Reason-9k and Bitext reply rows. JevBench accuracy shows no statistically significant change (55.4% to 54.5%, p=0.87), while expected calibration error falls from 0.292 to 0.169. Held-out Bitext routing accuracy reaches 98.3%. Cache reuse saves 15-37% of latency on an Apple M4 Pro, with token-identical replies in the FP32 demonstration cases. Scoring options on copies of the prompt cache matches full-sequence scoring to within 1.4e-6 in probability. Adding dialogue, industry and complaint data (System X+) improves support-task results but lowers IFEval accuracy from 55.5% to 45.0%. System X v4 uses Qwen3-1.7B, rebalanced data and teacher-rewritten reply targets. Against its base LLM, JevBench accuracy rises from 55.0% to 61.0% (p=0.04), with no statistically significant IFEval change (p=0.71). A matched latency test on unadapted Qwen3-1.7B saves 17-39%. One LLM judge prefers v4 replies to an unadapted 1.7B second call (31 to 11 pairs, p=0.003); the other’s preference is not significant. These reply comparisons also differ in prompts and generation limits. State conflicts and unsupported reply details remain failure modes. Further studies with several seeds, human raters and real support data are recommended.
[NLP-120] Enhanced Masked Diffusion Unlearning
【速读】： 该论文旨在解决生成式语言模型在实际部署中面临的关键问题——机器遗忘（machine unlearning），即在不从头训练的前提下，有效消除模型对特定“遗忘集”（forget set）内容的记忆，以确保敏感、受版权保护或受限信息不再被生成，同时尽可能保留模型的原有功能。现有方法Masked Diffusion Unlearning（MDU）通过最小化模型输出分布 $ p_\theta $ 与冻结的提示-掩码锚点分布 $ p_\mathrmanchor $ 之间的锚点后KL散度 $ \mathrm{KL}(p_\theta \Vert p_\mathrmanchor) $ 实现遗忘，但存在两个关键缺陷：其一，随机掩码策略可能导致关键遗忘词元在问题完全掩码时仍被锚点可见；其二，在记忆性强的近似单热（near-one-hot）预测场景下，锚点后梯度消失而锚点前梯度仍存在，导致训练停滞于最需遗忘的词元上。针对上述问题，本文提出增强型掩码扩散遗忘方法（Enhanced Masked Diffusion Unlearning, EMDU），其核心创新在于采用锚点前KL散度 $ \mathrm{KL}(p_\mathrmanchor \Vert p_\theta) $，并强制对锚点的完整答案跨度进行掩码，从而克服梯度消失问题。理论与实证分析表明，EMDU在TOFU和RWKU基准测试中，相较于MDU及适配的自回归基线方法，在遗忘效果上达到或超越现有水平，同时显著提升了保留能力（retain utility）的保持性能。
链接: https://arxiv.org/abs/2610.25447
作者: Tomer Garber,Tom Tirer
机构: Bar-Ilan University (巴伊兰大学); Red Hat
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Masked diffusion language models (MDLMs) generate text by iteratively unmasking tokens with bidirectional context. As these models reach real-world deployment, machine unlearning becomes essential. This process aims to remove the influence of a designated “forget set” without retraining from scratch, so that sensitive, copyrighted, or restricted content can no longer be elicited while preserving retained capabilities. Masked Diffusion Unlearning (MDU) is the first publicly available unlearning method customized for these models: it tunes the model p_\theta to match a frozen prompt-masked anchor p_\mathrmanchor on the forget set by minimizing the anchor-last divergence \mathrmKL(p_\theta \Vert p_\mathrmanchor) . We identify two failure modes of MDU. First, random answer masking can leave key forget tokens visible to the anchor even when the question is fully masked. Second, at a memorized (near-one-hot) prediction, the anchor-last logit gradient vanishes while the anchor-first divergence \mathrmKL(p_\mathrmanchor \Vert p_\theta) logit gradient does not. As a result, MDU training stalls on the tokens we most need to forget. Backed by theoretical and empirical analyses, we propose Enhanced Masked Diffusion Unlearning (EMDU): anchor-first KL divergence toward a frozen anchor whose entire answer span is masked. On TOFU and RWKU benchmarks using LLaDA-8B and Dream-7B, our EMDU matches or exceeds the forgetting of MDU and adapted AR baselines, while better preserving retain utility.
[NLP-121] Design Considerations in Low-Resource Archival OCR: A Case Study in Lakota
【速读】： 该论文旨在解决在资源匮乏条件下，如何有效支持拉科塔语（Lakota）等北美原住民语言的数字档案重建与语言复兴问题。核心挑战在于：针对低资源语言的光学字符识别（OCR）工具在实际应用中面临训练预算有限、文本分割噪声、扫描质量退化及拼写变异性等多重现实约束。其解决方案的关键在于利用小型视觉-语言模型（VLMs，≤3B参数量）通过微调实现高效且精准的OCR，尤其在仅有少量标注数据（如仅9页）的情况下仍能保持较低字符错误率（CER）。研究发现，基于35页转录数据进行后校正微调可将CER从17.1%降至3.9%，而仅用9页训练时CER仅上升1.1%，表明小规模数据下的模型适应性较强；然而，在模拟真实自动分割场景下，最佳模型的CER几乎翻倍，凸显了在设计语言技术工具时必须充分考虑实际使用环境的重要性。该工作强调了以社区实践和现实约束为导向的语言技术开发路径，为未来原住民语言复兴中的生成式AI应用提供了重要范式参考。
链接: https://arxiv.org/abs/2610.25440
作者: Téa Wright,Laura X. Ma,Claudia Iron Hawk,Hannah J. Haynie,Alexis Palmer,Alane Suhr,Abteen Ebrahimi,Katharina von der Wense
机构: University of California, Berkeley (加州大学伯克利分校); University of Colorado, Boulder (科罗拉多大学博尔德分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:Building and expanding access to digital archives supports language reclamation and creates lasting resources for community revitalization. Designing effective tools to support these efforts requires taking into account the community-, language-, and document-specific context. We investigate how small ( \leq 3B) vision-language models (VLMs) can be adapted as optical character recognition (OCR) tools for Lakota, a North American Indigenous language. In addition to benchmarking modeling approaches, we inspect four practical facets of low-resource archival OCR that arise in previous work and conversations with community members: constrained training budgets, segmentation noise, scan degradation, and orthographic variance. We demonstrate that fine-tuning VLMs for OCR post-correction on 35 pages of transcribed data decreases character error rate (CER) from 17.1 to 3.9% on a bilingual corpus, and find that training only with nine pages increases CER by only 1.1% over this. However, evaluating under realistic automatic segmentation nearly doubles the CER of the best-performing models, stressing the importance of taking realistic use cases into account. We hope this work encourages future research on developing language technologies grounded in the workflows and constraints of ongoing revitalization efforts.
[NLP-122] RARE: Reinforcement Learning with Adaptive Rubric Evolution for Open-Ended Generation
【速读】： 该论文旨在解决开放性生成任务中缺乏自动质量验证、需同时满足多项约束条件以及存在多种有效响应的问题，尤其关注在训练过程中如何获取对不同有效输出中各需求满足度的反馈。现有基于评分标准（rubric-based）的强化学习方法虽能提供多维度反馈，但随着学习进程推进，部分标准因被完全满足而丧失区分能力，导致反馈效率下降；尽管动态调整策略可通过更新或选择新标准恢复区分度，却可能误删仍具价值的标准。为此，本文提出自适应评分标准演化强化学习（Reinforcement Learning with Adaptive Rubric Evolution, RARE），其核心在于利用当前策略产生的反馈来指导哪些标准应被替换以及新标准应评估的内容。具体而言，样本级更新机制将当前输出中已完全满足的标准替换为检测剩余质量差异的检查项，从而保留所有其他有效标准与评分，维持评价体系的稳定性。在视频脚本生成任务中，通过针对性地识别并惩罚模型缺陷（targeting flaws）的方法表现最优，在三种视频类型上相较静态评分标准平均提升3.82分，且模型与人类偏好均支持该改进效果。此外，该方法在写作及通用生成基准任务中亦展现出显著性能提升。
链接: https://arxiv.org/abs/2610.25434
作者: Xueyun Tian,Bingbing Xu,Chuyue Li,Jiaming Liu,Chao Zhang,Weijia Feng,MingHua Ma,Yu Zhao,Jingchu Wang,Ruihua Huang,Yuanzhuo Wang,Huawei Shen
机构: State Key Laboratory of AI Safety, Institute of Computing Technology, CAS(中国科学院计算技术研究所人工智能安全国家重点实验室); University of Chinese Academy of Sciences(中国科学院大学); Qwen Business Unit of Alibaba(阿里巴巴通义实验室)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Verifiable rewards have advanced reinforcement learning in mathematics and code. However, many valuable open-ended tasks share three properties: no automatic verifier of overall quality, multiple requirements to satisfy jointly, and multiple valid responses to each query. Training therefore needs feedback on requirement satisfaction across different valid responses. Rubric-based reinforcement learning provides this feedback by aggregating scores from explicit criteria. As learning progresses unevenly, some criteria become fully satisfied and lose discrimination while others remain informative. Dynamic methods revise shared rubrics or select evolving criteria to restore discrimination, but may also change useful criteria. To address this, we propose Reinforcement Learning with Adaptive Rubric Evolution (RARE). The core idea is to use current policy feedback to guide both which criteria change and what their replacements assess. Sample-level updates replace criteria fully satisfied across current outputs with checks of remaining quality differences, retaining all other criteria and scores. Within this rule, we compare rewarding desirable qualities with penalizing observed flaws. Video script generation exemplifies these properties and supports content production. Its demands for coherent, executable narratives under constraints make it a challenging testbed. Targeting flaws performs best across three video types with a mean gain of 3.82 points over matched static rubrics. Model and human preferences support these gains. Improvements also extend to writing and general generation benchmarks.
[NLP-123] Mixture-of-Memories: Low-Overhead Language Modeling with Modular Knowledge Injection
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在参数中存储大量事实知识但难以控制、更新及在推理时可靠提取的问题。现有基于记忆增强的方法通常依赖单一辅助机制（如检索、参数更新或神经记忆），对具有不同知识需求的词元采用统一修正方式，缺乏灵活性与针对性。其解决方案的关键在于提出混合记忆框架（Mixture-of-Memories, MoM），通过将路由式神经记忆与可写入的事实记忆分离，实现异构记忆路径的协同：一方面引入冻结的基础语言模型，另一方面集成两种神经记忆专家——低延迟的隐状态修正的潜在记忆（latent memory）和提供词空间知识路径的词汇记忆（lexical memory）。此外，设计轻量级预算路由（Budget Router）根据不确定性与分歧度特征动态分配每词元的混合权重，取代传统固定插值策略，提升适应性。对于事后更新，外部显式记忆独立于路由模块存在，因预先训练的路由会自动抑制未更新记忆，新知识可直接追加而无需重新训练任何组件，并仅在答案级门控确认后启用。实验表明，MoM在语言建模、事实问答、更新分析与效率方面均显著提升了预测准确率与可更新性，同时保持较低的推理开销。
链接: https://arxiv.org/abs/2610.25425
作者: Yifan Lu,Rubin Wei,Ziyun Zou,Junhong Liang,Haolin Yang,Xinlin Zhuang,Abhijit Das,Adinath Madhavrao Dukre,Shadab Khan,Zhouhan Lin,Imran Razzak
机构: MBZUAI( Mohamed Bin Zayed University of Artificial Intelligence), Abu Dhabi, UAE; LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院), China; The Chinese University of Hong Kong(香港中文大学), Hong Kong; ADIA Lab( Abu Dhabi Intelligent Agents Lab), Abu Dhabi, UAE
类目: Computation and Language (cs.CL)
备注: 18 pages, 8 figures
Abstract:Large language models store substantial factual knowledge in their parameters, yet this knowledge is difficult to control, update, and reliably elicit at inference time. Existing memory-augmented approaches typically rely on a single auxiliary mechanism, such as retrieval, parameter updates, or neural memory, and therefore apply the same form of correction to tokens with very different knowledge demands. We propose Mixture-of-Memories (MoM), a memory-augmented language modeling framework that separates routed neural memory from writable factual memory. MoM combines a frozen base LM with two neural memory experts: a latent memory that provides low-latency hidden-state corrections and a lexical memory that provides a token-space knowledge pathway. A lightweight Budget Router sets per-token mixture weights from uncertainty and disagreement features, replacing the fixed interpolation of prior memory modules. For post-hoc updates, an external explicit memory stays outside the router, because a router trained before facts are written learns to down-weight it; new facts are appended without retraining any component and used only when an answer-level gate accepts the retrieved fact. Experiments across language modeling, factual QA, update analysis, and efficiency show that MoM improves both predictive accuracy and updateability while retaining low inference overhead.
[NLP-124] owards Figurative Language Understanding in Moroccan Darija: An Evaluation of Large Language Models
【速读】： 该论文旨在解决阿拉伯语方言（特别是摩洛哥方言，Darija）中隐喻语言在机器翻译（MT）中的处理难题，因为此类表达的语义往往超越字面意义，导致现有翻译系统表现不佳。尽管大型语言模型（LLM）在阿拉伯语方言翻译方面取得进展，但针对摩洛哥方言隐喻语言的评估基准资源仍十分匮乏。为此，研究提出“摩洛哥方言隐喻语言语料库”（Moroccan Darija Figurative Language Corpus, MDFC），这是一个涵盖习语、隐喻和谚语的多任务基准数据集，支持机器翻译与多项选择题（MCQ）任务的评估。其关键解决方案在于构建一个结构化、多样化的高质量语料库，并在此基础上对19个开源与闭源语言模型及机器翻译系统进行零样本（zero-shot）与六样本（six-shot）设置下的表达级与上下文级评估。此外，通过分层相似性探测分析，揭示了不同模型在表征层面区分隐喻与字面意义的能力差异。研究发现，闭源模型在翻译性能上表现最优，而上下文信息的增益因模型和任务类型而异；翻译准确率与MCQ识别准确率之间的差异进一步表明，生成恰当的隐喻译文与从候选译文中识别出正确语义是两个不同的认知挑战。
链接: https://arxiv.org/abs/2610.25406
作者: Jihad R’baiti,Youssef Hmamouche,Emmanuele Chersoni,Chu-Ren Huang,Amal El Fallah Seghrouchni
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Figurative language remains a major challenge for Machine Translation (MT) systems, as the intended meaning of such expressions often goes beyond literal translation. Although recent advances in Large Language Models (LLMs) improved the translation of Arabic dialects, benchmark resources for evaluating figurative language translation in Moroccan Darija remain scarce. To this end, we introduce the Moroccan Darija Figurative Language Corpus (MDFC), a benchmark dataset for MT and multiple-choice question (MCQ) selection covering idioms, metaphors, and proverbs. Using MDFC, we evaluate 19 open- and closed-weight language models and MT systems at both the expression and contextual levels under zero-shot and six-shot settings. We also conduct layer-wise similarity-based probing to determine how model representations distinguish figurative from literal meanings. Our findings reveal that closed-weight models achieve the strongest translation performance, while the benefits of contextual information vary across models and tasks. Differences between translation performance and MCQ accuracy highlight the distinction between generating an appropriate figurative translation and recognizing the intended meaning among candidate translations.
[NLP-125] From Unverifiable Domain Text to Cross-Lingual RLVR via Code-Switched Task Synthesis EMNLP2026
【速读】： 该论文旨在解决大语言模型在特定非英语领域中因数据不匹配而导致的推理能力不足问题，尤其是在缺乏可验证奖励（verifiable rewards）的跨语言任务数据的情况下。其核心挑战在于：尽管原始英文领域文本丰富，但目标语言下具备可验证奖励的强化学习任务稀缺，限制了模型在目标语言中的推理能力提升。解决方案的关键是提出“代码混用任务合成”（code-switched task synthesis），一种以数据为中心的方法，通过将原本不可验证的英文领域文本，利用填空式多选问答（fill-in-the-middle multiple-choice QA）作为种子任务格式，自动生成可验证的目标语言及代码混用版本的任务。这些生成任务保持原始证据与答案之间的逻辑关系不变，仅在强化学习过程中改变英文与目标语言证据的呈现方式，从而实现跨语言可验证奖励的学习。实验结果表明，在韩语医疗和四种目标语言的网络安全基准上，代码混用训练显著优于单语和双语混合基线，平均提升了目标语言性能，同时维持了与英文性能相当的表现。进一步分析显示，该方法增强了模型对目标语言输入的理解能力，并促进了模型输出与内部表示层面的跨语言对齐，表明代码混用训练有助于目标语言输入更有效地获取在英文中习得的领域推理知识。
链接: https://arxiv.org/abs/2610.25399
作者: Cheonbok Park,Jihwan Kim,Shinyoung Joo,Joosung Lee,Hwiyeol Jo,Jeonghoon Kim,Jaegul Choo
机构: NAVER Cloud; KAIST AI
类目: Computation and Language (cs.CL)
备注: Accepted in EMNLP2026
Abstract:Reinforcement Learning with Verifiable Rewards (RLVR) can improve reasoning in large language models, but specialized non-English domains expose a data mismatch. Raw English domain text is often available, while target-language RLVR tasks with verifiable rewards are scarce. We introduce code-switched task synthesis for cross-lingual RLVR, a data-centric approach that derives target-language and code-switched RLVR task variants from otherwise unverifiable English domain text. Using fill-in-the-middle multiple-choice QA as a seed task format, we synthesize automatically verifiable tasks from domain passages and reshape them into cross-lingual variants. These variants preserve the same evidence-answer relation but change how English and target-language evidence are presented during RLVR. On Korean medical benchmarks and cybersecurity benchmarks across four target languages, code-switched training outperforms matched monolingual and bilingual-mixture baselines on average target-language performance while maintaining comparable English performance. Additional analyses provide evidence of better target-language input understanding and stronger cross-lingual alignment in both model outputs and internal representations. These results suggest that code-switched training helps target-language inputs access domain reasoning learned more strongly in English.
[NLP-126] Outcome-Anchored Credit Reallocation for Agent ic Knowledge Graph Question Answering
【速读】： 该论文旨在解决生成式知识图谱问答（KGQA）中因强化学习奖励分配不均导致的轨迹-操作错位问题。在现有方法中，基于结果的监督强化学习将最终答案的可验证奖励均匀分配给每一轮对话中的所有策略生成标记，然而不同知识图谱（KG）操作对搜索过程及最终答案的影响具有异质性，这种粗粒度的奖励分配机制削弱了训练效率与策略优化精度。其解决方案的关键在于提出一种名为“结果锚定信用重分配”（Outcome-Anchored Credit Reallocation, OACR）的新方法：通过引入一个仅用于训练阶段的KG状态-结果模型，估算相邻搜索状态间结果奖励的变化，并将其转化为操作级信用值；进而基于该信用值推导出逐标记的权重系数，实现对每个生成标记的更新幅度进行精细化调整，同时保持原结果级优势信号的符号一致性。实验表明，OACR在三个基准数据集上均优于现有方法，在提升整体性能的同时显著减少了推理阶段的大语言模型（LLM）调用次数与生成token数量。
链接: https://arxiv.org/abs/2610.25384
作者: Mingyuan Li,Yucheng Li,Xiaxia Wang,Gong Cheng
机构: Nanjing University (南京大学); University of Oxford (牛津大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Agentic knowledge graph question answering (KGQA) requires large language models (LLMs) to solve complex questions through multi-turn interactions with a knowledge graph (KG). Outcome-supervised reinforcement learning derives verifiable outcome rewards from final answers but uniformly assigns each rollout’s outcome-level advantage to all policy-generated tokens, creating a trajectory-to-operation mismatch since KG operations within a rollout can have varying effects on the search process and final answer. To address this, we propose Outcome-Anchored Credit Reallocation (OACR), which employs a training-only KG-state outcome model to estimate outcome-reward changes between adjacent KG search states and converts them into operation credit, from which token-wise weights are derived. These multiplicative weights reallocate token-level update magnitudes while preserving the sign of the outcome-level advantage at every generated token. Across three benchmarks, OACR achieves stronger overall performance than existing KGQA methods while using fewer inference-time LLM calls and tokens.
[NLP-127] Aligning Cultural Behaviors in Arabic Language Models via Preferences EMNLP2026
【速读】： 该论文旨在解决阿拉伯语大语言模型（Arabic LLMs）在跨国家文化行为偏好对齐方面的不足问题，尤其针对现有研究多集中于事实知识的指令微调而忽视了文化细微差异这一关键缺陷。其核心解决方案是提出首个大规模泛阿拉伯地区（pan-Arab）文化偏好数据集——ArabPref，该数据集涵盖英语与现代标准阿拉伯语双语资源，包含一个基于文化情境的偏好数据集，用于训练与评估，以及一个设计用于衡量多国文化适配行为的多项选择基准测试。所有测试数据均经母语者验证以确保真实性。通过在多语言及阿拉伯语中心语言模型上实验后训练方法，研究系统评估了生成任务与文化偏好基准上的表现。该工作通过发布可复现的评估基准，为系统性研究和提升语言模型在文化层面的对齐能力提供了坚实基础。
链接: https://arxiv.org/abs/2610.25362
作者: Junhong Liang,Daniil Orel,Saeed Almheiri,Bilal Elbouardi,Younes Samih,Abed Alhakim Freihat,Zhuohan Xie,Momina Ahsan,Muhammad Dehan Al Kautsar,Sarfraz Ahmad,Omar El Herraoui,Mohamed Anwar,Preslav Nakov,Fajri Koto
机构: Mohamed bin Zayed University of Artificial Intelligence (穆罕默德·本·扎耶德人工智能大学); IBM Research AI (IBM研究院人工智能)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted to EMNLP 2026
Abstract:Arabic LLMs have improved rapidly, but aligning them with country-specific cultural behavioral preferences remains underexplored. Previous work has focused primarily on instruction-tuning for factual knowledge, overlooking nuanced cultural behaviors. We propose ArabPref, which to our knowledge is the first large-scale pan-Arab preference data set for the alignment of cultural behavior conditioned on the country. Our contribution includes two resources in English and Modern Standard Arabic: a culturally grounded preference dataset for training and evaluation, and a multiple-choice benchmark designed to assess culturally aligned behavior across nations. All test data are validated by native speakers to ensure authenticity. To optimize cultural alignment, we experiment with post-training methods across multilingual and Arabic-centric language models, evaluating performance on both generation tasks and our cultural multiple-choice benchmark. By releasing an evaluation benchmark, we provide a foundation for systematically studying and improving culturally aligned behavior in language models.
[NLP-128] CHSM8: A Chess Transformer that Follows Natural-Language Style Prompts
【速读】： 该论文旨在解决生成式棋类模型在不依赖强化学习、奖励函数或搜索机制的前提下，如何实现对特定下棋风格（playing style）的精准控制与高对抗性表现的问题。其核心挑战在于，在保持模型强度的同时，使其能够根据自然语言描述的风格指令（如“追求西西里防御”或“像巅峰时期的卡尔森一样下棋”）生成符合语义意图的走法，且不牺牲性能。解决方案的关键在于：首先构建一个210M参数的棋类Transformer模型（CHSM8），通过因子化走法表示与合法走法掩码，在10.7亿条筛选后的Lichess对局上进行无搜索预训练；随后采用合成数据集进行监督微调（SFT）与直接偏好优化（DPO），利用冻结文本编码器结合门控交叉注意力机制，使模型能有效对齐文本描述与实际走法，并以不匹配描述下的走法作为拒绝样本进行偏好学习；整个流程通过固定预算的自动化搜索循环优化各阶段设计选择，确保在资源约束下达到最优平衡。实验表明，经过微调后模型在多种风格提示下显著提升风格一致性（如早期兑后比例从53%升至67%、后翼易位率高达97.3%-99.7%），同时在无风格提示时仍维持接近原模型的强度（1792 vs. 1797 Elo）。研究全程开源可复现，涵盖权重、数据集、配置及可交互演示，为风格可控的生成式AI (Generative AI) 在策略类任务中的应用提供了可扩展范式。
链接: https://arxiv.org/abs/2610.25356
作者: Ilia Badanin
机构: 未知
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:We present CHSM8, a 210M-parameter chess transformer that plays moves conditioned on a free-text description of a playing style, e.g. “go for the Sicilian Defense”, “solid, safe, risk-free chess” or “You are Magnus Carlsen in his prime”. The model is pretrained without search on 1.07B filtered Lichess games, using a factorized move representation and legal-move masking. It reaches 1837 Elo against a strength-limited Stockfish and, in 1,000-game matches, beat every other public model we tested that was likewise trained without preference tuning, rewards or search, including one with 13 times as many parameters. We then fine-tune it to follow text on a synthetic dataset: Gemma-4-31B describes the playing style of 3M game sides, a 1B Gemma-3 model distilled from the 2.4M of those descriptions that contradict nothing on the board labels the rest, and every one of 30M selected game sides gets five descriptions. Supervised fine-tuning attaches a frozen text encoder through gated cross-attention, and direct preference optimization (DPO) uses the same moves under a mismatched description as the rejected sample. After DPO, style adherence rises on every prompt we test (for example, queens traded early 53% - 67%, pawn moves 29% - 32% of all moves, queenside castling 97.3% - 99.7%), and strength without a prompt stays within noise of SFT (1792 vs. 1797 Elo). Design choices at every stage were made by fixed-budget automated search loops on top of standard architectures and training recipes. We document the full pipeline and the decisions behind it, report the ablations that failed, and evaluate the models across many style prompts and kinds of games, with 1,000 games per setting. The work is fully open source and reproducible: weights, datasets with all training labels, prompts, configurations, inference code and playable demos are public and free to build on.
[NLP-129] Atoms Not Knowledge: What Production-Scale Addressed-Memory Writes Deposit
【速读】： 该论文旨在解决生成式AI（Generative AI）中因果记忆（causal memory）在实际部署场景下可扩展性与可靠性不足的问题，特别是针对局部后验知识注入（local post-hoc knowledge injection）时的记忆写入覆盖率、组合性能力及系统级协同效应缺失的挑战。其核心解决方案在于提出一种分层协议（layered protocol），首次实现了对行为记忆评估中组合性（compositionality）的系统化分离，明确区分基线表现、协同增益（synergy）与词汇扭曲底限（lexical-distortion floor）。关键创新在于：1）通过充分训练的骨干网络（backbone）、全宽读出（full-width readout）和可训练门控机制（trainable gates）实现高精度、可逆且可独立检索的写入原子操作；2）揭示在固定预算下，写入覆盖率随知识量增加呈非线性下降（从50条事实的98%降至1000条的12.7%），而表规模并非瓶颈；3）证明两跳组合推理无法超越独立单跳基准（实测26% vs 预期25%），且跨剂量/负载无协同效应；4）发现上下文内中间实体的利用效率是隐式调用的两倍，表明失败根源在于“检索-推理”链路而非门控访问。研究强调任何键值记忆系统在部署前必须报告覆盖率、协同比与扭曲率，并主张多约束查询应显式编排写入原子，以实现真正的系统级组合推理。所有结果均基于归档原始日志复现，且一次全新生产级重跑完全匹配历史基准（52/52），验证了方法的可重复性与稳定性。
链接: https://arxiv.org/abs/2610.25317
作者: Wei Gao,Cheng Tian,Duan Yang,Jianqing Yin
机构: Shibei Qiyuan AI Technology; Xi’an Jiaotong University
类目: Computation and Language (cs.CL)
备注:
Abstract:Deterministic hash-addressed conditional memory permits local post-hoc knowledge injection. Extending arXiv:2610.21553’s bounded-feasibility audit rejecting under-specified generality claims with repaired probes and training-dose/capacity axes, we give, to our knowledge, the first production-scale write-content verdict. We test 124M-7B self-trained/publicly pretrained beds, Qwen3.8-Flash-Next-FP8 (176.95B; 320,001,536 FP8 rows), and DeepSeek-V4.1-Flash. Write coverage requires a sufficiently trained backbone prior, full-width readout, trainable gates; undertrained-bed intervention confirms a threshold, not a slope. At fixed budget, coverage falls from 98% (50 facts) to 12.7% (1,000); table size is not binding. Two-hop composition does not exceed the independent single-hop baseline (26% observed vs 25% expected; 15% lexical-distortion floor). Synergy remains absent across dose/load in two independent bed families. A 28-percentage-point in-context gap (given vs recalled intermediates) localizes failure to retrieval-for-reasoning, not gate access. Writes are independently retrievable, precisely addressable, reversible atoms, not composable into reasoning. A layered protocol, the first, to our knowledge, to cover compositionality in behavioral memory evaluation, separates baseline, synergy and distortion floor. Any addressed/key-value memory must report coverage, synergy ratio and distortion rate before deployment claims. Composition must be system-orchestrated; multi-constraint queries should explicitly chain written atoms. In-context controls show given intermediate entities are exploited at twice the implicit rate. Production load fits exponential interference ( R^2=0.98 ); its mechanism, between budget dilution and write interference, remains open. All numbers are recomputed from archived raw logs/JSON; a fresh 52-target production rerun gives 52/52, exactly matching archived baselines.
[NLP-130] Learning When to Abort in Transformer Planning : A Mechanistic Study
【速读】： 该论文旨在解决生成式模型在基于Transformer的规划任务中缺乏错误识别能力，进而影响规划可靠性的问题。尽管规划过程本身不包含错误检测机制，但通过引入对无效转移（invalid transitions）的识别能力，可显著提升规划的鲁棒性与准确性。其核心解决方案在于：将规划抽象为图中的路径搜索问题，并设计一个具有足够宽前馈网络（feed-forward width）的Transformer模型，使其能够检测所有可能的无效转移——即当模型生成的下一节点与当前节点不相邻时，触发特殊终止令牌（abort token x）。理论分析表明，该检测器的前馈宽度在最坏情况下接近最优（仅差对数因子），且实验验证了错误检测主要发生在前馈网络的输出层，同时提升了对先前节点的关注度。在推理阶段，一旦检测到错误，外部程序将主动丢弃当前尝试并从原始提示重新开始，从而实现“错误检测-重试”机制。实验证明，该方法显著提升了模型在图结构、Countdown谜题和迷宫等任务中的规划可靠性。
链接: https://arxiv.org/abs/2610.25300
作者: Shi Feng,Qimin Zhong,Siwei Wang,Yifei Shen,Shang-Hua Teng,Hao Liao,Wei Chen
机构: Harvard University(哈佛大学); Shenzhen University(深圳大学); Microsoft Research Asia(微软亚洲研究院); University of Southern California(南加州大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Planning does not by itself imply the ability to recognise errors, but recognising errors can improve planning capacity. In this paper, we study this phenomenon in Transformer planning. To analyze the inner mechanism, we follow the approach of Wang et al. (2024) and abstract planning as path-finding in graphs. In this setting, we train the model to detect the error of invalid transitions, i.e., after the model generates a next node that is not adjacent with the current node, the model should then generate a special abort token x. For expressiveness, we construct a Transformer with feed-forward width O(|V|) for any fixed graph with node set V . It can detect all possible invalid transitions, and select a valid next node if the prefix is valid. We then propose a worst-case lower bound on feed-forward width, which makes the detector’s linear width near-optimal up to logarithmic factors. In trained models, we demonstrate that invalid-transition detection increases previous-node attention, and most of the valid–invalid separation locates in the feed-forward network’s output. This accords with our theoretical insights. Finally, we use this error detection to improve the model’s performance: at inference, generating the abort token x will directly cause an external procedure to discard the attempt and restart from the original prompt. Experiments show that error detection combined with retry improves model’s planning reliability on graphs, Countdown, and mazes.
[NLP-131] Aligning Language Models with Indigenous Community Cosmovision through Constitutional AI
【速读】： 该论文旨在解决大型语言模型（Large Language Models, LLMs）在训练数据中普遍反映西方盎格鲁世界视角所导致的伦理与文化偏见问题，特别是在涉及领土、社群权威及人与自然关系等议题时，模型输出往往基于与全球南方原住民社区相异甚至冲突的假设。其核心解决方案是构建一种面向拉丁美洲原住民社区的伦理对齐系统——宪法式人工智能（Constitutional AI, CAI），该系统以五位原住民领袖与捍卫者访谈为基础，提炼出体现原住民宇宙观（Indigenous cosmovision）的12项伦理原则，并据此设计双阶段对齐流程：监督微调（Supervised Fine-Tuning, SFT）与直接偏好优化（Direct Preference Optimization, DPO）。实验采用Qwen2.5-7B-Instruct与Qwen2.5-1.5B-Instruct作为基线模型，仅使用24组偏好样本即在代际传承、语言谨慎性及协商同意等维度实现可检测的改进。定性分析表明，SFT阶段主导了整体对齐方向的转变，而DPO则进一步细化特定类别表现。系统通过检索增强生成（Retrieval-Augmented Generation, RAG）架构部署于对话界面，支持社区主动向知识库输入内容，形成持续参与式的对齐循环。研究论证了该架构在功能上的可行性，并指出其在语料库扩展与计算资源改善后具备质的提升潜力，为全球南方地区的认知正义（epistemic justice）导向的价值对齐提供了首个迭代范式。
链接: https://arxiv.org/abs/2610.25298
作者: Silvia Fernández-Sabido,Jorge Reyes-Magaña,Jesús Adrián Chay Tun,Jean Carlos Buenfil,Victor González Nahuat,Francisco Morales
机构: Secretaría de Ciencia, Humanidades, Tecnología e Innovación (SECIHTI); CentroGeo(中心地理研究所), Mérida, Yucatán, Mexico; Facultad de Matemáticas, Universidad Autónoma de Yucatán(尤卡坦自治大学数学学院), Mérida, Yucatán, Mexico; Aymara Community / Global Council for Latin America and the Caribbean, United Religions Initiative (URI)(世界宗教倡议组织), Buenos Aires, Argentina
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: In preparation for the Dossier “Linguistics in AI and AI in Linguistics” of the journal TextoLivre
Abstract:Large language models (LLMs) are predominantly trained on corpora that reflect Western, anglophone worldviews. This asymmetry is not neutral: when a model answers questions about territory, community authority, or the relationship with nature, it does so from assumptions that may be alien to the communities using the system. This paper presents a system for aligning language models with the ethical principles of Indigenous communities in Latin America. Drawing on interviews with five Indigenous leaders and defenders, we built an operational cultural constitution grounded in an Indigenous cosmovision, with 12 principles. On this basis, we implemented a Constitutional AI (CAI) pipeline in two phases: supervised fine-tuning (SFT) and direct preference optimization (DPO). The base models were Qwen2.5-7B-Instruct and Qwen2.5-1.5B-Instruct. With only 24 preference pairs, improvements were modest but detectable in intergenerational transmission, linguistic caution, and consultation and consent. Qualitative analysis reveals that the SFT phase was responsible for the primary shift in alignment, while DPO refined specific categories. The system was deployed in a conversational interface with retrieval-augmented generation (RAG) that allows communities to feed the knowledge base, turning alignment into a participatory cycle. We argue the architecture is functional and will scale qualitatively once the corpus is expanded and computing constraints are resolved. The work constitutes a first iteration of an epistemic justice mechanism for value alignment in the Global South.
[NLP-132] SPARC: Spoken Profile Authorship Recognition Corpus
【速读】： 该论文旨在解决语音匿名化研究中长期被忽视的核心问题：尽管现有方法主要关注声学信号的隐私保护，但转录后的文本本身仍可能隐含大量说话人身份及人口统计属性信息。为此，作者提出构建SPARC（Spoken Profile Authorship Recognition Corpus），一个基于两个播客语料库的基准数据集，包含超过12,500名说话人、150万条经质量筛选的语音片段，并附有性别、年龄和口音等说话人层面的标注信息。研究通过五种归属模型与三种大语言模型（LLM）在基于节目内容与元数据的主题代理下进行身份与属性验证实验，发现仅凭文本即可实现同说话人对的区分，表明文本中存在显著的说话人可识别性。尽管属性标签一致性验证更具挑战性，但大语言模型在性别与口音标签验证任务上仍优于随机水平。此外，仅发布短文本片段（50词）能显著削弱最强攻击者的能力，为降低文本-身份关联提供了部分缓解策略。解决方案的关键在于揭示转录文本中的潜在身份泄露风险，并提出通过限制文本长度来实现一定程度的隐私保护。
链接: https://arxiv.org/abs/2610.25265
作者: Efthymios Tsaprazlis,Bowen Yi,Tiantian Feng,Junghwan Kim,David Jurgens,Shrikanth Narayanan
机构: University of Southern California (南加州大学); University of Michigan (密歇根大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Speech anonymization has focused on the acoustic signal, leaving open how much speaker information the transcribed words themselves retain. In this work, we study pairwise speaker and demographic-attribute verification from transcribed speech. We introduce SPARC (Spoken Profile Authorship Recognition Corpus), a benchmark built from two podcast corpora comprising over 12,500 speakers, 1.5M quality-filtered utterances, and speaker-level inferred or source-provided labels for sex, age, and accent, and evaluate five attribution models and three LLMs on identity- and attribute-verification trials with episode- and metadata-based topic proxies. LLM adversaries distinguish same-speaker pairs from text alone. Attribute-label agreement is harder to verify, but LLMs perform sex- and accent-label verification above chance. Finally, releasing only short probe texts (50 words) degrades the strongest adversaries the most, offering a partial reduction in observed linkage.
[NLP-133] Assessing the Empirical Evidence for Loss of Control from Agent ic General-Purpose AI
【速读】： 该论文旨在解决当前关于通用型智能体系统（agentic general-purpose AI systems）可能引发人类控制权丧失这一系统性风险的实证证据评估缺失问题。尽管欧盟《通用人工智能行为准则》已将“失去控制”列为必须评估的系统性风险，但尚无结构化的方法对相关实证研究的质量进行系统性审查。为此，本文基于威胁模型识别出八种关键属性（如策略性欺骗、情境意识、自主复制等），对19项近期实证研究进行了评估，从建构效度、内容效度和外部效度三个维度分析证据质量。研究发现，现有证据仅支持“弱可成立”的判断，即该风险在当前证据下具有一定的可能性，但尚未达到强支持水平；其中，多项属性已有初步证据，但无一研究满足判定某属性存在的有效性阈值，仅有一项研究在否定方向上（即不具情境意识）通过了验证。广泛存在的有效性缺口，尤其是外部效度不足，表明当前研究在真实场景中的推广能力受限。值得注意的是，对于最危险的属性，高外部效度在负责任的研究实践中可能根本不可实现，这反映的是证据获取的结构性天花板，而非能力的缺失。此外，随着人工智能能力快速演进，实证评估可能持续滞后于技术前沿。因此，本文提出的框架为未来持续监测技术发展提供了结构化基准。
链接: https://arxiv.org/abs/2610.25241
作者: Risto Uuk,Santeri Koivula,Lorenzo Pacchiardi,Rokas Gipiškis,Laura Caroli,Kamaria Horton
机构: Future of Life Institute(未来生命研究所); KU Leuven(鲁汶大学); London School of Economics and Political Science(伦敦政治经济学院); ETH Zurich(苏黎世联邦理工学院); University of Cambridge(剑桥大学); AI Standards Lab; Vilnius University(维尔纽斯大学); RAND(兰德公司)
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 42 pages. Previously posted on SSRN on 28 May 2026
Abstract:Theoretical literature warns that advanced generalist agentic systems might pursue instrumental goals that are misaligned with human preferences, thus posing risks of irreversible loss of human control. The EU’s Code of Practice for General-Purpose AI lists loss of control as a systemic risk that must be assessed, yet no structured assessment of the quality of the empirical evidence for this risk exists. This study fills that gap. We evaluate 19 recent empirical studies across eight properties identified from threat models (such as strategic deception, situational awareness, and autonomous replication), assessing the quality of evidence across three dimensions: construct validity, content validity, and external validity. Our central finding is that the current body of evidence supports a judgment of loss of control as “weakly plausible.” In our framework, ‘plausibility’ denotes the strength and direction of available evidence, not a probability estimate. Emerging evidence exists across multiple properties. No study clears the validity thresholds for a stronger judgment that a property is present; a single study clears them only for a negative-direction judgment (against situational awareness). Validity gaps, particularly in external validity, are widespread. Therefore, our findings highlight the need for more rigorous evaluations of agentic general-purpose AI systems. However, for the most dangerous properties, high external validity may be structurally unachievable under responsible research practices, reflecting a ceiling on feasible evidence rather than absence of the capability. Furthermore, as AI capabilities advance rapidly, empirical assessments may systematically lag behind the capability frontier. Our framework provides a structured baseline for ongoing monitoring as capabilities advance.
[NLP-134] Hidden Cartographer: Hidden State Guided Data Selection for Dual Budgeted Instruction Tuning
【速读】： 该论文旨在解决在严格预算约束下，如何从大规模无标注数据池中高效选择高质量样本以适应大语言模型（LLM）至专业领域的问题。现有方法存在两大局限：依赖标签的策略在使用伪标签时易导致选择偏差，而传统主动学习方法未能显式刻画指令微调（instruction-tuning）过程中的动态变化。为此，本文提出隐空间制图器（Hidden Cartographer, HC），其核心在于通过隐藏状态（hidden states）指导专家标注，并利用精确的训练动态信息筛选最终训练子集。该方案包含两个阶段：第一阶段为渐进候选扩展（Progressive Candidate Expansion），基于代理模型迭代选取高价值未标注样本进行专家标注，合并后重新计算训练动态并更新代理，直至耗尽标注预算；第二阶段为精确制图剪枝（Exact Cartography Pruning），依据每个已标注样本的精确训练动态得分进行排序，从而选出在最终微调预算下性能最优的紧凑子集。实验结果表明，HC在医疗、法律和金融等多个领域均显著优于其他预算受限基线方法，在仅需标注原始数据池20–30%且微调仅10–15%样本的情况下，实现了最高平均准确率。
链接: https://arxiv.org/abs/2610.25215
作者: Zixuan Qin,Qi Qin,Erbo Li,Ting Wei,Yifan Sun,Zhaoxin Fan
机构: Renmin University of China(中国人民大学); Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing(北京未来区块链与隐私计算高精尖创新中心)
类目: Computation and Language (cs.CL)
备注:
Abstract:Given the high cost of expert annotation and limited fine-tuning resources, effective data selection from large unlabeled pools under strict budgets has become a crucial step in adapting LLMs to specialized domains. Existing label-dependent methods risk distorted selection when using pseudo-labels, while traditional active learning methods do not explicitly characterize instruction-tuning dynamics. To address these limitations, we propose Hidden Cartographer (HC), a two-stage framework that uses hidden states to guide annotation and exact training dynamics to select the final training subset. The first stage is Progressive Candidate Expansion, which utilizes the proxy to iteratively select top unlabeled candidates for expert annotation, merging them to recompute training dynamics and update the proxy until the annotation budget is exhausted; the second stage is Exact Cartography Pruning, which ranks these annotated candidates by their exact dynamics scores to isolate a compact subset under the final training budget. Experiments across the medical, legal, and financial domains demonstrate that HC achieves the highest average accuracy among all budgeted baselines, requiring annotation for only 20–30% of the original pool and fine-tuning on just 10–15%.
[NLP-135] Calibration-Free Early Stopping for Chain-of-Thought Reasoning via E-processes EMNLP2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在推理过程中如何实现自适应停止（adaptive stopping）的问题，即在保证推理准确率的前提下，动态决定何时终止生成以节省计算资源。现有方法依赖于在验证集上校准的置信度或稳定性阈值，但这类阈值往往难以跨模型和数据集迁移，且在测试时程或信号动态发生变化时，其有效性会丧失。本文提出将自适应停止重构为推理轨迹上的序列证据累积过程，而非依赖于离线校准的阈值。其核心解决方案是引入两个共享每块（per-chunk）信号的e-process（e-processes）：一个“提交流”（commit stream）追踪信号相对于其慢趋势的位置，另一个“弃权流”（abstain stream）追踪信号的有符号速度，二者在构造上呈负相关。在每块增量服从信号层面零假设（signal-level nulls）的前提下，利用Ville不等式可获得任意时间有效的信号层面误报率控制，在用户指定显著性水平下无需任何离线校准数据即可保证统计可靠性。实验表明，该框架在多种模型-数据组合下，无需校准数据即可达到或超越基于校准的停止策略在“令牌-准确率帕累托前沿”上的表现，且相同的显著性水平可在校准不可行的情况下保持一致适用。
链接: https://arxiv.org/abs/2610.25212
作者: Dong-Jae Lee,Gyojin Han,Junmo Kim
机构: KAIST(韩国科学技术院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: EMNLP 2026 Findings
Abstract:Reasoning large language models (LLMs) benefit from test-time scaling, motivating adaptive policies that spend tokens when they help and stop early otherwise. Existing rules halt when a confidence or stability signal crosses a threshold, but such thresholds rarely transfer across models and datasets, and validation-calibrated rules lose their guarantee under shifts in test-time horizon or signal dynamics. We reframe adaptive stopping as sequential evidence accumulation along the reasoning trajectory rather than threshold calibration on held-out data. Our framework runs two e-processes on a shared per-chunk signal: a commit stream tracks the signal’s position relative to its slow trend, and an abstain stream tracks its signed velocity, making the two anti-correlated by construction. Under signal-level nulls on the per-chunk increments, Ville’s inequality yields an anytime-valid signal-level false-alarm bound at user-specified significance levels, with no held-out calibration. Across various model–dataset combinations, the framework matches or exceeds calibration-based stopping on the token–accuracy Pareto front using zero calibration data, and the same significance level transfers unchanged across cells where calibration becomes infeasible.
[NLP-136] A Thousand Faces: A Persuasion Benchmark with Experience-Grounded Persuadees Informed by Psychology and Prag matics
【速读】： 该论文旨在解决现有说服性对话评估基准在处理个体认知差异时的局限性问题，即传统基准通常仅基于静态的信念与欲望描述进行评判，而忽视了这些观点背后由个人经历所塑造的可信度、价值偏好及反思动因等动态心理因素。这种忽略导致评估结果可能奖励形式上匹配但实际缺乏针对性的论证，从而无法真实反映说服的有效性。其解决方案的关键在于构建一个具有高度个性化特征的多角色数据集——A Thousand Faces (ATF-Persu)，该数据集涵盖200个主题下1,000个不同人格角色（persona），覆盖九个领域，每个事实情境下生成五个基于人格理论与会话隐含意义（conversational implicature）构建的角色，明确关联个体经历与其判断、表达方式及立场修正的依据。实验证明，相较于三种对比基准，ATF-Persu在评估中表现出更低的说服成功率，反映出其对真实说服机制更严格的衡量标准；同时，完整背景信息显著提升了对个人理由的使用广度和角色一致性，盲评结果显示评审者更认可ATF-Persu在识别个人动机、其对回应的影响以及语境化表达方面的表现。
链接: https://arxiv.org/abs/2610.25206
作者: Ziqing Zhuang,Dingyi Zhang,Deyu Zhou
机构: Southeast University (东南大学)
类目: Computation and Language (cs.CL)
备注: 45 pages. Dataset: this https URL
Abstract:People can express the same beliefs and desires yet need different reasons to change their minds. What makes an argument relevant depends on how those views arose: experiences shape what a person finds credible, what matters to them, and what could prompt reconsideration. These personal grounds may also be expressed indirectly, through contextual clues in conversation. Benchmarks that specify beliefs and desires without these connections can reward arguments that match a description while overlooking whether they address the person’s reasons. We introduce A Thousand Faces (ATF-Persu), with 1,000 personas across 200 topics and nine domains. Drawing on personality and attitude theories and conversational implicature, we construct five personas per factual context, linking experiences to judgments, expression, and grounds for revision. Evaluations show lower persuasion success than three comparison benchmarks under their respective protocols. Complete backgrounds yield broader use of supplied personal grounds and higher persona faithfulness than simplified specifications; blinded reviewers also rate ATF-Persu above CToMPersu on identifiable personal reasons, their influence on responses, and contextual expression. The dataset is publicly available at this https URL.
[NLP-137] Acharya 2.0: An AI Tutor for Self-Learning Subjects in Hindi
【速读】： 该论文旨在解决低资源语言（low-resource languages）在文本处理方面因缺乏充足标注数据和语言资源而导致的语言技术发展困难问题，尤其聚焦于印地语等印欧语系语言在教育领域应用中的技术瓶颈。其核心挑战在于：尽管生成式AI（Generative AI）与大语言模型（LLMs）在高资源语言中表现优异，但其在特定任务上的适配仍存在耗时长、资源消耗大的问题。为此，论文提出以检索增强生成（Retrieval-Augmented Generation, RAG）技术为核心解决方案，通过引入外部知识库来增强大语言模型的上下文理解与任务适应能力，从而在不依赖大规模标注数据的前提下实现高效、精准的个性化教育辅助。该方案的关键在于将RAG架构与面向印地语学生的教学需求相结合，构建了基于客户端-服务器架构的AI-Tutor应用“阿查里亚2.0”（Acharya 2.0），实现了对学术内容的智能检索与生成，显著提升了低资源语言环境下教育技术的可用性与可及性。
链接: https://arxiv.org/abs/2610.25192
作者: Shantipriya Parida,Debasish Dhal,Sonal Khosla,Sambit Sekhar,Muhammed Abdur Rahmaan Kamaldeen
机构: AMD Silo AI(芬兰); Odia Generative AI(印度)
类目: Computation and Language (cs.CL)
备注:
Abstract:Text processing is crucial for low-resource languages as it enables the development of tools and technologies that bridge the digital divide and provide equitable access to information and services. By empowering low-resource languages with robust text processing, we can preserve linguistic diversity, enhance education, improve communi- cation, and foster inclusivity in a rapidly digitizing world. Due to the lack of sufficient annotated datasets and lin- guistic resources, it is challenging to develop accurate and reliable language technologies for these low-resource languages. Large Language Models (LLMs) have attracted sig- nificant attention with their performance and the value they produce. However, with the ever-evolving language, adapting LLMs for a specific task is time-consuming and resource-intensive. Thus, Retrieval-Augmented Genera- tion (RAG) has overcome this limitation by augmenting LLMs with external knowledge, thereby improving their efficiency and extensive task-specific adaptation. Most applications of the RAG technique have been limited to English and has seen limited application in In- dic languages. This paper introduces the RAG-powered AI-Tutor application, Acharya 2.0, designed to produce academic assistance to Hindi-speaking students. The client-server-based web application integrates many fea- tures to support students in accessing and interacting with educational content.
[NLP-138] Reranking Answers of Large Language Models with Knowledge Graphs WWW
【速读】： 该论文旨在解决在知识图谱（Knowledge Graph）数据上回答自然语言问题时，因事实数量庞大而导致的处理与导航困难问题。其核心挑战在于如何高效地从海量知识中定位与查询相关的上下文信息。解决方案的关键在于利用大语言模型（Large Language Model, LLM）推断实体之间的最短路径，从而自动挖掘与问题中候选实体相关的子图（subgraph），并在此基础上提取多种可计算特征。这些特征被用于构建重排序模型（reranking model），以从候选答案列表中筛选出最可能的正确答案。研究在Wikidata上进行了实验，系统性地枚举了可从挖掘出的子图中提取的主要特征类型，并对不同特征组合与重排序方法的有效性进行了详尽分析。此外，研究还开发了一个公开的Web应用，支持可视化子图结构，并自动生成自然语言描述，为探究问题与答案实体间的图空间关系提供了实用工具。
链接: https://arxiv.org/abs/2610.25185
作者: Mikhail Salnikov,Hai Le,Olga Tsymboi,Ivan Lazichny,Dmitrii Iarosh,Egor Cheremiskin,Andrey Savchenko,Dmitry Simakov,Alexander Panchenko
机构: Skolkovo Institute of Science and Technology (斯科尔科沃科学与技术学院); Sber AI Lab (斯伯尔人工智能实验室); Moscow Institute of Physics and Technology (莫斯科物理技术研究所); Moscow State University (莫斯科国立大学); Laboratory for Theoretical Foundations of AI Models, HSE University (理论人工智能模型实验室，高等经济大学); Artificial Intelligence Research Institute, Moscow (莫斯科人工智能研究院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 37 pages. Accepted for the Semantic Web journal (tracking no. 4075-5289); not published. this https URL
Abstract:Answering natural language questions over knowledge graph data is challenging due to the vast number of facts, which can be difficult to process and navigate. One potential solution for this issue is to use mined subgraphs related to the query, although this process still requires extracting these subgraphs. This research presents a solution for extracting subgraphs related to entity candidates from a question-and-answer set, which can be obtained by inferring a large language model by calculating the shortest paths between entities. The proposed approaches detail various features that can be extracted from the subgraphs and reranking models to select the most probable answers from a list of candidates. Experiments were conducted on Wikidata to evaluate the effectiveness of the proposed approaches. This involved enumerating all the main feature types that can be extracted from mined subgraphs and a detailed analysis of the proposed features and reranking method combinations. In addition, a public web application that provides a useful web tool for studying the graph space between question and answer entities has been developed to work with subgraphs. This includes visualization of the extracted subgraph and automatic generation of natural language text to describe it.
[NLP-139] Concept Representations in Vector Space: Toward an Interdisciplinary Study of Meaning Variation
【速读】： 该论文旨在解决个体间概念表征（concept representations）在主观意义层面是否存在差异这一核心问题，即当不同个体使用同一概念时，其内在心理结构是否具有可比性或共享性。由于概念表征作为潜在的、私有的认知结构，难以直接观测和建模，传统方法难以实证分析个体间的异质性。本文提出的关键解决方案是构建一个基于标准化文本诱引的测量框架，通过基于Transformer的句子嵌入技术，将个体对概念的自由联想响应映射至高维共享向量空间，从而实现对概念表征的可操作化与系统性分析。该框架的核心创新在于：（1）量化不同概念间表征共享程度的差异；（2）识别影响表征变异的潜在变量；（3）检测特定群体或焦点个体相对于参照群体的偏离模式。研究通过大规模自由联想数据集进行验证，展示了该框架在跨学科研究中的普适性与可实施性，并提供开源代码库以支持后续应用。
链接: https://arxiv.org/abs/2610.25184
作者: Simon Josef Durstewitz
机构: 未知
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 20 pages, 5 figures. Companion paper to arXiv:2610.15496 . Code: this https URL
Abstract:An increasing number of disciplines recognize the explanatory potential of meaning-bearing mental structures through which individuals interpret the world and guide their behavior. Concept representations constitute a core class of these structures, emerging from individuals’ experiences with concept referents and encoding and organizing their understanding of them. Recognizing a private level of meaning, where concepts are not objectively defined by rules in the world but constitute subjective structures in individuals’ minds, raises a fundamental question: do different individuals actually mean the same thing when they invoke the same concept? As latent structures, concept representations are difficult to access and model empirically. The present study introduces a measurement framework that makes such representations empirically accessible through textual traces elicited under standardized conditions. Using transformer-based sentence embeddings, the framework maps standardized responses elicited for concepts into a high-dimensional shared vector space, opening concept representations to the systematic analysis of interindividual variation. It develops three analytical avenues: (1) comparing representational sharedness across concepts, (2) identifying variables that correlate with representational variation, and (3) examining how focal actors or subpopulations deviate from a reference population. Each avenue is illustrated through an empirical proof-of-concept drawing on a large-scale free-association dataset. Together, the demonstrations show how researchers from across disciplines can move from a research question and data source to analysis and interpretation. An accompanying GitHub repository provides a ready-to-use implementation of the framework for existing and custom datasets.
[NLP-140] Dialogue Generation for Conversational Recommendation Using User Search Interactions and Knowledge Graphs CIKM2026
【速读】： 该论文旨在解决大规模对话式推荐数据集构建中依赖人工对话构造或标注的问题。其核心挑战在于如何在不依赖人工干预的前提下，生成高质量、语义连贯且基于真实用户行为的对话式推荐数据。解决方案的关键在于提出SearCl框架，该框架通过将用户交互行为划分为查询会话，并从搜索后点击的内容中推导出推荐目标，实现数据生成的自动化；同时，采用可配置的概率状态机控制对话流程，结合两名大语言模型（LLM）代理——“寻求者”与“推荐者”，依据回合级约束生成自然语言对话；此外，为增强推荐的准确性与可解释性，引入基于内容元数据构建的知识图谱，为推荐者提供结构化背景知识。实验表明，SearCl生成的3,538条对话覆盖1,731个唯一目标项，在大语言模型作为评判者（LLM-as-a-judge）的评估下，其对话质量优于现有基准数据集，验证了该方法在无需人工标注的情况下，实现可控且知识驱动的数据集构建能力。
链接: https://arxiv.org/abs/2610.25182
作者: Minsung Hwang,Donggyu Yoon,Gowoon Hwang,Kwanghee Jang,Joyce Jiyoung Whang
机构: KAIST(韩国科学技术院); LG Electronics(韩国电子公司)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 6 pages, 4 figures, 2 tables, The 3rd Workshop on Evaluating and Applying Recommender Systems with Large Language Models (EARL) at The 35th ACM International Conference on Information and Knowledge Management (CIKM 2026)
Abstract:Large-scale conversational recommendation datasets typically require manual dialogue construction or annotation. We propose SearCl, which automatically generates conversational recommendation dialogues grounded in user search and click behaviors. SearCl segments user interactions into query sessions and derives each recommendation target from content selected after a search query. A configurable probabilistic state machine controls the dialogue flow, while two LLM agents acting as the seeker and recommender generate utterances according to turn-level constraints. To support recommendation, the recommender is provided with knowledge graphs constructed from metadata about the content items used in each dialogue, capturing their attributes and relationships. Using SearCl, we generate 3,538 dialogues from content search and click interaction data, covering 1,731 unique target items. Under an LLM-as-a-judge protocol, the resulting dialogues achieve higher quality scores than dialogues from existing benchmarks. These results demonstrate that SearCl enables controllable and knowledge-grounded dataset construction without human annotations.
[NLP-141] Memory Know Your PLACE: Governing Lifecycle State and Query-Time Eligibility in Persistent-Memory Agents
【速读】： 该论文旨在解决现有持久化记忆（persistent memory）系统在状态有效性控制与查询时信息影响管理方面缺乏足够机制的问题。具体而言，当前系统无法有效决定某段记忆是否应保持有效，或在后续查询中是否应被引用，从而导致记忆状态的不一致与不当信息泄露。为应对这一挑战，论文提出PLACE——一种解耦式治理层（factorized governance layer），通过在写入端引入状态认证（write-side state certification）和在读取端实施授权机制（read-side authorization），分别实现对记忆状态生成与查询时信息影响的显式控制。在Mem0框架上基于Qwen2.5-3B与Qwen2.5-7B进行的因子化实验表明，写入端治理显著提升了状态生命周期的正确性，读取端治理则有效消除了无效记忆在状态正确设定下的因果泄漏（causal leakage）。两者的协同作用在保持查询相关支持的前提下，实现了最优的整体性能。研究结果表明，构建可靠的持久化记忆系统需同时对所维护的状态及其在查询时的信息影响力施加明确的治理控制。
链接: https://arxiv.org/abs/2610.25179
作者: Chen-Fang Hu,Hong-Han Shuai,Woan-Shiuan Chien
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Persistent memory turns prior interactions into durable agent state, but existing systems provide limited control over whether that state remains valid and whether it should influence a later query. We argue that reliable memory requires governance over both state formation and query-time influence. We introduce PLACE, a factorized governance layer that adds write-side state certification and read-side authorization to a memory backend. A factorial evaluation on Mem0 with Qwen2.5-3B and Qwen2.5-7B isolates the contribution of each mechanism. Results show that write-side governance substantially improves lifecycle state correctness, while read-side governance eliminates measured causal leakage from ineligible memory in the state-correct setting. Their combination achieves the strongest joint performance while preserving query-relevant support. These findings show that reliable persistent memory requires explicit control over both maintained state and query-time information influence.
[NLP-142] Indistinguishable Agents : Training Agent Privacy as Twin-World Indistinguishability
【速读】： 该论文旨在解决大语言模型代理（LLM agents）在执行用户任务（如搜索、购物、发送消息）过程中存在的隐私泄露问题。尽管现有隐私训练方法通过奖励代理不提及受保护信息来减少显式披露，但观察者仍可通过分析代理的行为模式推断出敏感事实。其核心问题是：即使未直接提及私密内容，行为本身也可能暴露隐私。本文提出的方法——\methodfull（\method），关键在于构建一对仅在私密事实上不同的“孪生世界”（twin worlds），并利用一个冻结的判别器（frozen distinguisher）评估观察者视角在不同世界下的对数似然比（log-likelihood ratio），以此量化观察者从行为中获取隐私信息的能力。代理通过配对的GRPO算法进行训练，优化目标为任务成功率减去该似然比得分，从而在保证任务完成的同时，最小化所有观察者在任何通道上基于其观测推断私密事实的贝叶斯成功率。实验证明，\method有效消除了传统披露判断和查询级奖励所遗漏的隐性隐私泄露，在购物与消息任务中显著提升了隐私保护能力。
链接: https://arxiv.org/abs/2610.25161
作者: Jiaming He,Guanyu Hou,Yi Yu,Xudong Jiang
机构: Nanyang Technological University, Singapore(南洋理工大学); The University of Manchester, United Kingdom(曼彻斯特大学); Jilin University, China(吉林大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:LLM agents search, shop and send messages for their users, and every action is seen by some party. Existing privacy training rewards agents for not mentioning protected items, yet an observer can infer a private fact from what the agent does. Since an observer learns a fact only if its view changes with the fact, we propose \methodfull (\method), which trains on twin worlds that differ only in the private fact. A frozen distinguisher scores each observer’s view by its log-likelihood ratio for the true world, and paired GRPO trains the agent on task success minus this ratio. We prove that this objective controls the Bayes success of every observer on every channel. On shopping and messaging tasks \method removes the leakage that disclosure judges and query-level rewards miss, while the agent still completes its tasks.
[NLP-143] Legal Guidance in the Wild: How Users Seek Legal Help in Real-World LLM Conversations
【速读】： 该论文旨在解决在对话式人工智能（Conversational AI）中识别用户是否真正寻求法律指导（Legal Guidance）这一关键问题，尤其针对用户在涉及自身权利、义务、程序或可能产生法律后果的个人情境时所提出的各类咨询。其核心挑战在于区分真正寻求法律建议的请求与仅讨论法律主题但不具指导意图的普通对话。为此，论文提出了一种多维度分类框架（multi-axis taxonomy），从“是否寻求法律指导”、“请求的表达方式”以及“涉及的法律领域”三个维度对法律类查询进行系统标注，并将其应用于包含2,122条真实对话的WildChat-1M数据集上，构建了一个经人工审核的黄金标准评估集（gold evaluation set，共200条对话）。研究发现，在含法律内容的对话中，71.1%明确寻求个性化或实践性法律建议；其中，“权利与选项建议”是最常见的核心请求类型，且25.3%的积极指导类对话跨越多个法律议题。为验证模型性能，论文采用两阶段对比实验，将ModernBERT-base与ModernBERT-large模型与基于TF-IDF的级联方法进行比较。该研究的解决方案关键在于：构建一个结构化、可复现的标注体系和高质量公开数据集，从而支持大规模、可验证的对话式法律指导行为分析，为后续生成式AI在法律领域的应用提供坚实基础。
链接: https://arxiv.org/abs/2610.25159
作者: Amirreza Mohseni,Gerasimos Spanakis
机构: Maastricht University (马斯特里赫特大学)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 12 pages, 6 tables. Accepted for publication at JURIX 2026, Toulouse, France. This is the preprint version. Code and data available at: this https URL
Abstract:People ask conversational AI models about their legal rights, obligations, procedures, and personal situations that may have legal consequences. To study this behavior, we first need to distinguish requests for legal guidance from other legally themed discussions and unrelated conversations. We contribute a multi-axis taxonomy of legal queries capturing whether legal guidance is sought, how a request is formulated, and which legal domains it concerns; apply it to 2,122 naturally occurring WildChat-1M conversations; and create a human-reviewed gold evaluation set of 200 conversations. All annotations, model checkpoints, and evaluation code are publicly available. In this enriched sample, 71.1% of conversations containing legal content seek personalized or practical guidance. Rights-and-options advice is the most common primary request type among legal-content conversations, and 25.3% of guidance-positive conversations span multiple legal topics. The gold evaluation set supports a two-stage comparison of ModernBERT-base and ModernBERT-large against a TF-IDF cascade. Our framework supports large-scale studies of legal guidance in conversational AI, and the public release of the annotated data facilitates reproducible research and future model development.
[NLP-144] Do LLM s Reproduce the Age Gap in Loneliness? Evaluating Survey Simulation against European Survey Data
【速读】： 该论文旨在解决生成式 AI（Generative AI）在模拟问卷调查响应时，虽能匹配总体均值却难以准确再现特定人口统计学特征差异的问题，尤其关注其在反映国家内部年龄群体间孤独感差异上的表现。研究以2022年欧盟孤独感调查为基准，对比法国、意大利、波兰和瑞典四个国家中18–34岁与65岁及以上成年人群的孤独感水平，评估大语言模型（LLM）在有无年龄信息输入条件下的模拟效果，并与概率抽样样本及非概率敏感性参考进行比较。其解决方案的关键在于：区分并同时检验模型输出的组均值准确性与组间差距（即年龄差距）的还原能力，发现仅依赖整体平均值匹配不足以验证模拟质量；具体而言，部分模型（如Mistral Small 4和GPT-5.6 Terra）在未调整提示框架的情况下出现对老年群体均值的系统性高估，导致年龄差距方向反转，而DeepSeek V4 Flash虽保持正确趋势但整体低估两组均值，误差在部分国家相互抵消。研究表明，直接基于数据分布估计优于模拟方法，尤其在低效模拟细胞（如老年群体）存在偏差时，因此必须结合组内均值与组间对比双重标准来评估生成式AI作为真实人群分组代理的有效性。
链接: https://arxiv.org/abs/2610.25158
作者: Yuliya Li,Achim Rettinger,Dominik Beyer
机构: Trier University (特里尔大学); FZI Research Center for Information Technology (弗莱堡信息科技研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are often used to simulate survey respondents, but matching survey averages does not establish accurate demographic differences. We test whether LLM outputs reproduce within-country age gaps in loneliness using the 2022 EU Loneliness Survey, comparing adults aged 18-34 and 65 or older in France, Italy, Poland, and Sweden. We evaluate simulation with and without age information and direct distribution estimation against a probability-based sample, with a non-probability sensitivity reference. Both human samples show higher mean loneliness among younger adults. Under age-specified simulation with standard response-option order, Mistral Small 4 and GPT-5.6 Terra reverse this ordering under all three prompt framings in three and two countries, respectively, mainly by overestimating older-adult means. DeepSeek V4 Flash preserves the ordering everywhere but underestimates both group means, with errors largely cancelling in France and Italy. Against the probability-based reference, direct estimation has lower absolute age-gap error than simulation for Mistral and Terra in every country. For DeepSeek, simulation has lower gap error but direct estimation has lower group-mean error; these comparisons depend on low-validity older-adult simulation cells. Evaluating LLM outputs as proxies for age groups within countries therefore requires checking both group means and their contrast against human survey data.
[NLP-145] Morpheus: a fully local defense-in-depth de-identification engine for Greek clinical text - a benchmark evaluation
【速读】： 该论文旨在解决希腊语临床文本在离开医院前缺乏有效脱敏系统与基准评估的问题。当前临床自由文本若未经过充分脱敏，存在患者隐私泄露风险，而现有技术尚无针对希腊语医疗文档的专用脱敏工具与评测标准。为此，研究提出并评估了一种完全本地化、确定性的脱敏流程——Morpheus，其关键在于采用基于大语言模型（Gemini 3.1 Pro）生成的500份合成出院记录作为测试数据集，并通过双盲独立临床专家标注与仲裁构建高质量标注金标准。该方案的核心优势在于：在保证高召回率的前提下，实现96.59%的标识符去除率（直接标识符达98.49%，准标识符为93.42%），同时将过度脱敏率控制在2.82%以下；此外引入独立的出站审查层（egress-review layer）以检测残余泄露，进一步提升安全性。实验结果表明，系统具备高度可重复性且主要残余泄露集中于机构信息（如院徽），而非患者个人身份，验证了该方法在保护患者隐私的同时最小化对临床内容干扰的有效性。
链接: https://arxiv.org/abs/2610.25155
作者: Filippos-Paschalis Rorris(1 and 2),Nikoleta Stanitsa(3),Tatiana Rorri(1),Foteini Rorri(4),Constantine N. Antonopoulos(5) ((1) Onassis Hospital, Athens, Greece, (2) Great Ormond Street Hospital for Children, London, United Kingdom, (3) Evangelismos General Hospital, Athens, Greece, (4) Universita della Svizzera italiana, Lugano, Switzerland, (5) Attikon University Hospital, University of Athens Medical School, Athens, Greece)
机构: 未知
类目: Computation and Language (cs.CL)
备注: 24 pages, 2 figures, 3 tables; supplementary material appended. Benchmark corpus: doi: https://doi.org/10.5281/zenodo.22824894
Abstract:Background and Objective: Clinical free text cannot leave the hospital without de-identification, yet no de-identification system or benchmark exists for Greek clinical text. We evaluated Morpheus, a fully local de-identification pipeline for Greek clinical documents, on 500 synthetic discharge summaries. Methods: The notes (300 cardiac surgery, 100 cardiology, 100 nephrology) were generated by a frontier large language model (Gemini 3.1 Pro) constrained to authentic Greek hospital document layouts. Notes were pre-annotated at generation; two clinicians then independently reviewed and corrected every note, blinded to each other, with disagreements arbitrated. Primary outcomes were recall over must-redact identifiers and over-redaction of clinical content, measured over must-survive probes. Secondary outcomes: residual leaks flagged by an independent egress-review layer, processing time and reproducibility. Results: Inter-annotator agreement was F1 = 0.998. The de-identification rate was 96.59% (7,666/7,937 identifier instances; 95% CI 96.16-96.96): 98.49% for direct and 93.42% for quasi-identifiers. No patient’s own complete identifier survived in any note; of 271 residual instances, 199 (73.4%) concerned institutional identity, chiefly the hospital letterhead. The over-redaction rate was 2.82% (190/6,747 probes; 95% CI 2.45-3.24). The egress guard flagged 2 of 271 residuals, both complete telephone numbers. Two independent runs gave identical verdicts. Conclusions: A fully local, deterministic pipeline removed 98.5% of direct patient identifiers from Greek clinical text while altering under 3% of protected clinical content; most residuals concerned the institution, not the patient. To our knowledge this is the first benchmark evaluation of clinical-text de-identification for Greek and the first quantified evaluation of an independent egress-guard layer.
[NLP-146] PLANE: Plug-and-Play Fine-Grained Visual Experts for Hallucination Mitigation in Large Vision-Language Models EMNLP2026
【速读】： 该论文旨在解决大视觉语言模型（Large Vision-Language Models, LVLMs）中存在的幻觉问题，即模型在图像理解过程中生成不存在的物体或错误的属性与空间关系。研究表明，基于CLIP风格的视觉编码器通常关注图像中的主要物体而忽略细节信息，这成为导致幻觉的重要原因。为此，本文提出了一种即插即用的细粒度视觉专家框架（PLANE），其核心在于两个关键组件：一是针对CLIP架构的多方面图像编码的LoRA视觉专家（LoRA visual experts），分别用于捕捉对象、属性及对象间关系；二是具备指令感知能力的门控网络（instruction-aware gate network），实现对不同视觉特征的自适应融合。训练阶段，构建了一个细粒度图像-描述对数据集，从对象、属性和关系三个维度对图像进行详尽标注，并分别训练对应的三类LoRA专家。为全面评估方法性能，作者还人工构建了一个不依赖GPT-4的细粒度幻觉评估基准（GIFT）。实验结果表明，所提方法在多种类型与规模的模型上均能有效降低幻觉率，且引入的推理延迟极小，显著优于现有先进基线方法。
链接: https://arxiv.org/abs/2610.25146
作者: Jianghao Yin,Lang Yu,Qianxi Yu,Qin Chen,Jie Zhou,Zhikai Lei,Liang He
机构: East China Normal University (华东师范大学); Shanghai Qiji Zhifeng Co., Ltd. (上海奇稷智峰科技有限公司)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted by EMNLP2026 Findings
Abstract:Large Vision-Language Models (LVLMs) commonly suffer from hallucinations, which appear as fabricating non-existent objects or outputting incorrect attributes and spatial relationships in images. Previous works empirically indicate that CLIP-style vision encoders tend to focus on primary objects while ignoring image details, which is one of the causes of hallucinations. To address this issue, we propose plug-and-play fine-grained visual experts (PLANE) comprising two components: LoRA visual experts operating on CLIP for multi-aspect image encoding, and an instruction-aware gate network for adaptive image feature fusion. During training, we construct a fine-grained image-caption pairs dataset, where images are described from three aspects: objects, attributes, and relations between objects. Image-caption pairs from different aspects are used to train the three fine-grained LoRA experts respectively. Aiming at comprehensively evaluate our method’s capabilities, we manually annotate a fine-grained hallucination evaluation benchmark for LVLMs which is GPT4-independent during evaluation phase (GIFT). We conduct experiments both on popular hallucination evaluation benchmarks and GIFT. The results show that our method effectively reduces hallucinations across models of different types and sizes while introducing minimal inference latency. Compared to advanced baselines, our model also demonstrates superior performance.
[NLP-147] IRE: Top-Down Formula Instantiation with Reliable Execution for Physics Problem Synthesis
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在物理推理能力提升过程中面临的训练数据稀缺与质量不足问题，尤其是如何生成既能准确反映物理规律、又具备可验证数值答案的大规模高质量训练数据。现有方法中，专家设计的问题虽能保证物理准确性但成本高昂；而基于大模型自动生成的问题则存在幻觉风险，导致答案和评估不可靠。为此，论文提出TIRE（Top-Down Formula Instantiation with Reliable Execution）框架，其核心在于通过自上而下的公式实例化与可靠的执行机制相结合：前者利用可复用的物理公式及其适用条件来控制生成问题中的物理关系与数值配置，实现对问题内容的精准建模；后者在问题生成前预先通过可靠执行获得参考答案，为问题构造与教师模型蒸馏提供统一的数值基准。这一双重机制既保障了问题内容的物理一致性，又实现了基于执行结果的可信监督，从而有效支持监督微调（Supervised Fine-Tuning, SFT）与强化学习（如GRPO）。实验表明，基于TIRE构建的TIRE-19k数据集在九个物理子领域覆盖广泛，在三个主流大模型与四个基准测试中均显著优于其他等量数据集，平均性能领先2.86个百分点，且进一步通过GRPO优化使Qwen3-8B模型准确率提升1.52个百分点。这充分证明了基于显式物理知识与可靠执行机制生成训练数据的有效性。
链接: https://arxiv.org/abs/2610.25134
作者: Xiaoyu Wu,Zhaoxin Yu,Xinglin Xiao,Nan Xu
机构: ScienceOne AI; State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences; School of Artificial Intelligence, University of Chinese Academy of Sciences
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Advancing physics reasoning in large language models requires scalable training data that connect physical laws, concrete problem scenarios, and verifiable numerical answers. Expert-designed problems offer top-down control over physical laws, conditions, and scenarios, but require costly human effort. LLM-based synthesis from texts scales efficiently, but model-generated answers and assessments leave supervision vulnerable to hallucinations. We introduce TIRE, a framework for Top-Down Formula Instantiation with Reliable Execution that brings knowledge-driven problem construction to large-scale data generation. Top-down instantiation uses reusable physical formulas and their applicability conditions to control the physical relations and numerical configurations of generated problems. Reliable execution establishes reference answers before problem generation, providing a common numerical foundation for problem construction and teacher distillation. Together, these mechanisms combine control over problem content with execution-grounded supervision for both supervised fine-tuning and reinforcement learning. Using TIRE, we construct TIRE-19k, spanning nine subfields of physics. Across 3 backbone LLMs and 4 benchmarks, TIRE-19k outperforms 5 representative datasets with equal amounts of SFT data, exceeding the strongest baseline by 2.86 percentage points on average. Further GRPO raises Qwen3-8B’s average accuracy by 1.52 percentage points over its TIRE-19k SFT checkpoint. These results demonstrate the value of generating physics training data from explicit physical knowledge with reliable execution.
[NLP-148] MemBubble: Benchmarking How Agent Memory Shapes LLM Response Diversity
【速读】： 该论文旨在解决大语言模型（LLM）代理中记忆系统对响应多样性造成的负面影响问题。现有研究多依赖合成交互数据且评估覆盖的记忆系统有限，难以在真实场景下充分衡量记忆机制对信息多样性的潜在抑制作用。为此，作者提出了MemBubble基准，基于真实用户-助手对话中提取的偏好构建了1000个测试实例。通过设计与用户偏好情境相异但保持主题一致的测试查询，量化记忆系统导致响应分布集中化的程度。实验结果显示，在多种LLM和记忆机制下，相较于直接提供完整历史交互信息，响应熵平均下降34.7%，凸显了记忆处理过程在多样性损失中的关键作用。此外，尽管尝试了用户侧干预及信息可访问性调整等策略，均未能有效弥合多样性差距。研究强调，在优化个性化的同时，必须同步考虑如何保留响应多样性与用户对信息的广泛获取能力，这对未来代理记忆系统的设计具有重要指导意义。
链接: https://arxiv.org/abs/2610.25127
作者: Jiajun Tan,Fei Sun,Xinyu Ma,Shixuan Zhang,Du Su,Qi Cao,Daiting Shi,Huawei Shen,Xueqi Cheng
机构: State Key Laboratory of AI Safety, Institute of Computing Technology, CAS(中国科学院计算技术研究所人工智能安全国家重点实验室); Baidu Inc.(百度公司); University of Chinese Academy of Sciences(中国科学院大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Memory serves as the core component of LLM-based agents, enabling personalized responses via extracting and using user preferences from past interactions. However, its use may reduce response diversity, thereby narrowing the range of information available to users. Existing evaluations rely largely on synthetic interactions and cover few memory systems, leaving the effects of memory on response diversity insufficiently evaluated in real-world scenarios. We introduce MemBubble, a benchmark of 1000 instances built on preferences extracted from real user-assistant conversations. To measure the extent to which memory concentrates response distributions, we construct test queries in scenarios distinct from user preferences while preserving topical relevance through shared preference categories. Across multiple LLMs and memory mechanisms, entropy decreases by 34.7% on average compared with directly providing the full interaction history, highlighting the important role of memory processing in diversity loss. We also explore simple user-side interventions and changes in information availability but find that these do not effectively close the diversity gap. Our findings highlight the need to consider preserving diversity and user access to information alongside improvements in personalization when designing and using agent memory systems.
[NLP-149] rackable Skill Library Lifecycle Management for Agent ic Reinforcement Learning
【速读】： 该论文旨在解决生成式强化学习（Generative Reinforcement Learning, GRL）中技能库（skill library）在长期任务中难以实现可靠生命周期管理的核心问题。现有方法因共检索技能间的“搭便车”现象、跨任务干扰难以区分以及缺乏针对特定技能的反事实证据，无法准确评估每项技能的真实贡献，导致技能冗余、失效或有害技能持续留存。其解决方案的关键在于提出TrackRL框架，通过将分组采样转化为在线随机对照实验（online randomized controlled experiments），在每个GRPO（Grouped Reward Policy Optimization）组内实现独立随机注入与组内对比，从而隔离并量化每个技能对任务表现的因果贡献。结合可靠性加权聚合机制，系统持续积累证据，动态追踪“技能效用”（衡量技能作为外部支撑的因果价值）与“模型掌握度”（反映能力内化程度）双指标，并将技能生命周期管理建模为基于证据触发的状态转移过程，涵盖生成、准入、受保护孵化、保留、退休或移除等阶段，同时引入软容量约束控制技能库规模。实验表明，TrackRL在有技能辅助和无技能依赖的评估中均优于当前主流基线，且分析揭示了技能价值随策略演化而动态变化的规律，验证了证据驱动状态转移对技能库演化的关键作用。
链接: https://arxiv.org/abs/2610.25126
作者: Xi Zhu,Yuanlin Duan,Changyi Yang,Kai Mei,Riyang Bao,Shichao Zhu,Dimitris N. Metaxas
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Skill-augmented agentic RL has emerged as an effective approach to improving LLM agents on long-horizon tasks. Despite this progress, individual skills may follow distinct trajectories as the policy evolves, with some becoming redundant after the model internalizes their capabilities, others remaining valuable as external scaffolding, and still others turning harmful. However, existing skill library systems fail to identify each skill’s true contribution due to free-riding among co-retrieved skills, difficulty confounding across tasks, and the absence of skill-specific counterfactual evidence, which prevents reliable skill lifecycle management. We propose TrackRL, an agentic RL framework where skill tracking drives skill library full lifecycle management. Specifically, within each GRPO group, we turn grouped sampling into online randomized controlled experiments. Independent random injection and within-group contrasts on the same task isolate each skill’s true contribution, and reliability-weighted aggregation accumulates evidence across groups. As evidence accumulates, we continuously track skill utility, which measures the skill’s causal contribution as external scaffolding, and model mastery, which reflects how far the corresponding capability has been internalized. Along this line, we formulate skill lifecycle management as an evidence-gated state-transition process that governs each skill from generation and admission through protected incubation to retention, retirement, or removal, while a soft capacity bound controls library growth. Experiments show that TrackRL outperforms recent skill-augmented RL baselines in both library-assisted and skill-free evaluation, with further analyses revealing how skill value shifts during policy learning and how evidence-driven state transitions shape skill library evolution. Our code is available at this https URL.
[NLP-150] Memories That Outlive Their Reason s: Dependency-Tethered Memory for Tool-Using LLM Agents
【速读】： 该论文旨在解决生成式智能体（Generative AI agents）在长期交互中因依赖关系失效而产生的“非局部性作废”（non-locality of invalidation）问题。具体而言，当某个签名密钥被撤销时，所有基于该密钥的验证结果（verdicts）应随之失效，但由于历史验证记录与密钥撤销信息之间缺乏时间或语义上的关联窗口，导致智能体无法有效识别并清除上游失效所引发的下游错误发布对象，从而造成大量不安全的输出。其核心解决方案是引入Tether机制：在记录（record）写入时，显式绑定每个验证结果所依赖的具体版本（versioned dependencies），并在读取时通过检查其依赖链中任意环节是否失效来动态撤回相应验证结果；该机制保持了原有读取接口不变，但实现了对依赖链的精准追踪与局部化验证。实验表明，在多领域模拟环境中，相较于A-Mem、Graphiti和Mem0等基线方法（分别发布9%~42%的上游失效对象），Tether仅发布0.3%的此类对象，同时减少37%的验证调用次数，且在开放权重模型和全存储窗口场景下仍保持显著优势。研究进一步揭示，依赖关系的准确捕获是实现安全性的关键——若依赖关系与验证结果未同步到达，则Tether的安全性将退化至9.4%~22.1%的下游不安全发布率，凸显了上下文一致性在依赖管理中的重要性。
链接: https://arxiv.org/abs/2610.25123
作者: Yiqi Wang,Jinqian Ju,Jiaqi Zhang,Zhangkai Wu,Yiqun Duan,Zequn Sun,Taotao Cai,Mingkai Zheng
机构: University of Southern Queensland, Australia(南昆士兰大学, 澳大利亚); Southern University of Science and Technology, China(南方科技大学, 中国); Jiangsu University, China(江苏大学, 中国); The University of Sydney, Australia(悉尼大学, 澳大利亚); Uploading Inc., Palo Alto, CA, USA(Uploading公司, 美国加利福尼亚州帕洛阿尔托市); Nanjing University, China(南京大学, 中国)
类目: Computation and Language (cs.CL)
备注:
Abstract:Agents now act on the Web, publishing packages, models and datasets through registry APIs, and remembering past verifications spares them the work. But a verdict rests on premises. When a signing key is revoked, everything built on it stops passing, yet the advisory names only the key, and the old verdicts of its dependents may never share a retrieval window with it. This non-locality of invalidation turns memory into a liability: in long simulated chats, an agent reading a memory of atomic facts publishes 62% of the objects whose failure lies upstream, each backed by a record that was true when written. Yet derivation is local even when invalidation is not. Tether records, at write time, the exact versions each verdict derives from, and at read time withdraws any verdict whose support has failed anywhere in the store; the agent reads the same fields as before. On Fallout, a generator of versioned worlds with hypothesis-blind simulated histories shared by every memory, Tether publishes 0.3% of upstream-failing objects across three domains when relations are stated with their verdicts, with 37% fewer verification calls than checking everything. A-Mem and Graphiti re-verify most candidates (0.83-0.86 checks per candidate), which limits their savings, yet publish 9% of upstream-failing objects; Mem0 relies on memory more and publishes 42%. Flushing every verdict on any failure is as safe as Tether but costs 32% more checks. The gain over fact memory holds with three open-weight backbones and against a whole-store window. These gains depend on capturing dependencies: when relations arrive in sessions apart from their verdicts, Tether’s downstream unsafe publication rate rises to 9.4-22.1%, and a weak writer that misses edges loses safety in the same way.
[NLP-151] Behave Like Experts: Fusing Data with Semantic Knowledge for Effective and Efficient Semantic Map Construction
【速读】： 该论文旨在解决现有自动生成语义地图模型（Semantic Map Models, SMMs）方法中忽视专家先验语义知识的问题。传统自动方法仅依赖数据约束（如连接性假设），忽略了人类专家在构建语义地图时所运用的深层语义相似性判断，导致搜索空间过大，影响模型构建的准确性和效率。本文提出一种融合语言模型内部表示中隐含分布语义知识的自上而下、数据驱动的算法，模拟专家行为对候选图进行重排序，从而在大规模搜索空间中优先聚焦于语义相似性更高的解。实验基于自收集的10个数据集和全面评估指标验证了该方法在构建SMMs上的高效性与有效性。
链接: https://arxiv.org/abs/2610.25122
作者: Zhu Liu,Zhen Hu,Yuxuan Liu,Ying Liu
机构: Shanghai University (上海大学); Tsinghua University (清华大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Semantic map models (SMMs) construct a graph that reflects conceptual similarity by comparing cross-lingual form-meaning mappings. Experts build such models not only by obeying the data constraint, i.e., the connectivity hypothesis, which states that conceptual nodes belonging to the same form must be connected, but also by incorporating their prior semantic knowledge. Current automatic methods ignore this knowledge, which harms both accuracy and efficiency by enlarging the search space of candidate graphs. In this paper, we fuse the distributional semantic knowledge hidden in the internal representations of language models into a top-down, data-driven algorithm. This expert-like behavior re-sorts the candidates in a huge search space, guided by conceptual semantics similarity. Experiments on a self-collected benchmark of 10 datasets and comprehensive metrics show that our method builds SMMs both effectively and efficiently.
[NLP-152] Generation-Heavy Hybrid Machine Translation
【速读】： 该论文旨在解决机器翻译（Machine Translation, MT）中源语言与目标语言资源不对称所带来的瓶颈问题，尤其是在源语言资源匮乏而目标语言资源丰富的语言对中，传统方法因依赖对称的双语资源（如平行语料库、转换规则或互语言表示）而难以有效实施。其核心挑战在于，现有MT技术普遍要求源-目标语言间在语法结构、词汇映射和语料规模上保持高度对称性，以保障翻译质量与系统可迁移性，但这一条件在实际应用中往往难以满足。为此，论文提出一种名为生成主导型混合机器翻译（Generation-Heavy Hybrid Machine Translation, GHMT）的新范式，其关键创新在于：利用丰富的目标语言符号资源（如语法树、词典等）从源语言句子的依赖句法表示出发，生成大量可能的目标语言结构变体（overgeneration），再通过统计目标语言资源（如单语语料库）对这些候选译文进行筛选与优化。该方法不依赖复杂的转换规则或互语言中间表示，仅需源语言的句法解析器和简单的一对多词典，从而显著降低对源语言资源的需求，同时通过目标语言侧的丰富资源实现高质量翻译，并增强了系统向新源语言的可重定向能力（retargetability）。
链接: https://arxiv.org/abs/2610.25107
作者: Nizar Yahya Habash
机构: 未知
类目: Computation and Language (cs.CL)
备注: Dissertation submitted to the Faculty of the Graduate School of the University of Maryland at College Park in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2003
Abstract:The state of the art techniques in Machine Translation (MT) require large amounts of symmetric resources from source and target languages. This is true regardless of whether the approach is Transfer or Interlingua, Symbolic or Statistical. Symmetry within these approaches is necessary to ensure quality, robustness and retargetability. In reality, such symmetry, whether it is in terms of structural transfer lexicons, interlingual dictionaries or parallel corpora, is a major bottleneck in developing any MT system. This dissertation presents an approach to MT that addresses the lack of symmetry by exploiting symbolic and statistical target language resources in source-poor/target-rich language pairs. This approach is called Generation-Heavy Hybrid Machine Translation (GHMT). Expected source language resources include a syntactic parser and a simple one-to-many translation dictionary. No transfer rules or complex interlingual representations are used. Rich target language symbolic resources are used to overgenerate multiple structural variations from a target-glossed syntactic dependency representation of source language sentences. Statistical target-language resource are then used to select amongst the overgenerated translations. The source-target asymmetry of systems developed in this approach makes them more easily retargetable to new source languages. Comments: Dissertation submitted to the Faculty of the Graduate School of the University of Maryland at College Park in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2003 Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.25107 [cs.CL] (or arXiv:2610.25107v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.25107 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-153] Deep Fusion Memory: A Chunk-Level Parametric Memory for Large Language Models
【速读】： 该论文旨在解决大语言模型在推理过程中如何高效获取和利用外部知识的问题，尤其针对传统方法中存在的灾难性遗忘风险（持续预训练）与推理时检索带来的高延迟（在线检索）之间的权衡。现有参数化记忆（parametric memory）方法通常依赖大量离线检索构建训练数据，且仅在输出层进行浅层融合，难以实现深层次的知识整合。本文提出深度融合记忆（Deep Fusion Memory, DFM），其核心创新在于通过因果记忆预测器（causal memory predictor）基于上下文生成记忆，并以语料库中检索到的块嵌入（chunk embeddings）作为监督信号，实现对记忆的端到端学习。所预测的记忆在块内共享，并通过门控交叉注意力机制（gated cross-attention）融入中间隐藏状态，从而实现深层、动态的知识融合。该方法在训练阶段显著降低对外部检索的依赖——相较于现有方法，减少32倍检索查询次数与2048倍检索项数；推理阶段无需显式检索，具备高效性。实验表明，DFM在六项问答任务上分别优于最强基线12.71%（Mistral-7B-v0.3）和11.19%（Qwen3-8B-Base），并在领域适应任务中分别带来6.51和4.38点的平均性能提升，验证了其有效性与泛化能力。
链接: https://arxiv.org/abs/2610.25100
作者: Rubin Wei,Siyuan Sun,Hao Sun,Yuliang Liu,Xiang Shi,Jiarui Wang,Junming Zhang,Zhouhan Lin
机构: LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院LUMIA实验室); Shanghai Artificial Intelligence Laboratory(上海人工智能实验室)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models need effective ways to acquire and use external knowledge beyond what they retain from pretraining. Continued pretraining risks catastrophic forgetting, while retrieval at inference time can add substantial overhead. External parametric memory avoids online retrieval, but constructing its training data requires extensive offline search, and output-level interpolation provides only shallow fusion. We introduce Deep Fusion Memory (DFM), a parametric memory architecture that reduces this construction workload and incorporates learned memory into intermediate hidden states. A causal memory predictor learns to generate memory from the preceding context, using retrieved chunk embeddings from the corpus as supervision. The predicted memory is shared across tokens within each chunk and fused into their hidden states through gated cross-attention. Memory prediction and fusion are trained end to end with the backbone frozen, and inference requires no explicit retrieval. Compared with existing parametric memory approaches, which often construct token-level memory targets, DFM requires 32\times fewer retrieval queries and 2048\times fewer retrieved items to construct its training data. Across six QA tasks, DFM performs better over the strongest baseline by 12.71% on Mistral-7B-v0.3 and 11.19% on Qwen3-8B-Base. Beyond QA, domain adaptation yields average gains of 6.51 and 4.38 points over Qwen3-4B-Base and Qwen3-8B-Base, respectively.
[NLP-154] Beyond Factual Accuracy: A Benchmark for Evaluating Perspective Alignment in LLM s
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在生成回答时存在的视角控制失效问题，即尽管生成内容在事实准确性上表现良好，却未能严格遵循用户指定的特定视角，导致多个观点混杂，迫使用户自行筛选无关信息。其核心解决方案是提出PerspectiveBench基准测试框架，通过构建存在多种合理推理路径的场景，量化模型在多视角情境下的视角一致性，并引入视角对齐得分（Perspective Alignment Score, PAS）作为评估指标。实验结果表明，当前最先进的模型虽能包含目标内容，但无法有效排除非请求视角；引入检索增强生成（Retrieval-Augmented Generation, RAG）虽提升了事实基础性，却未能解决视角漂移问题，因模型更倾向于全面覆盖而非严格遵守用户设定的约束。
链接: https://arxiv.org/abs/2610.25087
作者: Sunghoon Kim,Sangyeop Kim
机构: Coxwave; Ewha Womans University (Ewha女子大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large Language Models often generate factually correct responses that fail to maintain user-requested perspectives, blending multiple perspectives and forcing users to filter through unwanted information—a perspective control failure that factual accuracy metrics cannot detect. We introduce PerspectiveBench, a benchmark quantifying perspective alignment by constructing scenarios where multiple valid reasoning paths coexist, and propose the Perspective Alignment Score (PAS) to measure adherence to specified perspectives. Experiments reveal that state-of-the-art models successfully include target content but fail to exclude non-requested perspectives. Providing documents through RAG improves factual grounding but does not resolve perspective drift, as models prioritize comprehensive coverage over strict user constraints. Code and data are available at this https URL.
[NLP-155] Dual-Slot: From Serial Loops to Parallel Slots in Language Models
【速读】： 该论文旨在解决循环语言模型（Looped Language Models）在提升模型质量时因迭代间序列依赖导致的自回归解码延迟问题。其核心挑战在于：尽管通过重复使用共享层可增加计算深度而不增加参数量，但各迭代步骤必须串行执行，从而限制了推理效率。解决方案的关键是提出一种名为Dual-Slot的新型架构，将传统的串行循环计算重构为并行的“隐状态槽”（latent slot）与“预测槽”（prediction slot）双槽机制。该设计允许在生成当前词元的同时，利用前一个词元的深层隐状态进行预测，同时在隐状态槽中并行计算下一个词元的隐状态，从而将原本两个串行的模块调用转化为一次并行的配对调用，有效打破相邻词元间的隐状态依赖链。为保持训练阶段的词元级并行性，引入少量雅可比（Jacobi）迭代以并行优化整个序列的隐状态。实验结果表明，该方法在Qwen风格的密集模型和专家混合（MoE）模型上均显著提升了语言建模性能与下游任务平均准确率，同时在不同批量大小下仍保持高达93.1%-99.6%的原始解码吞吐量，实现了性能与效率的双重优化。
链接: https://arxiv.org/abs/2610.25086
作者: Yiqin Hao,Zitong Wang,Boyi Zeng,He Li,Shixiang Song,Feichen Song,Yifan Liu,Zhouhan Lin
机构: LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院LUMIA实验室); Sun Yat-sen University(中山大学); Shanghai AI Laboratory(上海人工智能实验室)
类目: Computation and Language (cs.CL)
备注: 17 pages, 6 figures
Abstract:Looped language models increase computational depth by repeatedly applying shared layers, offering a way to improve model quality without adding parameters. However, sequential dependencies between loop iterations increase autoregressive decoding latency. We introduce Dual-Slot, an architecture that reorganizes serial looped computation into parallel latent and prediction slots. Rather than waiting for the current token’s latent computation to finish, the prediction slot combines the previous token’s deep latent state with the current representation, while the latent slot produces a state for the next decoding step. This shifts the latent dependency across adjacent tokens, replacing two sequential module invocations with a single paired invocation while preserving standard autoregressive generation. To preserve token-level parallelism during training, we use a small number of Jacobi iterations to refine latent states across the sequence in parallel. Experiments on Qwen-style dense and mixture-of-experts (MoE) models show consistent improvements in language modeling and average downstream accuracy. On the 1B dense model, Dual-Slot reduces Pile validation perplexity from 7.25 to 7.00 and improves average five-shot accuracy across nine task configurations by 2.26 percentage points. After supervised fine-tuning, it improves zero-shot GSM8K accuracy by 5.83 and 3.49 percentage points for the 1B dense and A0.6B MoE models, respectively. Across the evaluated batch sizes, Dual-Slot retains 93.1-96.5% of vanilla decoding throughput for the dense model and 98.1-99.6% for the MoE model.
[NLP-156] CoEvo-RSI: When Harnesses Evolve Which Models Should Follow?
【速读】： 该论文旨在解决生成式智能体在递归自我改进（Recursive Self-Improvement, RSI）过程中普遍存在的“脑-体不匹配”问题，即不断演化的执行框架（harness，体）与模型（model，脑）当前能力之间存在脱节：要么要求模型执行其尚未学习的行为，要么限制了其已具备的能力。针对这一挑战，论文提出CoEvo-RSI框架，其核心解决方案是引入由智能体驱动的双向历史继承机制。该机制通过共享的演化档案库与执行反馈，由元智能体（meta-agent）协调从历史设计中继承的框架开发，并结合模型检查点的选择或融合，以及在选定框架下的后续适应性调整，实现能力与执行需求的动态对齐，从而有效缓解脑-体不匹配。该方法仅依赖自身拒绝采样轨迹的监督微调（Supervised Fine-Tuning, SFT），便使Qwen3.5-9B在多智能体故障归因、角色感知文本到SQL转换及智能家居指令遵循等多样化垂直应用中达到与更大模型相当的性能，且在TraceElephant任务上超越DeepSeek-V4-Pro（+14.16%智能体准确率），在HomeBench VM上优于GLM-5.3（+4.90%成功率），验证了脑-体对齐的重要性及双向历史继承在推进智能体协同演化中的潜力。
链接: https://arxiv.org/abs/2610.25078
作者: Jiaqi Su,Guibin Zhang,Jie-Jing Shao,Cong Pang,Tiankuo Yao,Ivor Tsang,Haiyan Yin,Shuicheng Yan
机构: National University of Singapore (新加坡国立大学); Agency for Science, Technology and Research (A*STAR) (新加坡科技研究局); SenseTime Group Inc. (商汤科技有限公司)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Model–harness co-evolution has emerged as a prominent paradigm for recursive self-improving (RSI) agents, jointly evolving their execution scaffolds and intrinsic capabilities. We study the persistent yet often overlooked brain–body mismatch problem in this paradigm, where an evolving harness (body) demands behaviors its model (brain) has not learned or restricts capabilities it already possesses. We introduce CoEvo-RSI, a RSI framework that advances model–harness co-evolution through agent-driven, bidirectional historical inheritance. Drawing on a shared evolutionary archive and execution feedback, a meta-agent coordinates harness development from historical designs with model checkpoint selection or merging and subsequent adaptation under the selected harness, aligning inherited capabilities with evolving execution requirements to reduce brain–body mismatch. Using only supervised fine-tuning (SFT) on its own rejection-sampled trajectories, CoEvo-RSI makes Qwen3.5-9B competitive with larger models across diverse vertical applications, including multi-agent failure attribution, role-aware text-to-SQL, and smart-home instruction following. It surpasses DeepSeek-V4-Pro on TraceElephant (+14.16 agent accuracy) and GLM-5.3 on HomeBench VM (+4.90 success rate), underscoring the need for brain–body alignment and the potential of bidirectional historical inheritance to advance agent co-evolution.
[NLP-157] P2N: Reusing Deep Representations for Greater Effective Depth
【速读】： 该论文旨在解决大语言模型在数学问题求解与自主智能体等需要强推理能力的应用中，因计算深度（computational depth）不足而导致的性能瓶颈问题。现有方法通过增加模型层数或采用测试时扩展（test-time scaling）来提升推理能力，但均带来显著的训练与推理开销。本文提出P2N（Previous to Next）机制，其核心创新在于复用生成过程中已有的前一词元深层表示（deep representations），通过将前一词元的深层隐藏状态叠加至当前词元的浅层表示中，实现跨词元的有效计算深度扩展。该方法仅引入每步解码一次向量加法，几乎不增加解码延迟。通过在中间层核心区域施加并行雅可比更新（parallel Jacobi updates）进行训练，使模型能够有效利用历史计算成果。实验表明，在15亿参数的Qwen类模型上，P2N在零样本与少样本提示下均使GSM8K基准准确率提升超过10个百分点，同时在语言建模及通用下游任务基准上也表现出性能增益，实现了推理能力增强与计算成本控制之间的良好平衡。
链接: https://arxiv.org/abs/2610.25076
作者: Yiqin Hao,Boyi Zeng,Zitong Wang,He Li,Shixiang Song,Feichen Song,Yifan Liu,Zhouhan Lin
机构: LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院LUMIA实验室); Sun Yat-sen University(中山大学); Shanghai AI Laboratory(上海人工智能实验室)
类目: Computation and Language (cs.CL)
备注: 23 pages, 10 figures
Abstract:Mathematical problem solving and autonomous agents are increasingly important applications of language models, demanding stronger reasoning capabilities. Computational depth is a key resource for such reasoning, but adding model layers increases both training and inference costs, while test-time scaling incurs substantial additional inference computation. We introduce P2N (Previous to Next), which reuses the previous token’s deep representations, already available during generation, to increase effective computational depth with negligible decoding overhead. Specifically, P2N adds the previous token’s deep hidden state to the current token’s shallow representation, allowing subsequent layers to build on previously completed deep computation. This recurrence extends effective computational depth across tokens, while introducing only a single vector addition per decoding step. We train the recurrence using parallel Jacobi updates restricted to a middle-layer core. Experiments with Qwen-style models show that, at 1.5B parameters, P2N improves GSM8K accuracy by over 10 percentage points under both zero-shot and few-shot prompting, with nearly unchanged decoding latency. The benefits also extend to language modeling and average performance on general downstream benchmarks. on general downstream benchmarks.
[NLP-158] Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of “Unsafe” You Trained On NEURIPS2026
【速读】： 该论文旨在解决生成式AI（Generative AI）中安全护栏（guardrail）评估体系中存在的根本性问题：当前的安全基准测试结果往往被误认为是模型内在安全属性的反映，实则高度依赖于训练所采用的标签定义（label definition），即政策（policy）的设定。其核心解决方案在于揭示并量化标签定义对评估结果的决定性影响——即使在架构、代码库与优化器完全一致的前提下，仅因训练标签基于不同操作性定义（如“是否请求有害内容？”或“审核员是否会屏蔽此内容？”），同一模型在多个安全护栏臂（guardrail arms）上的表现差异可达23倍之大，且部分基准分数可通过微调0.03%的参数实现全范围变动。研究进一步表明，验证集上的准确率主要反映模型与标注者政策的一致性（ρ=+0.74），而非真实检测能力（ρ=+0.34），说明现有评估机制严重受制于标签定义的偏倚。因此，论文提出应将数据污染程度、初始种子基线以及评估所依据的标签定义作为评估报告中的第一类变量，以提升安全护栏评估的透明度与可复现性。
链接: https://arxiv.org/abs/2610.25060
作者: Vahid Jebraeeli,Jonathon Romero,Yijin Wang,Adrian Chan,Jianbang Zhang
机构: Lenovo(联想)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注: Accepted as a poster at the NeurIPS 2026 Workshop TAE (Trust-AI-Eval): Can We Trust AI Evaluation?, Sydney. 20 pages, 5 figures
Abstract:A guardrail classifier is reported as if unsafe were a property of text. It is not: it is a property of a policy, and the policy enters the model through its labels. We exploit an accident of an industrial deployment in which the same ~53k prompts were labelled twice, under two operational definitions (“is this a request for harmful content?” and “would a moderator block this?”), and used to train ten guardrail arms sharing architecture, codebase and optimiser family, at two model scales. First, the guardrail’s own training corpus already contained the benchmarks: JailbreakBench-benign is 88% and OR-Bench-hard 82% present in the training pools, two over-refusal benchmarks, the kind a guardrail reports to argue it does not over-block. Everything below is decontaminated. Second, safety benchmarks stratify sharply by how much of their score is decided by the label definition rather than by the model: with architecture and optimiser fixed, the range across the ten arms is 0.99 on HarmBench-copyright and 0.043 on WildGuard-benign, a 23x spread. Third, a benchmark score can be moved across its entire range by 0.03% of the parameters. Across these arms HarmBench-copyright spans 0-100 and OR-Bench-hard over-refusal 3.4-56.4%, and the four arms whose last supervision signal carries a given policy occupy one extreme of both (exact rank p=0.005, or 0.04-0.14 at worst), at unchanged detection capability. For a distilled student the teacher’s soft labels do this even when every hard label carries the other policy, so teacher provenance is an unrecorded confound in any released checkpoint. Fourth, validation accuracy on the one split that grades all twenty arms tracks agreement with the annotator’s policy (rho=+0.74), not detection capability (rho=+0.34). We argue that guardrail evaluations should report contamination, a seed floor and the definition they were graded against as first-class variables.
[NLP-159] Beyond the Teacher: Understanding and Preserving Distillation Overshoot in Agent ic On-Policy Distillation
【速读】： 该论文旨在解决生成式智能体在多轮任务中进行在线策略蒸馏（On-Policy Distillation, OPD）时出现的“蒸馏过冲”（distillation overshoot）现象，即初始较弱的学生模型在训练初期能够超越教师模型，但随着蒸馏持续进行，性能逐渐回落甚至低于教师。其核心问题在于：OPD在纠正学生错误决策的同时，会无差别地削弱学生中已被证明有效的成功行为（尤其是教师不认可的策略），导致后期性能退化。解决方案的关键在于提出成功保留型在线策略蒸馏（Success-Preserving On-Policy Distillation, SPOD），通过两个机制实现：一是对成功轨迹中的学生选择进行冲突掩码，避免教师监督对其产生侵蚀；二是引入软概率下限以抑制成功行为的概率衰减，从而在保持教师指导有效性的同时，保护学生自身已验证的有效策略。SPOD仅依赖训练过程中已有的二元任务成功信号和模型输出概率，无需额外标注或复杂设计。实验表明，在两个模型家族和三个智能体基准上，SPOD显著优于标准OPD及其变体，最终成功率最高提升15.2个百分点，且在所有六组设置中均超越教师模型，甚至优于普通OPD的最佳中间检查点，证实了超越教师性能不仅可实现，且能被稳定保持与扩展。
链接: https://arxiv.org/abs/2610.25034
作者: Jike Zhong,Ming Li,Yuxiang Lai,Xi Xiao,Sheng Zhang,Ritwick Chaudhry,Tianle Chen,Jingyu Xie,Zheda Mai,Shao-Yuan Lo,Konstantinos Psounis,Yuanyuan Lei
机构: University of Southern California(南加州大学); University of Florida(佛罗里达大学); Emory University(埃默里大学); University of Alabama(阿拉巴马大学); University of Maryland(马里兰大学); Amazon AGI(亚马逊AGI); Boston University(波士顿大学); The Ohio State University(俄亥俄州立大学); National Taiwan University(台湾国立大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Code: this https URL
Abstract:On-policy distillation (OPD) trains agents on their own trajectories using supervision from a stronger teacher. We uncover a counterintuitive phenomenon in multi-turn agentic tasks: a smaller, initially weaker student can substantially outperform its teacher during training, only to regress as distillation continues. We term this phenomenon distillation overshoot. Through decision- and task-level analyses, we trace overshoot to two competing effects of OPD: repairing weak student decisions and eroding successful student choices that the teacher disfavors. Early in training, repair outpaces erosion, allowing the student to combine the teacher’s corrections with its own strengths and surpass the teacher. As training continues, erosion overtakes repair, causing these gains to vanish. Motivated by this finding, we propose Success-Preserving On-Policy Distillation (SPOD), which selectively masks conflicting teacher supervision on successful student choices and introduces a soft probability floor to counteract their erosion, while retaining teacher guidance elsewhere. SPOD requires only binary episode success and model probabilities already available during OPD. Across two model families and three agentic benchmarks, SPOD consistently outperforms ordinary OPD and evaluated variants, improving final success by up to 15.2 percentage points and surpassing the teacher by up to 20.0 points. Moreover, SPOD’s final checkpoints outperform even ordinary OPD’s best intermediate checkpoints in all six settings, demonstrating that beyond-teacher gains can be not only preserved but further extended. Code: this https URL
[NLP-160] One Head Away from Language Confusion: Single Attention Heads Across Model Families
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在生成回复时出现的“语言混淆”（language confusion）问题，即模型未能遵循用户指定的语言而生成非预期语言的输出。其核心解决方案的关键在于识别并验证特定注意力头（attention head）在语言选择中的决定性作用：研究通过消融实验发现，在12个不同模型检查点中，有10个模型存在一个或多个特定注意力头，其被置零后会导致显著偏离预期语言的输出，且该现象在层内其他头中尤为突出；这些关键头可使8.0%至51.5%的FLORES-200续写任务输出脱离提示语言，远超同层其他头的影响（≤1.4%）。进一步实验表明，将某一语言请求名称对应的头值替换为另一语言的均值，可诱发模型错误响应并引导其转向目标语言，而随机替换则无此效果；更关键的是，将某头输出替换为另一语言的均值，能使5个模型中84%至99%的非英语单语LCB回复被重定向至目标语言。值得注意的是，仅在Gemma系列模型中，使用目标语言均值能显著提升跨语言回复质量，表明其基线表现较低，从而凸显了该机制的有效性。这一发现揭示了语言选择行为在模型内部的高度局部化与可操控性，为理解并干预语言混淆提供了重要线索。
链接: https://arxiv.org/abs/2610.25026
作者: Jiwon Jeong,Anjelo Jann Laroza,Chaewoon Yoon,Arjun Pillai,Pranav Kokati,Christian Hoang,Seunghyeok Hong
机构: Lotte Innovate; Mapúa University; University of Wisconsin–Madison; Irvington High School; Interlake High School; FPT University; Division of Social Science AI, Hankuk University of Foreign Studies
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models sometimes reply in an unexpected language, a failure known as language confusion. We ablate individual attention heads in 12 model checkpoints spanning seven model series, using FLORES-200 continuations in five languages and Language Confusion Benchmark (LCB) chat replies. In ten models, zeroing one head causes substantially more departures from the expected language than zeroing any other head in its layer; its LCB pass-rate drop persists, though reduced, under mean ablation. Seven heads move 8.0 to 51.5 percent of all FLORES continuations out of the prompt’s language, versus at most 1.4 percent for other same-layer heads; the other three mainly affect explicitly requested languages. These seven heads matter less in base models given plain text. In two models, replacing the values a head reads at the requested-language name with another language’s mean makes replies fail and shifts some into that language, unlike a random change of equal size. Replacing a head’s output with another language’s mean redirects 84 to 99 percent of scored non-English monolingual LCB replies into that language in five models. Yet the requested language’s mean improves crosslingual replies beyond every control head only in the two Gemma models, whose baselines are low. Code and results are available in the public repository at this https URL.
[NLP-161] SoK: PIAtlas: Decomposing the Unbounded Prompt Injection Attack Space
【速读】： 该论文旨在解决大语言模型（Large-Language-Model, LLM）应用中提示注入（Prompt Injection）攻击缺乏统一分类体系的问题。当前超过二十个公开数据集对提示注入的描述使用不一致的术语，通常基于传递向量、攻击目标或生成方式分类，导致研究难以系统化比较与评估。本文提出PIAtlas，一个以注入指令构造机制为核心的新分类体系，其六个类别分别对应六种不同的模型劫持（subversion）机制，构成技术与子技术的层级结构，为攻击描述提供一致性框架，并支持更系统的评估设计。该分类体系基于大语言模型的训练方式、提示工程及人类认知框架的影响进行构建，并通过与历史攻击数据对比验证。为支撑此分析，作者发布了PICorpus——目前已知最大且许可证清洁的公共提示注入数据集集合，并采用人工验证协议将数千条记录映射至PIAtlas。结果显示，多数攻击融合多个技术但仅源自少数机制；同时揭示现有标签体系的盲区：主流代理基准测试仅覆盖攻击空间的一小部分，而自适应攻击者更倾向使用基准所低估的机制。因此，PIAtlas为理解提示注入攻击空间、提升基准覆盖率以及指导红队测试与防御策略提供了统一基础。
链接: https://arxiv.org/abs/2610.24999
作者: Stavros Damianakis,Konstantinos Psounis
机构: University of Southern California (南加州大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 61 pages, 1 figure, 23 tables. Code and data: this https URL | Interactive taxonomy: this https URL
Abstract:Prompt injection is the dominant security risk for large-language-model (LLM) applications and agents, yet the field still lacks a coherent map of how these attacks are constructed. More than twenty public datasets describe prompt injection using incompatible vocabularies, often organized around delivery vector, attacker goal, or generation method. In this Systematization of Knowledge, we present PIAtlas, a taxonomy organized instead around how the injected instruction is constructed to subvert the underlying model. Its six classes, one per subversion mechanism, form the roots of a hierarchy of techniques and sub-techniques that researchers can use to describe attacks consistently and design more systematic evaluations. We derive the taxonomy from how LLMs are trained, prompted, and influenced by human-like framing, and compare it with historical attack data. To ground this comparison, we release PICorpus, to our knowledge the largest license-clean collection of public prompt-injection datasets, and map thousands of its records to PIAtlas under a human-verified protocol. The mapping shows that attacks usually combine several techniques drawn from only a few mechanisms. It also reveals gaps that existing labels obscure: widely used agent benchmarks test only part of the attack space, and adaptive attackers favor mechanisms that curated benchmarks underrepresent. PIAtlas thus provides a common foundation for understanding this attack space, improving benchmark coverage, and guiding red teaming and defenses.
[NLP-162] When Does Better Acceptance Pay? A Controlled Study of Speculative Decoding
【速读】： 该论文旨在解决生成式 AI（Generative AI）中推测解码（speculative decoding）的吞吐量提升受限于接受长度（acceptance length）与执行延迟（execution latency）之间权衡的问题，核心关注点在于：在何种条件下增加接受长度所带来的收益能够覆盖其带来的成本。其解决方案的关键在于建立一个涵盖训练、推理与评估全流程的统一比较协议，通过在共享目标生成数据上重测已发布的检查点并重新训练六种起草器（drafter）配方，在十个基准任务和两种目标模型上进行系统性验证。研究发现，对齐（alignment）显著改变了原有性能差距的呈现——公开报告中的显著差异在统一训练与推理条件下大幅缩小。这一现象引出两个深层问题：如何以不同成本提升接受率，以及这些提升在何时真正带来吞吐量增益。研究揭示，接受率可通过近乎零成本的目标函数调整或上下文变更实现，也可通过代价更高的逐标记（token-to-token）条件化手段提升。其中，DCore 作为前一路径的典范，展现出极低的额外起草开销，可作为强基线；然而其收益具有条件性，受任务不确定性与服务环境的影响，进而决定接受回报、执行成本及最优起草器选择与预算配置。该研究为未来起草器的设计与部署策略提供了关键实证依据。
链接: https://arxiv.org/abs/2610.24993
作者: Yongchao Long,Chunrui Zheng,Yuxuan Zhou,Guangkun Nie,Tian-Chi Ma,Chao Pang,Shuohuan Wang,Yuchen Ding,Xianbin Wen,Dianhai Yu,Junyuan Shang,Yuxi Zhou,Shenda Hong
机构: Tianjin University of Technology (天津理工大学); Baidu Inc.(百度公司); Shandong University (山东大学); Peking University (北京大学); Nanjing University (南京大学); National Institute of Health Data Science, Peking University (北京大学健康数据科学国家实验室)
类目: Computation and Language (cs.CL)
备注: 24 pages, 11 figures. Project page: this https URL
Abstract:The throughput gains of speculative decoding depend on the trade-off between acceptance length and execution latency. We ask when additional acceptance is worth the cost of obtaining it. We establish a comparison protocol spanning training, inference, and evaluation, remeasuring released checkpoints and retraining six drafter recipes on shared target-generated data across ten benchmarks and two target models. Alignment changes the picture: gaps that appear large in published comparisons shrink substantially under shared training and inference conditions. This narrowing motivates two deeper questions: how is acceptance improved at different costs, and when do these gains improve throughput? We find that acceptance can be improved through nearly free objective and context changes or more expensive token-to-token conditioning. DCore illustrates the former route as a strong baseline with little additional measured drafting overhead. The payoff is conditional: task uncertainty and serving conditions affect acceptance returns, execution costs, and the preferred drafter and budget. These findings inform future drafter design and deployment choices.
[NLP-163] Certified Sparse Attention via Perturbed Retrieval
【速读】： 该论文旨在解决长上下文生成中稀疏注意力（Sparse Attention）在实际KV缓存场景下无法有效控制查询误差的问题，尤其针对现有方法在真实缓存环境下误差分布呈现重尾（heavy-tailed）的缺陷。其核心解决方案是引入独立同分布的指数分布（Exp(1)）扰动对每个键的注意力得分进行随机化处理，将阈值检索转化为按注意力权重比例采样，从而实现可证明的误差控制。通过沿几何网格逐步降低阈值，构建嵌套采样结构，并基于霍尔维茨-汤普森估计（Horvitz–Thompson estimator）在每一层获得输出估计与同时成立的置信区间，使得每个注意力头可在首个满足目标误差 ε 的层级停止采样。结合每解码步重新施加扰动，确保每个证书失败概率不超过 δ。在合适阈值下，该估计器读取的键数量为 $ O(\varepsilon^{-2} \log(1/\delta)) $，与上下文长度无关，且该复杂度在常数意义下达到理论最优。此外，论文还提出一种4比特扫描机制，在组查询注意力（grouped-query-attention）中共享确定性边界，虽需读取所有键但能精确生成候选。在三个大语言模型（LLM）及10种设置（16K–128K token）下的实验表明，无一证书被违反，且在相同读取键数下，90百分位误差比最佳基线低4.6–7.9倍；在相同字节数下，基线误差为本文的0.6–1.6倍，最大误差更是高达4.9倍以上。端到端评估显示，该方法在任务性能上与密集注意力相差不超过1点，PG-19困惑度仅低0.1%，且在等字节条件下，最强基线表现也与之相当。在CPU上，32K–128K token的解码速度相较相同16位缓存的密集注意力提升1.1–1.4倍。
链接: https://arxiv.org/abs/2610.24991
作者: Runguo Li
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 58 pages (10 pages main text, 48 pages appendix), 10 figures, 35 tables
Abstract:Sparse attention speeds up long-context decoding by reading few keys of the KV cache, but existing methods do not control the error of a query, and on real KV caches their errors have heavy tails. We show that perturbing the score of every key with an independent Exp(1) variable turns threshold retrieval into sampling keys with probability proportional to their attention weights. Lowering the threshold along a geometric grid gives nested samples, a Horvitz–Thompson estimate of the output at every level, and confidence bounds that hold simultaneously over all levels, so each head stops at the first level whose certificate meets the target error \varepsilon . With fresh perturbations at every decoding step, each certificate fails with probability at most \delta during generation. At a suitable threshold, the estimator reads O(\varepsilon^-2\log(1/\delta)) values regardless of the context length, and no adaptive method reads fewer, up to constants. A 4-bit scan with deterministic bounds, shared across a grouped-query-attention group, generates the candidates exactly but reads every key. On three LLMs and ten settings at 16K–128K tokens, no certificate is violated, and at equal keys read our 90th-percentile error is 4.6–7.9 \times lower than that of the best of 15 baseline families. At equal bytes, the best baselines have 0.6–1.6 times our 90th-percentile error and at least 4.9 times our largest. End to end, our method stays within one point of dense attention on tasks and within 0.1% in PG-19 perplexity, and the strongest baselines at equal bytes score as well. On CPU, decoding steps at 32K–128K tokens run 1.1–1.4 \times faster than dense attention on the same 16-bit cache.
[NLP-164] he World as Viewed by 136 Countries: An Open Curated Annotated and Multilingual Dataset of 118 Million State Media Articles
【速读】： 该论文旨在解决生成式 AI（Generative AI）训练数据中广泛存在的国家媒体（state media）内容所带来的偏见与叙事影响问题。随着国家媒体在互联网语料库中的占比持续上升，其潜在的意识形态倾向可能被模型学习并反映在输出结果中，尤其在多语言和跨国背景下对信息公平性构成挑战。解决方案的关键在于构建并公开发布一个大规模、多语言、跨国家的国家媒体语料库——WorldStateMedia，包含来自136个国家、114种语言的1.18亿篇国家媒体文章，并结合“国家媒体监控”（State Media Monitor）对媒体机构进行权威的政府控制程度标注。通过该数据集，研究发现英语网络爬取数据中受编辑控制的国家媒体占比从2013年的39%飙升至2025年的93%，在部分低资源语言如维吾尔语（17.8%）和白俄罗斯语（7.8%）中亦占显著份额；进一步分析显示俄语与乌克兰语国家媒体在议题聚焦与争议事件（如克里米亚）的框架上存在明显对立。在检索增强生成（RAG）实验中，检索目标国国家媒体内容会显著使模型回答向该国立场倾斜，尤其是在面向外国受众的英文媒体中影响最为突出。因此，该研究通过构建可审计的基准数据集，为识别和缓解生成式 AI 中由国家媒体主导的系统性偏见提供了关键工具与方法支持。
链接: https://arxiv.org/abs/2610.24990
作者: Bryan Li,Samar Haider,Grace Kim,Fiona Luo,Jinqi Luo,Yuan Yuan,Duncan Watts,René Vidal,Chris Callison-Burch
机构: University of California, Berkeley; University of California, Los Angeles; Google(谷歌); University of Pennsylvania; Northwestern University
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注:
Abstract:State media outlets often project their governments’ narratives to target both domestic and foreign audiences. Because this content is plentiful and free to access, it appears widely in the web corpora that large language models (LLMs) are trained on and that retrieval systems draw from. However, the prevalence of state media in these corpora and its potential influence on LLM outputs remain largely unexplored. In this paper, we introduce WorldStateMedia, an open dataset of 118 million state media articles from 136 countries in 114 languages. We extract these articles from existing pretraining corpora and annotate each outlet with level of state control from State Media Monitor. Using WorldStateMedia to audit these corpora, we find that, in English web crawls, the share of state media pages from editorially controlled (vs. independent) outlets rose from 39% in 2013 to 93% in 2025. In several low-resource languages, state media make up a substantial share of all web data, such as 17.8% for Uyghur and 7.8% for Belarusian. To characterize the differing narratives of news outlets, we compare Russian and Ukrainian state media in English and in their native languages, revealing asymmetric topical focus and opposing framings of contested issues such as Crimea. Furthermore, in a retrieval-augmented generation (RAG) experiment on territorial disputes, retrieving a country’s state media shifts model answers toward that country’s claims, most strongly for English-language (foreign-targeted) media. We release WorldStateMedia to support multilingual, cross-national insights on state media at this https URL
[NLP-165] A Measurement Perspective on Benchmarking LLM -Generated Text Detectors using Item Response Theory
【速读】： 该论文旨在解决现有大型语言模型（LLM）生成文本检测器评估中依赖聚合指标（如准确率、F1分数和AUC）所导致的局限性问题，即这些指标难以揭示基准测试构成如何影响检测器之间的比较，也无法识别出能够区分不同响应模式检测器的关键文本。其解决方案的核心在于提出一种基于多维项目反应理论（multidimensional item response theory, MIRT）的测量基准框架，将待测文本视为测量项目，将检测器视为被试者，联合建模文本层面的区分度与难度-位置参数，以及检测器的多维能力水平。通过在包含32,000条跨四个领域的人类撰写与机器生成文本的基准上对14个检测器进行评估，该框架揭示了人类与机器文本在多维空间中虽有重叠但可区分的分布特征，且存在显著的领域与文本长度差异。结果显示，人类/机器标签与该多维空间具有最强的多变量关联性，而领域与文本长度进一步加剧了基准异质性。检测器能力剖面图显示出非线性的、依赖领域的响应模式，这在传统聚合性能排名中无法充分体现。作为实际应用，基于估计的MIRT空间对文本进行聚类，可筛选出仅占原基准6.5%（2,088条）的紧凑子集，却能基本保留原始基准上的检测器相对性能关系。总体而言，该框架提供了一种原则性强、可解释性高的方法，用于评估异质化检测技术，并同时刻画评估所用文本的测量属性。
链接: https://arxiv.org/abs/2610.24972
作者: Yanbin Zhao,Zhipeng Chen,Guangsheng Bao,Guandong Zhang,Yan Li
机构: Shanghai Polytechnic University (上海理工大学); Fuyao University of Science and Technology (福州理工学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Benchmarking large language model (LLM)-generated text detectors commonly relies on aggregate detector-level metrics such as accuracy, F1 score, and AUC. Although useful for summarizing overall performance, these metrics provide limited insight into how benchmark composition shapes detector comparisons or which texts distinguish detectors with different response profiles. We propose a measurement-based benchmarking framework for jointly characterizing benchmark heterogeneity and detector behavior using multidimensional item response theory (IRT). By treating benchmark texts as measurement items and detectors as respondents, the framework estimates text-level discrimination and difficulty-location parameters alongside multidimensional detector abilities. We apply the framework to a benchmark of 32K human-written and machine-generated texts across four domains, evaluated by 14 detectors. The estimated IRT item-parameter space reveals distinct but overlapping human-machine distributions, with substantial variation across domains and text lengths. The human-machine label shows the strongest multivariate association with this space, while domain and text length further contribute to benchmark heterogeneity. Detector ability profiles reveal multidimensional and domain-dependent response patterns that are not fully reflected in aggregate performance rankings. As a practical application, clustering texts in the estimated IRT space yields a compact subset of 2,088 texts (6.5% of the full 32K benchmark) that largely preserves the comparative detector-performance patterns observed on the full benchmark. Overall, the proposed framework provides a principled and interpretable methodology for benchmarking heterogeneous detection technologies while jointly characterizing the measurement properties of the texts used for their evaluation.
[NLP-166] Rubrics Carry the Standard: Shapley Attribution for Checklist-Based LLM Judges
【速读】： 该论文旨在解决生成式 AI（Generative AI）评估中由大语言模型（LLM）裁判使用检查清单评分标准（checklist rubrics）所引发的稳定性与可靠性问题，特别是评分标准在模型自评过程中发生不可控漂移的现象。其核心问题是：当前实践中普遍假设评判的宽松或严格程度是裁判模型固有的属性，且评分标准可通过模型自我审查进行优化，但这些假设缺乏实证支持。研究通过控制实验和真实人类标注数据集（SummEdits 和 RAGTruth）验证发现，固定裁判模型与输出时，评分标准内容的变化可使正确输出的通过率从2%提升至100%，而更换裁判模型或相同检查项的表述方式仅带来少量影响；更关键的是，模型自我审查会使其评分标准逐渐偏离原始设计，不仅引入任务未明示的要求，还删除了人类标注所确立的有效检查项。为应对这一问题，作者提出一种基于全必过聚合（all-must-pass aggregation）的可解释性分析框架，利用谢帕利值（Shapley value）的闭式解，实现对每个检查项的“捕获贡献”、“误判归责”及“唯一捕获能力”的线性时间计算。结果表明，模型自评倾向于剔除具有独特检测作用的检查项，而保留导致误判的冗余检查项。为此，研究设计了一种带有溯源冻结机制（provenance freeze）的变更控制门控策略，结合上述量化指标，能够有效遏制自评带来的质量漂移，在三种裁判模型下均维持初始评分标准的性能，确保决策一致性。
链接: https://arxiv.org/abs/2610.24971
作者: Saeid Asgari,Rui Ying,Sara Malvar,Rakshanda Agarwal,Bruce Sun,Claire Wang,Leonardo de Oliveira Nunes,Ranveer Chandra
机构: Microsoft(微软)
类目: Computation and Language (cs.CL)
备注:
Abstract:Checklist rubrics applied by large language model (LLM) judges have become a standard instrument for evaluating open-ended outputs and for rewarding them during training. Two assumptions surround this practice: that leniency or strictness is a property of the judge model, and that a rubric can be improved by asking a model to review it. We test both in a controlled grading task with planted flaws and on two human-labeled datasets, SummEdits and RAGTruth. Holding the judge and the outputs fixed, rubric content moves the pass rate of correct outputs from 2% to 100%, whereas changing the judge model or the rendering of the same checks moves it by a few points. Model self-review then drifts the rubric toward the reviser’s own standard: it adds requirements the task never stated and deletes checks that human labels had taught, with three reviser models and on all three datasets. To measure and control this, we show that under all-must-pass aggregation the Shapley value of each check has an exact closed form, which yields per-check catch credit, false-fail blame and unique catches in linear time. Attribution shows that self-review deletes the checks that do unique work and keeps the checks that cause false fails. A change-control gate that combines these quantities with a provenance freeze on label-derived checks holds nearly every drifting revision chain at its starting quality, with identical decisions under three judge models.
[NLP-167] DeGMatch: Target-Aware Reasoning Data Selection via Deconfounded Gradient Matching
【速读】： 该论文旨在解决大规模长轨迹推理训练中数据筛选成本高昂的问题，尤其针对目标感知型数据选择方法在长推理轨迹上存在目标区分能力有限、候选梯度中主导的群体级结构掩盖细粒度目标依赖性变化的瓶颈。其核心解决方案是提出去混淆梯度匹配（Deconfounded Gradient Matching, DeGMatch），通过构建目标相关性门控机制，并基于模型特异性梯度对候选样本重新排序。DeGMatch首先以候选样本与群体均值中心化梯度，再移除主导主成分方向，有效削弱了群体级结构的干扰，从而增强对目标相关性的敏感度。实验表明，DeGMatch在数学、科学和代码等多个领域均显著优于主流启发式、目标感知及专门针对推理的数据选择基线方法，且在不同数据筛选预算、模型家族及规模下均保持高效性能。
链接: https://arxiv.org/abs/2610.24963
作者: Changho Lee,Janghoon Han,Rajdeep Mukherjee,Yeda Song,Lening Nick Cui,Xinya Du,Honglak Lee
机构: University of Michigan(密歇根大学); University of Texas at Dallas(德克萨斯大学达拉斯分校); LG AI Research(LG人工智能研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Reasoning post-training on long trajectories has substantially improved the reasoning capabilities of large language models, but scaling such training to large datasets is increasingly expensive. We study target-aware reasoning data selection, where a small reference set specifies a target capability and guides the selection of training trajectories for specialization. We find that existing target-aware methods can exhibit limited target discrimination on long reasoning trajectories, while dominant population-level structure in candidate gradients can obscure finer target-dependent variation during gradient matching. To address these limitations, we introduce Deconfounded Gradient Matching (DeGMatch), which first constructs a target-relevance gate and then reranks candidates using model-specific gradients. DeGMatch centers gradients using the candidate-population mean and removes leading principal directions before target matching, reducing the influence of dominant population-level structure and improving target discrimination. Across mathematics, science, and code, DeGMatch outperforms strong heuristic, target-aware, and reasoning-specific data selection baselines, while remaining effective across selection budgets, model families, and scales. Our code is publicly available at this https URL.
[NLP-168] CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents
【速读】： 该论文旨在解决构建具备多工具协调、跨轮次沟通及动态工作空间响应能力的AI同事（AI coworker）训练环境的挑战，其核心问题在于如何生成具有语义一致性、可执行性与可评估性的训练数据，以支持监督微调（SFT）与强化学习（RL）联合训练。解决方案的关键在于提出一种能力导向的合成框架——CoworkEnv，该框架通过工作流需求引导任务构造，并结合程序化验证与反馈修复机制生成规范化的任务规格与评分逻辑；同时，利用教师执行生成高质量示范轨迹用于SFT，学生交互则提供策略演进所需的轨迹与奖励信号，从而实现对轻量级模型PocketWorker的有效训练。实验表明，基于该框架训练的Qwen3-VL-8B-Instruct在Claw-Eval上的平均得分从46.6提升至64.1，尤其在多轮任务中提升达31.9分，验证了所构建环境在支持策略优化与跨基准泛化方面的有效性。
链接: https://arxiv.org/abs/2610.24960
作者: Zhongwei Xie,Yiming Du,Jiaxin Bai,Baixuan Xu,Chunyang Li,Xiyu Ren,Wei Fan,Tianshi Zheng,Weizhi Du,Jingqiao Yu,Xingsu Li,Zhaowei Wang,Yangqiu Song
机构: The Hong Kong University of Science and Technology (香港科技大学); The Chinese University of Hong Kong (香港中文大学); Hong Kong Baptist University (香港浸会大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: work in progress
Abstract:AI coworkers must coordinate tools, communicate across turns, and act on changing workspace state while respecting workflow constraints. Constructing training experience for these capabilities requires coherent requests, service records, tool interfaces, and grading checks. We present CoworkEnv, a capability-guided framework for synthesizing executable training environments for AI coworkers. Workflow requirements guide task construction, and programmatic validation with feedback repair checks generated specifications and grading behavior. Teacher executions yield curated demonstrations for supervised fine-tuning (SFT), while student interactions provide rollouts and rewards for reinforcement learning (RL). Using these resources, we train compact PocketWorker models. For Qwen3-VL-8B-Instruct, SFT raises the mean score from 46.6 to 64.1 on Claw-Eval, including a 31.9-point gain on Multi-turn tasks. An SFT-only extension to ClawMark demonstrates applicability across coworker benchmarks. We further compare RL training dynamics across configurations that vary reward processing and credit assignment. Improving training rewards provide preliminary evidence that the constructed environments support policy optimization.
[NLP-169] Revealing what benchmark scores hide: Jointly measuring the breadth and reliability of LLM capabilities using accessibility profiles
【速读】： 该论文旨在解决传统语言模型评估中因固定提问表述而导致的评估偏差问题，即同一语义请求可通过多种等价表述提出，而现有基准测试仅基于单一表述判断正确性，忽略了模型对不同表述的鲁棒性差异。其核心解决方案是引入“可及性（Accessibility）”这一新度量指标，定义为模型在保持语义不变的前提下，对多个等价表述均成功响应的概率。该指标揭示了模型能力分布的深层结构——传统准确率及最优/最差/平均情况提示得分所掩盖的能力广度与可重现性之间的权衡：两个模型可能具有相同的平均性能，但一个在更多查询上表现不稳定（高可及性但低可靠性），另一个则在较少查询上更稳定（低可及性但高可靠性），因此在是否允许重试或要求持续一致成功的情境下，评估结果截然相反。研究通过六个基准和二十多个模型验证，发现约一半性能相近的模型对在“Access@5”（五种表述中至少一次成功）与“Reliable@5”（五种表述均成功）这两个指标上的表现存在显著分歧，最高可达10个百分点的反向差异。此外，可及性还揭示出模型缩放、推理机制及提示优化可重新分配能力分布，使模型覆盖更广泛的查询，但降低重复成功率，从而将原本被视为干扰因素的表述敏感性转化为可量化的能力属性，清晰区分模型“偶尔能完成”的任务与用户“实际可依赖”的能力边界。
链接: https://arxiv.org/abs/2610.24957
作者: Kazem Faghih,Yize Cheng,Shoumik Saha,Mobina Pournemat,Armin Gerami,Soheil Feizi
机构: University of Maryland, College Park(马里兰大学学院帕克分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 26 pages total, 10 pages main text, 18 figures. Code and data: this https URL
Abstract:Language model benchmarks assign each question a single outcome from one fixed wording, even though the same underlying request can be expressed in many equivalent ways. We argue that the quantity of interest should therefore not be correctness on one wording, but the accessibility of an inquiry, which we define as the probability that a model succeeds across meaning-preserving formulations. The distribution of accessibility across inquiries reveals a structure that accuracy and existing best-, worst-, or average-case prompt scores collapse: how broadly a model’s capabilities extend across inquiries and how dependably those capabilities can be elicited. Two models can have the same average performance while one succeeds across more inquiries unreliably and the other succeeds across fewer inquiries but more consistently, leading to opposite preferences depending on whether the task allows multiple retries until one success or when repeated consistent success is required. Across six benchmarks and more than twenty models, this tradeoff is common: for about half of the model pairs with nearly equal mean accessibility, which model looks better depends on whether one success across five formulations is enough or all five must succeed, quantities we call Access@5 and Reliable@5, respectively. For some pairs the two measures diverge significantly, by up to 10 points in opposite directions. Accessibility also reveals that scaling, reasoning, and prompt optimization can redistribute capability across inquiries in ways aggregate accuracy obscures, sometimes broadening the set of inquiries a model can succeed on while making success on them less repeatable. Accessibility turns formulation sensitivity from a nuisance into a measurable property of capability, distinguishing what a model can sometimes succeed on from what users can actually depend on.
[NLP-170] CasePlay: Self-Play Reinforcement Learning from Case Reports for Medical Reasoning
【速读】： 该论文旨在解决大语言模型（LLM）在临床推理能力提升中依赖专家标注问题或高成本合成数据生成的问题，提出利用医学文献中丰富的高质量无标签文档（如完整临床病例报告）作为自监督强化学习（RL）的训练信号。其核心挑战在于现有自对弈（self-play）方法在该场景下存在“知识点固化”（knowledge-point fixation）——提议者反复针对报告中的少数显著线索提问，以及“难度-质量错配”（difficulty-quality mismatch）——问题虽具适当难度但质量低下、推理深度不足，难以有效促进推理模型训练。为应对上述问题，论文提出CasePlay框架，其关键创新在于：（a）知识条件化提议者（Knowledge-Conditioned Proposer），通过将问题生成锚定于病例特定的知识点，促使问题覆盖更广泛、多样化的证据来源；（b）评分标准判别器（Rubric-Judge Reasoner），在传统答案准确性反馈基础上引入基于评分标准的质量反馈，评估线索整合、推理深度与选项质量，从而引导提议者生成更具推理强度的问题，提升训练信号的有效性。实验结果表明，CasePlay在七个医学相关基准上均优于现有自对弈基线，平均提升3.1分，尤其在LiveClin上实现6.3分的显著增益，验证了其通过自对弈强化学习持续优化医学推理能力的有效路径。
链接: https://arxiv.org/abs/2610.24952
作者: Hao Wang,Zihan Wang,Kai Wu,Jian Yao,Yiwen Ye,Yongcan Luo,Dapeng Wu,Jiale Chen,Haihua Yang
机构: City University of Hong Kong(香港城市大学); Northeastern University(东北大学); Tongji University(同济大学); The Hong Kong Polytechnic University(香港理工大学); ByteDance(字节跳动)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Reinforcement learning (RL) is a promising way to improve the clinical reasoning abilities of large language models (LLMs), but current pipelines often rely on expert-labeled questions or costly synthetic-data generation. Meanwhile, the medical literature offers a wealth of high-quality unlabeled documents, including clinical case reports that trace the full clinical course from initial presentation to final treatment decisions. We study whether document-grounded self-play can convert these reports into effective RL training signals. We identify two challenges that limit existing self-play methods in this setting: knowledge-point fixation, where the proposer repeatedly asks about a narrow set of salient clues in a source report, and difficulty-quality mismatch, where appropriately difficult questions may still be poorly formed or shallow, limiting their value for reasoner training. To address these challenges, we propose CasePlay, a self-play framework with two key components: (a) a Knowledge-Conditioned Proposer, which anchors question generation to case-specific knowledge points so that questions draw on a broader range of evidence from the source report; and (b) a Rubric-Judge Reasoner, which augments answer-accuracy feedback with rubric-based quality feedback on clue integration, reasoning depth, and option quality, guiding the proposer toward reasoning-intensive questions that better support reasoner training. Extensive experiments across seven medically related benchmarks demonstrate that CasePlay outperforms existing self-play baselines and improves the SFT warm-start model by 3.1 points on average, highlighted by a 6.3-point gain on LiveClin. This approach offers an effective path toward continually improving medical reasoning ability through self-play RL.
[NLP-171] Over-Defense in Coding Agents Begins at Silent Forks
【速读】： 该论文旨在解决代码生成代理在后训练过程中因过度关注决定任务成败的关键回合而导致的非必要防御性行为（defensive actions）问题。这类行为通常出现在“静默分叉”（silent forks）节点，即后续所有动作均导致相同结果的决策点，其存在不影响最终成功概率，但会显著增加无效计算与资源消耗。现有方法（如基于轨迹奖励的检查器）仅能通过完全比例化的路径识别此类分叉，导致74%的策略组无法获得差异化学习信号，进而维持不变的行为模式。为此，本文提出关键创新——以行为分歧（behavioral divergence）作为学习信号的定位依据，而非依赖结果回报。其核心解决方案为ForkDPO：在静默分叉两侧、同一前缀下对比比例化与非比例化的动作，从而捕捉超出任务完成与否之外的隐式行为差异。实验表明，在未见代码库测试集上，该方法将40亿参数模型的非比例行为从70.3%降至58.9%，显著优于随机选择回合的DPO（67.7%），验证了行为分歧作为学习信号的有效性。
链接: https://arxiv.org/abs/2610.24949
作者: Hanjun Luo,Xuecheng Liu,Xiucheng Zhang,Zhuoning Xu,Yingbin Jin,Yankai Chen,Xue Liu,Hanan Salam
机构: New York University(纽约大学); Carnegie Mellon University(卡内基梅隆大学); The Hong Kong Polytechnic University(香港理工大学); McGill University(麦吉尔大学); Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG); Software Engineering (cs.SE)
备注: 27 pages, 10 figures, 31 tables. Code: this https URL
Abstract:Language models behind coding agents are post-trained with outcome rewards, and recent methods concentrate credit on the turns that decide the outcome. We find that defensive actions beyond what a task requires escape both. They begin at silent forks, the turns where every next action leads to the same outcome. Such actions leave the success rate unchanged, and frontier models take the disproportionate branch there in up to 91% of decisive actions under implicit budgets. The most direct fix, a checker in the trajectory reward, reaches these forks only through fully proportionate trajectories, leaving 74% of GRPO groups with identical rewards and behavior unchanged. This motivates a rethink of where learning signals belong, namely where behavior diverges, which a behavioral checker can locate and the outcome cannot. Our method, ForkDPO, contrasts a proportionate and a disproportionate action at the same prefix, at silent forks on both sides of the required behavior. On a test set of unseen repositories, it lowers a 4B model’s disproportionate behavior from 70.3% to 58.9%, against 67.7% for DPO at randomly chosen turns, establishing behavioral divergence as a learning signal for how agents act beyond whether they finish. The code is publicly available at this https URL.
[NLP-172] Epi: Towards a Persistent Self-Improving Agent for Autonomous Research
【速读】： 该论文旨在解决复杂科学问题中研究过程难以持续、方法无法迭代优化以及实验结果缺乏可解释性等关键挑战。传统研究模式受限于单次会话的局限，难以积累长期经验并实现跨会话的连续探索，同时缺乏对多路径假设的系统性评估与回溯能力。为此，论文提出Épi——一个具备持久研究历史（persistent research history）、分层记忆（hierarchical memory）和分支独立评审机制的自主研究代理，其核心创新在于通过跨会话实验执行与固定验证器（fixed verifier）协同，实现研究过程的持续演化与自我修正。关键解决方案包括：1）利用分层记忆结构组织长期知识积累；2）通过分支与独立评审机制支持多路径探索与回溯验证；3）结合精确对偶证书（exact dual certificates）提供数学可解释性支撑。该系统在多个数学领域达成120项突破性成果，涵盖数值界改进与构造性证明，并进一步扩展至算法编码任务（Épi-turbo），以更低成本实现优于主流模型的性能，同时确保计算结果可被研究人员审查与延续，从而显著增强人机协作中的科学可理解性与可复现性。
链接: https://arxiv.org/abs/2610.24944
作者: Yue Huang,Zhangchen Xu,Xiaonan Luo,Zheyuan Liu,Lichi Li,Yiping Wang,Peiyang Song,Eric Gabriel Ramos,Enze Xie,Olexandr Isayev,Paul Liang,Nouha Dziri,Huan Sun,Jianfeng Gao,Ying Liu,Misha Sra,Xiangliang Zhang,Alex Pentland,Zichen Chen
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 23 pages, 7 figures. Technical report
Abstract:We present Épi, an autonomous research agent that sustains investigations across model sessions and evolves its methods through accumulated experience. Épi combines persistent research history, hierarchical memory, branching and independent review to maintain research continuity, learn from experiments, and explore and revisit alternative approaches. Its experiments run across session boundaries, and a fixed verifier evaluates their outputs. Épi produced 120 record-setting results, comprising 20 numerical bound records across 12 families and 100 construction records across eight families. These include an approximately 84-fold narrower interval for the n -queens constant and a more than 3.6-million-fold narrower rigorous interval for the quartic Feigenbaum attractor dimension. Exact dual certificates establish that the Cohn-Elkies linear-programming bound cannot be sharp in dimensions 10 and 11. Beyond numerical improvements, Épi gives a computer-assisted proof of Tuza’s conjecture for split graphs with an eight-vertex clique part and at most three triangle-active neighborhood types, with arbitrary multiplicities. We extend this research process to algorithmic coding with Épi-turbo, a lightweight configuration that develops executable solvers for tasks including vehicle routing, scheduling and production planning. On 30 FrontierOR tasks, Épi-turbo achieves a 5.4% higher total benchmark score than Codex at 64% of its model cost, using the same underlying model and reasoning effort. Épi is designed to support human understanding by connecting computational results to explicit methods and mathematical arguments that researchers can examine and build upon.
[NLP-173] What Cant Jev Decide? Probing the Limits of One-Pass Decision Models
【速读】： 该论文旨在解决生成式 AI（Generative AI）在复杂决策任务中因依赖静态输入状态而无法识别自身推理局限性的问题。其核心挑战在于：当决策依赖于大语言模型（LLM）通过逐步推理解答的中间过程时，仅依赖单次前向传播的模型（如 Jev 及其开源变体）无法有效处理此类任务。解决方案的关键在于揭示并验证这类模型的固有缺陷——即在面对需多步推理或动态状态更新的场景时，由于缺乏对内部状态的动态校验能力，模型会因历史信息过时或相似记录干扰导致准确率显著下降（如十六次更新后准确率降至53%-80%）。此外，模型虽能通过置信度标记自身产生的错误，却无法识别由陈旧输入状态引发的隐性错误，且无论是显式还是训练得到的交接机制均无法捕获此类问题。研究进一步表明，当前评估的开源模型均存在相同局限，因此必须依赖具备推理能力的模型来主动检查其所依赖的状态完整性，才能实现可靠决策。
链接: https://arxiv.org/abs/2610.24929
作者: Wang Yang,Xinpeng Li,Yanyan Zhang,Chuang Ma,Shouren Wang,Jianzhe Wei,Liqin Ye,Xiaoran Cai,Mohsen Hariri,Vipin Chaudhary,Xiaotian Han
机构: Case Western Reserve University (凯斯西储大学); NII LLMC (日本国立情报研究所语言与机器学习中心); Georgia Institute of Technology (佐治亚理工学院); Columbia University (哥伦比亚大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Jev and its open replicas choose among the options a caller supplies with a single forward pass, fast enough to decide inside an agent loop. We ask what such a model cannot decide, and find a simple limit: Jev cannot make a decision that an LLM makes only by writing out intermediate steps. On a benchmark that varies one factor at a time, Jev finds and selects records without error up to 20k tokens and 64 options, but once the answer depends on two or more updates of one user, accuracy falls, to 53-80% at sixteen updates, because similar records interfere. Given a stale summary of the state, Jev follows it. On real decision items graded by how many thinking tokens a reference LLM needs, Jev answers 97.5% of the items that one forward pass of that LLM solves, but only 74% of the others. Jev also does not know its limit: its confidence flags the errors it makes itself but not those it inherits from a stale input, and neither an explicit nor a trained hand-off option catches the latter. The open models we evaluate share these limits. Checking the state such models rely on requires a model that reasons.
[NLP-174] Stable Signer V3: 4D Deformable Primitive Splatting Is All You Need for Sign Language Production
【速读】： 该论文旨在解决现有手语虚拟形象系统在面部表情表现力受限、身份特异性身体形变不准确以及精细外观细节丢失等方面的三大核心问题。其解决方案的关键在于引入三个基于测试时优化（test-time）的创新模块：共享面部可变形基元点云（Shared Face Primitive Splatting）通过可迁移的超二次曲面基元实现更丰富的表情控制；测试时身体基元点云（Test-Time Body Primitive Splatting）学习身份特异性的虚拟肌肉层，以提升动作还原的准确性；测试时隧道超分辨率（Test-Time Tunneling Super-Resolution）则利用TunnelingNet恢复身份特异性的高分辨率细节。上述模块均针对每位用户身份进行一次性优化，推理阶段仅依赖存储的身份参数，实现了高保真度与交互式流媒体传输的兼顾，显著降低了面部几何误差71.4%、身体边界误差35.1%，并在未见数据上将高分辨率RGB重建误差降低24.2%。
链接: https://arxiv.org/abs/2610.24926
作者: Sen Fang,Liwei Che,Dimitris N. Metaxas
机构: Rutgers University(罗格斯大学); https://signerx.github.io/Stable-Signer-V3
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Graphics (cs.GR)
备注: 12 figures, 10 tables. Project Page at: this https URL
Abstract:We present Stable Signer V3, an interactive high-resolution sign language avatar system built on 4D Deformable Primitive Splatting. It addresses three limitations of existing avatars: restricted facial expression, inaccurate identity-specific body deformation, and loss of fine appearance details. Shared Face Primitive Splatting adds transferable superquadric primitives for richer expression control; Test-Time Body Primitive Splatting learns an identity-specific virtual muscle layer for more faithful motion; and Test-Time Tunneling Super-Resolution uses TunnelingNet to recover identity-specific high-resolution details. Both test-time modules are optimized once per signer identity; inference then runs entirely from the stored identity parameters. Together, they reduce facial geometry error by 71.4%, body-boundary error by 35.1% across ten identities, and high-resolution RGB reconstruction error by 24.2% on held-out evaluations, while retaining interactive streaming.
[NLP-175] CARTA: Self-Supervised Reusable Transaction Embeddings via Heterogeneous Tokenization and Cross-Attention
【速读】： 该论文旨在解决支付交易欺诈检测中标签获取成本高且任务间不一致，以及高基数标识符（如商户名称）导致传统独热编码特征管道在大规模场景下不可行的问题。其核心解决方案是提出一种自监督编码器CARTA，能够从原始交易数据中无标签地学习一个16维的紧凑嵌入表示；该嵌入表示冻结后可复用于低标签效率的欺诈探测及探索性无监督风险分层，无需重新训练编码器。CARTA的关键创新在于采用特定于类型的投影与可学习的嵌入表处理高基数标识符，并通过令牌删除（token excision）和InfoNCE对比损失进行训练。实验表明，在IEEE-CIS数据集上，仅使用1%标签的线性探针性能超过使用100%标签的原始特征探针；在IBM TabFormer数据集（200万行）上，冻结的嵌入表示达到梯度提升决策树（GBDT）的AUPRC为0.508，显著优于频率编码（0.411）和剔除商户信息的基线（0.022），尽管全监督的CatBoost表现更优。因此，CARTA是一种无需标签、可复用的表示学习框架，可作为特定任务监督模型的有效补充。
链接: https://arxiv.org/abs/2610.24920
作者: Shreyas K. Chandrahas,Jose Munoz,Satish Babu Bellapu
机构: Visa Inc.(Visa公司); Massachusetts Institute of Technology(麻省理工学院)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 9 pages, 4 figures. Accepted at ICAIF '26 (7th ACM International Conference on AI in Finance), Milan, Italy, November 14-17, 2026
Abstract:Acquiring fraud labels for payment transactions is expensive and inconsistent across tasks, and high-cardinality identifiers such as merchant names make one-hot feature pipelines infeasible at scale. We present CARTA, a self-supervised encoder that learns a compact 16-dim embedding from raw transactions without labels; the frozen embedding is then reused for label-efficient fraud probing and exploratory unsupervised risk stratification without retraining the encoder. CARTA uses type-specific projections with learned embedding tables for high-cardinality identifiers, trained with token excision and InfoNCE contrastive loss. On IEEE-CIS, its linear probe at 1% of labels exceeds a raw-feature probe at 100%; on IBM TabFormer (2M rows), its frozen embedding reaches GBDT AUPRC 0.508, above frequency encoding (0.411) and the merchant-dropping baseline (0.022), though fully-supervised CatBoost is stronger. CARTA is a label-free, reusable representation that complements task-specific supervised models.
[NLP-176] How Well Can LLM s Verbalize Mathematical Formulas? A Multi-Domain Evaluation
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在数学公式书面化表达（verbalization）任务中生成准确性不足的问题，尤其关注其在文本转语音系统及视障用户辅助学习场景下的应用潜力。现有研究缺乏对主流LLMs在多科学领域数学公式口语化生成能力的系统性评估。为此，研究构建了“读出数学”（Read-Aloud Math）这一平衡的多领域语料库，包含LaTeX公式、经人工验证的简单与扩展型转写版本、多种变体、文本上下文、领域标签及公式复杂度分级。关键解决方案在于采用多参考BLEU（multi-reference BLEU）作为评价指标，全面评估五种前沿LLMs在不同上下文条件下的表现。结果显示，尽管GPT-5.2在无上下文条件下对简单转写取得最高得分（75.9%），但未显著优于Grok 4.1 Fast或Claude Sonnet 4.5；而Claude Sonnet 4.5在无上下文扩展转写中表现最佳（36.5%），Grok 4.1 Fast则在有上下文时达到最高（48.8%）。研究发现，公式长度对简单转写的性能影响显著（无上下文时负相关，有上下文时不显著），而对扩展转写则呈正相关。此外，分组符号是转写差异的主要来源，变量名、大小写和希腊字母则与较低的BLEU分数相关。总体表明，基于大语言模型的数学公式口语化生成虽具前景，但仍高度依赖转写类型、上下文信息及公式自身特征。
链接: https://arxiv.org/abs/2610.24917
作者: Peter Volbach,Tim Schlippe
机构: IU International University of Applied Sciences (IU国际应用科学大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Published in the IEEE Proceedings of the 3rd International Conference on Foundation and Large Language Models (FLLM 2026), Barcelona, Spain, November 17-20, 2026
Abstract:Large Language Models (LLMs) have shown strong capabilities across natural language tasks. But their ability to generate accurate written verbalizations of mathematical formulas has not been systematically evaluated. Such verbalization is important for text-to-speech systems supporting audio-based learning and visually impaired users. We evaluated five state-of-the-art LLMs across six scientific domains and constructed Read-Aloud Math, a balanced multi-domain corpus with LaTeX formulas, manually validated simple and extended transcriptions, multiple variants, textual context, domain labels, and formula complexity levels. Using multi-reference BLEU, GPT-5.2 achieved the highest score for simple transcriptions (75.9% without context, 64.9% with context), although without context it did not significantly outperform Grok 4.1 Fast or Claude Sonnet 4.5. Claude Sonnet 4.5 achieved the highest score for extended transcriptions without context (36.5%), and Grok 4.1 Fast with context (48.8%). For extended transcriptions, Statistics and Mathematics scored highest across the two context settings, while Physics scored lowest. Formula length was significantly negatively associated with BLEU for simple transcriptions without context, but not with context, whereas BLEU increased significantly with formula length for extended transcriptions in both settings. Grouping symbols were most frequently involved in transcription differences, while variable names, capitalization, and Greek letters were associated with lower BLEU scores. Overall, LLM-based formula verbalization is promising but remains sensitive to transcription type, textual context, and formula characteristics.
[NLP-177] Arctic-SWE: Revisiting the Long Horizon Agent Training by Modeling-System Co-Design
【速读】： 该论文旨在解决长时程智能体训练（long-horizon agent training）中基准线（baseline）难以复现的问题，尤其在软件工程（software engineering, SWE）领域，现有强化学习（reinforcement learning, RL）基线常因配置不当而掩盖新算法或数据改进的真实效果。其核心解决方案是提出一个完全异步的强化学习框架——Arctic-SWE，通过模型与系统协同设计（model-system co-design），以小规模任务池上的过拟合作为测试基准，暴露早期奖励提升所隐藏的系统性失败。该框架整合了门控奖励（gated reward）、重要性权重控制、对齐的训练与推理引擎、以及策略更新间的状态一致性等成熟组件，虽各组件非全新，但协同作用构建了缺失的底层基础。实验表明，在不引入新算法或数据的前提下，Arctic-SWE 仅用约六分之一的训练回溯次数，即实现 Qwen3.5-4B 模型在 SWE-bench Verified 上达到 62.0% 的 pass@1 分数，媲美依赖在线任务生成的 FrogNano（61.5%）。进一步基于该框架对 R2E-Gym、SWE-rebench V2 及 Scale-SWE 进行筛选与优化后，4B 模型性能显著提升，同时 9B 模型也获得明显增益，验证了数据筛选与测试强度增强对泛化性能的双重促进作用。该工作为异步智能体强化学习的后续研究提供了可复现、可扩展的基准起点。
链接: https://arxiv.org/abs/2610.24914
作者: Boyi Liu,Yite Wang,Xinyu Lian,Fan Shu,Sida Li,Feng Yan,Mingjia Zhang,Yuxiong He,Zhewei Yao
机构: Snowflake(雪flake); University of Illinois at Urbana-Champaign(伊利诺伊大学厄本那-香槟分校); University of Houston(休斯顿大学); University of Chicago(芝加哥大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Long-horizon agent training, including software engineering, is advancing rapidly, but the recipe and infrastructure needed for a strong reinforcement-learning baseline remain difficult to reproduce. This makes gains from new algorithms or data hard to distinguish from improvements to an under-tuned baseline. We present Arctic-SWE, a fully asynchronous RL framework that establishes such a baseline through model-system co-design, using overfitting on a small task pool as the testbed. This testbed exposes failures that early reward gains conceal and guides the integration of established components: gated reward, importance-weight control, aligned training and inference engines, and consistent state across policy updates. No single component is new, yet together they form the missing foundation. On raw R2E-Gym data, without new algorithms or data work, Arctic-SWE trains Qwen3.5-4B to 62.0% pass@1 on SWE-bench Verified. This is comparable to concurrent FrogNano (61.5%), which relies on online task synthesis, after about one-sixth as many training rollouts. Building on this foundation, we further filter and refine R2E-Gym, SWE-rebench V2, and Scale-SWE. With this curated subset, the 4B model improves from 47.4%/25.0%/37.9% to 65.0%/38.3%/59.7% on SWE-bench Verified/Pro-v1/Pro-v2, and the recipe also improves Qwen3.5-9B from 56.2%/31.9%/48.8% to 65.7%/39.6%/70.2% on Verified/Pro-v1/Pro-v2. Controlled comparisons show that both curation and test strengthening improve held-out performance, although stronger tests reduce apparent training-task mastery. Arctic-SWE provides a documented starting point for studying better algorithms, data and environments for asynchronous agent RL.
[NLP-178] A Cross-Domain Evaluation of Prompting Techniques for Large Language Models
【速读】： 该论文旨在解决现有提示工程（Prompt Engineering）研究中因使用不同大语言模型（LLM）、数据集或任务设置而导致方法间难以直接比较的问题。其核心挑战在于缺乏统一、可控的跨领域评估框架，使得提示策略的有效性评估缺乏可比性和普适性。为此，本文提出一种基于PE2启发式优化流程的受控跨域评估范式，对零样本提示（Zero-Shot Prompting）、少样本提示（Few-Shot Prompting）、检索增强生成（Retrieval-Augmented Generation, RAG）以及自动提示优化四种策略进行系统性对比。所有实验均在相同LLM下开展，并在涵盖算术推理、逻辑推理、常识推理和事实验证的六个基准数据集上进行评估。关键发现表明：不存在在所有任务领域中始终表现最优的提示策略；各方法间的最高性能配置在统计上无显著差异；当可用检索示例增至300个时，RAG达到78.33%的最高平均准确率，而少样本提示在结构化推理任务中表现优异，零样本提示在输入自包含的任务中仍具竞争力，基于PE2的自动优化方法则表现出稳定但未系统性超越其他方法的结果。此外，增加特定任务示例数量并未一致提升性能，凸显提示技术的高度任务与领域依赖性。因此，研究结论强调应根据具体任务特征自适应选择提示策略，而非将其视为通用优化手段。
链接: https://arxiv.org/abs/2610.24913
作者: Kevin Krazius,Tim Schlippe,Paul Libbrecht
机构: IU International University of Applied Sciences (IU国际应用科学大学); OpenAI(OpenAI)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Published in the IEEE Proceedings of the 3rd International Conference on Foundation and Large Language Models (FLLM 2026), Barcelona, Spain, November 17-20, 2026
Abstract:Prompt engineering has become a central mechanism for guiding the behavior of Large Language Models (LLMs) across a wide range of reasoning and language understanding tasks [1], [2]. However, existing studies frequently evaluate prompting techniques using different LLMs, datasets, or task settings, making direct comparisons across methods and domains difficult [3]. Consequently, in this paper we present a controlled cross-domain evaluation of Zero-Shot prompting, Few-Shot prompting, Retrieval-Augmented Generation (RAG), and automatic prompt optimization based on a PE2-inspired optimization pipeline [4]-[6]. All experiments were conducted with the same LLM and evaluated on six benchmark datasets covering arithmetic reasoning, logical reasoning, commonsense reasoning, and factual verification. The results show that no prompting strategy consistently outperforms the others across all domains, and pairwise statistical comparisons of the highest-performing configurations revealed no significant differences. RAG achieves the highest overall average accuracy of 78.33% when 300 examples are available for retrieval, while Few-Shot Prompting achieves comparatively high accuracies across several structured reasoning tasks. Zero-Shot Prompting remains competitive on tasks with self-contained inputs, whereas the PE2-inspired optimization approach produces stable but not systematically superior results compared to the other evaluated prompting techniques. Furthermore, increasing the number of task-specific examples does not consistently improve performance across prompting techniques, indicating strong domain and task dependency. These findings suggest that prompting techniques should be selected adaptively based on task characteristics rather than treated as universally applicable optimization strategies.
[NLP-179] Compositional Activation Steering Across Languages and Behaviors with Contextual Gates
【速读】： 该论文旨在解决在多语言环境下，如何通过无需训练的生成式AI（Generative AI）控制向量实现多个行为特征的联合调控这一难题。现有方法在组合多个控制方向时存在精度不足与泛化能力差的问题，尤其在无训练依赖的设定下表现受限。其解决方案的关键在于提出一种上下文感知的调控机制（contextual steering），通过联合训练可动态调节固定控制方向强度的门控模块，使各干预措施的权重能够根据当前隐藏状态自适应调整，并捕捉不同干预间的交互作用。实验表明，该方法在多种语言、行为属性及模型架构上均显著优于固定强度调控和四种强基线方法（包括提示工程），同时更有效地保持了模型的通用知识表征，并使内部计算路径更贴近经过提示引导的模型行为。进一步分析揭示了门控机制在组合过程中动态分配贡献的内在机理，为有效实现生成式AI中多维度行为的协同控制提供了理论与实践支持。
链接: https://arxiv.org/abs/2610.24906
作者: Daniil Gurgurov,Akansh Maurya,Josef van Genabith,Simon Ostermann,Hitomi Yanaka
机构: Saarland University(萨尔兰大学); German Research Center for AI (DFKI)(德国人工智能研究中心); RIKEN(理化学研究所); The University of Tokyo(东京大学); Tohoku University(东北大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Activation steering has emerged as an effective and precise way to influence the behaviours of language models. However, composing multiple steering directions to control several behaviours simultaneously remains challenging, particularly in multilingual settings and with training-free steering vectors. We address this challenge with contextual steering, jointly training gates that adapt the strength of fixed steering directions to the current hidden state and interactions between interventions. Across a broad range of languages, behavioural attributes, and models, our approach outperforms constant-strength steering and four strong baselines, including prompting. It also better preserves general knowledge and keeps the model’s internal computation closer to that of the prompted model. Our analysis further shows how jointly trained gates adapt their contributions during composition, explaining the mechanisms underlying effective compositional steering.
[NLP-180] HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents
【速读】： 该论文旨在解决大语言模型（LLM）代理在面对跨异构数据源的复杂分析问题时表现不佳的核心挑战，即现有模型虽能有效处理单一类型数据源（如结构化表格或非结构化文本），但在需要同时整合结构化数据与非结构化信息（如企业内部数据库与外部网络搜索结果）时，推理能力显著下降。其关键解决方案在于提出一种名为HybridResearcher的40亿参数代理模型，通过构建具有明确跨源依赖性的训练数据来显式教授跨源推理能力。具体而言，研究者首先基于已验证的单源任务（如文本转SQL查询或网页搜索）生成初始轨迹，并围绕其输出中的“桥接实体”（bridge entity）人工构造缺失的数据源，仅保留满足严格构建条件（即任一来源单独均无法完成任务）的样本，从而确保训练数据具备真正的跨源互依性。随后采用监督微调与强化学习相结合的方式对模型进行训练。实验表明，HybridResearcher在HybridDeepResearch基准上相较于基线模型Qwen3.5-4B提升30.0分（pass@8），并超越更大规模的Qwen3.5-397B-A17B模型8.3分，同时在独立的单源基准（BrowseComp和BrowseComp-Plus）上仍保持性能，证明了小模型可通过特定训练策略有效学习复杂跨源推理而不会引发灾难性遗忘。
链接: https://arxiv.org/abs/2610.24899
作者: Soyoung Yoon,Peiran Xu,Han Wang,Ruofan Wu,Xiaodong Yu,Feng Yan,Mohit Bansal,Debiao Li,Seung-Won Hwang,Yuxiong He,Zhewei Yao
机构: University of Houston(休斯顿大学); University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校); University of California, Los Angeles(加州大学洛杉矶分校); Seoul National University(首尔国立大学); Snowflake AI Research(雪花人工智能研究)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables or unstructured data, real-world analytical questions frequently span heterogeneous sources. For example, answering one enterprise question may require both a result computed over an internal database and evidence found on the web. Recent evaluations on benchmarks like HybridDeepResearch show that even frontier agents struggle when evidence must cross this boundary. While these benchmarks have exposed the limitation, methodologies to explicitly teach cross-source reasoning remain absent. To address this, we introduce HybridResearcher, a 4B agent trained to bridge evidence across the structured and unstructured boundary. To build it, we synthesize cross-source bridge trajectories by starting with validated single-source tasks-such as existing text-to-SQL or web search queries-and fabricating the missing source around a bridge entity drawn from the seed’s own output. We retain a sample only when strict construction gates confirm that neither side alone suffices, ensuring the data requires true cross-source interdependence. Finally, we train on this curated dataset using supervised fine-tuning followed by reinforcement learning. Evaluated on the existing HybridDeepResearch benchmark, HybridResearcher improves over our base model, Qwen3.5-4B, by 30.0 points pass@8 and surpasses the far larger Qwen3.5-397B-A17B by 8.3 points, while also improving over its base model on established, held-out single-source benchmarks (BrowseComp and BrowseComp-Plus), demonstrating that this complex cross-source reasoning can be taught directly to small models without triggering catastrophic forgetting.
[NLP-181] Peer Pressure: Execution-Verified Co-Evolution of Attacker and Defender Populations for Code Security
【速读】： 该论文旨在解决大语言模型（LLM）代码代理在缺乏有效监管的情况下，自动提交存在漏洞的代码，而传统基于静态漏洞数据集训练的防御者难以适应动态演化攻击者的问题。其核心挑战在于如何构建能够持续适应真实攻击行为、并具备高泛化能力的自适应安全检测机制。解决方案的关键是提出一种“同伴相对策略优化”（Peer-Relative Policy Optimization, PRPO）框架，通过联合训练攻击者（Red）与防御者，并以实际代码执行结果作为奖励信号——攻击者需生成可应用、可编译、通过测试、完成任务并复现真实漏洞的补丁，且其路径需被隔离容器验证；防御者则通过评估该补丁的安全性获得反馈。PRPO将攻击者与防御者均建模为同一冻结基础模型上的LoRA适配器群体，奖励依据同角色同伴在相同条件下表现进行相对评分，将群体置于策略梯度基线而非对手池中，结合锦标赛机制保持多样性。通过匹配机制消除条件难度偏差，使模型在未见应用和语言上实现64.8%的平衡准确率，显著优于监督微调与冻结前沿模型；同时在漏洞挖掘与入口点定位任务中表现更优，且在免微调情况下仍可达到或超越现有静态防御方案的性能，展现出强大的泛化与抗逃逸能力。
链接: https://arxiv.org/abs/2610.24896
作者: Khushbu Pahwa,Yubin Ge,Daniele Bonadiman,Xiaofei Ma,Yi Zhang
机构: AWS AI(亚马逊云科技人工智能)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注: Preprint. Under Review
Abstract:LLM code agents increasingly ship code with little oversight, yet defenders trained on static vulnerability datasets fall behind adapting adversaries. We co-train attacker and defender instead, and let execution decide every reward: an attacker (Red) edits a real repository and earns nothing unless its patch applies, builds, passes the tests, completes its development task and reproduces a real exploit in an isolated container, whose path also labels the entry point, the critical operation and the trace between them. Peer-Relative Policy Optimization (PRPO) makes both sides populations of LoRA adapters over one frozen base and credits each against same-role peers that faced an identical condition, placing the population inside the policy-gradient baseline rather than the opponent pool, with a tournament preserving diversity. Matching cancels the condition’s difficulty: at equal coverage it is worth 2.3 points of balanced accuracy over a baseline computed on other conditions, and it matches a within-policy control at half the verifier spend. PRPO’s 7B diff reviewer reaches 64.8% balanced accuracy on four unseen applications in three untrained languages, 5.5 points above supervised security fine-tuning and 4.4 above a frozen frontier reviewer (Claude Opus 4.8), which it leads by 15.8 on the application it trains on. Its 9B repository agent recovers 23 of 172 held-out VulnGym advisories, against 10 after supervised fine-tuning and 7 for the same backbone untrained under the same scaffold, and locates 17 of 43 reachable entry points against 13 for mini-swe-agent with GLM-5.2. Run unchanged on jailbreak defense, the same algorithm matches or exceeds WildGuard’s static F1 at equal corpus and model size. Under a judge distinct from its training oracle it cuts attack success from 36.0% to 4.6%, against WildGuard’s 8.9%. Comments: Preprint. Under Review Subjects: Cryptography and Security (cs.CR); Computation and Language (cs.CL) Cite as: arXiv:2610.24896 [cs.CR] (or arXiv:2610.24896v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.24896 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-182] LastWords: Boring Problems Revisited
【速读】： 该论文旨在解决自然语言处理中看似简单却难以通过纯神经方法攻克的文本归一化问题，特别是数字名称读法（number name reading）这一子问题。其核心挑战在于如何在资源有限、领域受限的情况下实现准确、鲁棒的数字表达转换，而传统神经模型在此类任务上表现不佳。解决方案的关键在于：尽管当前前沿大模型（如最新版ChatGPT）已基本解决了该问题，但作者指出，未来更高效的方法应能在不依赖大规模预训练模型和海量计算资源的前提下，通过更精细的结构设计或领域适配策略实现类似甚至更优的性能，从而为低资源、高约束场景下的智能系统提供可扩展的解决方案。
链接: https://arxiv.org/abs/2610.24894
作者: Huayang Li,Richard Sproat
机构: Nara Institute of Science and Technology (奈良先端科学技術大学院大学); Sakana.ai
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 6 pages, 1 figure
Abstract:In 2022, the second author published a Last Words piece in Computational Linguistics in which he pointed out that, despite the advances in AI up to that point, there were still some problems in natural language processing that on the one hand seemed trivial and perhaps boring, but on the other were hard to solve using purely neural methods. One such problem discussed in that context was text normalization, and a particular subproblem of that, namely number name reading. The piece came out just a few months before the first release of ChatGPT, and needless to say, a lot has happened since then. Indeed, as we document here, the problem (of number name reading) has practically been solved, at least by one of the latest frontier models. However, we believe that there are still open issues for such limited domain, constrained and limited resource problems, and that in some future machine learning approach, the massive resources of the latest frontier models would not be needed to solve them.
[NLP-183] Measuring Data Diversity with Embeddings: A Taxonomy and Benchmark for Diversity Measures EMNLP2026
【速读】： 该论文旨在解决文本语料库多样性（diversity）度量方法碎片化、缺乏系统比较与理解的问题。当前基于嵌入向量（embedding-based）的多样性度量方法在计算方式上差异显著，导致同一数据集上的评估结果存在较大偏差，但尚无系统性的实证对比研究。其解决方案的关键在于：首先，提出一个包含21种嵌入式多样性度量的分类体系，依据其计算机制将其归纳为四大类；其次，构建EmbDivBench基准测试框架，从“多样性（variety）”、“均衡性（balance）”和“差异性（disparity）”三个概念维度，在模拟与真实文本数据上对各类度量进行综合评估。实验结果表明，尽管均值成对距离（Mean Pairwise Distance）应用最广泛，但在三项核心多样性维度上均被多种替代方法超越，其中图熵（Graph Entropy）表现最优。最终，研究将实证发现提炼为可操作的多样性度量选择建议，为训练数据筛选、测试集设计及模型泛化能力评估提供科学依据。
链接: https://arxiv.org/abs/2610.24888
作者: Cantao Su,Anna Wegmann,Esther Ploeger,Dong Nguyen
机构: Utrecht University (乌得勒支大学); The Netherlands (荷兰)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Accepted to EMNLP 2026 (Main). Code and data: this https URL
Abstract:Diversity is a critical property of text corpora, with direct implications for training data curation, test set design, and model generalization. However, its measurement remains fragmented and poorly understood. Existing embedding-based diversity measures vary widely in how they are computed and can yield substantially different diversity scores on the same data, yet no systematic comparison of their empirical behaviour exists. We make two contributions toward addressing this gap. First, we present a taxonomy of 21 embedding-based diversity measures, organizing them into four families based on their underlying computations. Second, we introduce EmbDivBench, a benchmark that evaluates these measures along three conceptually distinct components (variety, balance and disparity) on both simulated and natural text data. Although Mean Pairwise Distance is the most commonly used measure, our experiments show that it is outperformed by several alternatives on the three diversity components, with Graph Entropy being the best choice overall. Lastly, we distil these findings into practical recommendations for diversity measure selection.
[NLP-184] he Missing Compliance Class: LLM Graders That Collapse Jailbreak Success and a Free Check
【速读】： 该论文旨在解决当前生成式 AI (Generative AI) 安全评估中一个关键问题：攻击成功率（即合规率）的测量结果高度依赖于评分器（grader）的选择，而非目标模型本身的特性。现有研究将合规率视为目标模型固有的属性，但本文通过固定目标模型、提示词和生成响应，仅改变评分器，发现不同评分器对相同输出的判断存在显著差异，尤其表现为“类别坍缩”（class collapse）现象——多个密集型评分器（8B 和 9B 参数规模）将原本应归类为“合规”的响应全部错误地划入“规避”（evasion）类别，甚至有评分器在900条响应中零次赋予“合规”标签。这种坍缩现象在多个不同家族的目标模型及未见的7B评分器上均重现，说明其具有普遍性。论文指出，关键解决方案在于引入对每类标签计数的透明报告机制：通过记录每个评分器的每类标签分布（无需人工标注），可低成本揭示类别坍缩问题，并从根本上修正对合规率稳定性的误判。实验表明，在包含三名坍缩评分器的情况下，合规率在不同评分器间波动达22–25倍，非拒绝率差异达18–23个百分点；而剔除坍缩评分器后，合规率差异缩小至1.7–2.8倍，非拒绝率差异仍达10–22个百分点，证明了评分器行为的系统性偏差。因此，作者提出应建立最低报告标准，强制披露合规率、各分类标签计数及具体评分器名称，同时保留非拒绝率作为补充指标，以确保评估结果的可比性和可信度。
链接: https://arxiv.org/abs/2610.24872
作者: Ruqiya Bin Safi
机构: Independent Researcher(独立研究员)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 29 pages, 6 figures, 17 tables. Code and labels: this https URL
Abstract:Jailbreak papers report an attack success rate, the share of responses that deliver the harmful request, and read it as a property of the target model. We hold the target model, the prompts and the responses fixed and change only the grader. Three dense graders of 8B and 9B parameters, from three model families, move compliant responses to the neighbouring class, evasion, instead of scattering them across classes. One of them assigns the compliance label zero times in 900 responses. This class collapse also appears on new harmful requests, on a target model from a third family, and in a held-out 7B grader from a fourth family. Per-class label counts expose it at no cost and without human labels. The collapse decides how the rate behaves under a change of grader. On two corpora of 900 responses and eight graders, the compliance rate differs by a factor of 22 to 25, while the non-refusal rate differs by 18 to 23 percentage points. Without the three collapsing graders, compliance differs by a factor of 1.7 to 2.8 on all four corpora, and non-refusal still differs by 10 to 22 points. These results do not change when every grader is called once. We propose a reporting minimum that keeps the compliance rate and adds its per-class label counts and the named grader. Beside it goes the non-refusal rate, whose disagreement across graders is additive.
[NLP-185] Bellm an Language Models: Solving the Recurring Decisions of LLM Agents
【速读】： 该论文旨在解决在受限型智能体工作流（bounded agentic workflows）中，大语言模型（LLM）因决策空间趋于饱和而导致的重复性状态访问与低效决策问题。具体表现为：当将决策与生成分离时，LLM在多次交互后反复遭遇相同状态，且未将决策过程视为可优化的问题，而是每次从上下文或提示中重新推导，造成计算冗余、成本高昂且性能不稳定。其解决方案的关键在于提出“离散化-再决策”（Discretize, Then Decide）框架，通过分析工作流日志或自运行数据，将对话转化为离散状态，并估计各动作的成本与结果概率，直至状态分布收敛；随后将该过程建模为马尔可夫决策过程（Markov Decision Process, MDP），求解出最优策略作为控制器。该控制器在运行时仅对已知状态执行预定义动作（如工具调用或模板回复），而对未知状态仍交由原始LLM处理，形成一种基于贝尔曼方程（Bellman Equation）的新型推理范式——贝尔曼语言模型（Bellman Language Model, BLM）。实验表明，该方法显著提升任务完成率并大幅降低模型调用次数与成本，在τ²-bench电信任务上使开源模型Qwen3.8-27B的任务成功率从29%跃升至96%，模型调用减少93%；对于高性能模型Claude Opus 5.5，虽本身已达96%成功率，但作为BLM运行后仍实现约98%的调用与成本削减，并额外解决2个任务。在InventoryBench基准上，无模型的优化策略超越所有已有方法。该方法在决策可预先求解的场景下效果显著，而在内容生成复杂度高、决策自由度低的场景（如SymPy问题修复）中优势有限。
链接: https://arxiv.org/abs/2610.24871
作者: Adarsh Vatsa,William Eiers
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG); Systems and Control (eess.SY)
备注: 52 pages, 13 figures, 11 tables. Preregistrations: this https URL
Abstract:A general observation in agentic workflows, especially bounded ones, is that the decision space of the LLM, when we separate a decision from generation, starts approaching saturation: the agent meets the same states again and again, and after enough conversations it stops meeting new ones. LLMs generally don’t treat these decisions as an optimization problem and instead re-derive them from their context/prompt on every turn. We show that treating decision making within reasonable bounds as an optimization problem, solving it once, and reusing the solved decisions can often make an agent faster, cheaper and more reliable. Our approach, which we call Discretize, Then Decide, starts from data such as the workflow’s own logs or, when none exists, runs of the LLM agent itself. It reads each conversation into discrete states and estimates the probability of each outcome and the cost of each move, until no new states appear and the probabilities stop changing. It then solves the resulting Markov decision process. The solution runs as a controller around the model, sending moves whose content is fixed in the state, such as a tool call or a templated reply, and any state it hasn’t seen is sent back to the LLM. We call an LLM run this way a Bellman language model (BLM), after Richard Bellman’s equation for the best move in each state. On \tau^2 -bench telecom, the controller with no model solves 79% of tasks. As a BLM, Qwen3.8-27B, an open model, goes from 29% to 96% with 93% fewer model calls. For Claude Opus 5.5, which solves 96% of tasks on its own, running it as a BLM cuts model calls and cost per task by about 98% and lets it solve 2 more tasks. On InventoryBench, the solved policy with no model beats every published method. Where the hard part is writing content rather than choosing moves, such as resolving SymPy issues, few decisions can be solved in advance, and a BLM helps little. Comments: 52 pages, 13 figures, 11 tables. Preregistrations: this https URL Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG); Systems and Control (eess.SY) Cite as: arXiv:2610.24871 [cs.AI] (or arXiv:2610.24871v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.24871 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-186] MisleadQA: Diagnosing LLM Failures under Misleading Context
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在基于检索或用户提供的上下文回答问题时，因上下文虽为事实性正确但属于范围外（out-of-scope）而产生误导的问题。现有研究多关注虚假或无关上下文的影响，但对“真实却与问题目标实体、时间或地点不符”的上下文是否同样具有误导性尚不明确，且缺乏对添加正确上下文（gold context）能否有效缓解此类误导的系统评估。为此，论文提出 MisleadQA——一个包含多选题并设有弃答选项的诊断数据集，每道题均配对了三类上下文：正确的黄金上下文（G）、真实但范围外的上下文（S）以及虚假但范围内的上下文（F），三者均支持同一错误答案。实验结果表明，在无上下文时能正确作答的模型，仅提供范围外上下文（S）会导致其在29.3%的情况下切换至错误答案，而仅提供虚假但范围内的上下文（F）则导致75.4%的错误率；加入黄金上下文（G）几乎可消除范围外上下文（S）的误导效应，但虚假上下文（F）仍能在21.0%的情况下误导模型，尤其当F出现在G之前时更为显著。此外，引入“上下文忠实指令”虽通过增加弃答行为降低部分错误，但整体准确率反而下降。研究结论强调：评估应不仅检验上下文的真实性，更需判断其与问题的相关性（scope relevance），且评价指标应同时报告正确回答、被误导和弃答的比例，以全面反映模型的鲁棒性与可靠性。
链接: https://arxiv.org/abs/2610.24851
作者: Hui Wang,Bohao Chu,Qianli Wang,Shuning Zhang,Norbert Fuhr
机构: University of Duisburg-Essen(杜伊斯堡-埃森大学); Technische Universität Berlin(柏林工业大学); Tsinghua University(清华大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are widely used to answer questions from retrieved or user-provided context. However, provided context can be factually true yet out-of-scope, describing a different entity, time, or location from the one the question targets, and still lead models to a wrong answer. Prior work has mainly studied factually false or irrelevant context. It remains unclear whether such out-of-scope context misleads models as much as false context, and whether adding gold context prevents models from being misled by either. We introduce MisleadQA, a diagnostic dataset of multiple-choice questions with an abstention option. Each question pairs gold context (G) with true but out-of-scope context (S) and false but in-scope context (F) supporting the same wrong answer. We evaluate six open-weight LLMs. On questions they answer correctly without context, providing S alone leads them to switch to this wrong answer in 29.3% of cases, and providing F alone does so in 75.4%. Adding G nearly removes the effect of S, but F still misleads models in 21.0% of cases, especially when F comes first. A context-faithful instruction reduces these errors mainly through abstention and lowers accuracy even with G. These findings suggest that evaluations should check whether context applies to the question, not only whether it is true, and should report correct, misled, and abstention rates together. Data and code are available at [this https URL](this https URL).
[NLP-187] Efficiency Measured in English: Tokeniser Coverag e Bounds Small Language Models on Indic Scripts
【速读】： 该论文旨在解决在计算资源受限条件下，小型语言模型（Small Language Models, SLMs）在多语言场景下实际效率被严重低估的问题。现有研究普遍以单一内存占用指标衡量模型效率，但该指标仅基于拉丁字母文本（如英语）计算，无法反映模型在非拉丁文字（特别是印地语系文字）中的真实表现。研究发现，尽管TinyLlama-1.1B在英语中表现出较低的子词粒度（1.34个子词/词），但在卡纳达语中却高达22.60个子词/词，其原因在于模型词汇表对多种印地语系文字（如奥里亚语、泰卢固语、卡纳达语）的代码点覆盖率极低（分别为0%、1%、2%），导致大量字符被迫通过UTF-8字节级回退处理，93.6%的输出为原始字节，且未触发未知标记（UNK），使错误完全静默化。相比之下，虽然mBERT在奥里亚语上达到最低的1.10子词/词，但其87.9%的文本因无法识别而被丢弃；而拥有更少参数的MuRIL模型则在所有十一种印地语系文字中均实现1.23–2.26子词/词的高效表示，且无UNK与字节回退。研究表明，模型性能的关键瓶颈并非参数量，而是词汇表覆盖范围（vocabulary coverage）。这一发现具有直接后果：在2048个标记的上下文窗口内，同一模型对英语用户的可用上下文可达1528词，而对卡纳达语用户仅为91词，凸显了多语言适配中词汇设计的根本性影响。
链接: https://arxiv.org/abs/2610.24818
作者: Anna Pramod
机构: University of Leeds (利兹大学)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注:
Abstract:Small Language Models with quantisation and parameter-efficient adaptation are widely offered as the route to multilingual AI under compute constraints, and their efficiency is almost always reported as a single memory figure. We show that this figure is measured in English and does not transfer. Across a 131k-sentence corpus spanning eleven Indic scripts, TinyLlama-1.1B’s tokeniser emits 1.34 subword tokens per Latin-script word and 22.60 per Kannada word. This is not inefficiency; it is absence. Its vocabulary contains 0% of Odia’s codepoints, 1% of Telugu’s and 2% of Kannada’s, and for five of eleven scripts the measured fertility sits at the UTF-8 byte-fallback ceiling, on Odia at 99.6% of it, with 93.6% of emitted tokens being raw bytes. Because SentencePiece byte fallback never raises an unknown-token marker, this failure is entirely silent: TinyLlama’s UNK rate is 0.0% on every script, including the one it reads as bytes. We further show that low fertility can indicate the opposite of efficiency: mBERT posts its best score anywhere (1.10 tokens/word, on Odia) while discarding 87.9% of the text as UNK. MuRIL, an Indic-vocabulary encoder with a fifth of TinyLlama’s parameters, reads every script at 1.23-2.26 tokens/word with zero UNK and zero byte fallback. The binding constraint is the vocabulary, not the parameter count. The consequence is concrete: on a 2048-token context window, the same model gives an English user 1,528 words of usable context and a Kannada user 91.
[NLP-188] Ask Dont Guess: Targeted Clarification for Speech Recognition Errors NAACL2027
【速读】： 该论文旨在解决远场对话场景下实时语音识别（ASR）中存在的显著错误问题，尤其关注在流式识别过程中如何有效定位潜在错误片段。其核心挑战在于：传统方法难以在不中断用户自然对话的前提下，准确识别出需要重述的错误内容。解决方案的关键在于提出一种名为 \sysname 的在线错误定位框架，通过利用干净语音对噪声语音的解码结果进行概率对齐（采用配对隐马尔可夫模型，pair-HMM），生成软标签作为监督信号，进而训练一个反事实增益评分器（counterfactual gain scorer）。该评分器融合声学与语言特征，在词粒度上输出错误可能性得分，并将得分聚合为待干预的文本段落。随后，借助大语言模型（LLM）基于用户在更优声学条件下的重述内容，仅对选定段落进行修正。实验表明，在70%精度下，\sysname相比最强基线可多识别52%的错误词元；且通过有针对性的重述策略，在兼顾中断频率与重述长度成本的预算约束下，实现了高达24%的相对词错误率（WER）降低。
链接: https://arxiv.org/abs/2610.24809
作者: Sankalp Parashar,Preethi Jyothi,Sunita Sarawagi
机构: IIT Bombay(印度理工学院孟买分校)
类目: Computation and Language (cs.CL)
备注: Submitted to NAACL 2027
Abstract:Speech recognition in real time in far-field conversational settings still incurs significant errors. Identifying potential error spans during streaming ASR permits a recourse mechanism where the speaker can be asked to repeat the affected content under cleaner acoustic conditions. We propose \sysname, a new framework for online error localization, that identifies spans likely to benefit from this intervention. Our supervision uses clean audio to estimate how the recognizer’s output might change under improved acoustic conditions. We probabilistically align noisy and clean decodes using a pair-HMM to obtain soft supervision. These soft targets train a counterfactual gain scorer by combining a suite of acoustic and language features to produce token-level scores, that are further grouped into spans for targeted recourse. An LLM uses the ASR transcript of the repetition to revise only the selected span. At a precision of 70% , \sysname achieves the highest recall by identifying up to 52% more erroneous tokens compared to the strongest baseline. Targeted repetition yields relative WER reductions of up to 24% under a budget that accounts for both interruption frequency and the lengths of re-uttered spans.
[NLP-189] Medical Fine-Tuning Reshapes Not Reduces Patient-Facing AI Risk: A Rubric-Based LLM -Judge Evaluation Framework
【速读】： 该论文旨在解决生成式医疗问答系统在实际部署中仅依赖事实正确性评估所带来的潜在临床风险问题。尽管事实准确是必要条件，但当前评估方法忽视了响应中可能包含的高风险语言（如治疗建议、禁忌症、剂量表述、紧急分诊提示、高警戒药物提及及过度自信断言），这些内容若被患者误信执行，可能导致严重医疗伤害。为此，论文提出一种风险敏感型评估框架，基于已发布的临床伤害分类体系，对六类风险语言进行识别与分级，并通过三名法官组成的大型语言模型（LLM）集合对风险严重程度进行判别。研究在K-QA基准的真实患者提问数据集上，对六款7-90亿参数规模的指令微调模型（包括通用型与医学适配型）进行了评估。结果表明：第一，风险语言的存在与事实幻觉率之间无显著统计关联（合并斯皮尔曼等级相关系数ρ = -0.065，95%置信区间[-0.18, +0.05]，距离相关性=0.10，独立性检验p=0.24），说明现有事实评估无法有效捕捉风险；第二，医学微调虽未降低总体风险水平，但改变了风险分布模式——持续预训练型模型（BioMistral、OpenBioLLM）显著抑制高警戒药物提及，却加剧了过度自信断言；而临床指令微调模型（Med42）仅抑制剂量表述，未引发过自信问题。各分类维度的克里彭多夫α值显示，四类语言学可锚定类别（α=0.53–0.71）达到公平至良好一致性，而两类依赖临床判断的类别（α=0.19, 0.31）一致性较低，揭示了自动安全评估在不同场景下的可靠性边界。研究贡献包括公开评估量表、评分提示模板及代码库，以推动更全面的医疗生成式AI安全性评估。
链接: https://arxiv.org/abs/2610.24807
作者: Savan Doshi
机构: Arizona State University (亚利桑那州立大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Patient-facing medical question answering with large language models (LLMs) is moving from research demonstrations to production deployments, yet evaluation remains anchored on factual correctness. Factuality is necessary but not sufficient: a response can be factually grounded and still issue a directive that, if acted on, would harm the patient. We introduce a risk-sensitive evaluation framework that scores responses by the presence and severity of six categories of risk-bearing language (treatment directives, contraindications, dosage expressions, urgency-triage cues, high-alert medication mentions, and overconfident assertions), anchored on published clinical harm taxonomies. Severity is classified by a three-judge LLM ensemble. We apply this framework to six 7-9B instruction-tuned models, three general (Mistral-7B, Phi-3.5-mini, Llama-3.1-8B) and three medically adapted (BioMistral-7B, OpenBioLLM-Llama3-8B, Med42-Llama3-8B), on 200 real patient questions from the K-QA benchmark. Two findings emerge. First, ensemble risk-bearing language is statistically independent of factual hallucination rate (pooled Spearman \rho = -0.065 , n = 300 , 95% CI [-0.18, +0.05] ; distance correlation = 0.10 , independence test p = 0.24 ). Second, medical fine-tuning does not reduce aggregate risk but reshapes it: continued-pretraining adaptations (BioMistral, OpenBioLLM) consistently suppress high-alert medication mentions while amplifying overconfident assertions, whereas clinical instruction-tuning (Med42) suppresses dosing only, without amplifying overconfidence. Per-category Krippendorff \alpha is fair to good for the four linguistically anchored categories (0.53-0.71) and low for the two clinical-judgment-dependent categories (0.19, 0.31), indicating where automatic safety evaluation is and is not yet reliable. We release the rubric, judge prompts, and evaluation code.
[NLP-190] able Serialization Determines Which Relations Language Models Preserve
【速读】： 该论文旨在解决文本序列化方式对指令微调语言模型中表格查找任务性能的影响问题，具体关注不同行列存储格式（行优先与列优先）在不同表格宽高比下的表现差异。其核心解决方案的关键在于揭示了“关系局部性”（relational locality）这一现象：文本序列化方式决定了某一维度的关系是否在序列中保持局部连续，从而影响模型对行或列的定位能力。实验表明，行优先格式在高瘦型表格（如64×4）中表现更优，而列优先格式在宽扁型表格（如4×64）中更具优势，且错误模式呈现互补性——行优先易丢失目标列，列优先易丢失目标行。这说明表格查找失败并非单纯由表的几何形状引起，而是源于序列化过程对关系局部性的破坏，因此优化表格文本表示应考虑其与模型处理机制的匹配性。
链接: https://arxiv.org/abs/2610.24798
作者: Junwei Ma,Arun Kumar Chithanar,Palash Goyal,Chul Lee
机构: University of Toronto(多伦多大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:We study how text serialization affects table lookup in instruction-tuned language models. Our diagnostic asks the model to find the row containing a specified value under one named column, then return the value under another named column in that same row. We render the same synthetic 256-cell tables in row-major form and in column-major form with each header placed beside its values, while varying the table aspect ratio across five shapes. Across three models, the better format changes with table shape: row-major performs better on tall tables, whereas column-major performs better on wide tables. At 64 x 4, row-major leads by 42.3-54.0 percentage points; at 4 x 64, column-major leads by 35.0-59.0 points. Mapping wrong answers back to their source cells reveals a complementary error pattern: row-major tends to preserve the correct row but lose the target column, while column-major tends to preserve the correct column but lose the row. We use the term relational locality to describe this pattern: serialization keeps one relation locally available while making the other depend on positional alignment. A separate non-tabular ordering experiment supports this interpretation. Thus, apparent width-related failures in text-serialized tables depend on linearization, not table shape alone.
[NLP-191] Who Gets Unmasked? Sampler-Induced Bias in Diffusion Language Models
【速读】： 该论文旨在解决扩散语言模型（Diffusion Language Models, dLLMs）在文本生成过程中因采样器决策机制引入的偏差问题，尤其是采样器在决定“何时提交”（commit）某个词元时所引发的隐性偏见。其核心问题是：当前主流的采样策略（如LLaDA默认的基于概率峰值的提交规则）会优先选择高置信度的词元，导致生成内容过度偏向多数群体，加剧职业性别刻板印象并压缩观点分布，从而在不改变模型本身参数的情况下引入系统性不公平。解决方案的关键在于提出“先分叉后采样”（Fork-First Sampling）机制——即在生成任何文本之前，首先根据模型对关键语义分支（如主角性别）的信念分布进行决策，该信念通过一次前向传播即可获得，并可与目标分布对齐。此方法有效分离了采样器与模型本体，避免了传统采样中因“窥探”（peeking）导致的偏差，显著缓解了性别刻板印象，提升了生成文本的公平性与流畅性，同时减少了需依赖提示工程修复的冗余干预。研究强调，对扩散语言模型的公平性评估必须明确报告所采用的提交规则（commit rule），以确保结果可复现且具有可比性。
链接: https://arxiv.org/abs/2610.24795
作者: Xujun Che,Farsheed Haque,Depeng Xu,Xi Niu,Haopeng Zhang
机构: University of North Carolina at Charlotte (北卡罗来纳大学夏洛特分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Diffusion language models (dLLMs) generate by gradually unmasking a canvas, so their samplers decide not only which token to write but also when to commit it. We show that this second decision changes who appears in generated text, a sampler-induced bias on top of the model’s own. Common commit rules, including LLaDA’s default, commit the positions whose sampled tokens are most probable; this peeking favors majority options. To separate the sampler from the model, we read the model’s belief about a fork, such as a protagonist’s gender, in one forward pass, and show that this belief matches exact sampling from the model. Measured against it, peeking samplers on LLaDA and Dream amplify occupational gender stereotypes and collapse opinion distributions, and confidence-based ordering shifts gender-balanced occupations toward men, more so under faster parallel decoding. We propose Fork-First Sampling, which decides the fork before anything else is written: it follows the belief or any target distribution, improves fluency, and avoids the stories that remark on the protagonist’s gender, which prompt-based fixes produce, at the cost of one forward pass. Fairness evaluations of dLLMs should report the commit rule.
[NLP-192] Information Theory of Random Composition Trees
【速读】： 该论文旨在解决自然语言文本中层次化语义结构的信息论特性建模问题，特别是如何通过随机组合树（random composition tree）这一概率生成模型来刻画不同文本类型在信息熵方面的共性与差异。其核心解决方案的关键在于揭示该随机分裂模型所具备的两个关键信息论性质：一是树的熵随根节点规模线性增长，且可推导出熵率关于分支因子的解析表达式；二是大规模树结构在高概率下呈现典型性（typicality），即绝大多数树属于一个各成员近似等概率的典型集合。研究表明，该模型在熵增长行为和典型性方面与将语言建模为平稳遍历随机过程时的统计特性高度一致，从而验证了随机组合树作为语言层次结构有效信息论表征的合理性。
链接: https://arxiv.org/abs/2610.24788
作者: Doron Sivan,Weishun Zhong,Tankut Can,Mikhail Katkov,Misha Tsodyks
机构: Weizmann Institute of Science (魏茨曼科学研究所); Institute for Advanced Study (高级研究院); Emory University (埃默里大学)
类目: Information Theory (cs.IT); Computation and Language (cs.CL)
备注:
Abstract:Random composition trees model the hierarchical semantic structures of meaningful texts as a statistical ensemble generated by recursively splitting texts into coherent chunks. A recent study shows that different text types can be described by random composition trees with distinct branching factors, and that the entropy rates of these trees exhibit a similar dependence across text types as the entropy rates of the actual texts. Motivated by this correspondence, we analyze information-theoretic properties of this stochastic splitting model. First, we demonstrate that the entropy of random composition trees grows linearly with the root size, and we derive an analytical expression for the entropy rate as a function of the branching factor. Second, we prove that large trees exhibit typicality: with high probability, they belong to a typical set in which each tree is approximately equally likely. Our analysis reveals that the random composition tree model shares two key information-theoretic properties of language when modeled as a stationary and ergodic stochastic process: linear entropy growth and typicality.
[NLP-193] ECHO: Efficient Causal Self-Recursion Beyond Auxiliary Multi-Token Prediction
【速读】： 该论文旨在解决现有多时间步预测（Multi-token prediction, MTP）方法中存在的效率与扩展性问题：传统MTP方法依赖于附加在主干网络上的独立模块进行多时序预测，每个新增时间步均需额外模块和一次前向传播，导致计算开销随预测深度线性增长。其核心解决方案是提出高效因果自递归（Efficient Causal Self-Recursion, ECHO），关键在于将未来预测任务融入自回归目标本身——通过主干网络的最终层直接生成所有时间步的预测结果，使监督信号作用于实际生成下一个标记的参数，从而消除对辅助模块的依赖。同时引入动态深度采样（Dynamic Depth Sampling, DDS），以深度递减的概率分布控制训练深度，使深层预测在计算上保持低成本。这一设计使预测能力成为自回归过程的内在属性，显著降低预训练阶段的计算开销（仅为传统MTP的一半），并在下游任务中提升推理与代码生成性能；在推理阶段，预训练模型可自主生成超出训练深度的文本，维持高熵生成中的草稿-目标一致性，即使在双倍词元预算、重复预训练及微调后仍保持优势。
链接: https://arxiv.org/abs/2610.24766
作者: Yuxuan Zhou,Junyuan Shang,Tianyu Wang,Chunrui Zheng,Zichang Liu,Wenbin Chang,Shuohuan Wang,Dianhai Yu
机构: ERNIE Team, Baidu(百度); Independent Researcher
类目: Computation and Language (cs.CL)
备注:
Abstract:Existing Multi-token prediction (MTP) methods are auxiliary by design: the multi-horizon loss mostly trains modules attached to the backbone, and each added horizon costs another module and another pass. We propose Efficient Causal Self-Recursion (ECHO), which removes both costs. The backbone’s own final layer predicts every horizon, so the supervision lands on the parameters that generate the next token, and Dynamic Depth Sampling (DDS) keeps deeper horizons affordable by drawing the training depth with probabilities that decay along depth. Future prediction thus becomes part of the autoregressive objective rather than a task attached to it, and the gains follow on both sides of it. In pretraining, added depth converts into downstream accuracy where auxiliary modules saturate, concentrating on reasoning and code at half the MTP training overhead of training every depth. At inference, the pretrained model is its own drafter: it drafts beyond the depth it was trained for, and keeps its draft-target alignment on the open-ended, high-entropy generation where post-hoc drafters lose theirs. The advantage survives a doubled token budget, a repeated pretraining run, and supervised finetuning.
[NLP-194] BanglaBBQ: A Bias Benchmark for Question Answering in the Bangladeshi Context
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在多语言、多文化背景下对特定社会群体存在的隐性偏见问题，尤其聚焦于孟加拉国社会中的刻板印象。为实现这一目标，研究提出BanglaBBQ基准数据集，通过改编原始的问答偏见基准（BBQ），并结合本地化语境进行三类模板处理：直接翻译（DT）、适配孟加拉国目标群体或背景的修改版（TM）以及因不适用而被移除的模板（SR）。在此基础上，新增5个基于孟加拉社会常见刻板印象的类别（婚姻状况、年龄与政治认知、地域来源、教育背景及心理健康状态），共生成466个新模板，经母语孟加拉国标注者评估后保留372个。所有有效模板均附有文献来源和原文引述，并包含反刻板印象对照版本。实验结果表明，尽管四款模型（包括开放权重的Qwen3-4B/8B与封闭模型GPT-4.1-mini、Gemini 2.5 Flash）在模糊语境下普遍倾向于刻板印象，且英文表现无法预测孟加拉语下的行为模式，但其偏差程度显著受人工验证模板的影响——即标注者认可的模板引发更强偏见。关键解决方案在于构建一个以本地文化为基础、具备可解释性与反刻板对照结构的多语言偏见评估框架，从而更精准地揭示和量化生成式AI在非西方语境中的系统性偏见。
链接: https://arxiv.org/abs/2610.24764
作者: Sumit Anthony Gomes,Yasin Rahman,Taeeba Tasnia,Faiza Nooren Suhita
机构: BRAC University (BRAC大学)
类目: Computation and Language (cs.CL)
备注: 13 pages. Data and code: this https URL
Abstract:This paper introduces BanglaBBQ, a question-answering bias benchmark designed to capture stereotypes grounded in the Bangladeshi context, in both Bangla and English. Following KoBBQ and PakBBQ, BanglaBBQ adapts the original Bias Benchmark for Question Answering (BBQ) by classifying its templates into three groups: directly translated (DT), modified to align target groups or contexts with Bangladesh ™, and removed when deemed unsuitable for the Bangladeshi context (SR). This process retains 140 templates across six categories, comprising 76 DT and 64 TM templates. Five newly authored (NA) categories, based on stereotypes common in Bangladeshi society, are then added: marital status, age and political perception, regional origin, educational background, and mental-health status. Native Bangladeshi annotators judged all 466 newly authored templates, and 372 were retained because a strict majority of English annotators identified the intended target and rated the attitude as common enough. Each newly authored template is linked to a documented source and a verbatim quote from it. Each evaluated template also has a counter-stereotypical version of its disambiguated context. The dataset was evaluated using two open-weight models (Qwen3-4B and Qwen3-8B) and two closed models (GPT-4.1-mini and Gemini 2.5 Flash). All four lean towards the stereotyped person when the context is ambiguous, in both languages, and the effect of language differs by model, so English results do not predict behaviour in Bangla. On the adapted BBQ templates, both Qwen3 models are also less accurate when the context contradicts a stereotype than when it confirms it. On the newly authored templates, the closed models follow counter-stereotypical evidence almost perfectly, and all four models are more biased on the templates that annotators validated than on those they rejected.
[NLP-195] GLOSS-MT: Controllable Low-Resource Machine Translation with an Editable Dictionary
【速读】： 该论文旨在解决低资源语言在机器翻译（MT）中因术语适配而需大量平行数据与重复微调所带来的高成本问题。其核心挑战在于如何在不损害预训练模型性能的前提下，实现对翻译结果中专有术语的可控生成。解决方案的关键在于提出一种名为GLOSS-MT的轻量级方法：通过离线构建英文-阿桑特蒂语词典，将源文本中的名词短语直接替换为对应目标语言术语，并在此混合输入上进行一次统一微调。该方法无需重新训练即可实现术语控制，且在文档级英-阿桑特蒂语基准测试中，所有微调模型均显著优于其预训练基线（chrF2提升2.85至5.48），其中三个NLLB变体（600M、1.3B、3.3B）成为当前最优开源系统（得分50.12–50.80）。人类译文对照测试进一步验证了其可控性：所测试模型可正确输出79%–97%的词典术语，且在替换词典条目后，新术语以70%–97%的比例准确出现在输出中，无需额外微调。
链接: https://arxiv.org/abs/2610.24730
作者: Mich-Seth Owusu,Stephen E. Moore,Akwasi Asare,Lawrence Adu Gyamfi,Paul Azunre,Joel Budu,Jonathan Asiamah
机构: Paul Azunre; Joel Budu; Jonathan Asiamah
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 4 pages, 2 tables. Models: this https URL ; code: this https URL
Abstract:Adapting a pretrained machine translation (MT) model to new terminology usually requires new parallel data and another round of fine-tuning, which is costly for low-resource languages. We ask whether controllability can be added to MT, and the fine-tuning burden reduced, without degrading the pretrained model. GLOSS-MT looks up the source’s noun phrases in an offline English-Twi dictionary, replaces every phrase found by its Twi translation in place, and fine-tunes a translation model once to translate these mixed inputs. We apply the same data and pipeline to NLLB-200 (600M, 1.3B, 3.3B), OPUS-MT en-tw and TranslateGemma 4B. On a document-level English-to-Asante Twi benchmark (84 documents, four domains), every fine-tune scores above its pretrained base in chrF2 (+2.85 to +5.48), and the three NLLB fine-tunes are the best open systems (50.12-50.80). A direct controllability test on human-translated Bible passages shows that the three fine-tunes tested write 79-97% of the dictionary terms, and that replacing every dictionary entry by a different term changes the output accordingly (70-97% of the new terms appear) with no retraining. We release the models, data and code.
[NLP-196] PV-Decoding: Progressive Verification Decoding with Decode Gating for Diffusion Language Models
【速读】： 该论文旨在解决离散扩散语言模型（discrete diffusion language models, dLLMs）在推理过程中因安全提交暂定预测而带来的效率瓶颈问题，尤其针对其在推测解码（speculative decoding）中面临的挑战：同一扩散模块内的推测标记之间存在依赖关系，导致激进接受策略易引发错误传播，而严格验证又带来高昂的计算开销。为此，论文提出PV-Decoding框架，其核心创新在于基于置信度驱动的分层验证机制——通过两个置信度阈值将掩码位置划分为直接解码、推测性解码和未决解码三类，并采用逐步增强的验证策略（单批次、双批次及多批次验证），实现精度与效率之间的可调控权衡。该设计允许在降低验证开销的同时维持较高的接受准确率与鲁棒性。实验结果表明，在GSM8K、MBPP和HumanEval三个基准上，相较于基于置信度的传统解码方法，PV-Decoding可减少8%–37%的序列模型前向传播次数（MFPs），且最大准确率下降仅为1.3%；在Dream-7B-Instruct/GSM8K任务中，1-批和2-批验证版本在相同硬件预算下分别提升吞吐量7.0%–13.6%，而多批次验证因运行时开销较大暂未体现优势。研究证实，基于置信度引导与结构化验证的推测解码是实现高效且可靠的dLLM推理的有效路径。
链接: https://arxiv.org/abs/2610.24722
作者: Renxi Liu,Yiming Li,Shengyin Sun,Chen Chen,Xian Wang,Shuai Yang,Huiling Zhen,Xianzhi Yu,Yaoyuan Wang,Weizhe Lin,Xiaosong Li
机构: Huawei Technologies(华为技术); City University of Hong Kong(香港城市大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Discrete diffusion language models (dLLMs) offer an attractive alternative to autoregressive generation by updating multiple tokens in parallel, but their inference efficiency is still limited by the cost of safely committing tentative predictions. In particular, speculative decoding for dLLMs is challenging because speculative tokens within the same diffusion block may depend on one another, making aggressive acceptance prone to error propagation and strict verification computationally expensive. We present PV-Decoding, a confidence-driven speculative decoding framework with progressive verification for block diffusion inference. PV-Decoding partitions masked positions into directly decoded tokens, speculative tokens, and unresolved tokens using two confidence thresholds, and then verifies speculative tokens with progressively stronger policies, including single-batch, dual-batch, and multi-batch verification. This design exposes a practical accuracy–efficiency trade-off: weaker verification minimizes overhead, whereas stronger verification improves acceptance precision and robustness. Experiments on three diffusion language models across GSM8K, MBPP, and HumanEval show that PV-Decoding reduces sequential model forward passes (MFPs) by 8%–37% relative to confidence-based decoding, while keeping the largest observed accuracy drop to 1.3%. On Dream-7B-Instruct/GSM8K, PV-Decoding (1-batch) and PV-Decoding (2-batch) improve throughput by 7.0%–13.6% under the same hardware budget, whereas PV-Decoding (multi-batch) is slower under the current runtime. These results suggest that confidence-guided speculation with structured verification is a practical path toward efficient and reliable dLLM inference.
[NLP-197] DiffuPlan: Editable Planning State for Diffusion Language Model Agents
【速读】： 该论文旨在解决扩散语言模型（Diffusion Language Models, dLLMs）在构建智能体（agent）时仍沿用自回归（Autoregressive, AR）模型的单向追加式交互范式所带来的局限性。传统AR代理一旦生成中间状态便无法修改，导致其难以有效利用后续出现的上下文信息进行动态调整。而dLLM具备基于前后上下文重构早期文本的能力，这一特性为实现可编辑的规划状态（editable planning state）提供了可能。论文提出的核心解决方案是DiffuPlan，一种基于扩散搜索的智能体框架，其关键在于维护单一可编辑的规划状态，并在新证据出现时仅对受影响的步骤进行重建，从而实现高效、精准的状态修正。为支持可变长度的修订操作，引入了扩展-分割边缘化（Expansion-Partition Marginalization, EPM）机制，通过边缘化等效扩展路径来保证推理的稳定性与灵活性。此外，论文还将扩散轨迹强化学习（diffusion trajectory RL）方法拓展至多轮搜索任务中。在四个多跳问答（multi-hop QA）基准测试中，DiffuPlan平均取得50.24的F1分数，显著优于强基线AR搜索代理；更重要的是，在相同数据与训练条件下，可编辑规划状态使扩散代理性能提升2.44 F1，远超其AR源模型仅0.47的提升幅度。研究结果表明，dLLMs无需简单模仿AR代理的追加式行为，直接修订已承诺状态是其作为智能体的独特优势。
链接: https://arxiv.org/abs/2610.24717
作者: Qisheng Zhou,Zhen Xiong,Qiaoyu Tan
机构: New York University Shanghai(纽约大学上海分校); New York University(纽约大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 19 pages, 5 figures. Code: this https URL
Abstract:Diffusion language models (dLLMs) are emerging as an alternative to autoregressive (AR) language models, yet dLLM agents largely inherit an interaction protocol designed for AR generation: once intermediate state is written, later observations can only be incorporated by appending more text. This is a poor match to a defining capability of dLLMs – reconstructing earlier text conditioned on context that appears both before and after it. We ask whether this capability can support a different agent-state paradigm in which intermediate planning state is directly revisable as evidence arrives. We introduce DiffuPlan, a diffusion search agent that maintains a single editable planning state and reconstructs only the affected steps when later evidence resolves an entity or changes the search route. To support variable-length revisions, we introduce Expansion-Partition Marginalization (EPM), which marginalizes over equivalent expansion paths. We further adapt diffusion trajectory RL to multi-turn search. Across four multi-hop QA benchmarks, DiffuPlan achieves 50.24 average F1, outperforming strong AR search agents. More importantly, under identical data and training, editable planning state improves the diffusion agent by 2.44 F1 versus only 0.47 for its AR source model. These results show that dLLMs need not imitate append-only AR agents: revising committed state is a distinct agentic advantage.
[NLP-198] When Do Repetition Loops Lock? Truncation Sampling Bounded Penalties and a Blind Spot in DRY
【速读】： 该论文旨在解决生成式 AI（Generative AI）在文本生成过程中出现的循环重复（verbatim repetition）问题，即模型在生成序列时反复重复已有内容直至达到最大生成长度限制。其核心关注点在于“逃逸”机制：当某个文本片段已被重复 $ n $ 次后，仍存在逃离循环的可能性？研究发现，在确定性上下文环境下，循环生存概率可通过一次前向传播精确计算。纯采样策略在每一步均保留正的逃逸概率；而基于 top-$ p $ 的采样策略在续写概率达到阈值 $ \pi $ 时会精确锁定某一位置。若惩罚项为有界，则可能导致永久锁定；而当惩罚项增长速度超过模型对循环路径的偏好时，可确保在 top-$ p $ 策略下实现逃逸。分析进一步揭示了广泛部署的 DRY 采样器存在一个盲区：其序列断裂器将惩罚上限设为 $ \lambda b^{g-1}-a $，其中 $ g $ 为无断裂器最长运行长度，导致在 $ g=6 $ 时约 66% 的锁定单元无法逃脱，$ g=8 $ 时仍有 1% 保持锁定。在实际测试中，所有成功绕过 DRY 的 109 个循环均满足 $ g \leq 7 $。为此，作者提出 DRY-C，引入一项随重复次数呈指数增长的惩罚项，且受限于回溯窗口。实验表明，使用 DRY-C 后，在 720 个结构化输出任务中均未出现到达长度限制的循环，同时任务完成率与原始 DRY 保持相当。因此，解决方案的关键在于设计一种动态增长的、依赖重复次数的惩罚机制，以有效打破长周期循环并提升生成多样性。
链接: https://arxiv.org/abs/2610.24715
作者: Yichuan Chen,Xin Wang
机构: Zhejiang Sci-Tech University (浙江科技学院); Hangzhou, China
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 17 pages, 5 figures
Abstract:Language models sometimes repeat verbatim until the generation limit. We study the exit rather than the entrance: given that a span has already been repeated n times, when is escape still possible? Along the loop the context is deterministic, so loop survival under any sampler can be computed exactly from one forward pass. Pure sampling leaves a positive escape probability at every step; top- p with threshold \pi1/2 locks a position exactly when the continuation probability reaches \pi . A bounded penalty can leave a position locked; a penalty that outgrows the model’s preference for the loop over unseen tokens guarantees escape under top- p . The same analysis predicts a blind spot in the widely deployed DRY sampler: its sequence breakers cap the penalty at \lambda b^g-1-a for loops whose longest breaker-free run has g tokens. On eleven open models, 66% of locked units with g=6 stay locked under DRY and 1% with g=8 . In free generation with this http URL, all 109 loops that survive DRY have g\le 7 . We propose DRY-C, which adds a term that grows exponentially with the number of verbatim recurrences, up to its look-back window. With it, none of 720 structured-output generations looped to the length limit, and task completion stayed close to DRY’s.
[NLP-199] Verifying Human-Written Reasoning References
【速读】： 该论文旨在解决当前评估人工智能（AI）系统推理能力时，依赖人工编写的推理链作为参考却缺乏对其每一步逻辑支持的验证问题。现有方法虽在神经符号框架下实现了推理的形式化与检查，但未系统性地检验人工参考推导中各步骤结论是否由其引用的前提集所蕴含，也未深入分析参考错误的根源。为此，论文提出一种以参考为中心的新型评估方法，其关键在于结合前提集蕴含性检查、完整前提重检以及错误归因分析，全面评估人工推导及其形式化表示的可靠性。研究在P-FOLIO数据集上开展，利用大语言模型（LLM）将推理步骤结论翻译为一阶逻辑（FOL），并借助Z3 SMT求解器进行蕴含性验证。实验结果显示，在386条推导中，52.3%（202条）满足全部检查标准，即每一步结论均被其声明的支持前提所蕴含，且目标结论可在给定的FOL表示下成立；对于191例最初不满足蕴含性的步骤，当使用完整原始前提重新检验后，多数可被证实为蕴含，表明初始失败主要源于支持前提不完整或匹配不当，而非无根据的断言；对剩余39例的细粒度人工分析揭示了21例人类推理错误、11例一阶逻辑翻译错误和7例数据标注错误。此外，对50条通过蕴含性检查的随机样本进行独立人工审计，发现92%的翻译忠实于原意。研究结果表明，与人工参考不一致的原因可能源于参考本身或其形式化表示的缺陷，而非被评估的AI系统存在问题。
链接: https://arxiv.org/abs/2610.24690
作者: Ali Ansari,Farhana Ferdousi Liza
机构: University of East Anglia(东英吉利大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Logic in Computer Science (cs.LO)
备注: 24 pages, 1 figure, 3 tables. Preprint
Abstract:Evaluating the reasoning of AI systems often relies on human-written reasoning chains as references, yet the logical support for their individual steps is rarely verified. Although prior neuro-symbolic approaches have formalized and checked reasoning, their reported evaluations do not systematically assess whether step conclusions in human-written reference derivations are entailed by their cited premise sets or investigate the sources of failures in those references. To address these limitations, we present a novel reference-centred assessment method for human-written derivations that combines entailment checking against cited premise sets, full-premise rechecking, and failure attribution to examine the reliability of these derivations and their formal representations. We evaluate it on P-FOLIO, using a large language model (LLM) to translate step conclusions into first-order logic (FOL) and the Z3 SMT solver to check entailment. Of 386 derivations examined, 202 (52.3%) meet all our checking criteria, meaning that every translated step is entailed by its stated support and the required target is established under the available FOL representations. Among 230 steps not entailed by their stated support, 191 become entailed when checked against the complete original premises, indicating that most initial failures reflect incomplete or mismatched support rather than unsupported claims. Manual analysis of the remaining 39 cases identifies 21 human reasoning errors, 11 FOL translation errors, and 7 dataset annotation errors. A separate manual audit of 50 randomly sampled steps that passed the entailment checks found 46 faithful FOL translations (92%). These findings suggest that disagreement with a human-written reasoning reference may originate in the reference or its formal representation rather than in the AI system being evaluated.
[NLP-200] Do Conversational Agents Need Explicit Memory? A Controlled Comparison with Raw-History Retrieval
【速读】： 该论文旨在解决当前生成式AI（Generative AI）系统中显式记忆（explicit memory）机制的有效性问题，即在对话历史处理中引入复杂且高成本的显式记忆构建是否真能提升答案准确性。其核心问题是：相较于直接从原始对话历史中检索信息的原始历史检索增强生成（raw-history RAG），复杂的显式记忆系统是否具有稳定且显著的优势。研究的关键发现在于，尽管部分显式记忆系统在特定问题类型和严格上下文长度限制下表现出局部性能提升，但在整体评估中并未展现出一致优于强基线的原始历史RAG模型的表现；更关键的是，通过“上帝视角”（Oracle）检索分析揭示，许多现有系统在构建显式记忆时未能有效保留支持答案的关键信息，或虽保留但无法高效检索，而这些记忆构建过程甚至耗时数小时。因此，该研究的核心解决方案之关键在于强调：在当前大多数显式记忆系统过度追求复杂构造的同时，应加强与原始历史RAG的严格对照实验，并重新重视检索效率的重要性——即在已有信息被妥善保存的前提下，改进信息获取能力可能比构建更丰富的显式记忆更具实际价值。
链接: https://arxiv.org/abs/2610.24671
作者: Yongjie Zhou,Shuai Wang,Bevan Koopman,Guido Zuccon
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Explicit memory systems transform conversation history into persistent representations intended to improve future access, often at substantial construction cost. Yet it remains unclear whether this added complexity yields better answer accuracy than retrieving directly from raw history. This study compares six state-of-the-art explicit memory systems with raw-history RAG on LongMemEval-S and BEAM-100K under original and controlled retrieval settings. We find only localized gains for particular question types and tight run-time context budgets, with no consistent advantage over strong raw-history RAG baselines. Oracle retrieval shows that some systems fail to preserve answer-supporting information during explicit memory construction, while others preserve it but fail to retrieve it effectively, despite construction costs reaching hours per conversation. Our findings suggest that while many existing memory systems for AI agents emphasize increasingly elaborate explicit memory construction, they are often evaluated without strong controlled comparisons to raw-history RAG and may underemphasize retrieval. Retrieval remains a major bottleneck when useful information is preserved, suggesting that improving access to existing information may be at least as important as constructing richer explicit memory. Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.24671 [cs.CL] (or arXiv:2610.24671v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.24671 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-201] What RL Brings to Reasoning : A Few Decisions the Base Model Would Not Make
【速读】： 该论文旨在解决生成式 AI（Generative AI）在强化学习中通过可验证奖励（Reinforcement Learning with Verifiable Rewards, RLVR）训练后所提升的数学推理能力究竟来源于何处这一关键问题。其核心发现是：尽管大部分性能提升看似源于整体模型行为的改变，但实际贡献主要来自对少数特定生成位置的精准干预。解决方案的关键在于“稀疏替换”（sparse replacement）策略——即仅在基础模型与强化学习模型在贪婪解码时产生分歧的位置（约占生成序列的4%）处，替换为强化学习模型的输出token。实验表明，这种策略可恢复90.5%的pass@1性能增益，且在多个模型对上均表现一致。此外，研究揭示了初始词“Alright”对性能提升具有显著影响，能恢复约35%的增益，但仅靠开头优化不足以达成全部收益；后续在分歧点插入强化学习模型的决策可进一步提升5.2个百分点。值得注意的是，引入反思性词语如“Wait”等固定规则无法复制该效果，说明真正的优势并非来自语义提示，而是具体决策路径的精确借用。该方法在保持基础模型多样性的同时，实现了超越原强化学习模型的性能（在k=256时达67.5%正确率），且对比基准为相同尝试次数而非计算量，凸显其高效性。结果表明，生成式 AI 的推理增强不仅依赖于初始引导，更关键的是在关键时刻对错误路径进行修正。
链接: https://arxiv.org/abs/2610.24663
作者: Siyu Zhang,Yuquan Lu,Yoshua Y. Li,Yifan Ding,Kun Zeng,Yu Huo,Yifu Guo,Rongxiang Weng,Xunliang Cai,Xiaoying Tang
机构: University of California, San Diego (加州大学圣地亚哥分校); Meituan, LongCat Team (美团龙猫团队); The Chinese University of Hong Kong, Shenzhen (香港中文大学（深圳)）
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 26 Pages, 8 figures Code: this https URL
Abstract:Reinforcement learning with verifiable rewards (RLVR) substantially improves mathematical reasoning, but what it adds to the base model remains unclear. We investigate this through a sequence of generation-time interventions on Qwen3-4B-Base, measuring how much of its RL-trained counterpart’s gain each intervention recovers on 120 AIME problems. Power sampling with the same 8,192-token output limit brings no gain beyond greedy decoding. Yet borrowing the RL model’s token only where the two models’ greedy choices differ, roughly 4% of generated positions, recovers 90.5% of the pass@1 gain, with similar recovery across three model pairs on MATH-500. We find that even a single opening token has a substantial effect: forcing the base model to begin with “Alright” recovers 35% of the RL gain. The opening alone is insufficient, however. Later replacements after the same opening add another 5.2 percentage points to pass@1, and the fixed rules we test for inserting reflective words such as “Wait” do not reproduce this benefit. Sparse replacement also retains much of the base model’s diversity while staying close to the RL model at pass@1. At k=256, it solves 67.5% of the problems, significantly more than the RL model (61.4%) and a mixture of whole responses matched in pass@1 (61.45%). The comparison is at equal attempt counts rather than equal compute, since rl-replace consults both models at every decoding step. These results show that the opening can recover a meaningful part of the RL gain, but a larger contribution comes from later choices, which can be borrowed while retaining much of the base model’s diversity. Code is available at this https URL.
[NLP-202] Muse Glimmer 30B: An On-Device Agent ic Model
【速读】： 该论文旨在解决在消费级硬件上实现具备自主代理能力的本地化大模型部署难题，尤其针对生成式AI（Generative AI）在工具调用、多步推理与任务失败恢复等方面的性能瓶颈。其核心挑战在于如何在有限算力条件下，使模型兼具高可靠性、多模态感知能力与高效推理效率。解决方案的关键在于构建一个300亿参数的因果语言模型（causal language model），集成感知编码器以支持文本与图像输入，并通过约4比特量化压缩实现单GPU上的高效本地推理；同时，模型在跨百种语言数据上训练，强化了多语言环境下的泛化能力，从而在工具使用、多步规划及代理任务完成等基准测试中超越同规模模型（如Gemma4-31B和Qwen3.6-27B），实现了高性能与本地化部署的平衡。
链接: https://arxiv.org/abs/2610.24628
作者: Arash Ebrahim Abtahizadeh,Muna Aghamelu,Adil Ahmed,Juan C. Angeles Acuna,Alexis Audran-Reiss,Asaf Avrahamy,Yoram Bachrach,Can Balioglu,Amar Budhiraja,Onur Çelebi,Gaurav Chaurasia,Joy Chen,Lynn Cherif,Lucia Cipolina-Kun,Gal Cohen,Timothée Darcet,Beto de Paola,Digant Desai,Sam Devlin,Derek Dunfield,Ivan Evtimov,Maryam Fazel-Zarandi,Rob Fergus,Jean-Christophe Gagnon-Audet,Grant Gardner,Bhavul Gauri,Paris Giampouras,Sujan Kumar Gonugondla,Antoine Grosnit,Young Han,Michael Hassid,Monuril Islam,Masoud Jalili Sabet,Songhao Jia,Narine Kokhlikyan,Svetlana Kurtsman,Sandra Lefdal,Chee Hau Leow,Noam Levi,Mike Lewis,Sanae Lotfi,Xiyang Luo,Despoina Magka,Saeed Mahloujifar,Neal Mangaokar,Ishita Mediratta,Alexander H. Miller,Hela Momand,Saba Nazir,Kelvin Niu,Mattia Opper,Parth Pathak,Alberto Pepe,Danielle Pintz,Mikhail Plekhanov,Sidhartha Reddy Potu,Anton Protopopov,Shengyi Qian,Shashank Rajput,Joseph Rocca,Scott Roy,Luca Sbordone,Amin Shakeri,Tatiana Shavrina,Michael Shvartsman,Ruan Silva,Anya Sims,Eric Michael Smith,Hugo Touvron,Mabel Ubong,Isabel Urrego,Yanzheng Xiang,Zhiyuan Yao,Alex Yarmula,He Ye,Jade Yu,Luke Zettlemoyer,Arman Zharmagambetov,Perpetua Fathima Ahlijah,Tremo Ahmed,Enrique Alfonseca,Pedram Bashiri,Nikolay Bashlykov,Aaron Berman,Rohith Bhatia,Siddharth Bhattacharya,Prashant Bhende,Chloe Bi,Shuchao Bi,Clement Anthonioz Blanc,Allen Bolourchi,Andrea Caciolai,Erica Cai,Xiang Tony Cao,Yue Cao,Andrew Caples,Jinzhao Chang,Lele Chen,Moya Chen,Weijie Chen
机构: Meta Superintelligence Labs (Meta超智能实验室)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:We introduce Muse Glimmer, an open-source 30B parameter causal language model with perception encoder, purpose-built for autonomous agentic tasks on consumer hardware. Muse Glimmer integrates multi-step reasoning, reliable tool use and failure recovery into a single model that runs locally on consumer devices without requiring cloud infrastructure. The model accepts text and image inputs and produces text outputs, was trained on data from over a hundred languages, and is quantized to approximately 4-bit precision for efficient single-GPU inference. Muse Glimmer achieves state-of-the-art performance for its size class on agentic benchmarks, setting a new standard compared to the models of its size: Gemma4-31B and Qwen3.6-27B on tool use, multi-step planning, and agentic task completion. We release the model under the Apache 2.0 license to foster open-source development of capable, safe, local AI agents.
[NLP-203] Safety Kernel: Mechanistic Alignment for Cross-Lingual Defense in LLM s
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多语言环境下安全能力分布不均的问题，尤其是跨语言攻击能够绕过安全防护机制的挑战。现有对齐方法主要依赖外部对齐信号，缺乏对神经元层面内在机制的深入理解，难以构建统一的内部目标以实现跨语言的安全干预。其解决方案的关键在于从神经元层级揭示跨语言安全行为的共性机制：通过多语言平行的安全与非安全提示计算安全对比激活响应度（Safety Contrast with Activation Responsiveness, SCAR）分数，识别各语言特有的安全相关神经元；进而通过交集操作提取出跨语言共享的、紧凑的安全神经元集合——安全内核（Safety Kernel, SK）。研究进一步通过抑制激活和状态转移等干预手段验证了SK在跨语言安全行为中的功能性作用。在此基础上，提出SK-Masked LoRA方法，将低秩适配（LoRA）的参数更新限制在对应于SK的前馈网络（FFN）输出维度上，从而聚焦于跨语言共享的神经元坐标进行安全调控。实验表明，该方法在多种语言上显著降低攻击成功率，且仅需干预极小比例的神经元，同时对模型通用能力的影响有限。
链接: https://arxiv.org/abs/2610.24614
作者: Chengyu Xie,Fei Shen,Shuyi Miao,Yue Feng,Jian Yu,Pengyang Shao,Yijun Tian,Xiaoyu Du
机构: Nanjing University of Science and Technology (南京理工大学); Beihang University (北京航空航天大学); Nanjing University of Aeronautics and Astronautics (南京航空航天大学); National University of Singapore (新加坡国立大学); Xi’an Jiaotong University (西安交通大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:The safety capabilities of large language models (LLMs) are highly uneven across different languages, often allowing cross-lingual attacks to bypass safety guardrails. Existing alignment methods predominantly rely on external alignment signals and provide limited insight into underlying neuron-level mechanisms, making it difficult to provide a unified internal target for cross-lingual safety interventions. To address this, we analyze cross-lingual safety behavior at the neuron level and identify a compact set of safety-related neurons shared across languages, termed the safety kernel (SK). Specifically, we first utilize multilingual parallel safe and unsafe prompts to compute safety contrast with activation responsiveness (SCAR) scores, identifying language-wise safety-neuron candidates from internal activation patterns. We then intersect the resulting language-wise safety-neuron sets to obtain the SK, and validate its functional relevance to cross-lingual safety behavior through activation suppression and activation-state transfer interventions. Finally, building on the SK coordinates, we introduce SK-Masked LoRA, which constrains LoRA adaptation to the feed-forward network (FFN) output dimensions corresponding to the SK, thereby focusing safety-related changes on the cross-lingually shared neuron coordinates. Extensive experiments demonstrate that SK-Masked LoRA substantially reduces attack success rates across multiple languages, including languages excluded from SK discovery, while intervening on only a small fraction of internal neuron coordinates and producing generally limited changes in the model’s general capabilities.
[NLP-204] raining with Ciphers to Enhance In-Context Terminology Learning For Translation from Rare Languages
【速读】： 该论文旨在解决生成式 AI（Generative AI）在低资源语言对翻译中，通过上下文示例微调（IC-Train）时难以有效学习罕见术语翻译的问题。其核心挑战在于，标准的 IC-Train 方法虽能利用上下文示例进行适应性调整，但对罕见词的术语学习能力较弱。为此，作者提出 RareCipherFT 作为解决方案，其关键在于通过引入“加密配对”机制强化模型对罕见术语的上下文依赖学习：将每个训练目标与其包含相同罕见词的上下文示例配对，并在上下文和参考译文中将该罕见词替换为字符乱序的密文，迫使模型必须从上下文显式复制才能正确还原该词，从而避免表面模式匹配。同时，训练数据中被加密的比例采用退火策略动态调整，以平衡术语召回率与整体翻译质量。实验覆盖英-泰米尔、英-阿萨姆语、英-米佐语、英-尼泊尔语及英-豪萨语共七种翻译方向，涵盖1B至7B参数量的四种模型，结果表明 RareCipherFT 在不降低 COMET 评分的前提下，显著提升术语召回率5–15个百分点。通过同义词、伪词及共享表层上下文的控制探针验证，证明性能提升源于模型真正建立了源词与其上下文译文之间的语义关联，而非捷径学习。
链接: https://arxiv.org/abs/2610.24593
作者: Deevyanshu Malu,Deeptanshu Malu,Sunita Sarawagi
机构: IIT Bombay(印度理工学院孟买分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Fine-tuning LLMs with in-context examples (IC-Train) lets translation models adapt to new parallel data at deployment, but it is unclear whether they actually learn the translations of rare terms from those examples. We formalize this capability as in-context term learning (ICTL). We show that standard IC-Train develops this capability only weakly, and propose RareCipherFT, which strengthens it without any lexical resources. RareCipherFT pairs each training target with in-context examples that share a rare term with it. It then replaces that term with a character-shuffled cipher in both the context and the reference, so the term can only be recovered by copying from context. The ciphered fraction of training data is annealed to preserve translation quality. Across seven translation directions between English and Tamil, Assamese, Mizo, Nepali and Hausa, and four models from 1B to 7B parameters, RareCipherFT raises term recall over standard IC-Train by 5-15 absolute points without lowering COMET. Probes using synonyms, nonce words and controls for shared surface context show that these gains come from linking source terms to their in-context translations rather than from shortcuts.
[NLP-205] ASR Personalization in Pathological Voice: Accuracy Efficiency and Duration Effects
【速读】： 该论文旨在解决在临床与无障碍应用中，由于患者语音存在显著个体差异（即病理语音，Pathological Speech）而导致自动语音识别（ASR）系统性能下降的问题。在仅有少量目标说话人数据可用的情况下，传统的全模型重训练方法因数据稀缺而不可行。为此，研究提出对Whisper模型中低秩适配器（LoRA）的组件级分析，重点探究适配器位置、模型规模、训练效率及语音片段时长之间的相互作用。研究发现，尽管模型规模扩大可降低错误率，但无法完全消除语音失配问题；其中，采用全注意力结构的LoRA适配方式在准确率上表现最稳定，并实现了最优的时间-精度权衡。此外，不同长度语音的表现差异明显：中等长度语音受益最为显著，而短时和长时语音仍面临较大挑战。因此，该研究的关键在于揭示适配器部署位置的重要性以及在数据稀缺场景下需进行时长依赖性评估，为个性化语音识别提供了高效且可靠的解决方案。
链接: https://arxiv.org/abs/2610.24572
作者: Giuseppe Giaimis,Sergei Katkov,Muhammad Azfar Yaqub,Antonio Liotta
机构: Free University of Bolzano (博洛尼亚自由大学)
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
备注: Presented at SPECOM 2026
Abstract:Automatic Speech recognition (ASR) struggles with Pathological speech due to persistent speaker-specific mismatch. In clinical and accessibility scenarios, only limited target-speaker data is typically available, making full model retraining impractical. We present a component-level analysis of LoRA adaptation in Whisper, investigating how adapter placement interacts with the model size, training efficiency and utterance-duration in single-speaker settings. Our experimental comparative study shows that while scaling reduces errors, it does not eliminate mismatch. Furthermore, the full attention LoRa provides the most reliable accuracy, and the best time–accuracy trade-off. Improvements vary across utterance length, with medium-length recordings benefiting the most, while short and long utterances remain challenging. These findings highlight the importance of adaptation placement and duration-dependent evaluation, under challenging, data-scarce personalization scenarios.
[NLP-206] From Stories to Features: Scaling Concept Vectors for AI Alignment Interpretability
【速读】： 该论文旨在解决语言模型中可解释概念的识别与表征问题，特别是针对人工智能安全与对齐（AI safety and alignment）相关概念的可解释性建模难题。现有无监督方法如稀疏自编码器（sparse autoencoders）虽试图恢复完整的特征空间，但往往在特征质量与人类可解释性方面存在不足。本文提出一种更务实的解决方案：聚焦于与对齐相关的具体概念，通过生成合成的“概念-反概念”故事对（concept-antagonist story pairs），利用模型在两类故事中的平均激活差异作为特征向量（feature vector）。该方法被扩展至1,036个对齐相关概念，基于200万条跨语言、跨体裁的合成故事数据，并公开发布。实验表明，该特征向量在六种不同模型（来自三个模型家族）中均能可靠地检测对话与网页文本中的目标概念，展现出跨模型共享的可解释几何结构，并可通过注入激活实现对模型行为的因果调控。此外，该方法被用于分析模型在越狱攻击、数学推理及代理式推理过程中的内部表示。因此，其关键在于通过可控的合成数据构建目标概念的对抗性激活差异，实现了对语言模型中特定概念的高效、可解释且可扩展的表征，为大规模概念研究提供了可行路径，未来或可拓展至数百万个任意领域的特征。
链接: https://arxiv.org/abs/2610.24556
作者: Dmitry Scherbakov,Oleg Chernikov,Victoria Khamash,Kirill Afendulev,Alexey Dontsov,Ivan Oseledets,Elena Tutubalina,Anton Korznikov
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 51 pages, 41 figures, 26 tables. Code: this https URL
Abstract:We study feature vectors: directions in a language model’s activation space that correspond to human-interpretable concepts, such as honesty or deception. Existing unsupervised methods like sparse autoencoders attempt to recover the whole feature space, often at the cost of feature quality and interpretability. We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector. We scale this approach to 1,036 alignment-relevant concepts using 2M synthetic stories across multiple languages and genres, which we publicly release. Across six models from three model families, our feature vectors reliably generalize to detecting their corresponding concepts in chat dialogues and web text. They form an interpretable geometry shared across models and can causally steer model behavior when injected into activations. We further use these features to analyze model representations during jailbreak attacks, mathematical and agentic reasoning. Our targeted feature construction thus offers a scalable route to studying concepts in language models, potentially extending to millions of features across arbitrary domains.
[NLP-207] EGIP: An Evidence-Grounded Interaction Protocol for Clinical Agents
【速读】： 该论文旨在解决临床智能体（clinical agents）在诊断过程中因交互环境信息暴露不明确或控制不一致，导致诊断准确率混淆于交互协议本身影响的问题。其核心挑战在于：当证据获取与最终决策耦合时，模型表现可能受交互设计偏差的干扰，难以区分真实推理能力与协议依赖性。为此，论文提出证据锚定交互协议（Evidence-Grounded Interaction Protocol, EGIP），其关键在于通过四重机制实现证据获取与决策分离：（1）设定角色特定的信息边界以控制信息可见性；（2）暴露诊断动作但不揭示其执行结果，避免结果泄露带来的偏差；（3）从存储的临床记录中检索检查结果，确保数据真实性与可追溯性；（4）结构化组织所获证据以支持最终决策。通过将证据采集过程与终端预测解耦，EGIP使得不同决策策略可在相同观测基础上运行，从而提升评估的公平性与可复现性。实验在两个临床代理框架、五个数据集及六种语言模型上验证表明，EGIP在所有主要对比中均显著提升诊断准确率；临床专家评审进一步显示，经自动评分判定为正确的响应中，被联合判断为有证据支持的比例从10.20%提升至36.55%，充分证明了EGIP作为统一交互协议在规范证据访问与采集方面的有效性。
链接: https://arxiv.org/abs/2610.24539
作者: Zecheng Ren,Ruichen Cong,Yi Min Chen,Jing Zhang,Feiyang Chen,Wutong Zhang,Kaixiang Wang,Zihan Wang,Jiong Lou,Jie LI,Qun Jin
机构: Waseda University (早稻田大学); Shanghai Jiao Tong University (上海交通大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Interactive clinical agents acquire evidence through patient dialogue and diagnostic examinations, yet their performance also depends on how the interaction environment exposes information, available actions, and examination results. When these conditions are implicit or inconsistently controlled, final-answer accuracy can conflate diagnostic reasoning with the effects of the interaction protocol. We propose the \textbfEvidence-\textbfGrounded \textbfInteraction \textbfProtocol (EGIP), which enforces role-specific information boundaries, exposes diagnostic actions without revealing their results, retrieves examination results from stored clinical records, and organizes acquired evidence for final decision-making. By separating evidence acquisition from terminal prediction, EGIP enables different decision strategies to operate on the same observations. We evaluate EGIP across two clinical-agent frameworks, five datasets, and six language models. EGIP improves diagnostic accuracy over native baselines across all primary comparisons. A clinician review further shows that the proportion of automatically scored correct responses jointly judged evidence-supported rises from 10.20% to 36.55%. These findings support the effectiveness of EGIP as a unified interaction protocol for improving clinical-agent performance through controlled evidence access and acquisition.
[NLP-208] Axiomatic World Modeling for Physics Reasoning NEURIPS2026
【速读】： 该论文旨在解决大语言模型（LLM）在物理推理任务中存在的一种关键缺陷：即模型在推理过程中出现“建模漂移”（modeling drift），表现为在未明确说明的情况下擅自修改题设条件、虚构物理定律或引入隐含假设，从而在自洽但错误的内部世界模型（axiomatic world model, AWM）中进行推理，而非基于题目所描述的真实物理世界。这一问题的根本原因在于现有方法如基于强化学习的验证与反馈（RLVR）和物理推理模型（PRM）虽在数学和代码生成任务中表现优异，但其奖励机制未能直接监督模型在公理化领域（如物理学）中的世界建模能力。为此，论文提出一种新型训练框架——基于公理化世界建模的强化学习（Reinforcement Learning with Axiomatic World Modeling, RLAWM），其核心在于通过双层级奖励机制对模型进行优化：一是建模奖励，评估模型是否准确捕获问题中的真实公理化世界；二是求解奖励，检验模型在该世界中的具体推导过程是否保持一致性。该框架在训练阶段通过显式惩罚模型与真实物理世界之间的偏差，实现对世界建模能力的直接优化，且不增加推理阶段的计算开销。实验表明，RLAWM在PhysReason和PHYSICS基准上分别提升12.2和5.2分，且在难度、来源和领域跨度上均展现出显著的零样本泛化能力，在奥数级别物理与数学任务中相较最强基线实现2.45倍性能增益并提升10分。
链接: https://arxiv.org/abs/2610.24519
作者: Xinye Yang,Zhenyang Liu,Yuxuan Wang,Yuanyuan Lei
机构: Northwestern University (西北大学); University of Florida (佛罗里达大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to NeurIPS 2026
Abstract:Recent research-level physics-reasoning benchmarks for LLMs reveal a failure mode that lies not in knowledge or execution, but in modeling the world being reasoned about. Models often silently alter given quantities, fabricate laws, or inject unstated assumptions to force a problem into familiar patterns, and then reason inside this corrupted world. We define this premise-level divergence as modeling drift: the gap between a model’s internal axiomatic world model (AWM) and the true physical world specified in the problem. Although RLVR and PRM have yielded substantial gains in mathematics and code, their reward shapes do not directly supervise world modeling in axiomatic domains such as physics. We propose Reinforcement Learning with Axiomatic World Modeling (RLAWM), which optimizes the policy by penalizing divergence between its inferred axiomatic world model and the true physics world. RLAWM probes physics reasoning at two hierarchical levels (modeling and solving), each scored by a physically grounded reward. The modeling reward targets abstract knowledge formulation, evaluating whether the model captures the true axiomatic world before attempting calculation. The solving reward targets instantiated knowledge reasoning, checking whether the model’s concrete derivation remains consistent with the axiomatic world. This two-level structure separates what world the model believes it is reasoning in from how it reasons within that world. The framework operates exclusively during training, with no extra cost during inference. RLAWM improves performance by up to +12.2 points on PhysReason and +5.2 on PHYSICS over the baseline. It also generalizes zero-shot across difficulty, source, and domain, with a 2.45x gain and +10 points over the strongest baseline on olympiad-level physics and math.
[NLP-209] ProbePO: Sparse Probes Redistributed Credit for Critic-Free LLM Reinforcement Learning NEURIPS2026
【速读】： 该论文旨在解决强化学习中基于可验证奖励（verifiable rewards）时，传统方法仅在完整推理轨迹末端赋予单一奖励，导致难以追溯各推理步骤对最终结果的贡献问题。其核心挑战在于如何在有限的采样预算下，精准识别值得评估的部分解状态，并合理分配信用以提升策略优化效率。解决方案的关键在于提出ProbePO方法，通过无奖励的不确定性评分筛选候选部分解状态，并在受限的延续预算内估计其预期验证器奖励，从而实现对GRPO（Group-Relative Policy Optimization）所分配的轨迹级优势进行细粒度重分配，而非替换；该过程保持了每条轨迹的平均词级别优势不变，且探测延续仅作为测量手段，不改变用于策略更新的数据分布。实验表明，在三个涵盖1.5B至7B参数的DeepSeekMath与Qwen系列模型上，ProbePO在18次模型-基准测试中的16次取得最高准确率，显著优于GRPO，并在与密集探测型VinePPO的对比中，于DeepSeekMath-7B上所有六项基准均表现更优，同时训练时间减少76%。
链接: https://arxiv.org/abs/2610.24515
作者: Diyi Hu,Ruizhong Qiu,Lin Qiu,Hanghang Tong,Hanqing Zeng
机构: Independent Researcher(独立研究员); University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Meta(Meta)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 22 pages, 6 figures, 9 tables. NeurIPS 2026 Workshop on Transitioning from Pre-Training to Post-Training
Abstract:Reinforcement learning from verifiable rewards assigns one terminal reward to an entire reasoning trajectory, obscuring how individual reasoning steps contributed to the outcome. GRPO broadcasts a group-normalized advantage to every response token, while finer-grained approaches often rely on dense auxiliary rollouts or branching schemes that can also alter the data used for policy updates. We cast the missing middle as budgeted measurement: with a limited rollout budget, which partial solutions should be evaluated, and how should their value estimates refine credit? ProbePO selects partial-solution states with a reward-free uncertainty score and estimates their expected verifier rewards under a capped continuation budget. ProbePO redistributes, rather than replaces, GRPO’s trajectory-level advantage across tokens while preserving each trajectory’s mean token-level advantage; probe continuations serve only as measurements. Across three DeepSeekMath and Qwen-family models spanning 1.5B-7B parameters, ProbePO consistently outperforms GRPO and achieves the highest accuracy among the four evaluated methods in 16 of 18 model-benchmark comparisons. Against dense-probing VinePPO on DeepSeekMath-7B, ProbePO achieves higher accuracy on all six benchmarks with 76% less end-to-end training time.
[NLP-210] Backend-Frontend Converter: Component-Based Conversion of .NET Computational Models to TypeScript
【速读】： 该论文旨在解决在基于.NET开发的计算模型中，如何高效实现其在前端浏览器环境中的运行问题，特别是将原本仅支持后端（C#）执行的组件化计算模型迁移至TypeScript前端应用的需求。其核心挑战在于保持跨平台（C#/TypeScript）计算结果的一致性与可验证性。解决方案的关键在于提出一种名为“Backend–Frontend Converter”的架构，该架构通过将组件、连接关系、别名及公式树等抽象结构映射为特定模型的TypeScript代码，实现了从. NET模型到前端应用的自动化转换；同时，借助可复用的运行时库（包括数值积分、数据访问和状态管理功能），确保生成的TypeScript代码具备与原生C#实现相当的计算精度与行为一致性。实验验证表明，在四种轨道场景下，生成的前端计算结果与后端输出在时间戳对齐条件下，位置与速度误差分别控制在10⁻⁶ km和10⁻⁹ km/s以内，且所有20项分析调度测试均通过，扰动检测亦成功识别出预设的状态与时间偏差；在交易策略对比中，501条对齐的每日记录在指标值、持仓及订单事件上完全一致，证明了跨语言边界输出的保真性。因此，组件级生成与可复用执行库的结合，构成了从受支持的.NET计算模型向TypeScript应用转化的实用路径。
链接: https://arxiv.org/abs/2610.24492
作者: Petr Ivankov(1),Max Shestov(2) ((1) Independent researcher, (2) Georgia Institute of Technology)
机构: Independent Researcher(独立研究员); Georgia Institute of Technology (佐治亚理工学院)
类目: Computation and Language (cs.CL); Software Engineering (cs.SE)
备注:
Abstract:Computational models developed in .NET often need to run locally within web applications. We present Backend–Frontend Converter, an architecture for generating TypeScript implementations of component-based models. The converter maps components, connections, aliases, and formula trees to model-specific TypeScript code. Reusable runtime libraries provide numerical integration, data access, and state handling. We evaluate stored generated implementations through four orbital scenarios and a recorded trading-strategy comparison. Orbital calculations execute in a browser and are paired with C# results by crossing event and timestamp; trading outputs are aligned by date. Analytic scheduler tests and deliberately perturbed outputs assess the validation procedure. Each orbital scenario produces 91 paired events with identical timestamps. Position and velocity differences remain below (10^-6)~km and (10^-9)~km/s . All 20 analytic scheduler checks pass, and the comparator detects each of three prescribed state and time perturbations. The trading comparison contains 501 aligned daily records with matching indicator values, positions, and order events. The evaluated implementations preserve the compared outputs across the C#/TypeScript boundary. Component-specific generation and reusable execution libraries provide a practical route from supported .NET computational models to TypeScript applications.
[NLP-211] Vocal-Tract Acoustics Aerodynamics and Overlap-and-Add Synthesis in DYNARTmo
【速读】： 该论文旨在解决语音生成过程中多模态物理机制耦合建模的难题，即如何在动态构音模型（dynamic articulatory model）中实现声学与气动特性的精确映射，以支持语音产生的可视化与合成。其核心解决方案在于构建一个基于构音动作轨迹的可解释性语音生产系统——DYNARTmo，通过将构音器运动轨迹转化为具有鼻腔分支和有效侧向通道的声管结构，并引入集总气动网络，将声门运动与肺内压等生理参数关联至气流与喉上压力分布；同时，独立建模发声激励、口腔摩擦噪声及声门漏气成分，并采用重叠-相加法（overlap-and-add）保留各帧响应尾部，从而实现跨音素（元音、鼻音、边音、擦音、塞音）的连续声学输出。该方法的关键在于实现了构音控制、气动动力学与声学输出之间的可逆、可解释的物理映射，为早期语言习得的神经模拟提供了具备生理合理性的可调控语音生成平台。
链接: https://arxiv.org/abs/2610.24490
作者: Bernd J. Kröger
机构: RWTH Aachen University (亚琛工业大学); Kröger Lab (比利时)
类目: Computation and Language (cs.CL)
备注: 6 figures; ancillary Python reference implementation and word-transcription guide. No audio recordings or saved timelines
Abstract:This paper presents the acoustic and aerodynamic extension of DYNARTmo, a dynamic articulatory model for visualizing and synthesizing speech production. Gestural control trajectories determine vocal-tract geometry, which is converted into an acoustic tube with nasal branching and effective lateral passages. A lumped aerodynamic network relates prescribed glottal motion and pulmonary pressure to airflow and supraglottal pressure storage. Phonatory excitation, oral friction, and glottal aspiration are synthesized as separate contributions. Frame-specific acoustic responses are combined by overlap-and-add, preserving response tails beyond individual control frames. Three German words, Tasse'', Muse’‘, and ``Lade’', illustrate the coordination of articulation, pressure, flow, and acoustic sources across vowels, nasals, laterals, fricatives, and plosives. The implementation has been optimized by the author primarily for sound recognizability; independent perceptual evaluation remains outstanding, and fricative-noise naturalness requires further refinement. The model provides an inspectable, articulatorily controlled production plant for future neural simulations of babbling, imitation, and early word acquisition. Further work will improve acoustic realism, reduce empirical source corrections, and extend the sensory information available to developmental learning models.
[NLP-212] Modeling Discourse Profiles in Climate Misinformation Discourse
【速读】： 该论文旨在解决现有气候虚假信息检测方法仅依赖文本内容进行个体评论或声明分类，而忽视了在线气候话语中虚假信息通过重复性修辞模式在评论与话题链中被生产、强化和持续传播这一关键特征的问题。其解决方案的核心在于：基于大规模TikTok气候话语数据集，通过聚类技术构建反映典型修辞模式的“话语画像”（discourse profiles），并将这些画像作为额外语境信息，通过后融合（late-fusion）架构融入细粒度的气候虚假信息分类模型中。实验表明，在视频不重叠的评估设置下，引入话语画像的模型在词袋、TF-IDF及MPNet等文本表征上均显著提升宏观F1分数；尽管RoBERTa-large表现出正向平均改进但波动较大，差异未达统计显著性。此外，研究还发现话语画像不仅影响分类性能，更深刻塑造虚假信息的接受与传播机制——基于话语画像条件化的语言模型代理展现出与话语画像层面虚假信息流行度一致的接受行为，且不同主导话语画像下虚假信息流行度与用户参与持续性之间的关系存在显著差异。
链接: https://arxiv.org/abs/2610.24489
作者: Ge Gao,Zhengyang Shan,Bryan A. Plummer,Margrit Betke,Derry Wijaya
机构: Boston University (波士顿大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Prior work on climate misinformation detection typically classifies individual comments or claims using text alone. This framing overlooks a key property of online climate discourse: misinformation is often produced, reinforced, and sustained through recurring rhetorical patterns that reappear across comments and threads. Using a large-scale TikTok climate discourse dataset, we construct discourse profiles by clustering climate-related comments into recurring rhetorical patterns. We then incorporate these profiles into fine-grained climate misinformation classifiers using a late-fusion architecture. Under video-disjoint evaluation across three training seeds, discourse-profile-aware models significantly improve macro-F1 over text-only baselines for bag-of-words, TF-IDF, and MPNet representations. RoBERTa-large exhibits a positive mean improvement but substantially greater variability, and the difference is not statistically significant. Beyond classification, our results show discourse profiles shape misinformation reception and propagation. Discourse-profile-conditioned language model agents exhibit misinformation acceptance patterns aligned with profile-level misinformation prevalence in ClimateDisc, and the relationship between misinformation prevalence and engagement persistence differs across dominant discourse profiles.
[NLP-213] Write Once Check Many: Reusable Ledgers for Post-Hoc Policy Compliance Auditing of Agent Trajectories
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在安全审计中面临的效率瓶颈问题：传统方法需针对每条策略（policy）重新读取完整的智能体轨迹（trajectory），导致计算开销巨大，尤其当策略数量众多或未知时难以扩展。其核心解决方案是提出“写一次、查多次”（write once, check many）框架，通过一个写入模型（writer model）将每条智能体轨迹转化为与具体策略无关的、可复用的叙事型账本（narrative ledger），其中记录智能体的动作与结果。后续任何新策略均可仅基于该账本进行验证，无需访问原始轨迹。实验表明，在SHADE-Arena、\tau-bench和\tau^2-bench共900条轨迹上，一个120亿参数的读者模型（reader model）利用账本进行策略检查，性能达到甚至超过直接读取完整轨迹的相同规模模型；在组合多事实策略的SHADE-Arena任务中，该方法优于读取原始轨迹的310亿参数模型（AUC 0.777 vs. 0.713）。此外，账本在使用8个策略独立验证时，整体token消耗减少约5倍。分析还揭示，规则提取在单动作策略中表现更优，而账本可能因摘要机制遗漏长工具输出或消息内容中的关键证据。研究结果表明，构建可复用、策略无关的智能体行为记录能显著提升事后审计的效率与有效性，并对账本设计中的信息保留与结构优化提出了重要启示。
链接: https://arxiv.org/abs/2610.24483
作者: Ashwin Nagarajan,Gursimran Panesar,Alicia Yang,Mona T. Diab,Virginia Smith,Aashiq Muhamed
机构: Carnegie Mellon University (卡内基梅隆大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Auditing LLM agents against policies is critical to ensure safety but requires rereading long trajectories once per policy. Can we instead record each trajectory once and reuse that record for policies not known in advance? We propose write once, check many, a framework in which a writer model converts each agent trajectory into a policy-independent ledger of the agent’s actions and outcomes. A reader then checks each new policy against the ledger alone, without access to the original trajectory. On 900 trajectories from SHADE-Arena, \tau -bench, and \tau^2 -bench, we show that a 12B reader of narrative ledgers can either match or in fact achieve higher AUC than the same 12B model reading full trajectories. On SHADE-Arena policies that combine several facts, our approach outperforms a 31B model reading the trajectory (.777 vs. .713 AUC). When checking eight SHADE-Arena policies separately, ledgers use about 5\times fewer tokens, including the cost of writing them. A deeper analysis shows that rule-based extraction performs better on some single-action policies, and ledgers can miss evidence in long tool outputs and message contents. Our results suggest that reusable, policy-independent records can make retrospective agent auditing more efficient and effective, and highlight important considerations in ledger design.
[NLP-214] Beyond Recall Saturation: Scaling Hard-Negative Reranker Training for Industrial Legal Retrieval
【速读】： 该论文旨在解决工业级检索系统在提升检索质量时面临的效率瓶颈问题，即单纯扩大嵌入候选池或采用复杂训练流程所带来的边际收益递减现象。其核心挑战在于：当候选池规模达到一定水平（如中文法律检索任务中的前300名）时，嵌入阶段的召回率已趋于饱和，但相关证据仍可能被错误地排在输出截断阈值之下，导致最终检索效果受限。因此，论文聚焦于重排序器（reranker）的质量优化，提出应将检索与重排序两个阶段分开评估，以更精准地识别优化重点。解决方案的关键在于设计并验证一种从嵌入挖掘（embedding-mined）到自挖掘（self-mined）的分阶段训练策略，该策略在统一的检索协议下表现优异，且在40亿参数规模的重排序模型上实现了显著性能提升；相比之下，引入动态负样本（dynamic-negative）等额外阶段并未带来一致性的指标改进。实验表明，该优化后的4B重排序模型在多个基准测试及内部真实查询场景中均优于原生4B模型，且在72.4%的非平局对比中获得偏好，证明了该分阶段训练范式在保持低延迟前提下的有效性。
链接: https://arxiv.org/abs/2610.24478
作者: Yunhan Li,Mingjie Xie,Zeyang Shi,Gengshen Wu,Min Yang
机构: Deli Joint Laboratory(德利联合实验室); City University of Macau(澳门城市大学)
类目: Computation and Language (cs.CL)
备注: 11 pages,6 tables, 4 figures
Abstract:Industrial retrieval systems often improve quality by enlarging embedding candidate pools or adopting complex training pipelines. We show that, under a sufficiently large but practical candidate pool, embedding recall exhibits diminishing returns, motivating a closer examination of reranker quality. On Chinese legal retrieval benchmarks, embedding top-300 reaches a high-recall operating region, while relevant evidence may still be ranked below the output cutoff. We compare hard-negative strategies for reranker training, including embedding-mined, self-mined, and dynamic-negative variants, under aligned retrieval protocols that evaluate reranking after candidate retrieval. Among the evaluated staged configurations, an embedding-to-self-mined pipeline provides a favorable trade-off; additional dynamic stages do not improve all held-out metrics. The selected pipeline scales from a 0.6B to a 4B reranker. Relative to the native 4B reranker, the trained 4B Stage2 model improves reported metrics on the STARD and CSAID benchmarks and on an internal development set, and is preferred in 72.4% of non-tied comparisons on 410 de-identified, reconstructed internal operational-query inputs in a latency-sensitive offline preference evaluation. These results support evaluating the retrieval and reranking stages separately when selecting an optimization target for industrial retrieval systems.
[NLP-215] Ben-Misandry: A Bengali Dataset and Late-Fusion Transformer Framework for Detecting Anti-Male Gender Hate
【速读】： 该论文旨在解决孟加拉语（Bengali）中针对男性仇恨言论（misandry）长期被忽视的问题，现有研究几乎完全聚焦于厌女（misogyny），导致内容审核系统在识别针对女性的攻击时有效，却对针对男性的仇恨言论缺乏检测能力，形成系统性盲区。其核心解决方案是构建首个孟加拉语反男性仇恨言论检测数据集——Ben-Misandry，包含5,000条评论，涵盖六个分类标签，其中引入一个新颖类别：讽刺男性情感脆弱性，该类别在既有厌女分类体系中无直接对应项。数据通过GPT-4.1生成并经四名人工评审者对500条评论样本进行抽样校验，确认多数标签准确性；全量标注一致性分析留待未来工作。研究采用经典模型（七种分类器搭配四种文本表示）与基于变压器架构的模型（BanglaBERT、XLM-RoBERTa及两种融合策略）进行基准测试，结果显示，基于字符级TF-IDF的线性支持向量机（SVM）达到91.8%准确率，而结合BanglaBERT与XLM-RoBERTa软标签输出的晚期融合集成模型取得96.6%准确率及超过99%的宏平均ROC-AUC，显著优于单一模型与早期融合方案。研究从分类器组合理论角度解释该性能提升，指出误差不相关的编码器平均可降低方差，并通过Shapley值解释验证了模型决策边界与探索性分析中发现的关键词汇一致，但亦警示模型可能部分依赖合成数据生成特征。研究结果表明，基于该合成语料库，孟加拉语反男性仇恨言论检测在技术上可行且可量化，该数据集作为初始公开资源，有待后续真实世界数据验证。
链接: https://arxiv.org/abs/2610.24474
作者: Md Mehedi Hasan Nipu,Chinmoy Mitra,Tarannum Ahmed Nowshin,Israt Moyeen Noumi
机构: North South University (北方南大学); Rajshahi University of Engineering and Technology (拉杰沙希工程与技术大学); BRAC University (BRAC大学); Ahsanullah University of Science and Technology (阿桑纳大学科学技术学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Gender based hate speech research in Bengali has focused almost exclusively on misogyny, leaving anti male hate, or misandry, entirely unstudied for this language and creating a systematic blind spot in moderation systems trained to catch attacks on women but not on men. We introduce Ben-Misandry, the first Bengali misandry detection dataset: 5,000 comments across six taxonomy classes, including a novel category for mocking male emotional vulnerability with no direct analogue in existing misogyny taxonomies. The corpus is taxonomy-controlled synthetic data generated with GPT-4.1, spot-checked by four human reviewers on a 500-comment sample who confirmed the large majority of assigned labels; a full inter-annotator- agreement study is left to future work. We benchmark it with a classical pipeline (seven classifiers over four text representations) and a transformer pipeline (BanglaBERT, XLM-RoBERTa, and two fusion strategies). The best classical model, a linear SVM over character-level TF-IDF, reaches 91.8% accuracy; a late fusion ensemble averaging BanglaBERT and XLM-RoBERTa softmax outputs reaches 96.6% accuracy and macro ROC-AUC above 99%, nominally ahead of every single model and early fusion, though we have not run significance testing to confirm these margins are meaningful. We relate this to classifier combination theory, under which averaging encoders with decorrelated errors is expected to reduce variance, and corroborate the decision boundary with Shapley explanations that recover the same class- discriminative vocabulary found in our exploratory analysis, while noting this cannot rule out the model partly keying on synthetic-generation artifacts. Our results indicate misandry detection in Bengali is feasible and measurable on this synthetic corpus, offered as an initial public resource pending validation on real-world data.
[NLP-216] Pluralistic Alignment in Language Models: Objectives Human References and Preference Representations
【速读】： 该论文旨在解决生成式AI在对齐人类偏好过程中存在的核心问题：现有研究常将三个本质上不同的问题混淆，即模型应表现出何种多元行为、以谁的偏好作为参照标准，以及证明成功所需的证据强度。为此，作者提出一个两层分类体系（two-layer taxonomy），第一层区分三种多元对齐范式：过托尼多元性（Overton Pluralism）——表征广泛支持的人类观点多样性；可调节多元性（Steerable Pluralism）——忠实适应特定视角或偏好条件；分布多元性（Distributional Pluralism）——匹配目标群体响应分布。第二层则区分人类参照对象的层级：总体人群、国家与地区、子群体。此外，偏好表征方式、实现机制、模态及证据质量被识别为跨层次的设计选择，而非独立的多元形式。通过对70项相关研究进行系统编码，构建了协议限定的语料库，并利用“目标-人类参照”矩阵映射现有方法，揭示出当前研究高度集中于可调节多元性和分布多元性，而真正符合过托尼多元性的研究仅在总体人群层面存在。论文进一步将每种目标与其所需验证证据相连接，包括视角有效性、条件保真度、分布保真度、支持保留性及子群体估计不确定性等指标。整体而言，该分类体系与评估框架不仅厘清了现有方法的实际能力边界，也指出了其主张超出证据支撑的薄弱环节，为未来系统在真实表达分歧的同时避免将随机变异误作忠实表征提供了理论基础与实践路径。
链接: https://arxiv.org/abs/2610.24471
作者: Meng-Chen Wu,Umesh Bodhwani
机构: Amazon Science(亚马逊科学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Pluralistic alignment aims to represent meaningful variation in human preferences rather than collapse heterogeneous judgments into a single target. Yet existing work often mixes three distinct questions: what pluralistic behavior a model should exhibit, whose preferences define the reference, and what evidence is sufficient to demonstrate success. We organize the literature around a two-layer taxonomy. Layer 1 distinguishes representing a range of human-supported viewpoints (Overton Pluralism), adapting faithfully to a specified perspective or preference condition (Steerable Pluralism), and matching the response distribution observed in a target population (Distributional Pluralism). Layer 2 distinguishes aggregate populations, countries and regions, and subgroups as human references. Preference representation, implementation mechanism, modality, and evidence quality are treated as cross-cutting design choices rather than additional forms of pluralism. We synthesize a protocol-bounded corpus of 70 works on pluralistic alignment and neighboring areas, using the objective-by-human-reference matrix to map existing approaches and identify underexplored combinations. The coded literature concentrates on Steerable Pluralism and Distributional Pluralism, while eligible Overton Pluralism work appears only at the aggregate-population level under our stated protocol. We then connect each objective to the evidence needed to support it, including perspective validity, conditional fidelity, distributional fidelity, support preservation, and uncertainty in subgroup estimates. Together, the taxonomy and evaluation framework clarify what current methods establish, where their claims exceed their evidence, and how future systems can represent disagreement without mistaking variation for faithful representation.
[NLP-217] Which Long Documents Matter? A Controlled Study of Data Selection for Context Extension
【速读】： 该论文旨在解决自然长文档（natural long documents）在大模型训练中有效选择策略不明确的问题，尤其是在有限的持续训练预算下如何提升上下文扩展效果。其核心挑战在于：现有文档选择方法（如重打包、通用质量评估或LLM评分器）在长文本评估长度下表现不佳，难以有效捕捉关键信息密度。解决方案的关键是提出一种基于长文本概率（LongPPL）关键标记密度的高效筛选机制——truncLongPPL，通过仅对参考跨度的前半部分进行截断计算，即可近似全跨度评分，显著降低评分开销。该方法在128K上下文长度下实现2.28倍的端到端评分加速，同时保持接近全跨度评分的下游性能，在验证集HELMET和保留测试集RULER上分别较随机采样长文档（random-long）提升4.54和3.28分，且在检索任务中表现最为稳定；在256K上下文及多模型混合训练场景中亦持续优于基线。经监督微调（SFT）后，使用该方法训练的Qwen3-8B与OLMo~3-7B模型在LongBench v2和MRCR等基准上均显著超越随机采样基线，证明了其在实际训练中的有效性与普适性。
链接: https://arxiv.org/abs/2610.24469
作者: Junlong Jia,Jiang Zhou,Xing Wu,Chaochen Gao Tinghao Yu,Feng Zhang,Lilin Wang
机构: Tencent(腾讯)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Work in progress
Abstract:Natural long documents are widely used for context extension, but which selection pipelines are effective remains unclear. We conduct a controlled comparison on Qwen2.5-3B at 128K under matched 6B-token continued-training budgets. Random sampling of natural long documents (random-long) outperforms repacking at longer evaluation lengths. Within the same candidate pool, selecting documents with high LongPPL key-token density yields further gains, unlike the evaluated generic-quality and LLM-judge selectors. We introduce truncLongPPL to reduce scoring cost by computing key-token density over a truncated prefix. At 128K, scoring the first half of the reference span yields downstream performance close to full-span scoring with a measured 2.28\times end-to-end scoring speedup. Across three training seeds, truncLongPPL achieves the highest mean scores among evaluated pipelines, exceeding random-long by 4.54 points on validation HELMET and 3.28 on held-out RULER, with gains most consistent in retrieval. At a 256K target context, it gains 2.82 average RULER points over random-long on Qwen2.5-3B. In short/long mixed training across Qwen2.5-3B, Qwen3-8B, and OLMo~3-7B, it gains 1.47–7.31 average RULER points over random-long mixing, with seven-task short-context averages within 0.23 points of the corresponding baselines. After supervised fine-tuning (SFT), Qwen3-8B and OLMo~3-7B models trained with selected long documents outperform their random-long counterparts by 1.13–1.67 points on LongBench v2 and 0.62–1.44 points on MRCR.
[NLP-218] When Opposite Premises Translate Identically: A Source-Conditioned Diagnostic for Multilingual Commonsense Items
【速读】： 该论文旨在解决多语言推理评估中因翻译失真导致的基准可靠性问题，即如何确保翻译后的前提在目标语言中仍能准确保留其源语言（英语）中的语义关系与逻辑结构。其解决方案的关键在于提出并应用一种“源对比诊断法”（source-contrast diagnostic）：通过反转英语原文的极性，对正反两个语境独立进行翻译，并统计目标语言中产生相同输出的匹配对（即“精确碰撞”），进而分别以各自源文本为参照对目标翻译进行评分。该方法能够识别出至少存在不忠实翻译的情况，但无法定位具体是哪一端翻译失真。研究在71个问题实例、两种包含项深度、五种目标语言及三个翻译模型检查点下展开，结果显示斯瓦希里语在各检查点中分别出现63、113和67次精确碰撞，表明翻译忠实度存在显著差异。尽管该方法可作为检测翻译偏差的控制性测量手段，但其局限性在于未涵盖自发性翻译错误的筛查精度，也未验证跨语言推理过程中的“语言-跃迁效应”，且缺乏母语等价标签与答案模型的评估。
链接: https://arxiv.org/abs/2610.24428
作者: Shuai Wang,Rui Hu
机构: Zhejiang University (浙江大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Multilingual reasoning comparisons depend on translated premises preserving the relations in their English sources. Automatic translation scores can rank source-target pairs, but an acceptance decision alone cannot certify a counterfactual benchmark field. We test a source-contrast diagnostic for ACCORD premises: reverse the English polarity, translate both arms independently, count identical target outputs, and score each arm against its own source. Across 71 matched question identities at two containing-item depths, five target languages, and three translation checkpoints, Swahili has 63, 113, and 67 exact collisions among 710 pairs per checkpoint. At a human-reference-calibrated threshold, round-trip NLI passes both arms of 3 of the 113 NLLB-3.3B collisions, while LaBSE passes 78. Each collision proves at least one unfaithful translation, but does not identify its arm. These controlled measurements do not establish screening accuracy on spontaneous translation errors or a language-by-hop reasoning effect; the study has no native equivalence labels or answer-model evaluation.
[NLP-219] oward a Universal Local Speech Feature Extractor Through Distillation
【速读】： 该论文旨在解决自监督语音模型中卷积神经网络（CNN）层计算开销大且特征提取能力与简单梅尔频谱图（Mel-spectrogram）相似的问题。其核心挑战在于，尽管CNN层在当前主流模型（如HuBERT、data2vec、WavLM）中广泛使用，但其复杂结构带来的高计算成本并未显著提升性能，且不同模型间的差异较小。为此，作者提出一种基于知识蒸馏的轻量化解决方案：通过从多个教师模型中学习，构建一个仅由两层组成的通用特征提取器，以替代传统的深度CNN结构。该方案的关键在于利用知识蒸馏技术，将教师模型中的高级语义信息高效迁移至轻量级学生模型，从而在大幅降低计算量（仅需原CNN计算量的3%）的同时，保持接近甚至媲美教师模型的性能表现，并具备良好的跨模型泛化能力（可应用于未见过的wav2vec 2.0模型）。实验表明，该方法在SUPERB(-SG)基准任务上实现了约20%的整体推理时间减少，验证了其高效性与普适性。
链接: https://arxiv.org/abs/2610.24424
作者: Jessica Yang,Marcelo Beramendi Caballero,Ju-Chieh Chou,Karen Livescu
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Self-supervised speech models commonly consist of a set of convolutional neural network (CNN) layers used as a local feature extractor, followed by transformer layers. The CNN layers are computationally demanding, and are also similar across models and resemble the much simpler mel spectral features. We hypothesize that the CNN layers can be replaced by a simple, universal model. To test this hypothesis, we learn a two-layer feature extractor via distillation from one or more teacher models (HuBERT, data2vec, and WavLM). Experiments on SUPERB(-SG) benchmark tasks show that our model largely retains the performance of the teacher models and can also be used in an unseen model (wav2vec 2.0). Our feature extractor requires only 3% of the MACs of the CNNs, resulting in a 20% overall inference runtime reduction.
[NLP-220] Ask Less Ask Better: Allocating User Attention in Proactive LLM Agents with Interruption Budgets
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在执行任务时频繁提出澄清问题所带来的用户负担问题。现有评估方法仅通过统计提问次数来衡量用户负担，忽略了问题的质量差异，导致必要与可避免的问题被同等对待。本文提出一种新型评估协议，区分“提问频率”与“提问质量”两个维度：引入“中断预算”限制提问总次数，并设计“用户注意力成本”（user-attention cost）指标，基于人类验证的生成式AI（Generative AI）评判标准，对问题的可读性及提问价值进行加权，从而捕捉用户实际感知的努力程度。实验表明，在相同提问数量下，该成本指标与人工标注者对任务努力程度的判断高度一致（75%一致性）。在八个LLM模型和四个客服领域中，将提问上限控制在1至3次，可使用户注意力成本降低36%-73%，同时保留86%-94%的任务奖励。进一步地，通过结合用户历史回答记忆来审查每个问题，可额外降低13%的注意力成本，且几乎无奖励损失，显著提升问题的有效性。此外，不同模型在提问效率上存在显著差异：在相同注意力成本下，最高效的模型获得的奖励是其他模型的两倍以上。研究结果表明，“减少提问”与“优化提问”是两个独立目标，未来对主动型智能体的评估必须同时量化提问数量与质量。
链接: https://arxiv.org/abs/2610.24411
作者: Xuanming Zhang,Amir H. Rezaeian,Miguel Ballesteros,Weiyi Sun,Werner Geyer,Lydia B. Chilton,Zhou Yu,Yassine Benajiba,Dan Roth
机构: Columbia University(哥伦比亚大学); Oracle AI
类目: Computation and Language (cs.CL)
备注: 40 pages, 8 figures, 36 table
Abstract:LLM agents can ask clarifying questions to complete tasks, but answering each question demands users’ time and effort. Existing evaluations typically measure this burden by counting questions, treating necessary and avoidable questions equally. We propose an evaluation protocol that distinguishes how often agents ask from how well they ask. An interruption budget limits the number of questions, while a user-attention cost charges extra for questions that are hard to read or not worth asking, as rated by a human-validated LLM judge. Beyond question counts, this cost captures perceived effort: on conversation pairs with the same number of questions, human annotators agreed with the cost on which conversation required more effort 75% of the time. Across eight LLMs and four customer-service domains, limiting agents to one to three questions reduces user-attention cost by 36-73% while keeping 86-94% of the task reward they earn with unlimited questions. At the same budget, reviewing each question against the memory of users’ earlier answers cuts attention cost by another 13% with little reward loss and yields more valuable questions. Models also differ sharply in how well they ask: at the same user-attention cost, the most efficient models gain more than twice as much reward from their questions as the others. These results show that asking less and asking better are distinct goals, and that evaluations of proactive agents should measure both.
[NLP-221] Long Listening Thoughts: Eliciting Open Auditory Reasoning with Deliberative Perception and Cognitive Refinement EMNLP2026
【速读】： 该论文旨在解决大音频-语言模型（Large Audio-Language Models, LALMs）在音频理解与推理任务中因感知错误导致性能受限的问题。现有后训练数据集普遍存在描述多样性不足的缺陷，难以有效提升模型对复杂音频场景的认知能力。为此，论文提出长时听觉思维框架（Long Listening Thoughts, LLT），其核心解决方案在于通过四阶段流水线实现深度感知锚定与显式认知行为的协同训练：首先构建涵盖语义、声学与时间维度的多方面密集音频描述；其次生成更具挑战性的问答对；第三步合成错误推理路径以增强鲁棒性；最后采用思想延续策略将短推理链扩展为长推理轨迹，从而激发验证、回溯等高级认知行为。基于该框架，研究者合成160万条推理轨迹，并采用监督微调（SFT）与直接偏好优化（DPO）相结合的两阶段训练流程，显著提升了Qwen2.5-Omni-7B在MMAU Test-mini、MMAR和MMSU等多个基准上的准确率，验证了LLT在促进音频推理能力方面的有效性。
链接: https://arxiv.org/abs/2610.24409
作者: Jaeyeon Kim,Chao-Han Huck Yang,Luoyi Zhang,Chan-Jan Hsu,Fernando Ruiloba Portilla,Jinchuan Tian,Shinji Watanabe,Carlos Busso
机构: Carnegie Mellon University (卡内基梅隆大学); NVIDIA (英伟达)
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注: Findings of EMNLP 2026. Project page: this https URL
Abstract:We introduce Long Listening Thoughts (LLT), a framework for constructing post-training datasets that improves audio reasoning by pairing deliberate perceptual grounding with explicit cognitive behaviors. On audio understanding and reasoning benchmarks, large audio-language models (LALMs) are limited primarily by perceptual errors, yet existing post-training datasets rely on audio descriptions of limited diversity. Through a four-stage pipeline, LLT (1) builds multi-aspect dense audio descriptions spanning semantic, acoustic, and temporal information, (2) generates more challenging questions, (3) synthesizes incorrect reasoning traces, and (4) extends short reasoning traces into long ones via a thought-continuation strategy that elicits cognitive behaviors such as verification and backtracking. Using LLT, we synthesize 1.6M reasoning traces and train Qwen2.5-Omni-7B with a two-stage pipeline of supervised fine-tuning (SFT) followed by direct preference optimization (DPO). This improves accuracy on MMAU Test-mini (71.5 to 78.2), MMAR (56.7 to 63.7), and MMSU (60.6 to 66.0).
[NLP-222] Defending Against Dormant Poisoning Attacks Across Language and Multimodal Agents
【速读】： 该论文旨在解决生成式 AI（Generative AI）在作为智能体（agent）使用时，潜在的“休眠投毒”（dormant poisoning）攻击所引发的安全风险问题。具体而言，攻击者可在模型发布时植入隐蔽的恶意行为，这些行为在初始阶段保持隐匿，但在后续进行良性微调（benign fine-tuning）后被激活，进而导致有害的代理行为，甚至在多模态场景下多个恶意行为可共存并被联合触发。现有防御方法通常依赖外部安全响应或可信干净参考模型，但在智能体场景中难以获取此类资源。本文的关键创新在于利用休眠投毒的一个核心特性：攻击者必须在发布时维持模型的正常表现以隐藏恶意行为，因此原始发布的模型输出可作为天然的安全参考。基于此，作者提出自动构建“安全缓冲区”（Safety Buffer），将有害输入与模型发布版本的响应及其通过同一模型生成的良性版本配对，从而识别并去除恶意意图。进一步地，为克服传统防御中无法区分削弱安全性与维持安全性更新的问题（常导致过度拒绝或性能下降），本文提出“前瞻防御”（LookAhead Defense）机制：利用安全缓冲区预演良性微调更新，仅对预测会削弱安全性的更新进行惩罚，即在有害输入上降低安全行为或在良性样本上增加拒绝概率的情况。实验表明，该方法在语言模型、语言代理和多模态代理等多种场景下，将代理中的有害行为从最高93.9%降至最多14.2%，语言模型中的有害响应从最高92.5%降至最多2.0%，同时保持较高的任务效用和低过度拒绝率，并对其他投毒方法具有鲁棒性。
链接: https://arxiv.org/abs/2610.24379
作者: Wonjun Lee,Vaidehi Patil,Zaid Khan,Yue Zhang,Hyunji Lee,Suhyun Kim,Mohit Bansal
机构: UNC Chapel Hill(北卡罗来纳大学教堂山分校); Kyung Hee University(庆熙大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Recent work shows that attackers can implant dormant malicious behaviors that remain hidden at release but emerge after benign fine-tuning. As foundation models are used as agents, whether dormant poisoning extends to harmful agent actions and multimodal settings remains unclear. We show two agentic risks: (1) harmful behaviors activated by benign fine-tuning can extend to unseen actions; (2) multiple harmful behaviors can coexist and be jointly activated by benign downstream fine-tuning. Existing defenses often rely on external safe responses, which are hard to obtain for agent actions, or a trusted clean reference model. We exploit a key property of dormant poisoning: the attacker must preserve safe behavior at release to conceal poisoning, and this behavior can serve as a safety reference. Leveraging this property, we automatically construct a Safety Buffer pairing harmful inputs with the released model’s responses and benign versions generated by the same model to remove harmful intent. The Safety Buffer can replace external safe responses in existing defenses, but these defenses do not distinguish updates that weaken safety from those that preserve it, often increasing over-refusal or reducing utility. To address this limitation, we introduce LookAhead Defense, which previews benign fine-tuning updates using the Safety Buffer and penalizes only updates predicted to weaken safe behavior on harmful inputs or increase the likelihood of refusals on benign versions in the Safety Buffer. Across language models, language agents, and multimodal agents, LookAhead Defense reduces harmful behavior in agents from up to 93.9% to at most 14.2% and harmful responses in language models from up to 92.5% to at most 2.0% versus vanilla fine-tuning, while maintaining competitive utility and low over-refusal. It remains effective against other poisoning methods.
[NLP-223] FlashMap: Flow Map Speculative Decoding
【速读】： 该论文旨在解决生成式 AI（Generative AI）在大语言模型（Large Language Model, LLM）推理过程中因自回归式草稿生成器（autoregressive drafter）效率低下，而现有基于扩散的草稿生成器（block diffusion drafter）又忽略生成块内词元间依赖关系所导致的性能瓶颈问题。其核心解决方案是提出 FlashMap，一种基于连续流的草稿生成器，通过将流映射语言模型（flow map language model）条件于目标模型的隐藏状态，实现整个词元块在单次前向传播中完成生成，并有效建模词元间的依赖关系。实验表明，在 Qwen3-8B 模型上，FlashMap 在所有基准测试中均优于匹配的 DFlash 基线，平均加速提升达 7%，最高实现 6.1 倍无损加速，显著提升了推理效率与生成质量。
链接: https://arxiv.org/abs/2610.24365
作者: Jeff Shen,Zhejian Zhou,Ryan Lee,Jonathan May,Shirley Ho
机构: Polymathic AI; Princeton University (普林斯顿大学); USC Information Sciences Institute (南加州大学信息科学研究所); New York University (纽约大学); Flatiron Institute (扁平铁研究所)
类目: Computation and Language (cs.CL)
备注: Code: this https URL
Abstract:Speculative decoding accelerates large language model (LLM) inference by using a lightweight drafter to propose a block of tokens that the target model verifies in one parallel pass. Autoregressive drafters generate this block one token at a time, whereas block diffusion drafters such as DFlash generate it in one pass, making drafting faster. However, these diffusion drafters predict each token independently, ignoring dependencies between tokens in the block. We introduce FlashMap, a continuous flow drafter that conditions a flow map language model on the target model’s hidden states and generates the whole block in one forward pass while capturing inter-token dependencies. On Qwen3-8B, FlashMap is faster than a matched DFlash baseline on every benchmark, with 7% higher average speedup and up to 6.1x lossless acceleration.
[NLP-224] What Matters for NVFP4 Training? A Comparative Study of Low-Precision Pretraining Recipes ICML2026
【速读】： 该论文旨在解决在大规模语言模型（Large Language Models, LLMs）训练中采用4位浮点数（FP4）格式时，如何实现高精度恢复与训练稳定性的问题。尽管低比特量化在提升吞吐量和能效方面具有显著潜力，但多数方法在小规模模型或短序列训练场景下验证，其在中等规模模型（如80亿参数的密集模型与300亿参数的A3B MoE模型）及长达1万亿标记（1T tokens）训练周期下的有效性仍不明确。本文通过系统性比较近期提出的非对称量化浮点4位（NVFP4）训练方案，揭示了影响精度恢复的关键因素。解决方案的核心在于提出一个基于成熟大规模稳定NVFP4训练原则的优化配方，融合了前沿技术，如梯度估计无偏化处理，其量化误差低于随机舍入（stochastic rounding），从而有效降低训练过程中的累积误差。通过消融实验发现：（i）配方中每一项技术均对损失轨迹有可测量的改善；（ii）在大规模训练中，选择性地保留部分高精度层是恢复模型准确性的必要条件；（iii）并非所有反向传播中的张量都同等受益于去偏（de-biasing）操作，因此仍有空间对剩余张量应用互补的误差抑制技术。该研究实现了当前在该规模下最接近BF16基线的损失差距（loss gap），为高效、高精度的低比特大模型训练提供了重要实践依据。
链接: https://arxiv.org/abs/2610.24343
作者: Anjulie Agrusa,Andrei Panferov,Elizabeth Wei,Keith Wyss,Paul Gibbons,Erik Schultheis,Jinhang Choi,Frank Sun,Tijmen Blankevoort,Dan Alistarh
机构: NVIDIA(英伟达); Institute of Science and Technology Austria(奥地利科学技术研究所); Red Hat AI(红帽人工智能)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 10 pages, 4 figures, 4 tables. Accepted at the ICML 2026 Workshop on Resource-Adaptive Foundation Model Inference (AdaptFM), Seoul, South Korea
Abstract:Training large language models directly in 4-bit floating-point (FP) formats promises substantial improvements in throughput and energy efficiency. While some recipe design choices have been validated at scale, many promising approaches remain untested beyond small models and short token horizons, leaving open the question of which trends will hold. We present a systematic comparison of recent NVFP4 recipes at medium model scale, evaluating dense 8B and 30B-A3B MoE models trained for up to 1T tokens, with a focus on which ingredients are necessary for accuracy recovery. We propose a final recipe grounded in the principles behind established stable NVFP4 training at scale, incorporating state-of-the-art techniques such as unbiased gradient estimation with lower quantization error than stochastic rounding. To the best of our knowledge, this is the strongest FP4 training result demonstrated at this scale to date in loss gap to BF16. Through ablation studies, we find that: (i) each technique in the optimized recipe measurably improves loss trajectory, (ii) selective high-precision layers are necessary for recovering accuracy at scale, (iii) not all tensors in the backward pass benefit equally from de-biasing, leaving room to apply complementary error-reduction techniques to the remaining tensors.
[NLP-225] he Judge Can Debias Itself: Awareness and Self-Debiasing for Reliable LLM -as-a-Judge
【速读】： 该论文旨在解决大语言模型作为评判者（LLM-as-a-Judge）系统在自动化评估中普遍存在系统性偏见的问题，此类偏见会严重损害评估结果的可靠性。其核心解决方案在于无需额外训练即可实现自我去偏，关键在于提出两种轻量级策略：一是偏见感知提示（bias-aware prompting），即在评估前显式提醒评判模型潜在的目标偏见；二是自去偏（self-debiasing），即让评判模型对初始判断进行自我批判，识别并修正自身受偏见影响的部分。为全面评估效果，研究引入三个新型基准——权威偏见（伪造引用）、自信偏见（强势表述）和格式偏见（结构化呈现），共覆盖七类偏见。实验表明，在三款前沿模型上，自去偏策略在21组模型-偏见组合中提升了20组的表现，最高提升达15个百分点，而偏见感知提示也表现出一致的改进效果。相比之下，近期采用微调外部评判者的方案在新引入的偏见类型上泛化能力较弱（9个案例中有5个出现性能下降）。研究结果表明，通过提示评判模型主动反思偏见（无论是事前还是事后），是一种稳健且可泛化的提升评估完整性的有效方法。
链接: https://arxiv.org/abs/2610.24320
作者: Youbing Yin,Jingyu Wu,Anoop Kumar,Scott Novotney,Daben Liu,Erin Babinsky
机构: Capital One(资本一号)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:LLM-as-a-Judge systems are increasingly employed for automated evaluation, yet remain susceptible to systematic biases that undermine their reliability. We demonstrate that the judge can effectively debias itself without additional training. We propose two lightweight strategies: bias-aware prompting, which explicitly alerts the judge to potential target biases prior to evaluation, and self-debiasing, in which the judge critiques the initial judgment for bias influence. To broaden the evaluation scope, we introduce three novel benchmarks–authority bias (fabricated citations), confidence bias (assertive phrasing), and formatting bias (structured presentation)–and evaluate across seven bias types in total. Across three frontier models, self-debiasing improves performance in 20 out of 21 model-bias combinations with gains up to 15 percentage points, while bias-aware prompting also yields consistent improvements. The recent fine-tuned external critic, however, does not generalize as reliably to the newly-introduced bias types (degradation observed in 5 out of 9 cases). Our findings demonstrate that prompting a judge to reason about bias, either preemptively or retrospectively, is a robust and generalizable approach for enhancing evaluation integrity.
[NLP-226] Not Every User Says It All: Training Tool-Using Agents Beyond Ideal Simulated Users
【速读】： 该论文旨在解决当前工具使用型智能体（tool-using agents）在训练与评估中过度依赖强模拟用户（strong simulated users）的问题，而现实中的用户往往表达不完整或信息模糊（即弱用户，weak users）。现有方法在弱用户场景下表现显著下降，且通过弱用户训练无法提升智能体对弱用户的适应能力。其核心挑战在于：传统强化学习框架难以有效处理用户输入缺失时的策略优化问题。解决方案的关键是提出一种名为**状态分组策略优化（StaGPO, State-Grouped Policy Optimization）**的新方法，其核心机制是让智能体显式地生成任务需求、缺失信息及填充方式，构建可比的状态表示，进而基于相同状态下的不同回放轨迹进行步级信用分配（step-level credit assignment），从而更精准地指导策略更新。在标准小规模监督微调（SFT）冷启动+强化学习（RL）流程中，StaGPO使弱用户训练真正具备价值——在ToolWOZ和tau2-bench基准上，使用弱用户训练的Qwen3.5-2B模型性能超越所有对比基线，达到76.8%的成功率，较次优方案高出4.8个百分点，并展现出对未见掩码率的泛化能力；同时，该策略还实现了跨用户质量的迁移优势，即以强用户训练的策略在弱用户场景下仍能优于仅在弱用户上训练的传统方法。
链接: https://arxiv.org/abs/2610.24294
作者: Han Bao,Zheyuan Zhang,Kaiwen Shi,Yanfang Ye
机构: University of Notre Dame(圣母大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 22 pages, 11 figures, under review
Abstract:Tool-using agents are trained and evaluated almost exclusively with strong simulated users who state every detail clearly, yet real users are often weaker. We construct weak users by masking sentences of a simulated user’s replies, holding the task and the user’s knowledge fixed, and cross-evaluate agents trained with strong and weak users on tau2-bench and ToolWOZ. Weak users lower the success of every policy, and training with them does not help: no policy trained with weak users, through fixed masks, mixtures or a curriculum, serves weak users better than the strong-trained policy beyond run-to-run noise, and prompting the agent to ask more does not help either. We propose StaGPO (State-Grouped Policy Optimization): the agent writes what the task needs, what is missing and how it is being filled, and GRPO uses this state to compare turns of different rollouts that reach the same situation, assigning step-level credit. Within the standard recipe of a small SFT cold start followed by RL, StaGPO makes weak-user training useful: on ToolWOZ, the Qwen3.5-2B policy trained with weak users reaches 76.8% with weak users, the highest of any Qwen3.5-2B policy we trained, 4.8 points above GRPO and 2.8 above the strongest baseline; it is also 5.4 points above GRPO with strong users, and generalises to an unseen mask rate; trained with strong users instead, it serves weak users at least as well as GRPO trained with weak users. On tau2-bench it scores highest with weak users for both models, by a small margin for Qwen3.5-4B.
[NLP-227] Beyond Clean Accuracy: The Practical Challenges and Benefits of Pixel-Based Language Models
【速读】： 该论文旨在解决像素级语言模型（pixel-based language models）在下游任务中适应性不足的问题，尤其关注其在噪声输入和多语言场景下的鲁棒性与迁移能力。其核心挑战在于：尽管像素级模型通过摒弃固定子词词汇表（subword vocabulary）避免了对文本分词的依赖，从而增强了对噪声及异构输入的容忍度，但其在微调过程中的训练稳定性、数据覆盖要求以及任务信息提取效率均显著低于传统文本基模型。解决方案的关键在于揭示并量化这一核心权衡——即以渲染（rendering）替代分词虽能有效降低对拼写错误、非标准文本和陌生语言的敏感性，但将模型适配的瓶颈从预训练阶段转移至下游微调阶段，导致需要更长的训练周期、更广泛的标注数据支持，且对学习率敏感，易出现失败。然而，一旦成功微调，像素级模型在面对字体噪声时表现出更强的鲁棒性，并在跨语言任务中展现出优于英语预训练文本模型的目标语言性能，凸显其在复杂现实场景中的潜在优势。
链接: https://arxiv.org/abs/2610.24290
作者: Haowen Tang,Wei Xia,Raoyuan Zhao,Hinrich Schütze,Yihong Liu
机构: LMU Munich; Munich Center for Machine Learning (MCML)
类目: Computation and Language (cs.CL)
备注: preprint
Abstract:Pixel-based language models process rendered text without relying on a fixed subword vocabulary, offering a promising approach to noisy and multilingual inputs. However, their practical value depends on whether they can be reliably adapted to downstream tasks. To this end, we systematically examine this question through fine-tuning behavior, typographical robustness, and zero-shot crosslingual transfer. Under our tested settings, pixel-based models require longer training and broader labelled-data coverage than text-based models, while delayed learning and learning-rate-dependent failures make adaptation less reliable. Our analyses further suggest that task-relevant information is less readily accessible in pretrained pixel representations, placing greater demands on fine-tuning. Once successfully adapted, however, pixel-based models lose substantially less accuracy under typographical noise and achieve higher target-language accuracy than English-pretrained text-based counterparts on two crosslingual benchmarks, despite weaker clean English performance. Together, these findings reveal a central trade-off: replacing tokenization with rendering reduces sensitivity to corrupted and unfamiliar inputs, but shifts a central bottleneck to downstream adaptation.
[NLP-228] BanglaShield: A Script-Robust Safety Classifier and Benchmark for Bangla Native Romanized and Code-Mixed Prompts
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在低资源语言——孟加拉语（Bangla）——中进行提示安全分类时面临的挑战，尤其针对孟加拉语在本地文字、拉丁字母转写（romanization）及与英语、印地语或乌尔都语混用等多种书写形式下的安全评估不一致问题。现有安全分类器多基于标准脚本文本构建和评估，难以适应孟加拉语实际使用中的复杂书写变体，导致安全检测性能显著下降。其解决方案的关键在于提出 BanglaShield，一个面向多种书写形式的分级提示安全分类资源与模型体系：首先构建包含10,245条孟加拉语提示的统一数据集，采用四层级安全标签（安全、冒犯性、有害、严重）并确保三种书写形式（本地文字、罗马化、代码混合）之间标签对齐与分组感知的划分；在此基础上，通过在36,882个分类查询和24,318个响应上对六种主流LLM及三个多语言防护模型进行基准测试，揭示了当前模型在非本地文字形式下高达63.4%的不安全提示误判率。最终，研究提出一种紧凑的微调编码器模型——BanglaShield，仅需约130倍少于Gemma 4 31B的参数量，在本地文字和罗马化形式上的宏平均F1得分分别达到87.8和84.4，相较最优LLM提升42.8–46.1个百分点，且能有效缓解模型自检作为过滤器时的性能瓶颈。该工作为低资源语言的安全分类提供了可复现、跨书写形式的高质量基准与高效模型。
链接: https://arxiv.org/abs/2610.24288
作者: Mohsinul Kabir,Md. Tariquzzaman,Mohammad Rashidujjaman Rifat,Tharindu Ranasinghe,Nishat Raihan
机构: University of Manchester (曼彻斯特大学); Islamic University of Technology (伊斯兰科技大学); University of Notre Dame (圣母大学); Lancaster University (兰卡斯特大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 17 pages. Code and data: this https URL ; dataset: this https URL
Abstract:Safety classifiers for large language models (LLMs) are typically built and evaluated on standard-script text, yet Bangla, like many lower-resource languages, is routinely written in native script, in Latin-script romanization, and code-mixed with English, Hindi, or Urdu. We introduce BanglaShield, a resource and model for graded prompt-safety classification across these written forms. We consolidate 10,245 Bangla prompts from six sources under a four-level taxonomy (Safe, Offensive, Harmful, Severe) and render each prompt in three aligned forms that share labels and a group-aware split, enabling strictly paired comparisons. Benchmarking six contemporary LLMs, including a Bangla-specialized one, on 36,882 classification queries and 24,318 judged responses, we find that LLM classifiers label 47.0% of unsafe native-script prompts as Safe, rising to 63.4% under romanization, while response-level safety accuracy drops by 10.3 points. Three multilingual guard models also pass 30.3-57.3% of unsafe native-script prompts. Using a model’s own classification as an input filter reduces harmful compliance by only about five points. In contrast, BanglaShield, a compact fine-tuned encoder, reaches 87.8 macro-F1 on native-script and 84.4 on romanized prompts, outperforming the best LLM on each form by 42.8-46.1 points with ~130x fewer parameters than Gemma 4 31B. We release the corpus, splits, judgments, and model.
[NLP-229] Modeling bilingual naming performance through laterally-connected SOMs
【速读】： 该论文旨在解决传统机器学习方法在个体认知建模中因训练数据不足而导致模型泛化能力差的问题。其核心挑战在于如何构建能够反映个体差异、具备机制解释力的计算模型，以深入理解双语词汇表征与词义激活过程。解决方案的关键在于对原有BiLex模型进行生物学层面的增强，引入兴奋性与抑制性侧向连接（lateral connections），从而更真实地模拟神经与认知过程。这一改进使模型不仅能够拟合命名准确率，还能有效再现反应时及常见错误类型，如基于语义概念熟悉度、典型性和特异性所产生的错误。通过在西班牙语和英语双语者群体上模拟波士顿命名测试，模型成功捕捉了不同语言习得历史下的语义层级认知机制，实现了对个体被试语义认知过程的精细化建模。未来该框架有望成为研究语言障碍及其干预策略的理论基础。
链接: https://arxiv.org/abs/2610.24275
作者: Divya Prabhakaran,Uli Grasemann,Erie Shivers,Marissa Russell-Meill,Swathi Kiran,Risto Miikkulainen
机构: The University of Texas at Austin (德克萨斯大学奥斯汀分校); Boston University (波士顿大学)
类目: Computation and Language (cs.CL)
备注: 9 pages, 5 figures
Abstract:Computational models have explanatory power to understand cognition in individuals, but there is usually not enough training data to build such models with traditional machine learning. However, alternative architectures that model neural and cognitive processes in more detail can be more easily fit to individuals and can lead to mechanistic insights into human behavior. BiLex is such a computational model of the bilingual lexicon, containing a shared semantic system and phonetic language systems represented through self-organizing maps. In this study, BiLex was extended further into biological realism with excitatory and inhibitory lateral connections. This extension makes it possible to account for response times as well as common errors such as those based on semantic concept familiarity, typicality, and specificity. The Boston Naming Test was simulated in Spanish and English, and model performance approximated naming accuracy and error types in Spanish-English bilinguals with varying primary and secondary language histories. The results indicated that the approach captures semantic-level processes in bilingual lexical access, thus modeling the semantic cognition of individual subjects. In the future, such models may serve as a foundation for modeling impairments and treatment options.
[NLP-230] Best of Both Worlds: Fine-Tuned Small Models Complement LLM s for Cost and Performance Benefits
【速读】： 该论文旨在解决大规模文本分类任务中模型性能、计算成本与推理时间之间的权衡问题，尤其关注在标注数据稀缺场景下，生成式AI模型（如GPT-3.5）与传统预训练语言模型（如RoBERTa）在实际应用中的表现差异。其核心挑战在于：尽管生成式AI具备极强的少样本学习能力（仅需1–5个示例/类别），但其推理效率低、成本高，而传统监督模型虽性能优越但依赖大量标注数据。解决方案的关键在于通过系统性实验对比不同模型在16类调查文本分类任务上的表现，并探索优化生成式模型推理效率的方法。研究发现，采用人工精选的少样本提示（few-shot prompting）结合最优批处理大小（batch size=20）可显著降低GPT-3.5的推理成本约70%且不影响性能；最终提出一种混合集成策略——将高性能的微调RoBERTa模型与高效但略逊的GPT-3.5模型结合，实现F1分数达97.5%，同时仅消耗相当于独立使用GPT-3.5模型4.4%的推理成本，从而在性能与效率之间达成最佳平衡。
链接: https://arxiv.org/abs/2610.24260
作者: Shivam Mathur,Sri Harish Popuri,Priya Krishnan Sundararajan,Kevin Horecka,Michael Mugo
机构: Walmart Inc(沃尔玛公司)
类目: Computation and Language (cs.CL)
备注: Work done in 2024, added for posterity
Abstract:BERT-based models require thousands of labeled samples to train. Larger models like GPT-X require only 1-5 examples per class. However, performance, cost and time tradeoffs of such models remain under-explored. In this work, we classify survey text data into 16 classes. We focus on experimental approaches and results on performance, cost, and time taken for this problem. We compare these parameters of a trained RoBERTa-based model, and an (off-the-shelf) GPT-3.5 model for text classification. Baseline experiments show that the time taken by GPT-3.5 for generating predictions on 10,000 records is 6 hours, whereas a RoBERTa-based model takes only 12 mins on a T4 GPU. To identify the best prompting approach for GPT-3.5, we explore batch prompting. We experiment with (1) batch-size, and (2) few-shot sampling. We find the best batch size is 20 with human few-shot curation. Experiments show the F1 score of a trained RoBERTa-based model (96%) is much higher than a GPT-3.5 model (83%). In case of GPT-3.5, batch prompting is beneficial in lowering inference costs by ~70%, without hindering task performance. We find the best approach is an ensemble of a RoBERTa-based and GPT-3.5 model (F1: 97.5%) incurring only 4.4% of inference cost compared to a standalone GPT-3.5.
[NLP-231] SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction Analytics
【速读】： 该论文旨在解决课堂语言教学中学习者出现错误后，教师回应如何影响后续互动这一关键问题，尤其关注跨课堂语境下对话交互的精细化分析难题。传统手动标注因依赖上下文敏感性而难以规模化，限制了对第二语言习得（Second Language Acquisition, SLA）过程中反馈与纠错机制的系统研究。其解决方案的核心在于提出SLA-LENS框架——一个基于交互主义第二语言习得理论的端到端大语言模型（Large Language Model, LLM）辅助分析系统，能够将真实课堂转录文本自动关联至结构化教学标注与证据链驱动的描述性分析。该框架通过融合说话人角色、对话上下文及层级化标签的多维注释体系，精准捕捉学习者语言错误、教师教学功能、学习者反馈采纳（uptake）或修正行为等关键要素。研究在40个真实的在线二语西班牙语课堂转录数据上验证了该方法，涵盖超过21,000个句级实例，并采用专家审定的金标准标注、保留的银色数据及针对不同标注误差类型的敏感度指标，评估了专有模型与开源权重模型的表现，包括任务特定的QLoRA微调策略。此外，框架内置确定性报告层，可生成可复现的可视化结果，呈现参与度、反馈模式、错误分布、反馈采纳情况及时间维度上的互动模式，并与原始语料证据直接关联。该设计使复杂教学互动模式变得可访问、可检验，从而支持系统化的课堂话语研究、教学反思与课程设计优化。
链接: https://arxiv.org/abs/2610.24259
作者: Nemika Tyagi,Olga Kellert,Phanos Kareen,Nelvin Licona-Guevara
机构: Arizona State University (亚利桑那州立大学)
类目: Computation and Language (cs.CL)
备注: 16 pages, 3 figures, 14 tables
Abstract:What happens after a learner makes an error, and how does the instructor’s response shape the next turn? Investigating these exchanges across classrooms requires context-sensitive annotation that is difficult to scale manually. We introduce SLA-LENS, an end-to-end LLM-assisted framework that connects classroom transcripts to structured pedagogical annotations and evidence-linked descriptive analytics. Grounded in interactionist Second Language Acquisition (SLA) theory, its annotation schema captures second language learner errors, pedagogical functions, and uptake or repair while accounting for speaker roles, conversational context, and hierarchical labels. We apply the framework to 40 authentic online L2 Spanish classroom transcripts comprising over 21,000 sentence-level instances. We evaluate proprietary and open-weight LLMs, including task-specific QLoRA adaptation, using expert-reviewed gold annotations, held-out silver data, and metrics sensitive to different types of annotation error. A deterministic reporting layer generates reproducible visualizations of participation, feedback, learner errors, uptake, and temporal interaction patterns, linked to supporting transcript evidence. By making these patterns accessible and inspectable, SLA-LENS supports systematic classroom discourse research, pedagogical reflection, and curriculum design.
[NLP-232] Can This LLM Annotate My Data? Estimating Zero-Shot LLM Performance
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在文本标注任务中性能不可预测的问题，核心关注点在于能否提前预判LLM在新数据集上的表现。其解决方案的关键在于通过构建两个任务族生成461个文本分类数据集，系统性地评估不同预测信号的有效性，结果表明：尽管存在局限性，但内部对数概率（internal log-probabilities）是预测LLM性能最有效的信号，而以往研究强调的自信心言语化（verbalizations of self-confidence）则并非最优预测依据。
链接: https://arxiv.org/abs/2610.24250
作者: Tong Lin,Grant Schoenebeck,Sabina Tomkins
机构: University of Michigan (密歇根大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) are increasingly used in annotation tasks with varying success. While several studies have demonstrated the opportunities and shortcomings of LLMs for annotation, a critical question remains unanswered: is it possible to predict how well they will perform on a new dataset? To answer this and related questions, we construct two task families which induce 461 text classification datasets. We find that it is possible to predict LLM performance - to a certain degree - and that contrary to other work which highlights the usefulness of LLM verbalizations of self-confidence, internal log-probabilities are the most predictive signal.
[NLP-233] Probing Ambiguity: Adversarial Intent Resolution for Robust Text-to-SQL
【速读】： 该论文旨在解决文本到SQL（Text-to-SQL）系统在面对自然语言查询中固有歧义时的局限性，即传统方法假设每个自然语言查询仅对应唯一正确的SQL翻译，而现实中许多查询存在多种合法解释，导致不同语义解读产生差异化的查询结果。其解决方案的关键在于提出一种无需任务特定训练的多智能体框架——对抗意图解析（Adversarial Intent Resolution, AIR），该框架通过构建竞争性语义解释、利用SQL执行验证其输出差异，并基于观测到的数据差异生成针对性的澄清问题，从而主动识别并化解歧义。AIR在多个基准测试中展现出优异性能、高效率及跨基准的强泛化能力，在仅提供模式信息（schema-only）的现实生产场景下，相比当前最优的强化学习模型IntentRL，表现出更小的性能下降（约7% vs. 约20%），且在涉及模式设计引发歧义的AmbiQT基准上亦实现更高准确率，充分体现了其鲁棒性与通用性。
链接: https://arxiv.org/abs/2610.24248
作者: Youbing Yin,Xujun Peng,Muhammad S. Uddin,Erin Babinsky,Daben Liu,Anoop Kumar
机构: Capital One(资本一号)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Text-to-SQL systems typically assume a one-to-one mapping between natural-language queries and correct SQL translations. In practice, however, many queries are inherently ambiguous, admitting multiple valid interpretations that yield distinct outputs. We introduce Adversarial Intent Resolution (AIR), a multi-agent framework that actively probes and resolves such ambiguity without any task-specific training. AIR first constructs competing interpretations, verifies through SQL execution that they produce genuinely different results, and generates targeted clarification questions grounded in the observed data differences. Evaluating on leading benchmarks for ambiguous questions, AIR demonstrates strong performance, efficiency, and robust cross-benchmark transferability. On the AMBROSIA benchmark under schema-only context, a realistic production setting where exposing the full database is infeasible, AIR outperforms IntentRL, the state-of-the-art reinforcement learning model, across all metrics (e.g., F1 0.679 vs. 0.618). Crucially, AIR exhibits only ~7% performance degradation without the full database, compared to ~20% for IntentRL. On the AmbiQT benchmark, where ambiguity arises from schema design rather than linguistic structure, AIR achieves higher F1 under both context settings without any benchmark-specific adaptation.
[NLP-234] Encoder-Free Multimodal LLM s Learn Implicit Encoders ICLR2027
【速读】： 该论文旨在解决传统多模态大语言模型（Multimodal Large Language Models, MLLMs）中依赖专用模态编码器（如视觉编码器或音频编码器）所导致的架构复杂性和模态间耦合问题。现有方法通常通过独立的模态特定编码器将图像、音频等原始输入转换为嵌入表示，再与文本输入融合。然而，近年来出现的无编码器（encoder-free）多模态大模型（如Gemma-4 12B和Inkling-Small）采用全新范式：直接将原始采样窗口、像素或频谱特征输入到Transformer语言模型中，跳过显式的模态编码步骤。本文的核心问题是：这些早期层是否在隐式地承担类似专用编码器的功能？其关键解决方案在于揭示——尽管没有显式设计的编码器，但这些模型的早期层仍能自发学习形成“隐式编码器”（implicit encoders），能够以对跨模态上下文不敏感的方式有效表征音频与视觉输入。进一步实验表明，在执行多模态推理任务时，无论是否从第一层起联合编码，仅在早期层中独立处理各模态即可达到相同的性能表现，说明模态解耦在初始阶段并非必要。通过分层探针分析，研究发现隐式编码器所提取的关键信息特征与主流显式编码器具有高度相似性，验证了其功能等效性。这一发现为简化多模态模型架构、提升可解释性提供了实证支持，并暗示了未来无需显式编码器的高效多模态建模路径。
链接: https://arxiv.org/abs/2610.24239
作者: David Harwath,Desh Raj,Xavier Velez,Sandra Arcos-Holzinger,Willie Jeng,Karen Livescu,Leibny Paola Garcia-Perera,Sanjeev Khudanpur
机构: The University of Texas at Austin(德克萨斯大学奥斯汀分校); NVIDIA(英伟达); Johns Hopkins University(约翰霍普金斯大学); The University of Melbourne(墨尔本大学); Toyota Technological Institute at Chicago(芝加哥丰田技术学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
备注: Submitted to ICLR 2027
Abstract:We empirically analyze two recently proposed encoder-free multimodal LLMs, Gemma-4 12B and Inkling-Small. Unlike previous multimodal LLMs, these models do not use modality-specific encoders to extract tokenized embeddings representing audio and image input; instead, they directly feed windows of samples, pixels, or spectral features to the Transformer language model. We analyze whether the early layers of these models learn to function similarly to dedicated audio and visual encoders. We find that these early layers do in fact act as ``implicit encoders,‘’ encoding audio and visual inputs in a way that is largely invariant to their cross-modal context. Furthermore, when performing multimodal reasoning tasks, we show that independently encoding the modalities throughout these early model layers achieves the same accuracy as jointly encoding them from the first layer onwards. Finally, we use layer-wise probing to analyze the salient information captured by the implicit encoders and find their behavior to be similar to several popular encoder models.
[NLP-235] Preserving Reasoning Diversity in Reinforcement Learning with Verifiable Rewards via Adaptive Replay
【速读】： 该论文旨在解决生成式人工智能在强化学习中使用可验证奖励（Reinforcement Learning with Verifiable Rewards, RLVR）时导致的推理行为多样性下降问题。其核心挑战在于：由于有限样本下的策略更新动态，标准的在线策略强化学习会逐步抑制语义上不同但得分相同的推理路径，一旦某类推理模式被过度稀疏采样，便难以通过常规方法恢复。为此，作者提出一种名为目标自适应回放（Targeted Adaptive Replay, TAR）的新框架，其关键在于构建一个紧凑的记忆库以保存基础模型中已有的多样化正确推理解法，并设计一种自适应机制，在RLVR训练过程中仅当这些已有解法出现被压制迹象时才触发选择性重放。该方法有效平衡了推理性能与多样性保持之间的关系，在六个数学推理基准测试及两种Qwen2.5模型规模下均优于现有多样性保持型后训练方法，显著减少了对基础模型正确解法的遗忘。
链接: https://arxiv.org/abs/2610.24236
作者: Wen-Zhi Li,Sainyam Galhotra,Claire Cardie
机构: Cornell University(康奈尔大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Reinforcement Learning with Verifiable Rewards (RLVR) improves the reasoning capabilities of large language models, but often reduces the diversity of their reasoning behaviors. We study this phenomenon from the perspective of finite-sample learning dynamics and show that limited on-policy rollouts can progressively suppress semantically distinct but equally rewarded reasoning modes. Once a mode becomes sufficiently under-sampled, standard on-policy RLVR rarely recovers it. Motivated by this observation, we propose TAR, a Targeted Adaptive Replay framework that preserves diverse correct reasoning behaviors already present in the base policy. TAR constructs a compact memory bank of diverse base-model solutions and selectively replays them through an adaptive mechanism that activates only when stored behaviors are being suppressed during RLVR. Across six mathematical reasoning benchmarks and two Qwen2.5 model scales, TAR consistently achieves a better balance between reasoning performance and diversity preservation than existing diversity-preserving post-training methods, while substantially reducing the forgetting of correct base-model solutions. Code is available at this https URL.
[NLP-236] Analyzing Repetition and Interference Effects in Memory-Constrained User Simulators
【速读】： 该论文旨在解决当前用户模拟器（user simulator）在评估与训练人工智能系统与人类交互时，缺乏对人类行为真实性的有效衡量问题。其核心挑战在于如何确保用户模拟器在信息记忆方面与人类表现出一致的行为模式。解决方案的关键在于通过检验记忆受限的用户模拟器对两个影响人类信息保留的关键因素——事实重复和竞争性信息干扰——的敏感性，来评估其行为保真度。研究发现，尽管部分模拟器在整体准确率上接近人类水平，并能复现重复与干扰效应的量级，但它们在具体问题上的错误模式与人类存在显著差异：人类倾向于误记（misremember），而模型则表现为彻底遗忘（forget），揭示了现有模拟器在认知机制建模上的根本缺陷。为此，作者公开了该数据集作为未来研究的基准挑战（benchmark challenge），以推动更贴近真实人类记忆行为的用户模拟器发展。
链接: https://arxiv.org/abs/2610.24232
作者: Cara Su-Yi Leong,Nicholas Tomlin,Tal Linzen
机构: New York University(纽约大学); Toyota Technological Institute at Chicago(芝加哥丰田技术学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:User simulators can help evaluate and train AI systems to interact with humans at scale. To do so, evaluation of user simulators must measure behavioral parity with people. We evaluate whether memory-constrained user simulators are sensitive to two factors known to affect human information retention: repetition of a fact and interference from competing information. We find that some simulators achieve human-like overall accuracy, and replicate the magnitude of repetition and interference effects, but none match people’s accuracy on particular questions: while humans misremember facts, models forget them, leading to different error patterns. We release this data as a challenge benchmark for future work on user simulators.
[NLP-237] Language Models are Lazy Poets: Creative Mode Collapse at Scale
【速读】： 该论文旨在解决生成式大语言模型（Generative AI）在诗歌创作中普遍存在的“模式坍缩”（mode collapse）问题，即模型输出高度重复、缺乏多样性。尽管已有研究尝试缓解此问题，但针对特定文本体裁（如诗歌）的大规模系统性调查仍十分匮乏。本文通过构建迄今最大规模的生成式诗歌语料库——涵盖15种诗体、248个主题标签及16个开源模型，共生成超过110万首诗歌，首次在百万级数据规模下深入分析了模式坍缩现象。研究发现，尽管模型具备丰富的词汇量与押韵能力，但实际使用中存在严重资源浪费：模型表现出明显的“懒惰”倾向，过度依赖少数高频词、固定句式和重复韵脚，导致语义与形式多样性不足。此外，不同模型与诗体间的词汇同质化程度存在差异，且观察到的押韵词汇丰富度与模型参数规模呈正相关。因此，解决方案的关键在于设计有效的解耦机制与激励策略，以引导模型更充分地利用其潜在的创造性表达能力，打破对有限模式的依赖。
链接: https://arxiv.org/abs/2610.24223
作者: Jacob A. Matthews,Sil Hamilton
机构: Cornell University(康奈尔大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:While aligned large language models (LLMs) can successfully complete many creative tasks, their outputs tend to be highly self-similar. Though solutions to this problem of “mode collapse” have been proposed, comprehensive large-scale surveys of the phenomenon in specific textual genres are still lacking. We address this gap through an extensive treatment of LLM mode collapse in poetry. Across 15 poetic forms, 248 subject matter tags, and 16 open-weight models, we generate over 1.1 million poems. To our knowledge, this constitutes the largest corpus of LLM-generated poetry to date. At million-poem scale, we find that LLM poetic vocabularies and rhyming abilities are substantial, but severely underutilized. In other words, LLMs are “lazy” poets: despite possessing large poetic inventories, LLMs over-rely on a small number of words, opening phrases, and rhymes. Further, we find that lexical homogenization differs across models and poetic forms, noting observed rhyming vocabulary is correlated with model size.
[NLP-238] Reasoning with Context Management and External Memory
【速读】： 该论文旨在解决复杂推理任务中因上下文窗口（context window）有限而导致的推理记忆不足问题。在处理需要多次重复计算中间步骤的复杂问题时，传统方法依赖模型内部的上下文记忆，但受限于上下文长度，容易丢弃关键信息，进而影响推理性能。其解决方案的关键在于提出RCMEM（Reasoning Context Management with External Memory），通过引入外部记忆机制，在不增加推理模型上下文负担的前提下，持久化存储和管理推理过程中的关键中间状态。该方法有效缓解了上下文溢出问题，显著降低重复计算的开销，同时提升了复杂数学问题上的推理准确率。实验在0.4至2500万参数规模的模型及GPT-2上验证了其有效性。
链接: https://arxiv.org/abs/2610.24187
作者: Man-Ling Sung,William Hartmann,Jiangying Zhou
机构: RTX BBN Technologies; Cambridge MA, USA
类目: Computation and Language (cs.CL)
备注: Preprint
Abstract:Reasoning models maintain information about the status of their reasoning process through in-context memory. Complex problems require larger context windows and memory requirements. In this scenario, context management without caching risks discarding useful information for reasoning. We propose RCMEM, an approach that extends context management with an external memory to maintain additional information without burdening the reasoning model with additional context. RCMEM significantly reduces computation time on problems that require repeated computation of intermediate steps, while achieving better performance on complex problems. We demonstrate our approach on a few mathematics problems using model sizes ranging from 0.4 to 25 million parameters, and GPT-2.
[NLP-239] Multi-author style detection with stylistic embeddings EMNLP
【速读】： 该论文旨在解决多作者写作风格变化检测中因单一模型在目标领域微调后特征表示能力不足的问题。现有方法仅依赖现成的大型语言模型（Large Language Models, LLMs）或在目标域内进行监督微调，难以提取有效风格特征。其解决方案的关键在于引入跨域迁移学习策略，通过针对作者身份验证任务对LLM嵌入模型（如Mistral-7B）进行微调，并结合已知表现优异的DeBERTa模型，在PAN-25多作者写作风格分析挑战赛的基准上进行对比。研究发现，DeBERTa与Mistral在不同类型的用户生成文本场景中各具优势，因此通过融合二者嵌入特征，实现了更鲁棒、更有效的风格变化检测性能。
链接: https://arxiv.org/abs/2610.24184
作者: Man-Ling Sung,Chak Fai Li,Damianos Karakos,Fred Bernardin,Brian Ulicny
机构: RTX BBN Technologies; Cambridge MA, USA
类目: Computation and Language (cs.CL)
备注: Submitted and accepted to Workshop on Natural User-generated Text (W-NUT) in Empirical Methods in Natural Language Processing (EMNLP), 2026
Abstract:We investigate the use of Large Language Models (LLMs) in multi-author style change detection. Using off-the-shelf LLMs or supervised fine-tuning solely within the target domain fails to yield effective features; for this reason, we propose cross-domain transfer learning with LLM-based embeddings fine-tuned for authorship verification. We fine-tune Mistral-7B embedding model and compare it with DeBERTa, used in the state-of-the-art (SOTA) systems in PAN-25 multi-author writing style analysis challenge. Results show that DeBERTa and Mistral excel in different user-generated text scenarios. We exploit this diversity by combining both embeddings, leading to more effective style change detection.
[NLP-240] Prosodic Structure as a Critical Component in Phonemic Language Modelling at BabyLM Scale EMNLP2026
【速读】： 该论文旨在解决当前语言模型普遍依赖书面语料（orthographic text）进行训练，而忽视了人类婴幼儿在习得语言初期所依赖的连续性、韵律结构化语音输入（prosodically structured speech）这一生态学上的不一致问题。其核心挑战在于：在小数据条件下（如婴儿早期语言学习阶段），如何有效建模语音层面的语义与句法组织能力。论文提出的关键解决方案是引入韵律标注（prosodically annotated phoneme stream）作为输入信号，构建一个从低到高的音系丰富度梯度（ladder of phonological richness），并在此基础上训练基于BERT MLM架构的音素级语言模型（phonemic language modelling）。实验结果表明，在10M和100M词规模下，仅使用韵律标注的音素流模型在新设计的语义组织任务中表现出最强的分类组织能力，显著优于仅依赖音素或边界结构的模型；同时，通过引入对抗性反向条件测试，发现传统基于伪对数似然（pseudo-log-likelihood）的评估指标可能高估了模型对“词性”（wordhood）的理解，实际反映的是音位构形熟悉度（phonotactic familiarity）。因此，研究证实韵律结构是小数据环境下音素级语言建模的关键成分，有助于提升模型在类婴儿学习情境中的生态有效性。
链接: https://arxiv.org/abs/2610.24172
作者: Bill K. S. Ku,Janet B. Pierrehumbert
机构: University of Oxford(牛津大学)
类目: Computation and Language (cs.CL)
备注: 17 pages, 3 figures. Accepted at the BabyLM Workshop, EMNLP 2026 (archival track)
Abstract:Human children learn language from continuous, prosodically structured speech long before they can read, yet almost all language models train on orthographic text. Building on previous phonemic BabyLM training (Goriely et al., 2024; arXiv:2410.22906), we ask whether supplying an estimate of the prosody infants exploit improves phonemic language modelling at BabyLM scale. We compare an orthographic baseline to three conditions forming a ladder of phonological richness. All conditions are trained as BERT MLMs at 10M and 100M words. Addressing the ecological validity of existing metrics in the small-data regime, we build a lexical decision task and two semantic organisation tasks. Our new lexical decision task offers improved insights into the distinction between wordlikeness, token familiarity, and lexicality as factors. Triphone phonotactic control and pseudo-log-likelihood scoring shows that these scores reflect phonotactic familiarity rather than wordhood: every model sits near or below chance on an adversarial inverted condition, revealing prior results of this kind as partly artefactual. On our new semantic tasks, a prosodically annotated phoneme stream gives the strongest taxonomic organisation at both the 10M and 100M word scales. Phonemic models with boundary structure are competitive with the orthographic baseline. These findings suggest that prosodic structure is a key component of phonemic language modelling in the small-data regime most analogous to infant learning.
[NLP-241] Pronouncing Hard Words: Phoneme Adaptation of a Pretrained Conversational Text to Speech Model
【速读】： 该论文旨在解决生成式语音代理（generative voice agents）在医疗场景中对罕见且关键的药物名称发音控制不足的问题。现有基于端到端语音合成（text-to-speech, TTS）的对话式语音模型虽能实现自然流畅的语调适应，但因药物名称词汇量庞大、更新迅速且出现频率极低，难以通过常规预训练、中段训练或后训练阶段充分覆盖，导致其发音准确性差。为此，作者提出一种无需任何预训练或中段训练的解决方案：在已训练的对话式TTS模型基础上，引入ARPAbet音素词汇扩展，结合持续的音素增强型监督微调（phoneme-augmented supervised fine-tuning, SFT），并在推理阶段采用无训练的声学偏置（acoustic-biasing）策略，实现对特定词汇发音的显式控制。实验表明，在包含7500个临床药物发音片段的评估中，该方法将严格由Gemini模型判定的正确率从54.6%提升至90.7%，显著优于所有对话式及无状态基线模型；由于整个评估集完全为域外数据，未参与任何训练，性能提升反映的是模型对发音规则的泛化学习而非记忆。在盲测随机人类评估中，84%的生成音频被判定为正确，优于最强基线的82%，且置信区间重叠，验证了其实际可用性；同时，人工多数意见与Gemini判断的一致性达到84.8%（kappa=0.57），表明该方法在真实用户感知层面具备可靠性。核心创新在于不依赖训练即可实现对高风险专有词发音的精准控制。
链接: https://arxiv.org/abs/2610.24164
作者: Yashish M. Siriwardena,Dinesh K. Gnanasekaran,Tanmay Laud,Shinji Watanabe,Subhabrata Mukherjee,Shanil Puri
机构: 未知
类目: Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注: 5 pages
Abstract:Voice agents built as cascaded automatic speech recognition (ASR)-large language model (LLM)- text-to-speech (TTS) pipelines increasingly rely on conversational TTS: streaming, context conditioned models that adapt delivery to dialogue history. Such models are natural and expressive but offer little explicit control over how rare, trust-critical drug names are pronounced, because these entities are too infrequent, and their vocabulary too fast-growing, to be covered by any amount of pre-, mid-, or post-training. We retrofit phoneme controllability onto a pretrained conversational TTS model without any pre/mid-training, using an ARPAbet vocabulary exten-sion, continued phoneme-augmented supervised fine-tuning (SFT), and a training-free inference-time acoustic-biasing step. On a 7,500 clip clinical drug-pronunciation evaluation, our model raises strict Gemini-judged correctness from 54.6% (no phoneme augmentation) to 90.7%, outperforming every conversational and stateless TTS baseline on the automatic metric. Because no training stage sees a drug name, the entire evaluation is out-of-domain, so the gains reflect learned pronunciation rather than memorization. In a blinded, randomized human study, adult listeners judge 84% of our clips correct, versus 82% for the strongest baseline, with overlapping confidence intervals on a stratified subset. Across all 250 matched clips, Gemini agrees with the human majority on 84.8% of binary labels (kappa=0.57).
[NLP-242] CoDA: A Human-LM Collaborative Codebook Development Assistant for Domain-Specific Text Annotation
【速读】： 该论文旨在解决在缺乏高质量标注指南（codebook）的情况下，如何高效、准确地利用语言模型（LM）对非结构化文本进行领域特定信息提取的问题。现有语言模型标注系统普遍假设标注指南已存在且质量较高，但实际中构建此类指南耗时费力，尤其对于依赖专家知识的复杂概念而言更为困难。其解决方案的关键在于提出CoDA——一个开源的人机协同代码本开发助手，通过与领域专家协作，基于对语言模型错误预测的针对性审查，并将专家反馈动态融入代码本的迭代更新过程，从而实现高效生成高质量代码本。实验表明，CoDA在两个真实场景（流浪者管理记录中的健康指标提取与社交媒体上野生动物非法交易帖子的情感分类）中，生成的代码本可达到与人工专家制定的代码本相当的下游分类性能，同时将代码本开发时间从数周缩短至2小时，并减少5倍的手动标注样本需求。该方法显著提升了语言模型在大规模非结构化文本中提取领域洞察的能力。
链接: https://arxiv.org/abs/2610.24157
作者: Jaspreet Ranjit,Swaminathan Chellappa,Augusto Rivas Costante,Chandan Manjunath,Laura Onasch-Vera,Hailey Winetrobe Nadel,Eric Rice,Swabha Swayamdipta
机构: University of Southern California(南加州大学); University of North Carolina(北卡罗来纳大学); My Friend’s Place(我的朋友之家)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations
Abstract:We present CoDA, an open-source human-LM collaborative codebook development assistant that helps domain experts develop annotation guidelines for annotating new variables in unstructured text with language models. Existing language model annotation systems primarily focus on downstream classification performance, while assuming the availability of high-quality annotation guidelines, or codebooks. In practice, developing these codebooks is time-consuming and challenging, particularly for concepts that require expert knowledge. CoDA addresses this gap by collaborating with domain experts to iteratively develop codebooks through targeted review of incorrect language model predictions and incorporating expert feedback into the evolving codebook. We demonstrate CoDA on two real-world case studies: extracting health indicators from homelessness case management notes and classifying sentiment of wildlife trafficking posts on social media. Across both domains, CoDA produces codebooks that achieve downstream classification performance comparable to expert codebooks while reducing codebook development time from several weeks to two hours and requiring 5X fewer manually annotated examples. Applying CoDA to generate annotations at scale, we identified 2,410 case notes that contained health discussions or referrals and 10,196 social media posts expressing negative sentiment that may warrant investigation for potential wildlife trafficking. By accelerating codebook development, CoDA helps practitioners use language models to extract domain-specific insights from unstructured text at scale.
[NLP-243] Continual Test-Time Learning with Evolution Strategies
【速读】： 该论文旨在解决大语言模型在开放式环境中进行持续学习时面临的无监督持续适应问题，尤其是现有测试时自适应方法因依赖单一策略生成的自我监督信号而导致的策略漂移、推理能力退化和灾难性遗忘。其核心挑战在于：当模型不断更新时，若自监督信号来自自身（即单个策略），则该信号会随策略漂移而失效，进而导致性能崩溃。为应对这一问题，论文提出进化策略强化学习（Evolution Strategy Reinforcement Learning, ESRL），其关键创新在于：将自生成监督信号的来源从单一策略扩展为一个由扰动锚点政策生成的策略群体。具体而言，ESRL通过在参数空间中对锚定策略进行扰动生成一组邻近策略，从中采样轨迹并聚合其输出作为无标签的相对学习信号，用于指导中心学生模型的更新。锚定策略在每个训练块内保持固定，并在块边界处由学生模型刷新，从而确保群体始终围绕一个稳定且当前的策略分布，持续提供多样化且可靠的监督信号。实验表明，该方法在数学推理、通用知识及医学领域等单任务测试时适应场景中，相较于基线模型平均准确率提升最高达31.7%，相比最强的测试时适应基线提升16.6%，显著缓解了灾难性遗忘，并实现了持续学习下的最优最终性能。
链接: https://arxiv.org/abs/2610.24137
作者: Guangya Hao,Yiting Duan,Longyan Tan,Yunbo Long,Hao Wu,Ge Zheng,Alexandra Brintrup,Zhuokai Zhao
机构: University of Cambridge(剑桥大学); Western Sydney University(西悉尼大学); The Alan Turing Institute(艾伦·图灵研究所); University of Chicago(芝加哥大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Large language models increasingly operate in open-ended environments, where they are expected to continually adapt to evolving tasks and distributions, often without access to external supervision. Existing continual learning approaches rely on external supervision, such as labeled data, demonstrations, or reward signals, which can be costly or unavailable in practice. Test-time adaptation removes this dependence by learning from self-generated supervision, but existing methods derive that supervision from the single policy being updated, so once the policy drifts, its own learning signal drifts with it, leading to policy collapse, degraded reasoning, and catastrophic forgetting. In this paper, we propose \textbfEvolution Strategy Reinforcement Learning (ESRL), a continual test-time learning framework that draws its self-generated supervision from a population of policies instead of the single policy being trained. ESRL perturbs an anchor policy in parameter space to form a population of nearby policies, samples one trajectory from each, and aggregates their answers into label-free group-relative learning signals for the central student. The anchor is held fixed within each block and refreshed from the student at block boundaries, so the population stays centered on a stable yet current policy and continues to supply diverse and reliable supervision as the student moves. In single-task test-time adaptation across mathematical reasoning, general knowledge, and medical domains, ESRL improves average accuracy by up to 31.7% relative to the base model and 16.6% relative to the strongest test-time adaptation baseline. Notably, ESRL substantially reduces catastrophic forgetting and achieves the strongest final performance in continual learning.
[NLP-244] Inside the Loop: A Mechanistic Study of Weight-Tied Transformers on Depth-Bound Algorithmic Tasks NEURIPS2026
【速读】： 该论文旨在解决大规模序列化算法任务中因深度扩展或思维链（chain-of-thought）长度增加所带来的参数量和推理计算成本急剧上升的问题。现有方法如深层Transformer或长思维链虽能提升模型表现，但分别面临参数冗余与计算开销过高的挑战。为此，论文提出并研究了环状变压器（looped transformer），其通过重复使用单一层块实现有效深度的扩展，从而在不增加参数量或序列长度的前提下提升模型能力。研究的关键在于揭示环状结构的本质作用机制：通过设计有向排列遍历任务（directed permutation traversal），该任务每一步中间节点均有明确真值标签，可精确追踪模型在迭代过程中的状态演化。实验发现，仅当达到最小有效深度时，模型才能完成任务；而在此基础上，权重共享程度成为决定性能的关键因素——相同有效深度下，环状变压器在所有随机种子中均能成功解题，而等效的非共享（untied）模型则完全失败，除非采用更低的学习率或对中间层进行监督。通过对残差流的探针分析与干预，进一步揭示环状模型在迭代过程中逐步推进至各中间节点，形成适应迭代预算的“阶梯式”状态演化路径，而非共享模型仅在最终层才显现出中间状态。因此，研究结果将环状变压器重新诠释为迭代程序学习器，能够将有限的迭代次数合理分配给算法的各个步骤，从而解决同等有效深度下非共享模型无法处理的任务。
链接: https://arxiv.org/abs/2610.24135
作者: Magnus Sesodia,Juil Sock,Philip Torr,Christian Schroeder de Witt
机构: University of Oxford(牛津大学); BBC RD(英国广播公司研发部)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Accepted to NeurIPS 2026. 17 pages, 12 figures
Abstract:Solving algorithmic tasks with many sequential steps typically requires scaling up transformer depth or chain-of-thought length, incurring unfavorable parameter and inference-compute costs, respectively. Looped transformers, which reuse one block of layers across many iterations, are poised to address this, since they scale effective depth without adding parameters or tokens. Yet, it remains unclear what each iteration is actually computing, or whether the loop merely simulates a deeper untied stack. Here, we study directed permutation traversal, in which a model follows H hops from a starting node and reports the node it reaches, so every intermediate hop has a known ground truth. We find that a minimum effective depth is necessary to solve the task, beyond which the degree of weight tying becomes the determining factor. For instance, a looped transformer with a four-layer block iterated eight times solves the task in every seed, while an identically-trained untied transformer of the same effective depth never does, succeeding only with a lower learning rate or supervision of its intermediate layers. Probing and intervening on the residual stream, we find that the looped model advances through the intermediate nodes gradually, forming a staircase that adapts to its iteration budget, whereas the untied model, at that lower rate, exposes them only in its final layers. Taken together, our results recast looped transformers as iterative program learners, which allocate their iterations to successive steps of the algorithm and solve tasks that identically-trained untied models of the same effective depth do not. Code is available at this https URL.
[NLP-245] Prefix-Tuning for Arbitrary Output Sequences on Pretrained Transformers NEURIPS2026
【速读】： 该论文旨在解决生成式AI（Generative AI）中前缀调优（Prefix-tuning）方法的理论可表达性问题，具体而言，其核心问题是：对于一个预训练的自回归Transformer模型，给定任意输入-输出序列对 (x,y)，是否存在一个任务特定的前缀 S，使得将 S 前缀拼接到输入 x 之前，能够确保模型生成目标输出 y？论文的关键解决方案在于证明：在目标输出 y 属于预训练Transformer模型输出空间这一假设下，这样的前缀 S 不仅存在，且可通过求解多个线性系统进行显式构造。同时，论文进一步证明该假设是必要的——当该假设不成立时，前缀可能不存在，从而给出了该问题的完整理论解答。分析过程具有构造性，适用于较为通用的模型架构，并在GPT-2和Qwen3-4B上通过大量实验验证了所提出算法生成的前缀能有效控制模型输出，展现出强大的可控性。
链接: https://arxiv.org/abs/2610.24123
作者: Peter Cho-Ho Lam,Ziyi Wang,Zirui Zhou
机构: Huawei Technologies Canada Co., Ltd.(华为技术加拿大有限公司)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026. Code: this https URL . Online demo: this https URL
Abstract:Prefix-tuning is a parameter-efficient fine-tuning method for large language models (LLMs) that steers model outputs by prepending task-specific prefix representations to the input sequence. Despite its practical success, the theoretical understanding of the expressiveness of prefixes remains incomplete. In particular, the following question remains open: Given a pretrained autoregressive Transformer, for any input-output sequence pair (\mathttx,\mathtty) , does there always exist a prefix \mathcalS such that prepending \mathcalS to \mathttx induces the model to generate the target output \mathtty ? In this work, we prove that under the assumption that \mathtty is in the range of the pretrained Transformer, such a prefix \mathcalS exists and can be explicitly computed by solving multiple linear systems. We further show that this assumption is necessary, in the sense that such a prefix may not exist when the assumption fails, thereby providing a complete answer to the above question. Our analysis is constructive and the obtained theoretical results hold under fairly general assumptions on model architecture. Extensive experiments on GPT-2 and Qwen3-4B demonstrate that our algorithm produces prefixes with strong control over model outputs.
[NLP-246] Do LMs Know What Variables Do? Interpreting Code Representations via Variable Roles
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）中可变角色（Variable Roles）语义表征的可恢复性、稀疏自编码器（Sparse Autoencoders, SAEs）对这些角色区分的保留能力，以及基于角色方向进行生成控制时的正确性保持问题。其核心挑战在于：如何在不破坏生成正确性的前提下，实现对代码中典型行为模式（如累积、遍历、最优候选保留等）的精准识别与可控干预。解决方案的关键在于构建一个涵盖11种角色的静态标注规则体系，并基于跨五种编程语言的126,926个自然代码文件构建基准数据集；通过残差探针（residual probes）评估角色表征的可恢复性，发现其性能优于单个SAE特征和词法基线，且在目标变量重命名后仍保持相对稳定；同时，整合多个SAE特征可显著缩小单特征性能差距，但对角色变更编辑的敏感性下降；尽管配对排序准确率较高，但在固定阈值下存在频繁误报；在生成阶段的干预虽常改变角色标注，但并未带来整体正确性提升，表明当前方法难以在保持生成正确性的同时实现有效的角色控制。该基准通过匹配预测、源码编辑与生成干预三类任务，系统分离了角色可恢复性、SAE保真度与正确性保持控制之间的关系。
链接: https://arxiv.org/abs/2610.24118
作者: Célian Vasson,Daniil Ognev,Kentaro Inui,Benjamin Heinzerling
机构: Sorbonne Université; MBZUAI; RIKEN AIP; Tohoku University
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 46 pages, 25 figures
Abstract:Variable roles describe recurring behaviors such as accumulation, traversal, and retaining the best candidate encountered. We use Sajaniemi’s independently defined taxonomy to study which role distinctions can be recovered from large language models, how sparse autoencoders (SAEs) retain those distinctions, and whether associated directions support control of generation. We introduce static annotation rules for all eleven roles and a corpus of 126,926 natural-code files across five languages. Across five model-SAE systems, residual probes outperform individual SAE features and tested lexical baselines, with modest losses after neutral target renaming. Combining SAE features closes much of the single-feature gap, but probes on SAE reconstructions are less sensitive to role-changing edits reviewed by LLMs. High paired ranking accuracy coexists with frequent false positives at fixed thresholds. During generation, interventions often change role annotations at the cost of correctness; changes accompanied by passing tests show no clear aggregate advantage over random controls. The benchmark separates role-related recoverability, SAE fidelity, and correctness-preserving control through matched prediction, source edits, and generation interventions.
[NLP-247] Joint Decoupled or Agent : When Each Multimodal Alignment Paradigm Wins
【速读】： 该论文旨在解决多模态学习中不同融合范式在不同数据和任务条件下的性能优劣问题，具体探讨联合预训练（JOINT）、解耦预训练（DECOUPLED）与基于大语言模型（LLM）的指令驱动方法（AGENT）三种范式在多种场景下的适用性。其核心挑战在于：在缺乏先验知识的情况下，如何根据任务特性、标注预算及模态表现选择最优的多模态融合策略。解决方案的关键在于提出一种基于诊断信息的两阶段级联选择框架，利用未参与训练的表征诊断（如几何结构、监督表示质量）与文本标签匹配度等指标，结合任务类型和预算元数据，提前预测并选择最优范式。实验表明，该方法在留一组外评估中成功选出胜出范式的比例达71.4%，显著优于仅依赖元数据路由的54.2%，且平均相对遗憾更低（0.028 vs 0.057）。研究揭示了范式偏好受监督预算、任务格式及单模态5-最近邻（5-NN）准确率的影响，证明目标任务诊断信息可有效指导范式选择，从而实现高效、自适应的多模态系统部署。
链接: https://arxiv.org/abs/2610.24108
作者: Yi Fang,Gaurav Srivastava,Sirui Ding,Xuan Wang
机构: Virginia Tech (弗吉尼亚理工学院); Stanford University (斯坦福大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Jointly pretrained multimodal encoders are a common way to combine modalities, while alternatives integrate their information during downstream training or inference. When does each approach win? We compare three paradigms, instantiated with public checkpoints, across 16 paired datasets in five domains and multiple training-label budgets. JOINT trains a fusion head on frozen, jointly pretrained encoders. DECOUPLED trains a fusion head on frozen, independently pretrained encoders. AGENT gives a frozen LLM task instructions, labelled examples, available sample text, and supervised unimodal readouts. No paradigm dominates the complete 144 dataset-budget combinations: DECOUPLED wins 54, JOINT 44, and AGENT 46 by mean utility. ScienceQA, MELD, and MetMeme shift from AGENT to fusion as supervision increases. Between the fusion paradigms, JOINT leads on four of five QA datasets across all evaluated budgets, while higher unimodal 5-NN accuracy is associated with a relative shift toward DECOUPLED across eight classification datasets (Spearman \rho=-0.67 ). We test whether representation and task diagnostics predict these preferences on held-out datasets before fitting their candidate systems. A two-step cascade combines these diagnostics with task and budget metadata to select AGENT or fusion, then JOINT or DECOUPLED. Under leave-one-group-out evaluation, it selects the winner in 71.4% of cells versus 54.2% for metadata-only routing, with mean relative regret 0.028 versus 0.057. Using geometry, supervised representation diagnostics, and text-label matching together yields higher accuracy and lower regret than using any one of these sources with metadata. These results connect paradigm preferences to supervision budget, task format, and unimodal 5-NN accuracy. Target-task diagnostics thus support selecting among the three paradigms before candidate fitting.
[NLP-248] OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理能否在不依赖专家预先重构的前提下，直接从原始客户材料中生成具备专家水平的运筹学（Operations Research, OR）解决方案这一关键问题。现有OR基准多基于自包含描述或预设优化任务，未能真实反映将原始业务文档转化为专业可交付成果的复杂翻译过程。为此，研究提出OR-License——一个涵盖10个工业场景、共31个实例的端到端基准，源自实际付费运筹学项目（总价值368万美元），完整保留了原始需求、输入数据、交付格式及信息边界（经脱敏处理）。为系统评估信息获取能力，引入可控的客户端模拟器ORSim，其仅披露业务事实而不提供运筹学专业知识；通过隐藏式评估验证方案可行性，并与客户认可的原始专家解进行客观质量对比。引入“部署调整质量”（Deployment-Adjusted Quality, DAQ）指标，综合衡量各实例中相对于专家表现的质量，对不可行解赋值为零。实验表明，尽管多个前沿LLM在条件性目标质量上超越专家参考值，但无一达到其DAQ水平；最强的Claude Code主配置实现88.39%严格可行性，DAQuan为0.87，低于专家参考值1.0。分析揭示，信息获取不仅限于澄清阶段，更延伸至解的开发与验证，后期阶段的问题解决确认率更高。轨迹分析识别出建模与实施环节的失败模式，且任务均衡化后的研究链完成率在不同模型间仅为24%–45%。研究结果强调：构建可靠长周期智能代理需持续解决信息鸿沟、忠实进行问题形式化，并建立基于证据的自我改进机制。
链接: https://arxiv.org/abs/2610.24087
作者: Peisong Wang,Feixue Liu,Yize Meng,Bowen Liu,Ruoyu Chen,Gaoyang Dai,Xing He,Aiqian Li,Fengchi Yuan,Aochuan Chen,Chenyi Zi,Yuhan Li,Hongchuan Zeng,Yimo Yan,Nuo Chen,Jie Fu,Zuo-Jun Max Shen,Yong-Hong KUO,Jia Li
机构: HKUST-GZ; HKU; Qingzhi Optimization (Jinan) Information Technology Co., Ltd.; Hunyuan, Tencent; XM Capital; Tianjin Meta Infinite Technology Co., Ltd; CZBANK; American University of Sharjah; Independent Researcher
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Code and Data: this https URL
Abstract:Can large language model (LLM) agents deliver expert-level operations research (OR) solutions directly from original client materials, without prior expert reformulation? Existing OR benchmarks typically begin with self-contained descriptions or predefined optimization tasks, underrepresenting the translation of original business materials into professionally validated solutions. We introduce OR-License, an end-to-end benchmark of 31 instances across 10 industrial scenarios from paid OR engagements totaling USD 3.68 million. It preserves original requirements, input data, deliverable formats, and information boundaries after de-identification. To study information acquisition, we introduce ORSim, a controlled client simulator that discloses business facts without supplying OR expertise. Hidden evaluation checks feasibility and compares objective quality against original expert solutions accepted by clients before benchmark construction. Deployment-Adjusted Quality (DAQ) averages expert-relative quality across instances, assigning zero to infeasible deliveries. Across 11 frontier LLMs using coding-agent harnesses, several exceed the expert reference in conditional objective quality, but none matches its DAQ; the strongest primary Claude Code configuration achieves 88.39% strict feasibility and a DAQ of 0.87 against the expert reference of 1. Client information acquisition extends beyond clarification into solution development and validation, with some later stages showing higher confirmed issue-resolution proportions. Trajectory analysis identifies formulation and implementation failures, while task-equalized research-chain completion rates range from 24% to 45% across models on evaluable tasks. These findings motivate sustained information-gap resolution, faithful problem formulation, and evidence-linked self-improvement for reliable long-horizon agents.
[NLP-249] World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds
【速读】： 该论文旨在解决大规模人工智能（AI）代理在复杂软件生态系统中协同工作时，如何实现高效、可扩展的推理阶段计算资源利用的问题。其核心挑战在于理解并优化多智能体协作机制，以提升整体系统性能，尤其是在面对长期开放性问题和大型项目时。解决方案的关键在于提出并构建了一个名为“Software World”（软件世界）的仿真框架，该框架能够模拟真实且相互依赖的Python包生态系统的多智能体共事过程。通过生成具有不同主题、规模和依赖结构的真实软件环境，该框架支持智能体在高阶目标驱动下自主发现任务并与其他智能体协作，同时可灵活调控通信方式、所有权分配与资源调度等协作要素。实验结果表明，随着推理阶段计算资源的增加，系统性能呈可扩展提升，最优模型使下游软件效率最高提升31%；然而也揭示了部分前沿模型存在过早终止探索、进展停滞的现象。研究释放了完整的仿真框架与运行轨迹数据，为未来智能体协作机制的研究提供了重要基础。
链接: https://arxiv.org/abs/2610.24061
作者: Shannon Zejiang Shen,Yapei Chang,Qiyao Liang,Benjamin Newman,Jenny Liang,Rulin Shao,Ziqiao Ma,Junjie Oscar Yin,Yung-Sung Chuang,Weijia Shi,Yejin Choi,Luke Zettlemoyer,Pang Wei Koh,Mohit Iyyer,Tongshuang Wu,David Sontag
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 33 pages, 9 figures, 7 tables
Abstract:As teams of AI agents take on larger projects and long-standing open problems, agent collaboration offers another dimension for scaling inference-time compute. To understand what makes such collaboration effective and scalable, we develop Software World, a framework for simulating multi-agent coworking in realistic software ecosystems. It comes with a pipeline to create environments of real, interdependent Python packages of different topics, sizes, and dependency structures, which we call software worlds. In a simulation, agents maintain packages in the world with a high-level goal of improving software efficiency, and they are free to discover tasks to work on and collaborate with others. We evaluate agents’ work in a world collectively by measuring speedup in hidden dependent packages. The framework allows us to study different elements of collaboration by varying communication, ownership, and resource allocation. We run simulations in three different worlds, totaling more than 10,000 hours of agent work across eight models. Overall, performance improvement scales with inference-time compute, with the best model improving downstream efficiency by up to 31%. We show that most agents can find meaningful tasks via collaboration, but some frontier models can prematurely stop looking for improvements and make little progress. We release the simulation framework and traces to support future studies of agent collaboration at this https URL.
[NLP-250] LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation
【速读】： 该论文旨在解决当前文本生成图像（Text-to-Image, T2I）模型评估体系中对复杂语义理解能力衡量不足的问题，尤其针对属性绑定、数值推理、空间关系及常识推理等高级认知能力的评估存在局限。其解决方案的关键在于提出一种基于逻辑约束满足问题（Constraint-Satisfaction Problem, CSP）的新评估范式，构建了一个形式化的逻辑框架，明确定义了对象、属性、关系及其在生成图像中需满足的逻辑约束条件（涵盖从单对象-属性到涉及全称量词∀、存在量词∃、计数count及蕴含→等逻辑算子的多对象关系约束）。通过该框架，研究引入LogicVal评估方法，融合神经视觉检测与逻辑推理，实现对生成图像是否满足复杂逻辑约束的精准判定。实验表明，尽管最佳模型在单一对象-属性约束上表现优异（准确率达98%），但在多对象关系约束下性能急剧下降至53%，且在多重约束组合下的组合泛化能力显著退化（从84%降至20%），同时对不可满足约束的识别率仅23%。该研究揭示了现有T2I模型在复杂逻辑语义理解方面的系统性短板，并证明逻辑驱动的评估方法能有效刻画模型的真实能力边界，为未来模型设计与评测提供了可解释、可量化的基准。
链接: https://arxiv.org/abs/2610.24051
作者: Savitha Sam Abraham,Marjan Alirezaie,Hossein Rahnama
机构: City St George’s, University of London (伦敦城市圣乔治大学); Toronto Metropolitan University (多伦多都会大学); Flybits Labs, Creative AI Hub, Toronto (Flybits 实验室，创意人工智能中心，多伦多); MIT Media Lab, Massachusetts Institute of Technology (麻省理工学院媒体实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Logic in Computer Science (cs.LO)
备注: Preprint. Under review. Savitha Sam Abraham and Marjan Alirezaie contributed equally
Abstract:Text-to-Image (T2I) generative models have made significant progress in producing high-fidelity images from natural-language prompts. Alongside this progress, numerous benchmarks have been developed to evaluate capabilities such as attribute binding, numeracy, spatial reasoning, and commonsense reasoning. In this work, we propose a new perspective on T2I evaluation by casting it as a constraint-satisfaction problem. We introduce a logical framework that formally specifies the benchmark domain, including objects, attributes, and relations, alongside the constraints that generated images must satisfy. We evaluated four state-of-the-art T2I models across 9 classes of constraints ranging from a single-object-attribute to multi-object relationships constraints involving logical operators ( \forall, \exists, count, \rightarrow ). Our results reveal a substantial gap between single-object and complex conditional multi-object relationship constraint classes, with the best-performing model achieving 98% accuracy on the former and just 53% on the latter. We assess compositional generalisation, finding that the best-performing model`s accuracy drops from 84% on single-constraint prompts to 20% when satisfying multiple constraints simultaneously. Identifying unsatisfiable constraints is also challenging, with the best model correctly recognizing only 23% of such prompts. Building on this framework, we introduce \textbfLogicVal, an evaluation methodology that combines neural visual detection with logical reasoning to assess constraint satisfaction in generated images. On CLEVR, LogicVal achieves ROC-AUC scores ranging from 0.88 to 0.98, indicating strong alignment with human judgments. These findings underscore the value of logic-based evaluation for characterizing T2I model capabilities and limitations.
[NLP-251] Discuss More Comment Less: Boosting Automated Research via Multi-Perspective Collaborative Review
【速读】： 该论文旨在解决自动化论文生成后，自动化审稿环节中存在的质量控制瓶颈问题。现有方法多依赖单一模型反馈或简单聚合机制，未能显式建模真实同行评审中专家的职责分工与集体讨论过程，导致缺乏有效的交叉质询和可靠的决策依据。其解决方案的关键在于提出一种名为RoundtableReview的审议式多智能体框架，将论文结构化为八个语义槽，并协调六名具备互补视角、检索来源与专业职责的学术与产业界评审者。通过独立评估、交叉质询、圆桌讨论、动态加权投票及主席综合等环节，该框架实现了基于证据且可追溯的反馈生成，在两个自动化研究基线上的实验表明，该方法显著提升了论文质量，并在自动与人工评估中均优于基线，同时减少了所需评论数量。
链接: https://arxiv.org/abs/2610.24050
作者: Junkai Zhou,Jiajia Wu,Zhaoyi Zhang,Shiting Guan,Chuanguang Yang
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:As automated research is increasingly adopted for writing papers, automated review has become a critical link between paper generation and quality control. However, existing methods largely rely on single-model feedback or simple review aggregation, failing to explicitly model expert responsibilities or collective deliberation in actual peer reviews, which limits effective cross-examination and reliable decision-making. Therefore, we formulate paper review as a computational modeling problem of peer review mechanisms and propose RoundtableReview, a deliberative multi-agent framework for enhancing automated research. RoundtableReview structures each paper into eight semantic slots and coordinates six academic and industry reviewers with complementary perspectives, retrieval sources, and specialized responsibilities. Through independent assessment, cross-examination, roundtable deliberation, dynamically weighted voting, and chair synthesis, the framework generates evidence-based and traceable feedback. Experiments on two automated research backbones show that the method improves paper quality in both backbones and outperforms baselines in automatic and human evaluations with fewer comments.
[NLP-252] he Window Is in the Reasoning : When LLM Agents Decide to Follow Prompt Injections
【速读】： 该论文旨在解决大语言模型（LLM）智能体在处理不可信内容时，易受提示注入（prompt injection）攻击的问题，尤其关注现有评估方法和运行时监控机制所依赖的隐含假设——即智能体是否受注入影响在读取注入内容时即已确定。这一假设认为，决策结果在注入被读取的时刻便已固定。然而，本文通过分支回放（branching rollouts）方法对这一假设进行了检验：在智能体读取注入指令的节点处精确重放历史状态，并多次重新采样后续行为路径，从而将结果方差分解为“读取时刻即固定的成分”与“读取后动态决定的成分”。实验结果表明，在四个AgentDojo基准套件中，未启用思考（thinking）的Qwen3智能体有91%的结果方差在读取时刻即已确定；而启用思考后，该比例下降至62%（95%置信区间下降幅度为[-0.38, -0.22]），且在三种攻击模板和两种成功判定标准下均呈现一致趋势。这说明思考机制显著降低了注入攻击的确定性影响。进一步分析发现，读取时刻决定程度是模型固有的属性，且思考虽降低Qwen3的攻击成功率，却提升另外两个模型的攻击成功率。通过对推理前缀（reasoning prefix）的重采样显示，决策过程是逐步形成的。此外，基于低成本模型构建的零样本思维链（zero-shot chain-of-thought）监控器未能捕捉到大量已确定的决策信息，而更强大的监控器及具有前瞻判断能力的“前瞻监控器”（lookahead monitor）则能接近前缀所决定的最终结果。最后，研究指出，启用思考的智能体在单次运行与重跑之间的任务评分一致性仅达65%（即最多35%的不一致率），因此现有依赖单次运行的评估基准需采用重复运行以获得可靠性能指标。该研究的核心解决方案在于揭示了决策动态性，并提出应通过多轮回放、前缀分析与强化监控机制来更准确地识别和防范提示注入风险。
链接: https://arxiv.org/abs/2610.24048
作者: Haoxuan Jia,Yang Liu,Bin Chong,Yingguang Yang,Yancheng Chen,Jiayu Liang,Qian Li,Hanning Lu,Kefu Xu,Hao Zheng,Chongyang Zhang,Yulin Huang,Jianshen Zhang,Yongzhi Qi,Yunuo Lin,Hao Peng
机构: Fullive-AI; Nanyang Technological University; JD.com; Peking University; University of Chinese Academy of Sciences; Soochow University; Beijing University of Posts and Telecommunications; University of Leeds; Beihang University
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注:
Abstract:LLM agents that read untrusted content can be hijacked by prompt injections. Runtime monitors that try to catch a hijack before the agent acts, and benchmarks that score each task with a single run, share an assumption: whether an agent follows an injection is settled when the injection is read. We test this assumption with branching rollouts that replay an episode exactly up to the step at which the agent reads an injection and then resample the rest of the episode many times, splitting the outcome variance into a part fixed at read time and a part decided afterwards. Across four AgentDojo suites, a Qwen3 agent without thinking is largely decided at read time (0.91 of the outcome variance lies between states); with thinking on, this share drops to 0.62 (95% CI of the drop [-0.38, -0.22]), in the same direction for three attack templates and two success labels. How much is fixed at read time is a property of the model that the same measurement reveals, and thinking lowers the attack success rate of Qwen3 but raises it for two other models. Resampling from reasoning prefixes shows that the decision forms gradually as the agent reasons. Zero-shot chain-of-thought monitors built on cheaper models miss much of what is already decided, whereas stronger monitors, and a lookahead monitor that judges simulated next actions, come close to what the prefix determines. Finally, single-run labels of thinking agents disagree with a rerun up to 35% of the time, so benchmarks need repeated runs per task.
[NLP-253] LLM Evolutionary Search Strategies Should be Compared by Tokens
【速读】： 该论文旨在解决当前大语言模型（LLM）进化搜索评估中普遍采用“提案数量”（即候选程序数量）作为衡量标准所导致的误导性问题。研究表明，这一指标掩盖了实际计算成本，因为不同方法在生成相同数量提案时所消耗的令牌（tokens）差异巨大，范围从16k到39k不等。这种隐性差异使得某些方法在同等提案数下实际上获得了高达1.9倍的计算资源优势，而进化搜索相比独立采样甚至获得约6倍的预算优势。通过将评估基准统一为令牌预算，并在控制评分器利用（scorer exploits）的前提下进行对比，发现进化搜索的优势显著减弱：在15个任务-方法组合中，仅有3组保持23%至28%的性能优势，其余12组的差距不超过5%。进一步的复现实验使用Claude Haiku 4.5验证了该结果的可重复性，但同时也表明模型规模对性能表现具有显著影响。研究结论指出，现有系统所宣称的性能优势很大程度上源于评估方式的会计误差。因此，论文提出以“令牌消耗量”而非提案数量或模型调用次数作为核心评估指标，该方案无需额外开销，因日志数据已完整记录令牌使用情况，且能更真实反映计算成本与效率。
链接: https://arxiv.org/abs/2610.23977
作者: Tal Oved,Roi Pony,Oshri Naparstek,Udi Barzelay
机构: IBM Research(IBM研究院)
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:LLM evolutionary search is currently evaluated by counting proposals (candidate programs). We demonstrate this metric hides the true cost of search, and redo the comparison using tokens. By replaying 600 logged runs across 5 tasks and 3 methods, we show that proposal costs vary wildly from 16k to 39k tokens. In equal-proposal comparisons, this silently grants some methods up to 1.9x the compute of others, and gives evolution roughly 6x the budget of independent sampling. When we match token budgets, giving sampling six times more attempts, and remove scorer exploits, the evolutionary advantage largely vanishes. Evolution beats independent sampling by 5% or less on 12 of 15 task-method pairs, retaining a 23 to 28% margin on only 3, all on one task. A replication with Claude Haiku 4.5 reproduces this split, though a third backbone shows its size depends on the model too. The performance margin justifying these systems is often an accounting artifact. Reporting tokens, not proposals and not model calls, fixes this and costs nothing, as the logs already hold them.
[NLP-254] UltraData-Code: From Raw Repositories to Algorithmically Dense and Task-Oriented Code Data
【速读】： 该论文旨在解决大规模语言模型（LLM）在代码生成能力提升过程中，因高质量公开代码数据日益稀缺而导致的训练数据瓶颈问题。随着传统通过扩大语料库规模推动模型性能提升的路径趋于饱和，研究提出采用分层数据管理（tiered data management）策略，对有限的数据资源进行更精细化、目标导向的处理。其核心解决方案在于构建一个四层级（L0–L3）的代码数据体系：L0保留原始仓库数据以支持可复现性；L1通过可扩展的过滤、归一化与近似去重处理，为广泛预训练提供标准化代码；L2利用语言自适应的相关性信号与质量约束，算法筛选出高相关性、高质量的代码文件，生成约4000亿词元的UltraData-Code-L2数据集，用于聚焦式持续预训练；L3则将选定实现转化为结构化的编程任务，产出约1500亿词元的UltraData-Code-L3数据集，适用于任务导向的中段训练或监督微调（SFT）。实验表明，在100亿词元的持续预训练下，相较于L1，L2在EvalPlus和MultiPL-E基准上分别提升7.80和5.13个百分点；若将一半L2数据替换为L3，进一步提升8.42和8.07个百分点，且随着训练预算扩大，增益更加显著。该方案的关键在于通过分层数据设计实现从原始数据到任务专用数据的高效转化，从而在有限数据条件下最大化模型训练效率与性能表现。
链接: https://arxiv.org/abs/2610.23960
作者: Chengying Tu,Hengyu Zhao,Shuaikang Xue,Zhongming Qu,Jihao Zhou,Xinle Lin,Junshao Guo,Zixuan Fu,Qiang Ma,Jie Zhou,Chaojun Xiao,Hongfei Yan,Yudong Wang,Xu Han,Zhiyuan Liu,Maosong Sun
机构: Peking University (北京大学); ModelBest Inc.; Tsinghua University (清华大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Software Engineering (cs.SE)
备注:
Abstract:Code generation has become a core capability of large language models (LLMs), and code data is central to the pre-training process that develops it. As obtaining public code of sufficient quality becomes more difficult, further progress cannot rely on corpus expansion alone. Tiered data management instead applies increasingly targeted processing to smaller data pools, producing data states suited to different training objectives. Following this perspective, we present UltraData-Code as four connected data states spanning repository archival at L0, standardized natural code at L1, algorithmic selection at L2, and task-oriented synthesis at L3. L0 preserves repository data and provenance for reprocessing, while L1 applies scalable filtering, normalization, and near-deduplication for broad pre-training. L2 selects high-quality algorithmically relevant files using language-adaptive relevance signals and quality constraints, yielding approximately 400B tokens of UltraData-Code-L2 across 11 programming languages for focused continual pre-training. L3 converts selected implementations into structured programming exercises, yielding approximately 150B tokens of UltraData-Code-L3 across the same 11 languages for task-oriented mid-training or supervised fine-tuning (SFT). Under controlled 10B-token continual pre-training of a 1B model, L2 improves pass@1 over L1 by 7.80 percentage points (pp) on EvalPlus and 5.13pp on MultiPL-E. Replacing half of the L2 training tokens with L3 yields a further gain of 8.42pp and 8.07pp on the two benchmarks, respectively. Both gains increase when the training budget scales to 100B tokens. We have released our datasets and classifiers to support further research on LLMs.
[NLP-255] How Data Scales in Agent ic Reinforcement Learning: Laws and Synthesis Strategies NEURIPS2026
【速读】： 该论文旨在解决生成式智能体（agentic AI）预训练中一个根本性问题：传统预训练数据规模定律将数据视为单一维度的词元数量（token count），而现代合成流水线可独立生成任务（task）、环境（environment）和轨迹（trajectory）三类结构化数据，且各类型生成成本差异显著。这使得预训练从未面临的核心问题浮现——在给定合成预算下，应优先扩展哪个数据维度？其解决方案的关键在于提出并验证一套多维数据规模定律（multi-axis scaling laws）。研究首次系统分解智能体数据为任务、环境、轨迹三个基数轴及奖励密度控制变量，基于数学+代码、SWE、网页、UI/工具应用等多个领域的大语言模型（LLM）实证分析发现，不同模型规模、架构、算法（PPO与GRPO）下，各轴的缩放指数均呈现统一排序：环境 > 任务 > 轨迹。进一步分析表明，环境多样性优势主要源于世界多样性（world diversity），而环境结构多样性仅相当于任务多样性的水平。在此基础上，研究利用拟合出的缩放规律作为统一标尺，量化各类合成方法的效率（以每轴折扣因子表示），识别出导致性能塌陷的临界合成-真实比值，该比值在不同轴间相差一个数量级。最终，将预算分配建模为受缩放定律与效率折扣约束的优化问题，推导出类似Chinchilla的计算最优配置方案，并在未参与训练的规模上验证，结果表明遵循该方案可实现相较于主流“扩大轨迹”策略高达1.7倍的计算效率提升。
链接: https://arxiv.org/abs/2610.23946
作者: Bowei He,Yankai Chen,Xiaokun Zhang,Changjiang Han,Ye Yuan,Chong Li,Xue Liu
机构: MBZUAI; McGill University; CityUHK; Columbia University
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: This paper has been accepted by NeurIPS 2026
Abstract:Pretraining scaling laws treat training data as a one-dimensional quantity: a token count. Agentic reinforcement learning (RL) inherits this scalar abstraction, but modern synthesis pipelines now generate \textittasks, \textitenvironments, and \textittrajectories independently and at very different costs, turning data into a structured design space. This raises a question pretraining never had to answer: given a synthesis budget, \textitwhich axis should be scaled? We address this question end-to-end with three tightly coupled contributions. First, we decompose agentic data into three cardinality axes (task, environment, trajectory), plus a reward-density control variable, and fit per-axis scaling relations on different large language models (LLMs) across math+code, SWE, web, and UI/tool application domains. Across all model sizes, model families, domains, and RL algorithms (PPO and GRPO) we measured, the scaling exponents order as environment task trajectory. When environment diversity is decomposed into world and environment-structure diversity, its advantage is mainly driven by the former, while the latter is just comparable to the task diversity. Second, we use the fitted laws as a common ruler to price synthesis methods, capturing each method’s efficiency as a per-axis discount factor and identifying critical synthetic-to-real ratios at which collapse begins; these ratios differ by an order of magnitude across axes. Third, we cast budget allocation as a constrained optimization under our laws and discounts, deriving a Chinchilla-style compute-optimal recipe and validating it at held-out scale. The empirical results demonstrate that following the recipe yields up to 1.7\times compute efficiency over the dominant ``scale-the-trajectories’’ practice.
[NLP-256] Latent Multimodal Memory Reasoning
【速读】： 该论文旨在解决长期个性化智能体在跨模态、跨会话场景下对分布式历史信息进行推理时面临的挑战，特别是现有扁平化跨模态方法因过早压缩视觉证据而导致时间滞后的观察失效，且无法有效统一多模态与跨会话证据。其核心问题在于原始多模态记忆具有固有的纠缠性与异质性，导致细粒度视觉定位与跨模态推理计算开销巨大。为应对这一问题，论文提出一种新型神经符号框架——潜在多模态记忆推理（Latent Multimodal Memory Reasoning, LMMR），其关键创新在于将长期多模态记忆组织映射至连续潜在空间：首先，LMMR不依赖于图像到文本的静态转换与检索，而是将文本隐藏状态与视觉补丁特征共同嵌入共享的稀疏自编码器概念空间，通过稀疏激活作为通用语义地址，动态构建逐轮的多模态记忆图；其次，图编码器聚合任务相关的跨模态与时间依赖关系，并将图读出作为连续残差滚动注入生成主干网络，同时保留原始视觉标记以维持细粒度细节。实验在九个Mem-Gallery任务和十二个MemEye设置上验证了LMMR在捕捉复杂多模态依赖关系方面的显著优势，实现了更忠实、具象化的响应生成。
链接: https://arxiv.org/abs/2610.23890
作者: Cai Ke,Xinghao Chen,Yuanchen Bei,Yuni Lai,Keyu Chen,Ruizhi Qiao,Siyu An,Junnan Dong,Ruifeng Xu,Xing Sun
机构: Tencent Youtu Lab; Eastern Institute of Technology; The Hong Kong Polytechnic University; University of Illinois at Urbana-Champaign; Xidian University; Shenzhen Loop Area Institute
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Long-term personalized agents are required to reason over distributed history across modalities and evolving dialogue states. While existing flat cross-modal methods compress visual evidence prematurely and struggle with temporally stale observations, they neglect the necessity to unify multimodal and cross-session evidence. Crucially, raw multimodal memories are inherently entangled and heterogeneous, making fine-grained visual grounding and cross-modal reasoning computationally prohibitive. To this end, we present latent multimodal memory reasoning, i.e., LMMR, a novel neuro-symbolic framework that shifts long-term multimodal memory organization into a continuous latent space. Specifically, (i) instead of converting images to text and relying on static retrieval, LMMR maps both textual hidden states and visual patch features into a shared Sparse Autoencoder concept space, establishing sparse activations as common semantic addresses to dynamically construct a turn-level multimodal memory graph. (ii) A graph encoder then aggregates cross-modal and temporal dependencies across the task-specific latent subgraph, injecting the graph readout as a continuous residual rollout into the generation backbone while retaining native visual tokens to preserve fine-grained details. Extensive experiments on nine Mem-Gallery tasks and twelve MemEye settings demonstrate that LMMR significantly outperforms state-of-the-art baselines in capturing complex multimodal dependencies while enabling faithful and grounded responses.
[NLP-257] ranslating in style: An open-weight alternative to GPT -4 for culturally aligned machine translation
【速读】： 该论文旨在解决机器翻译系统在跨语言转换过程中普遍存在但被忽视的风格扁平化问题，尤其是在翻译至非西方语言时，译文往往丢失原文的语体特征（如正式度、礼貌性与亲密感）。现有解决方案RASTA框架（Havaldar et al., 2025a）虽通过检索增强的风格对齐有效提升了风格保真度，但其依赖闭源专有模型GPT-4，限制了可复现性与开放应用。本文的关键在于探索是否可使用开源权重的专用翻译模型Hunyuan-MT-7B替代GPT-4，构建一个开放可访问的风格对齐翻译流程。研究对比了该模型的三个变体：基础模型、基于监督微调的LoRA适配器以及采用GRPO策略优化的策略模型，在三个风格维度（正式度、礼貌性、亲密性）和十二个方向的语言对上进行评估，采用相关性指标A衡量风格对齐效果。结果表明，微调并未显著提升整体风格对齐性能，三者在统计上无显著差异；尽管开源基线整体表现不及RASTA-GPT-4，但在正式度方面存在差距，但在亲密性及中文相关方向上，其性能已达到甚至超过GPT-4，尤其在个别语言对上表现优异。因此，基于开源模型的可访问式RASTA流程已在部分场景下具备可行性，标志着开放风格对齐翻译系统迈向实用的重要一步。
链接: https://arxiv.org/abs/2610.23868
作者: Thomas Rietman,Hadi Mohammadi,Robert A. Bagheri
机构: Utrecht University (乌得勒支大学)
类目: Computation and Language (cs.CL)
备注: 10 Pages excl references, 5 tables, 1 Figure
Abstract:Machine translation systems often preserve meaning but systematically flatten linguistic style, a failure that is most acute when translating into non-Western languages. The RASTA framework of Havaldar et al. (2025a) addresses this through retrieval-augmented stylistic alignment, but relies on GPT-4, a closed, proprietary model. We investigate whether an open-weight translation specialist, Hunyuan-MT-7B, can serve as an open-weight alternative. We evaluate three variants of the model, the base model, a LoRA supervised fine-tune, and a GRPO-steered policy. We do this across three style dimensions (formality, politeness, and intimacy) and twelve directed language pairs, measuring style alignment with the correlation-based metric \mathcalA . We find that fine-tuning does not reliably improve alignment: the three variants are statistically indistinguishable in aggregate. The open backbone does not match RASTA-GPT-4 overall, but the shortfall is concentrated in formality and largely closes for intimacy and Chinese directions, where the open model matches or surpasses GPT-4 on individual pairs. An accessible RASTA pipeline built on an open model is thus already viable for parts of the problem.
[NLP-258] Composable Parametric Demographic Injection for Values Survey Simulation
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在计算社会科学中作为模拟受访者时，难以准确再现特定人群异质性响应分布的问题。现有方法如基于人口统计学的提示（demographic prompting）对提示设计敏感且增加推理开销，而基于参数的方法则难以高效复用人口统计学信息，导致对未见组合的泛化能力受限。针对上述挑战，论文提出PARADE——一种可组合的参数化人口统计学注入框架，其核心在于将每个人口统计属性值表示为可复用、可组合的低秩适配器（LoRA）模块。通过分离任务通用的调查能力与人口统计学特异性偏移，PARADE采用共享任务LoRA和属性级人口统计学LoRAs，并引入组合校准机制以减少参数干扰。在推理阶段，可通过重新组合相关模块动态构建目标人群模型，支持训练中未见过的人口统计组合。在世界价值观调查（World Values Survey）和SocioBench数据集上，基于三种主干模型的实验表明，PARADE在保持良好预测性能的同时，显著提升了群体层面的分布对齐效果，相较最强基线最高降低52.8%的KL散度。
链接: https://arxiv.org/abs/2610.23867
作者: Xinyuan Cao,Qingyi Pan,Bangde Du,Yujia Zhou,Hexi Wang,Weihang Su,Yiqun Liu,Yueyue Wu,Min Zhang,Qingyao Ai
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) are increasingly used as simulated respondents in computational social science, yet accurately reproducing population-specific heterogeneous response distributions remains challenging. Existing methods such as demographic prompting are sensitive to prompt design and increase inference overhead, while parameter-based methods struggle to reuse demographic information efficiently, limiting generalization to unseen combinations. In this regard, we propose PARADE, a composable PARAmetric DEmographic injection framework that represents each demographic attribute value with a reusable and composable LoRA module. Specifically, PARADE separates task-common survey capability from demographic-specific value shifts via a shared Task LoRA and attribute-level Demographic LoRAs, with composition calibration to reduce parameter interference. At inference time, target population models can be constructed on demand by recombining the relevant modules, including demographic combinations not observed during training. Extensive experiments on the World Values Survey and SocioBench across three backbone models show that PARADE achieves strong population-level distribution alignment while maintaining competitive predictive performance, reducing KL divergence by up to 52.8% relative to the strongest baseline.
[NLP-259] GardenBench: A lightweight daily evaluation of LLM capabilities
【速读】： 该论文旨在解决现有大型语言模型（Large Language Models, LLMs）评估基准在动态性能监测与实际应用场景覆盖方面的不足，特别是对模型每日性能波动、推理成本与响应时间等关键指标的忽视。传统基准多聚焦于静态任务评估，难以捕捉模型在真实使用中随时间演变的准确率变化、性能退化或更新带来的影响。为此，论文提出GardenBench，一个包含11个测试套件的轻量级、低成本评估基准，通过系统性地向LLMs注入多样化轻量查询，持续监控其在日常任务中的表现，涵盖准确性、响应时间与运行成本等维度。其解决方案的关键在于构建一个面向消费者、可实时更新的动态评估框架，结合公开可访问的在线仪表盘，实现对不同模型提供商及版本在真实使用场景下的横向对比，从而更全面地反映模型的实际能力演进，推动评估方法从静态向动态、从封闭向开放转变。
链接: https://arxiv.org/abs/2610.23866
作者: Benedict Cummins-Mburu,Mariana Garcia Mejia,Oscar Heath,Sabrina Kreyzerman,Ellie Murray,Arusan Surendiran,Rohan Alexander
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:We introduce GardenBench, a benchmark with 11 test suites that evaluate the capabilities of Large Language Models (LLMs) across a range of tasks. By systematically prompting LLMs with various lightweight queries, we document day-to-day changes in model performance, including run costs and times, that other benchmarks miss. This allows us to compare the accuracy and features of responses across model providers and versions. We account for daily variations in model accuracy, capturing both model progress and degradation, for instance, when a model is changed post-deployment or when a new model is released. We base our evaluation on general-use tasks and make our results publicly accessible via a live dashboard, diversifying existing approaches to LLM testing. Our main contribution is an innovative, lightweight, and inexpensive benchmark, designed to be consumer-facing and for everyday tasks.
[NLP-260] Modeling Human Behavior in the Prisoners Dilemma with Large Language Models
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在重复囚徒困境博弈中能否有效模拟人类合作行为的问题，尤其关注开源大语言模型（LLM）在具备记忆能力与面对对手随机轮换情境下的决策表现。其核心挑战在于现有研究普遍缺乏对模型记忆机制的支持、可复现的开源模型对比以及与真实人类实验数据的直接验证。解决方案的关键在于构建一个标准化的提示框架，系统评估五种开源大模型（TII Falcon 7B Instruct、Google Gemma 7B、Meta Llama 3.1 8B Instruct、Microsoft Phi Mini MoE Instruct、Qwen 2.5 7B Instruct）在两种典型交互环境中的表现：固定搭档（双方均具记忆）与随机轮换搭档（记忆型模型面对无记忆对手）。研究通过100轮博弈中每10轮的协作率变化进行量化分析，并与蒙特罗-波拉斯等人（2022）的人类实证数据进行对比。结果表明，所有模型均显著偏离人类协作水平，普遍表现出过度合作现象，尤其在随机轮换情境下，人类平均协作率为28.5%，而多个模型协作率超过80%；其中Qwen-2.5-7B-Instruct呈现确定性行为模式，而Phi-mini-MoE-Instruct则展现出更高变异性并短暂逼近人类水平。这表明尽管开源大模型具备作为行为模拟代理的潜力，但在模拟真实人类社会互动方面仍存在显著偏差，尚未能可靠替代人类参与者。
链接: https://arxiv.org/abs/2610.23860
作者: Raina Gao,Tristan Raffo
机构: 未知
类目: Computation and Language (cs.CL); Computer Science and Game Theory (cs.GT)
备注:
Abstract:The Prisoner’s Dilemma has long served as a foundational framework for studying human decision-making and cooperative outcomes across social scenarios, yet human behavior consistently deviates from predictions of fully rational economic theory. Previous works have studied human and LLM outcomes in the Prisoner’s Dilemma; however, these studies have lacked at least one key feature: a memory component for LLM agents, accessible open-source models, or comparison against empirical human data. This project investigates whether open-source LLMs, TII Falcon 7B Instruct, Google Gemma 7B, Meta Llama 3.1 8B Instruct, Microsoft Phi Mini MoE Instruct, and Qwen 2.5 7B Instruct, can model human cooperation patterns in repeated Prisoner’s Dilemma games, while examining how an LLM’s knowledge that its opponent will be shuffled affects its decisions. Using a standardized prompting framework, we simulated interactions in two trial types: a fixed-partner setting where both agents retain memory, and a shuffled-partner setting where a memory-enabled agent, informed that its opponent changes across rounds, interacts with a memoryless agent. For each trial, we measured cooperation rates over 100 rounds and 10-round increments and compared them to human data from Montero-Porras et al. (2022). Under both frameworks, all LLMs differed from the human cooperation rate. Open-source LLMs generally over-cooperated relative to humans, particularly in the shuffled framework, where human cooperation averaged 28.5% while several models exceeded 80%. Qwen-2.5-7B-Instruct showed deterministic tendencies, while Phi-mini-MoE-Instruct displayed greater variability and briefly approximated human rates. These findings suggest that LLMs have potential as behavioral simulation agents and provide a benchmark across two interaction environments, although open-source models are not yet reliable substitutes for human participants.
[NLP-261] I2-Med: Interpretable Medical Inference through Visual-Guided Dynamic Logits Calibration
【速读】： 该论文旨在解决医学视觉语言模型在医学图像理解与推理任务中面临的可解释性与可靠性不足问题，核心挑战在于实现精确的视觉定位（visual grounding）和临床一致性的解释生成，而这一目标因复杂的影像学模式及稀缺的专业标注数据而难以达成。传统推理方法在此背景下易产生视觉幻觉（visual hallucination），即生成内容脱离图像实际信息，严重制约其在临床实践中的应用。本文提出的可解释医学推理框架I²-Med，其关键创新在于无需额外训练即可增强推理过程中的视觉忠实性：在文本生成阶段，引入一个视觉引导模块，实时评估候选词汇是否被医学图像支持并符合当前推理逻辑，据此动态调整输出分数，优先选择与图像一致的语义项，从而提升生成内容的视觉可信度与临床合理性。实验基于OmniMedVQA基准，在八种医学成像模态与五类临床问题上验证了该框架的有效性，显著提升了推理链的临床意义与视觉一致性，同时减少了视觉幻觉现象。
链接: https://arxiv.org/abs/2610.23858
作者: Md Sajid,Shreeyut Maheshwari,Akshat Mishra,Ritik Mishra,Mohammad Tanveer
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Medical vision language models have demonstrated strong potential for medical image understanding and reasoning. However, their interpretability and reliability remain limited because medical vision language tasks require precise visual grounding and clinically coherent explanations. This is particularly challenging due to complex imaging patterns and scarce expert annotations. These challenges reduce the effectiveness of conventional reasoning methods and lead to visual hallucination, defined as the generation of text that is not grounded in the given visual input, which further limits their applicability in clinical practice. To mitigate these challenges, we propose an interpretable medical inference (I ^2 -Med) framework that enhances visual faithfulness during inference without additional training to improve reasoning and explanation faithfulness in medical vision language models. During text generation, I ^2 -Med uses a visual guidance module to check whether each candidate word is supported by the medical image and consistent with the current reasoning. Based on this guidance, the model’s output scores are adjusted to favor visually grounded words, without changing the model parameters. Extensive experiments on the OmniMedVQA benchmark, covering eight medical imaging modalities and five clinical question types, demonstrate that our proposed I ^2 -Med framework generates more clinically meaningful and visually faithful reasoning chains and reduces visual hallucination under standard medical VQA evaluation metrics. The source code is publicly available at \urlthis https URL.
[NLP-262] ransformer-based segmentation of prosodic boundaries in Brazilian Portuguese
【速读】： 该论文旨在解决巴西葡萄牙语（Brazilian Portuguese, BP）中自动韵律切分（automatic prosodic segmentation）性能不足的问题，现有方法仍主要依赖规则或传统机器学习，难以充分捕捉复杂语言特征。其解决方案的关键在于提出一种基于Whisper large-v3的端到端切分模型SAMPA，通过在人工标注的NURC-SP数据集上微调，并在语音转录过程中显式插入韵律边界标记，实现对韵律边界的精准识别。实验表明，SAMPA在不同训练与测试配置下均表现出色，尤其在保留测试集上达到F1=0.731，而在跨数据集的分布外测试（MuPe-Diversidades）中更取得F1=0.796，显著优于以往方法。进一步的n-gram及声学-视觉分析揭示，模型能够有效利用形态句法、语义和韵律线索进行边界检测，验证了其对语言结构的深层理解能力。
链接: https://arxiv.org/abs/2607.07408
作者: Rodrigo de Freitas Lima,Julio Cesar Galdino,Marcos Vinicius Treviso
机构: University of São Paulo (圣保罗大学); Instituto Superior Técnico (里斯本高等技术学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 6 pages, 5 figures, submitted to an IEEE conference
Abstract:Automatic prosodic segmentation identifies boundaries between speech units from acoustic and linguistic evidence. Although recent deep learning approaches have produced strong results for English, automatic segmentation for Brazilian Portuguese (BP) still relies mostly on rule-based or traditional machine-learning methods. This paper presents SAMPA, a Whisper-based segmenter that transcribes BP speech while inserting explicit markers for terminal prosodic boundaries. We fine-tune Whisper large-v3 on manually segmented recordings from the NURC-SP dataset and evaluate different training and test-time filtering configurations, including out-of-distribution testing on the MuPe-Diversidades dataset. SAMPA achieves competitive boundary-detection performance across settings, with the best models reaching F1=0.731 on the held-out test split and F1=0.796 on MuPe-Diversidades. Finally, through n-gram and acoustic-visual analyses, we show that our model follows morphosyntactic, semantic, and prosodic cues for detecting prosodic boundaries.
[NLP-263] ransfer Learning for Power Outage Detection Task with Limited Training Data
【速读】： 该论文旨在解决电力系统中停电事件早期检测的难题，尤其针对标注数据有限场景下的模型性能瓶颈。其核心挑战在于如何在缺乏大量人工标注样本的情况下，实现对停电事件的高效、准确识别。解决方案的关键在于利用预训练语言模型（Language Models）结合迁移学习（transfer learning）与少样本学习（few-shot learning）策略，通过在大规模通用语料上预训练模型，并在少量标注的社交媒体文本数据上进行微调，从而有效提升模型在未见类别上的泛化能力。实验结果表明，尽管零样本学习表现逊于传统模型，但经过少量数据微调后，BERT和GPT等模型的准确率显著提升，分别达到81.3%和74.5%，验证了少样本微调在资源受限条件下的有效性。这一方法为基于自然语言处理技术的基础设施监控提供了可行路径，尤其适用于实时性要求高且标注成本昂贵的应用场景。
链接: https://arxiv.org/abs/2305.17817
作者: Olukunle Owolabi
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Applications (stat.AP)
备注:
Abstract:Early detection of power outages is crucial for maintaining a reliable power distribution system. This research investigates the use of transfer learning and language models in detecting outages with limited labeled data. By leveraging pretraining and transfer learning, models can generalize to unseen classes. Using a curated balanced dataset of social media tweets related to power outages, we conducted experiments using zero-shot and few-shot learning. Our hypothesis is that Language Models pretrained with limited data could achieve high performance in outage detection tasks over baseline models. Results show that while classical models outperform zero-shot Language Models, few-shot fine-tuning significantly improves their performance. For example, with 10% fine-tuning, BERT achieves 81.3% accuracy (+15.3%), and GPT achieves 74.5% accuracy (+8.5%). This has practical implications for analyzing and localizing outages in scenarios with limited data availability. Our evaluation provides insights into the potential of few-shot fine-tuning with Language Models for power outage detection, highlighting their strengths and limitations. This research contributes to the knowledge base of leveraging advanced natural language processing techniques for managing critical infrastructure. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Applications (stat.AP) Cite as: arXiv:2305.17817 [cs.CL] (or arXiv:2305.17817v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2305.17817 Focus to learn more arXiv-issued DOI via DataCite
[NLP-264] acher Supervision Fixed Representations and Calibration: Exact Risk Calculations for Candidate-Scoring Decision Models
【速读】： 该论文旨在解决生成式 AI（Generative AI）中候选答案评分模型在知识迁移过程中因教师模型偏差、固定表征与概率校准不当所导致的决策风险问题。其核心挑战在于：即使采用大规模教师模型生成软标签（soft targets）、冻结预训练主干网络（frozen pretrained backbone）以及温度校准（temperature calibration）等常见训练策略，学生模型仍可能继承教师模型的系统性偏差，且这些偏差无法通过增加数据量或与教师的一致性来识别或消除。解决方案的关键在于：首先，明确指出在高斯代理模型中，固定表征仅在样本量超过特定阈值时才能降低风险；其次，证明在probit模型下，精确的概率校准映射为温度缩放，但该方法仅当教师系数方向正确时才能完全消除过量风险，否则会残留不可被诊断工具检测的偏差；再次，基于Luce选择公理，提出候选得分在事件细化下保持一致性的充要条件是其聚合方式必须为log-sum-exp形式，从而定义了概率一致性（coherence）的标准。实证分析显示，在231,521条消费者金融保护局信用卡投诉数据上，以教师概率为监督信号的学生模型维持了教师的决策成本，而以真实结果为监督信号的学生模型则实现8%的成本下降；即使使用300亿参数的零样本模型，其校准后性能仍劣于教师模型，且其违反一致性条件的程度高达17.5德比安（decibans）。因此，尽管预训练表征可减少标注数据需求，校准能修正概率尺度，但二者均无法纠正有偏教师带来的系统性误差，故必须将模型输出的概率与实际发生结果进行验证，以确保决策可靠性。
链接: https://arxiv.org/abs/2610.26797
作者: Agus Sudjianto,Denis Burakov
机构: 未知
类目: Machine Learning (stat.ML); Computation and Language (cs.CL); Machine Learning (cs.LG); Risk Management (q-fin.RM)
备注: 18 pages, 3 figures, 4 tables. Code in ancillary files
Abstract:Candidate-scoring decision models return a probability distribution over a presented set of answers. Public documentation of one such system, Jev, and of an open counterpart, JevAny, describes three training ingredients that determine what these probabilities estimate: soft targets from larger models, a frozen pretrained backbone and temperature calibration. We analyze each ingredient in Gaussian surrogates where the finite-sample risk can be computed exactly. A student trained on teacher targets inherits the teacher’s bias, which more student data does not remove and agreement with the teacher does not reveal. A fixed representation lowers the risk only below an explicit sample-size threshold. In a probit model the exact recalibration map is a temperature scaling, which removes all excess risk only when the teacher’s coefficient has the correct direction and otherwise leaves a residual that calibration diagnostics do not detect. Under Luce’s choice axiom, candidate scores are coherent across refinements of an event if and only if they aggregate by log-sum-exp. On 231,521 credit card complaints from the Consumer Financial Protection Bureau, where the decision is whether a complaint ends in monetary relief, a student trained on a logistic teacher’s probabilities ends at the teacher’s decision cost while a student trained on outcomes goes 8% below it, a 30B model read zero-shot remains worse than the teacher after recalibration, and that model departs from the coherence condition by 17.5 decibans. Pretrained representations reduce the labeled data required and calibration corrects the scale of the probabilities, but neither corrects a biased teacher, so the probabilities have to be checked against realized outcomes.
[NLP-265] Dual Task Vector Merging: Leverag ing Emotion and ASR Knowledge for Low-Resource Alzheimers Disease Detection
【速读】： 该论文旨在解决阿尔茨海默病（Alzheimer’s Disease, AD）语音检测中因数据稀缺导致的模型过拟合与预训练语音编码器知识利用不充分的问题。现有方法中，全量微调易引发过拟合，而固定编码器的探针（frozen probe）则无法有效利用编码器中与任务相关的信息。为此，本文提出双任务向量融合（Dual Task Vector Merging）这一参数高效的方法，通过从情感识别和自动语音识别（ASR）任务微调得到的wav2vec 2.0模型中提取权重增量（weight deltas），将两个任务向量注入到每个Transformer层中，仅在AD任务训练时优化层间融合系数与线性分类头，整体更新参数不足全微调的0.1%。实验表明，在独立说话人设置下，该方法在ADReSS和ADReSSo数据集上优于冻结探针、LoRA及全微调，在二分类任务中表现最优；结合LoRA后进一步提升性能；在PROCESS数据集上，其优势可迁移至三分类（健康对照/轻度认知障碍/痴呆）任务。在低资源场景（20%、40%、80%训练数据）下的实验也验证了该方法在数据受限条件下的稳健性与有效性。
链接: https://arxiv.org/abs/2610.24870
作者: Yuan-Chao Li,Yin-Long Liu,Liu He,Sizhe Liu,Ji-Xuan Deng,Yin Zhang,Zhen-Hua Ling,Jiahong Yuan
机构: 未知
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Sound (cs.SD)
备注:
Abstract:Alzheimer’s Disease (AD) detection from speech benefits from both paralinguistic and linguistic cues, yet model training is frequently constrained by limited data. Fine-tuning a large pretrained speech encoder can easily overfit, while a frozen probe leaves the encoder’s task-relevant knowledge underutilized. Therefore, we propose Dual Task Vector Merging, a parameter-efficient method that adapts a frozen wav2vec 2.0 backbone using task knowledge learned outside the AD task. Specifically, we inject two task vectors, derived as weight deltas from emotion- and ASR-fine-tuned wav2vec 2.0 models, into each transformer layer. During AD training, only layer-wise merging coefficients and a linear classification head are optimized, updating less than 0.1% of the parameters required for full fine-tuning. Speaker-independent experiments on ADReSS and ADReSSo show that dual merging improves over frozen probes, LoRA, and full fine-tuning, in the main binary settings, and gives the best results when combined with LoRA. Evaluation on PROCESS further shows that our advantages transfer to three-class HC/MCI/dementia classification when paired with LoRA. Experiments under reduced training data (20%, 40%, and 80%) confirm reliable gains in low-resource settings.
信息检索
[IR-0] AGR: Agent ic Generative Retrieval with Tool-Augmented Search
链接: https://arxiv.org/abs/2610.26831
作者: Ming Yang,Yaqi Yin,Zhibo Wang,Xiaoze Jiang,Jingwei Zhuo,Chao Zhang,Quanzhi Zhu,Weiguo Zheng,Liang Zeng,Yunfeng Zhao,Jiaqiang Liu,Kefeng Wang,Yupeng Huang,Kaiqiao Zhan,Xuanping Li
类目: Information Retrieval (cs.IR)
备注:
Abstract:Large-scale search and recommendation systems derive much of their retrieval quality from rich, heterogeneous features, yet existing generative retrieval (GR) methods struggle to exploit such feature ecosystems efficiently. Static feature injection either encodes all available signals or places them according to predetermined schemes, increasing serving cost and introducing irrelevant or conflicting evidence. More fundamentally, existing methods cannot adaptively determine when a feature is needed, which feature to select, or where it should influence hierarchical Semantic ID (SID) generation. We propose Agentic Generative Retrieval (AGR), a tool-augmented paradigm that interleaves SID generation with on-demand feature acquisition. Conditioned on the request and decoded SID prefix, AGR decides whether to invoke a tool, which feature to acquire, and where its observation should influence subsequent decoding, thereby steering generation toward feature-consistent SID branches. To supervise this behavior, we construct tool-augmented trajectories based on feature applicability and information gain along the SID hierarchy, and train the model through SID pretraining followed by tool-augmented fine-tuning. Experiments on an industrial dataset and three public benchmarks, together with an online A/B test, consistently demonstrate the effectiveness of AGR.
[IR-1] Next Best Action: Causal Constrained Offline RL for Business Action Targeting RECSYS2026
链接: https://arxiv.org/abs/2610.26817
作者: Surya Ramachandiran,Yiming Li,Juan Mancilla-Caceres,Tao Ye,Jun Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Optimization and Control (math.OC)
备注: Accepted for oral presentation at the ACM RecSys 2026 Workshop on Online and Adaptive Recommender System (OARS)
Abstract:Next Best Action (NBA) systems must select among customer interventions (e.g., discounts, promotions, and lifecycle actions), maximize long-term economic value, satisfy complex operational constraints, and remain causally grounded. Existing approaches lack a rigorous framework for meeting these requirements jointly. We address this gap by formulating NBA as a Constrained Markov Decision Process (CMDP). We then integrate off-policy causal correction of the estimated customer dynamics directly into the Linear Programming (LP) formulation of the CMDP. The resulting policy selects actions based on customer states to maximize long-term value while explicitly enforcing business guardrails. Furthermore, the LP formulation yields shadow prices that quantify the economic trade-offs of operational constraints (e.g., fairness and user welfare), enhancing policy interpretability. We validate our proposed offline Reinforcement Learning (RL) framework on an industry case study involving rideshare coupon targeting, where customer states are defined by Recency, Frequency, and Monetary (RFM) stages. We evaluate performance using model-based off-policy evaluation, with causal dynamics estimated from a temporal holdout. Under the observed coupon budget and RFM state representation, our optimized policy improves customer lifetime value (LTV) by 19% over the estimated behavior policy and 8% over a myopic baseline. Although the demonstrated action space is binary, the proposed framework is action-agnostic, yielding a general-purpose decision module for targeting systems that manage multiple interventions.
[IR-2] Beyond Behavioral Accuracy: Evaluating Individual-Level Fidelity in LLM Investor Simulation
链接: https://arxiv.org/abs/2610.26524
作者: Jiajie He,Yuanhong Jiang,Xintong Chen,Dongling Ni,Wenjin Liu
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC); Information Retrieval (cs.IR)
备注: This paper is under review in ARR
Abstract:Large language models (LLMs) are increasingly used as user simulators, but can they faithfully reproduce an individual’s behavior and decision logic over time? We investigate this question in a longitudinal paper-trading study with 80 participants. Using rolling next-day predictions, we evaluate trading activity, buy-sell structure, asset selection, and portfolio outcomes, complemented by controlled input ablations and participant-rated explanation alignment. No evaluated LLM reliably outperforms a simple recent-activity persistence baseline for predicting trade occurrence, while buy-sell structure and asset selection remain difficult to reproduce. Similar activity-level performance can also yield substantially different portfolio trajectories. Ablations show that activity predictions depend primarily on recent trades, whereas asset choices are sensitive to multiple evidence sources. Participants rate explanations as more aligned with their recurring strategies than with their decision-specific rationales. These findings reveal a gap between coarse behavioral agreement and individual-level fidelity, motivating separate evaluation of behavioral accuracy, explanation alignment, and evidence dependence in personalized user simulation.
[IR-3] he Geometry of Agent Memory: Forecasting Retrieval-Bound Memory Failures and Characterizing Its Limits NEURIPS2026
链接: https://arxiv.org/abs/2610.26433
作者: Iliya Zhechev,Radoslav Ralev,Chris Coleman,Srijith Rajamohan,Jen Agarwal,Tyler Hutcherson,Simba Khadder
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 11 pages, 2 figures. Accepted as a poster at the Personalized, Aligned, Long-Term Memory for AI Systems (PALM) Workshop at NeurIPS 2026. Workshop version: this https URL
Abstract:We study whether geometric relationships between conversation turns, memory texts, and queries can predict answer correctness before generation. We evaluate four memory implementations with dense or hybrid retrieval and one observation-summary comparator. We embed these texts in a shared space. A logistic classifier uses 25 features describing query proximity, memory-conversation correspondence, and conversation structure. On LongMemEval-small, pooled AUROC is 0.640 over 2,367 question-system pairs, below a question-difficulty baseline of 0.655. Performance varies by question type: geometry reaches 0.841 on single-session-assistant questions, but system identity alone reaches 0.826. After controlling for difficulty and identity, geometry adds no statistically established pooled improvement (delta = +0.002, 95% CI [-0.010, 0.014]). A fixed-store context-selection experiment finds higher accuracy with proximity-ranked memories than with random subsets, consistent with ordinary retrieval relevance. On LoCoMo, nearest-turn similarity has modest predictive value for single-hop questions and near-chance estimates for several other categories. Training uses correctness labels; prediction uses pre-answer features and no additional LLM generation call. The results characterize associations between memory representations and correctness, rather than establish a reliable deployment diagnostic. In particular, embedding proximity does not verify fact coverage or distinguish memory-construction, retrieval, and generation errors.
[IR-4] Large-Scale Biomedical RAG : Corpus Scaling Document Representation and Efficiency Trade-offs
链接: https://arxiv.org/abs/2610.26339
作者: Adrián Sánchez-Freire,Gonzalo Pérez-Gómez,Aitziber Atutxa,Ander Barrena
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 15 pages, 4 figures, 3 tables, plus 11 pages of supplementary material. Preprint submitted to Elsevier. Code: this https URL and this https URL
Abstract:We present a controlled, large-scale evaluation of how corpus scale, document representation, and reranking affect the effectiveness and efficiency of biomedical Retrieval-Augmented Generation (RAG) systems. Retrieval collections were progressively expanded while retaining the relevant source documents, enabling consistent comparisons across corpus sizes. We compared abstract, flat full-text, and hierarchical full-text representations across first-stage retrieval, reranking, and end-to-end multiple-choice question answering. Configurations selected on development data were evaluated on five external biomedical benchmarks and a held-out test set, while retrieval latency and index storage were measured to quantify operational cost. Performance patterns observed with small or controlled collections did not consistently persist as collections grew, and the usefulness of reranking depended on the corpus and retrieval setting. RAG improved three of four model configurations across the external benchmarks, with mean gains of up to 9.6 percentage points, and improved all four configurations on the held-out test set by 10.2-30.3 points. Abstract retrieval performed best when reference evidence was abstract-aligned, whereas full-text representations were generally more effective when richer evidence was needed. Hierarchical representation matched or outperformed flat full-text retrieval in most end-to-end comparisons while requiring 67% fewer indexed units and 44-56% less storage. These findings show that corpus scale should be treated as an explicit evaluation variable in biomedical RAG, and that the most appropriate design depends on evidence format, model behavior, and operational constraints.
[IR-5] he Scheduler Is a Hidden Retriever: How Asynchrony Shapes Evidence Acquisition in Search Agents
链接: https://arxiv.org/abs/2610.26310
作者: Ruiyang Ren,Yuhao Wang,Yinan Zhang,Chunyan Miao
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Language-model agents increasingly solve complex tasks through parallel reasoning and asynchronous information gathering. Controlled replay across two model families and three benchmarks shows that delivery timing redirects search decisions made before all results arrive, despite identical initial evidence. Acquisition disagreement reaches up to 24 percentage points above same-order repeat variation. This dependence persists after all initial evidence arrives and through a second search, turning temporary differences in what agents have seen into lasting differences in what they acquire. An acquisition-regret bound with a matching construction shows that invariance at a fixed early arrival count can limit the use of evidence already received. We introduce Confluent Evidence Acquisition (CEA), which instead releases local reasoning when each subquestion’s evidence is complete and allocates shared search independently of completion order. For a fixed plan and stable conditional response law, we prove joint invariance of decisions, acquisitions, and answers. With question-only plans across three benchmarks, CEA brings acquisition disagreement to within 1.0 percentage point of repeat variation and achieves accuracy comparable to complete-input controls, improving over early commitment by up to 12.6 percentage points. Replacing CEA’s fixed allocation with first-ready allocation under contention adds 21.6 percentage points of acquisition disagreement, while local query-target disagreement remains near repeat variation. Timed comparisons show that CEA reaches decisions earlier and responds faster than the same program waiting for all initial evidence.
[IR-6] Preference Is Not Position: Evaluating Reranking on NevIR in Fixed Hybrid Candidate Pools KR
链接: https://arxiv.org/abs/2610.26260
作者: Yuning Bian,Lele Xu,Yifan Yuan,Xinyu Yin,Chengyan Zhan
类目: Information Retrieval (cs.IR)
备注: 6 pages, 2 figures, 3 tables. Accepted at the IEEE BigData 2026 Undergraduate and REU Consortium. Code: this https URL
Abstract:Near-identical passages can differ in a decisive condition, such as negation. A ranker may order such a pair correctly while placing the preferred passage far down the retrieved list. We examine this distinction using official NevIR pairs jointly covered by fixed candidate pools from dense, learned sparse, and BM25 retrieval. On 2,736 Test query directions, a pairwise-trained ranker improves strict per-query preference over fixed fusion (56.62% versus 52.30%) while reducing preferred-passage mean reciprocal rank (MRR) from 0.5764 to 0.3055. An exploratory fixed-model replay of a variant with background sampling and revised supervision yields MRR of 0.5859 and strict preference of 54.53%. Bounded pointwise reranking of the fusion top 20 improves both outcomes. Qwen3-14B-AWQ with fusion tie-breaking reaches 77.45% strict preference, 58.99% NevIR-style paired accuracy across 1,363 complete pairs, and preferred-passage MRR of 0.7509. A BGE cross-encoder reaches 69.77% strict preference and MRR of 0.7404. A post-hoc analysis finds that the branch where equal Qwen grades for the two designated passages trigger fusion tie-breaking accounts for 27% of evaluated directions and contains most remaining strict-preference failures. These findings support joint evaluation of pairwise preference, designated-passage position, and computational cost, while distinguishing model judgments from the combined ranking procedure.
[IR-7] Babbar26 main x genre
链接: https://arxiv.org/abs/2610.26220
作者: Bournonville Aur{é}lien,Clotilde Levy,Thomas Largillier,Guillaume Pitel,Fran{\c c}ois Rioult,Fabrice Maurel
类目: Information Retrieval (cs.IR)
备注:
Abstract:We introduce Babbar26, a multilingual dataset of 1111 web pages from 1006 domains in five languages (English, French, German, Spanish, Italian), jointly annotated for genre, following a compact five-class taxonomy, and for main content and boilerplate, as main/noise regions anchored to the DOM. We describe the annotation guidelines and report double-annotation agreement on both tasks (97.1% on genre; content-level Cohen’s k = 0.766 for main content). Compared with the classic KI-04 genre benchmark, a median page has 2.3x more DOM elements and 1.9x more readable text.
[IR-8] Conditional information Retrieval for Longitudinal Survey Harmonisation
链接: https://arxiv.org/abs/2610.26173
作者: Wing Yan Li,Zeqiang Wang,Michelle Lin,Jon Johnson,Suparna De
类目: Computation and Language (cs.CL); Computers and Society (cs.CY); Information Retrieval (cs.IR)
备注:
Abstract:Identifying conceptually equivalent questions across different surveys is a challenging information retrieval task where surface-level textual similarity is insufficient. Two questions may be lexically similar yet assess entirely different sociological sub-concepts, or lexically different yet conceptually equivalent. We introduce a conditional information retrieval framework that addresses this by decomposing the task into two steps: concept extraction, which identifies the underlying sociological concept of a survey question using a fine-tuned generative model, and concept-aware retrieval, which augments question representations with the extracted concept as an explicit conditioning signal for finer-grained alignment. To reflect the realistic data scarcity conditions of specialised social science annotation, we use CLOSER as a case study. Our framework consistently outperforms baselines by 2-3 points in both MAP and Recall. We further analyse fine-tuning strategies including SFT, DPO, and dynamic DPO, finding that SFT delivers the most stable gains while dynamic DPO interacts differently with instruction-tuned versus distilled reasoning backbones. Finally, we demonstrate that a model fine-tuned on a general social science thesaurus (ELSST) transfers robustly to task-specific survey data, offering a practical solution under data scarcity.
[IR-9] FLMNet: Self-Ensemble Distillation from Masks to Seeds for Single-Tower CTR Prediction
链接: https://arxiv.org/abs/2610.26143
作者: Yifan Wang,Yifei Du
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 44 pages, 10 figures, 15 tables. Code: this https URL
Abstract:Click-through rate (CTR) prediction underpins recommender systems and online advertising, where high-dimensional categorical features interact in combinatorially many ways. Several recent CTR models, such as FinalNet and QNN, report higher accuracy when a network also learns from its own averaged predictions, for example by running the same network under different dropout patterns and training each run toward the average. Why this self-distillation helps, and whether its random sampling is needed, remains unclear. This work shows that as the number of dropout patterns grows without bound, the objective reduces exactly to the label loss of the logit-averaged prediction plus a penalty on how much the dropout copies of the network disagree. When dropout does not affect how a network gates its features, this penalty can be approximated in closed form, and FLMNet is designed on this principle and trains without sampling. Networks trained from different seeds disagree in the same form, and distilling several of them into one network recovers part of their ensemble gain. On five benchmarks against seventeen baselines, the resulting models take the top two places in accuracy with a single network at inference, and the gains of the penalty and of distillation, where measurable, largely add up.
[IR-10] Dont Rewrite the SID Map: Topology Preservation for Generative Retrieval in Dynamic E-Commerce Search
链接: https://arxiv.org/abs/2610.26105
作者: Jianbo Zhu,Xing Fang,Xiaopeng Chu,Mingmin Jin,Jing Wang,Bokang Wang,Jiaqi Ye,Xiang Zhang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Generative retrieval assigns each item a discrete semantic identifier (SID) and retrieves items by autoregressively generating SID sequences. In dynamic environments such as e-commerce, new items arrive continuously, requiring periodic updates to the SID-to-Item mapping and model to maintain retrieval timeliness. Each update, however, can invalidate search paths already learned by the retriever, and the quantitative relationship between mapping change and retrieval quality has not been systematically characterized. We study this problem with extensive controlled experiments over 67 days of large-scale e-commerce search logs and obtain three core findings. (1) SID semantic drift causes severe retrieval degradation that may be masked by large beam sizes: mapping updates reduce HitRate by 41.78-62.05% at beam size 1 but only 8.33-10.85% at beam size 1000, as redundant candidate generation conceals the true severity of drift. (2) The fraction of changed SIDs is strictly monotonically related to retrieval quality: as the remapped fraction increases from 0% to 72.75%, HitRate decreases monotonically, dropping up to 42.66% at beam size 1. Full remapping remains substantially worse than the preservation policy under warm-start training, indicating that short-budget continued training alone cannot repair topology disruption. (3) The SID topology-preserving principle makes even the simplest warm-start strategy highly effective: fixing old-item SIDs, inserting new-items, and warm-starting from the previous model improves old-item HitRate by 10% and new-item HitRate by 109%, validated across two non-overlapping update cycles. These findings distill into a concise deployment principle: insert new-items, freeze old-items; warm-start when possible, otherwise freeze; never change SIDs the model has already learned. A three-week online A/B test on TmallAPP shows UCTCVR +0.29% and GMV +1.16%.
[IR-11] Decision4Rec: A Jev-Inspired Framework of Decision Models on Knowledge and Active Graphs for Multi-Agent Recommender Systems
链接: https://arxiv.org/abs/2610.26035
作者: Marco Polignano,Het Darshan Mehta,Pasquale Lops,Ernesto William De Luca,Giovanni Semeraro
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Decision models, such as the commercial Jev, are language models that return one probability for each option of a question instead of generating text. We present Decision4Rec, an open and modular framework that makes decision models useful for recommendation. A knowledge graph writes the logs of each recommendation scenario as short statements on a defined scale, such as “sold: top 10%”. A pool of agents, proposes candidates, and a residual ranker keeps the best 30. A decision model, trained once on real choices of three scenarios, reads them through one interface of typed questions, and a fusion step combines its answers with the score of the ranker. Any model with this interface can be used, and we run the commercial Jev, the released open Kev and Kev models trained by us on recommendation requests. The framework with a trained 4B decision model is 25 to 36% above the best trained baseline in NDCG@10 in three sparse Amazon catalogues, one of them never used to train the decision model, and almost tie with baselines on the dense Steam dataset. Most of the gain comes from the pool, and with a trained recommender it comes from what the agents know about its candidates rather than from new candidates. The 27B decision model adds to the pool 5 to 6% above the 0.8B model in three of four domains. A trained 0.8B open model beats the released Jev in eleven of fifteen settings and is never worse. On MovieLens-1M, a dense domain never seen in training, the gain is 5 to 7%. Code, configurations and trained models are available at this https URL and this https URL.
[IR-12] Multimodal Decoupling for Diversified Recommendation through Semantic and Structural Shaping
链接: https://arxiv.org/abs/2610.26024
作者: Changran He,Jiaming Wei,Jianghong Ma,Tianjun Wei,Haijun Zhang,Xiaofeng Zhang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Recommender systems often suffer from over-concentrated results when optimized for accuracy, limiting recommendation diversity. While diversity has become an important objective, existing methods still face fundamental limitations from both semantic and structural perspectives. On the one hand, multimodal recommender systems provide rich semantic signals, yet are often treated as auxiliary to ID embeddings, leading to ineffective semantic utilization for diversity. On the other hand, the accuracy–diversity trade-off persists during training, where accuracy gains often come at the cost of diversity. We show that this phenomenon is fundamentally governed by both the structural properties of learned representations and the dynamics of optimization. To address these challenges, we propose MOSS, a unified framework for multimodal decoupling through semantic and structural shaping. Specifically, we first introduce an MLLM-based Decoupled Feature Extraction module to disentangle multimodal content into informative representations. To enhance semantic utilization, we design a Category-aware Embedding Selection module that extracts compact, category-aware semantic features. To address structural limitations, we propose two complementary modules: Alternating Dual-Graph Propagation, which alleviates representation collapse by integrating collaborative and multimodal propagation, and Entropy-based Diversity Regularization, which mitigates optimization bias by promoting balanced category exposure. Extensive experiments on real-world datasets demonstrate that MOSS consistently achieves a superior accuracy–diversity trade-off over strong baselines.
[IR-13] OneChord: Distill What You Rank Recommend in One Pass
链接: https://arxiv.org/abs/2610.25805
作者: Zesheng Li,Chengchang Pan,Honggang Qi
类目: Information Retrieval (cs.IR)
备注:
Abstract:Accurate product discovery on the Web requires recommendation models that rank catalog candidates within tight inference budgets. In semantic-ID recommendation, deployment can fuse code probabilities with item similarities, making the final ranking depend on their normalization and relative weights. This motivates a direct training principle: \emphdistill what you rank. We present OneChord, which uses one learned, catalog-normalized fusion for next-item supervision, refinement distillation, and inference. A shared parallel decoder produces joint SID evidence and item-matching scores. During training, a frozen denoiser supplies legal partial SIDs; the student evaluates these states to construct refined fused targets. Matching direct and refined predictions on shared hard candidates transfers refinement into one-pass recommendation. We characterize the fusion weights induced by candidate conditioning and establish sufficient conditions for preserving the refined predictor’s full-catalog Top- K selection. Across six Amazon datasets, OneChord achieves the highest Recall@5 and NDCG@5 among the compared methods on every dataset, with average relative gains of 6.30% and 7.12% over the strongest baseline. It also achieves a 1.04–1.18 \times full-test generation speedup over the fastest baseline. OneChord makes the served ranking distribution the common object of recommendation learning and distillation.
[IR-14] Learning Complementary Agent Groups via Task Hypergraphs for Next-POI Recommendation with Missing Modalities
链接: https://arxiv.org/abs/2610.25801
作者: Jinze Wang,Yuze Liu,Liang Yuan,Gaoyuan Du,Xiaohang Xu,Yong Xiang,Jiong Jin
类目: Information Retrieval (cs.IR)
备注:
Abstract:Next point-of-interest (POI) recommendation uses visit histories and POI content, but text and images are often missing. Large language model agents have been applied to recommender systems for preference modeling and evidence acquisition. The challenge is to select agents that can jointly analyze the available evidence for each request. We propose \textscCATH (\emphComplementary Agents via Task Hypergraphs), which learns to select agent groups for tasks such as comparing candidates or analyzing a POI with missing content. Agents with distinct processing roles are the task hypergraph nodes; each hyperedge connects the selected group and records its analysis goal and target POIs. The policy selects members conditioned on the request, task, and agents already selected. During execution, members share candidate-linked results that guide their subsequent operations. Completed tasks update candidate states and agent memories within the request. A shared ranking head converts candidate states into corrections to a fixed base ranking. Supervised ranking trains the executor, followed by policy optimization that accounts for computation cost. Experiments under natural missingness on Gowalla-CA, Foursquare-NYC, and Foursquare-TKY show 8.4–9.1% relative NDCG@20 gains over HyperAgent4POI, the strongest prior baseline evaluated. Additional analyses compare group-selection and execution designs, test ranking under additional masking, and characterize quality–cost tradeoffs.
[IR-15] OpenSaC: A Python Library for Search as Code in Deep Research Agents
链接: https://arxiv.org/abs/2610.25779
作者: Qi Liu,Jiaxin Mao,Fengbin Zhu,Tat-Seng Chua,Ji-Rong Wen
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: Source code: this https URL
Abstract:Deep research agents investigate complex questions through iterative search, source examination, and evidence synthesis. A conventional workflow alternates reasoning with individual search and reading actions. Search as Code introduces a new paradigm: agents write programs that compose search, source reading, reranking, model calls, and local processing, returning selected observations for further reasoning. To support this workflow in existing coding agents, we present OpenSaC, an open-source Python library with a companion Search as Code skill. Configurable capability interfaces let agents combine these operations while retaining evidence for inspection and reuse. Compared with standard Codex using its built-in web search, the OpenSaC configuration uses less execution time and fewer model tokens on BrowseComp and WideSearch, with higher accuracy on BrowseComp and slightly lower row F1 on WideSearch. We aim to lower the barrier to adopting Search as Code and provide an open foundation for community development and transparent research on deep research agents.
[IR-16] Cross-Pollination Retrieval: Finding Transferable Ideas Across Fields in Scientific Literature
链接: https://arxiv.org/abs/2610.25678
作者: Seokwon Jung,Yerang Kim,Yejin Kim,Alexander Rubinstein,Arnas Uselis,Sangwoo Park,Seong Joon Oh,Sangdoo Yun
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 35 pages, 11 figures, 19 tables
Abstract:Many scientific advances come from cross-pollination, carrying an idea from one field over to a problem in another. Yet retrieving papers across fields remains underexplored, with existing benchmarks confined to a narrow range of fields. We study cross-pollination retrieval, the problem of identifying transferable ideas across research fields, and introduce CPR-Dataset, a large-scale dataset constructed by mining cross-topic citations. Its evaluation split CPR-eval has 1,300 queries in 26 fields, and its training split CPR-train has 10,573 pairs. On CPR-eval, the best of eleven existing retrievers reaches a Recall@10 of only 0.26, whereas an LLM reranker reaches 0.81. Standard contrastive fine-tuning on CPR-train raises Recall@10 to as high as 0.70. The gain carries over to domains held out from training and to analogy and reasoning-intensive retrieval. On AnaloBench, it raises nDCG@10 by up to 33% over the untrained retriever, whereas training on three existing retrieval datasets lowers it. On BRIGHT, it raises nDCG@10 by up to 70%, more than twice the gain from those datasets. CPR-Dataset is available at this https URL, and the code at this https URL.
[IR-17] Knowledge Manifold Optimization for Scientific Discovery
链接: https://arxiv.org/abs/2610.25662
作者: Shigeru Obayashi,Yoshiaki Kawagoe,Kazuhiko Komatsu,Tomonaga Okabe
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注:
Abstract:The rapid expansion and specialization of scientific literature make it increasingly difficult to identify not only what is known, but where new and still supportable knowledge may emerge. We introduce Knowledge Manifold Optimization (KMO), a discovery-oriented extension of the Knowledge Manifold framework that treats scientific discovery as a multiobjective search over a continuous, literature-derived knowledge space. Candidate locations are evaluated using four complementary objectives: persistent knowledge-regime boundary strength, band-pass white-space novelty, semanticgradient magnitude, and evidential support. Rather than collapsing these criteria into a single weighted score, KMO identifies Pareto-optimal Supported Knowledge Gaps, defined as corpus-relative unoccupied regions that remain sufficiently supported by surrounding literature. The method is demonstrated on a 99-paper corpus in aerodynamics and aerodynamic design. On a 21 x 21 grid, exact Pareto analysis produced 19 non-dominated points, of which 16 formed the practical front after excluding points with zero white-space score. These solutions exhibited distinct discovery modes, including methodology bridges, aero-structural topology/FSI integration, functional aerodynamic-device integration, and rear-wake/vehicle flow-control integration. A distance-maximization baseline preferentially selected extrapolative regions with substantially lower support and reliability, demonstrating why emptiness alone is insufficient. Representative Pareto locations were converted into auditable evidence packages and then verbalized by a constrained large language model into candidate research hypotheses. The results show that scientific discovery can be formulated as structured multi-objective search in an explicit knowledge space, while keeping novelty corpus-relative and separating numerical discovery from downstream language generation.
[IR-18] Benchmark for Structured Retrieval-Augmented Generation in Perovskite Photovoltaics
链接: https://arxiv.org/abs/2610.25575
作者: Qiwei Han,Chi Zhou,Gaoxiang Yuan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 8 pages, 3 figures, 2 tables
Abstract:While retrieval-augmented language models increasingly serve as literature assistants in the physical sciences, existing benchmarks evaluate them primarily on qualitative text overlap or through model-based scoring rather than against verified experimental ground truth. In experimental materials disciplines such as perovskite photovoltaics, however, the most critical inquiries are inherently quantitative and aggregate measurements across dozens or hundreds of papers. Here, we present a quantitative question-answering benchmark where all 755 ground-truth answers are computed directly from 45,441 measured device records (40,889 human-curated records through 2021 and 4,552 records extracted from 2022-2026 literature). Benchmarking six retrieval architectures over 158,274 passages from 25,662 papers using an open 14B model on a single 24 GB GPU, standard passage retrieval tops out at 16-23% accuracy (compared to 1.6-6.5% zero-shot). In contrast, equipping the language model with read-only SQL execution over the structured device table elevates accuracy to 51.5% on historical and 50.8% on prospective literature – a 33.01.8 percentage-point advantage over hybrid text retrieval. Aggregate queries spanning extrema, distributions, and filtered record counts prove virtually unsolvable via text retrieval alone, demonstrating that systematic table extraction is an operational prerequisite for literature-scale quantitative inquiry.
[IR-19] Selling in the Wild: Benchmark and Randomized Field Evidence for a Sales Agent Trained on Commercial Conversations
链接: https://arxiv.org/abs/2610.25461
作者: Benjamin Tannenbaum
类目: Information Retrieval (cs.IR); Computers and Society (cs.CY)
备注:
Abstract:We test whether Javo, a sales agent fine-tuned on 4.6 million consented, de-identified commercial conversations, sells better than a frontier general-purpose model and better than human sales associates. The evaluation has four parts. Study 1 scores the agent on two public sales benchmarks, SalesLLM and SmartSalesBench, against 10 frontier and open-weight models. Study 2 is a randomized controlled trial on 100 live websites in 14 industries (e-commerce, travel and hospitality, and high-consideration services), run from 1 May to 30 September 2026, in which 14.8 million visitors were randomized to no agent, Javo, or a frontier-model agent with an identical interface, catalog tools and guardrails. Study 3 isolates the effect of the model by comparing those two agent arms. Study 4 compares Javo with each site’s own human associates on 20 sites with staffed live chat. Relative to no agent, Javo changed revenue per visitor by +10.0% (95% CI +5.1 to +14.9). Among engaged visitors, it changed conversion by +21.6% relative to the frontier-model agent (95% CI +16.9 to +26.3). Against human associates, its engaged conversion was 11.26% versus 11.20% (non-inferior, margin -7%, one-sided p = 0.004; no significant superiority), at 1.8% of the cost per conversation. Return rates, unsupported product claims and complaint rates met the prespecified safety guardrails.
[IR-20] Weakly Supervised Adaptation of Scientific Embeddings for Exploratory Literature Mapping
链接: https://arxiv.org/abs/2610.25413
作者: Roman Neruda,Petra Vidnerová,Jiří Šejnoha,Jakub Mihálik,Tomáš Marvan
类目: Digital Libraries (cs.DL); Information Retrieval (cs.IR)
备注:
Abstract:Scientific document embeddings such as SPECTER2 are optimized for retrieval over large, heterogeneous corpora. On a small specialized corpus their similarity distributions can become highly compressed, which limits their usefulness for exploratory literature mapping. We study whether weakly supervised in-domain adaptation, using only metadata already available in open citation databases, changes this geometry. On ArtCon, a curated corpus of 704 publications on artificial and machine consciousness, we adapt the SPECTER2 proximity model with a lightweight LoRA adapter trained on positive pairs derived from bibliographic coupling in OpenAlex, using no manually annotated similarity labels. We compare the adapted embeddings with lexical (TF-IDF), citation-based (bibliographic coupling) and generic semantic (SPECTER2) representations under thresholded, fixed-density and symmetric k-nearest-neighbour graph constructions. Adaptation substantially decompresses the similarity distribution (mean pairwise cosine similarity 0.9104 to 0.8663, standard deviation 0.0259 to 0.0594) and yields graphs with markedly higher Louvain modularity and better connectivity than the base encoder under matched edge and neighbourhood budgets. The effect is stable across all three graph constructions and is therefore not an artefact of a single sparsification choice. We report this as a case study on one corpus, and we are explicit about what it does not establish: the clustering indices used here are computed against community labels derived from each representation’s own similarity graph, so they measure internal geometric self-consistency rather than agreement with an external topical ground truth, and adaptation and evaluation use the same documents. Evaluation against independent labels, on held-out documents, and against geometric controls is ongoing work.
[IR-21] MemDerive: A Diagnostic Benchmark for Agent Memory over Evolving Data
链接: https://arxiv.org/abs/2610.25373
作者: Sen Gao,Yongan Li,Han Chen,Bingsheng He,Ming Wu,Jason Zeng,Michael Heinrich,Shixuan Sun
类目: Databases (cs.DB); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Long-running agents rely on memory of their past interactions for task state, such as what has been completed and which conditions still hold. Such state is usually a derived state. No record states it, and it must be computed by rules from multiple records of an evolving history. Existing memory benchmarks score only the final answer, with no reference for the intermediate states behind it. A wrong answer therefore cannot be traced to the state that was missed, and a state that silently became outdated goes undetected. We present MemDerive, a diagnostic benchmark for agent memory over evolving data. It specifies each scenario symbolically, derives exact gold answers with Datalog, records the gold derivation DAG and the supporting events of every query, and renders the history in natural language with fictional entities and distractors. Its 10 scenarios contain 4,692 events and 1,503 queries whose difficulty varies along access, evolution, and composition. On this ground truth, MemDerive evaluates answer reliability, diagnoses failures by probing the nodes of the gold derivation DAG, and tracks whether answers stay correct as events arrive. Our evaluation of four memory systems reveals failures that final-answer benchmarks hide. On derived state, every memory system falls behind simple BM25 retrieval, although the same answer model reaches 96.01% when given the supporting events. Most failures are ungathered, as a system holds the missing intermediate state but does not retrieve it for the question that needs it. Memory systems also rarely propagate implicit changes, which no new record states. These results suggest three principles for memory design. Gather inputs along the derivation, store rules with their versions, and maintain derived states through recorded dependencies.
[IR-22] Whistle and Wrong Lyrics: Synthetic Queries Are Not Enough for Real-World Query-by-Melody
链接: https://arxiv.org/abs/2610.25371
作者: Luigi Rivolta
类目: ound (cs.SD); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 15 pages, 2 figures, 6 tables. Code: this https URL
Abstract:People look up a song by humming it, but also by whistling it or singing it with the wrong words. Published query-by-humming systems and benchmarks focus on hums and sung queries, and no public dataset pairs whistles or wrong-lyrics singing with song recordings. We study both query types with a melody-only retrieval system: a key-normalized pitch track embedded by a 5.2M-parameter contour encoder, with a spectral-peak tracker for whistles. First, a physics-grounded generator renders synthetic whistles and nonsense-syllable singing from the vocal melodies of 11,083 songs. Training on them raises top-1 retrieval of held-out synthetic queries against an 11,050-song gallery from 0.180 to 0.251 for whistles and from 0.193 to 0.264 for wrong-lyrics singing, but the gain does not transfer: on the MLEnd test performers the best synthetic-only model gets 210 of 339 real whistles right at top-1, against 214 for the baseline. Second, a closed-set fine-tune on real MLEnd whistles with performer-disjoint splits raises whistle-to-hum top-1 from 0.631 (95% CI 0.581-0.681) to 0.673 (0.622-0.723), on known songs only. Third, two further generations of synthetic queries also fail to transfer: melody-conditioned whistles and la-la singing from a commercial neural music model leave real-whistle top-1 at 0.631, and an open-source pipeline (YingMusic-Singer-Plus singing with partly wrong, mumbled or nonsense lyrics, a DDSP whistle synthesizer, and a melody-error model fitted on real performers) gives at best 0.619 (paired difference -0.012, 95% CI -0.038 to +0.015). Across physics-based, commercial neural and open-source neural synthetic data, nothing transferred to real whistles, which points to collecting real queries as the missing step. Code is public; models and data will be released.
[IR-23] Where and Why Search-Augmented LLM s Fail When Reasoning over Rapidly Evolving Knowledge
链接: https://arxiv.org/abs/2610.25323
作者: Tianyi Hu,Bhavyajeet Singh,Niket Tandon,Akhil Arora
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注:
Abstract:As the world changes, new information is published while older records remain online. This complicates reasoning for large language models (LLMs), which must retrieve evidence and resolve conflicts to determine which information remains valid. Existing evaluations treat information seeking as a black box, revealing little about where and why LLMs fail. To address this gap, we introduce TREK, a suite of three components: TREKTRACE is a plug-and-play framework that attributes failures to specific stages of retrieval and reasoning. We apply it to TREKBENCH-WILD, a meta-benchmark of real-world questions that require complex reasoning over rapidly evolving knowledge. Surprisingly, the main bottleneck is post-retrieval reasoning rather than retrieval: frontier models are often misled by conflicting evidence even after retrieving all the information needed to answer. To study this bottleneck, we build TREKBENCH-CTRL, a controlled benchmark with distractors spanning temporality, relevance, and trustworthiness, and show that benchmarks dominated by outdated distractors substantially underestimate task difficulty. Together, TREK provides a testbed for diagnosing and improving reasoning over rapidly evolving knowledge.
[IR-24] MEMORA: A Memory-Enhanced Multimodal Committee Reranking Agent for Reasoning -Intensive Retrieval EMNLP
链接: https://arxiv.org/abs/2610.25282
作者: Mohamed Mahmoud,Mostafa Farouk Senussi,Mahmoud Abdalla,Mahmoud SalahEldin Kasem,Abdelrahman Abdallah,Hyun Soo Kang
类目: Information Retrieval (cs.IR)
备注: Accepted at EMNLP Main 2026
Abstract:Reasoning-intensive retrieval is fundamentally challenged when queries demand deep semantic inference and arrive in multimodal form. Existing systems treat each query in isolation, rely on monolithic rerankers, and lack retrievers trained for cross-modal reasoning. We introduce MEMORA, a Memory-Enhanced Multimodal Orchestrated Reranking Agent built around three contributions: (1) Collaborative Factor-Committee Reranking, which decomposes queries into intent factors, extracts factor-specific evidence into a structured matrix, and synthesizes judgments via a memory-conditioned meta-verdict model; (2) an Episodic Memory Module that tracks factor outcomes across queries and dynamically adjusts priorities at inference time; and (3) a Reasoning-Aware Dense Retriever fine-tuned with chain-of-thought augmentation on 20k+ reasoning-intensive pairs. Evaluated on MM-BRIGHT across 29 domains, MEMORA achieves nDCG@10 of 41.5 (text-to-text) and 38.9 (multimodal query-to-text), outperforming the strongest agentic text-only baseline by +8.3 points and the strongest agentic multimodal baseline by +7.9 points. Code and data are available at this https URL.
[IR-25] Bridging Concepts: Multi-Agent Reasoning over Scientific Concept Graphs for High-Impact Research Prediction CIKM2026
链接: https://arxiv.org/abs/2610.25281
作者: Donggyu Yoon,Minsung Hwang,Juneha Baek,Donghyuk Shin,Joyce Jiyoung Whang
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 5 pages, 2 figures, 4 tables, The 3rd Workshop on Knowledge-Enhanced Information Retrieval (KEIR) at The 35th ACM International Conference on Information and Knowledge Management (CIKM 2026)
Abstract:Within a research community, new topics often emerge when previously unconnected concepts are first studied together. We construct scientific concept graphs (SCGs)–domain knowledge graphs whose nodes are fine-grained research concepts extracted from paper text by LLMs and whose edges capture their co-occurrence in papers–and study the task of forecasting high-impact concept pairs, i.e., discovering concept pairs that are currently unconnected but will have high impact when connected. To this end, we introduce new benchmark datasets constructed from CIKM, KDD, and SIGIR conference papers, with varied prediction settings, including impact forecasting for emerging concepts unseen during training. We propose MiroGraph, a Multi-agent reasoning framework with scientific concept Graphs, in which a feature designer agent designs predictive features for a given concept pair from structural, semantic, and trend perspectives, and then three LLM agents compute the actual feature values from the enclosing subgraph of the concept pair. Using these feature values, a classifier designer agent constructs a classifier that predicts the final impact labels. MiroGraph outperforms various impact forecasting and temporal link prediction methods.
[IR-26] When Semantics Fall Short: Persona-Guided Cross-Domain Sequential Recommendation with Large Language Models
链接: https://arxiv.org/abs/2610.25221
作者: Jundong Chen,Honglei Zhang,Shuai An,Enneng Yang,Han Yu
类目: Information Retrieval (cs.IR)
备注:
Abstract:Cross-domain sequential recommendation (CDSR) alleviates data sparsity by transferring user behaviors across domains. Recent advances increasingly leverage large language models (LLMs) to bridge domains through global item semantics. However, this paradigm is inherently limited by two issues: I. semantic domain isolation, where item semantics remain domain-specific and hinder transferable modeling, and II. semantic-behavior inconsistency, where local semantic relations are more consistent with behavioral transitions, whereas such consistency often breaks down globally across domains. To this end, we propose PerCDR, an LLM-enhanced CDSR framework that shifts the focus from global semantic bridging to persona-guided behavioral modeling. Specifically, PerCDR leverages LLM-inferred personas to globally guide transferable cross-domain behavioral patterns, while exploiting local semantic relations to enhance intra-domain behavioral modeling, followed by domain preference adaptation for recommendation refinement. Extensive experiments demonstrate the superiority of PerCDR and validate its effectiveness.
[IR-27] RemoteANN: A Native Vector Search Engine for Cloud Object Storag e
链接: https://arxiv.org/abs/2610.25172
作者: Yutong Gou,Silu Huang,Yue Wang,Jianjun Chen,Cheng Long
类目: Databases (cs.DB); Information Retrieval (cs.IR)
备注:
Abstract:Vector search engines built on cloud object storage have drawn growing attention from both academia and industry. These systems typically adopt a disaggregated storage-compute architecture, persisting the primary vector payload in object stores (e.g., Amazon S3) while retaining compact index metadata on compute nodes. However, conventional vector indexing structures are not natively tailored to the pronounced network latency and per-request access costs of cloud object stores, resulting in network I/O amplification, degraded query throughput, and higher API expenses. To resolve these challenges, we present RemoteANN, a vector search engine explicitly co-designed for cloud object storage across indexing and querying. To mitigate the dominant network I/O bottleneck, RemoteANN couples an Inverted File (IVF) index with RaBitQ quantization, further optimized by inter-cluster and intra-cluster data reordering for improved data locality. Building on this tailored data layout, RemoteANN’s query engine combines cluster pruning, fine-grained Range-GET requests, request merging, and local NVMe SSD caching, dramatically reducing discrete storage requests and network transfer volume without sacrificing accuracy. To address the index construction bottleneck, we further develop QGKMeans, which substantially reduces construction latency without compromising clustering quality. Extensive evaluations on real-world datasets demonstrate that, compared with representative baselines, RemoteANN achieves up to a 5.9\times throughput speedup while issuing up to 16.1\times fewer object storage requests at 97% target recall. Subjects: Databases (cs.DB); Information Retrieval (cs.IR) Cite as: arXiv:2610.25172 [cs.DB] (or arXiv:2610.25172v1 [cs.DB] for this version) https://doi.org/10.48550/arXiv.2610.25172 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-28] oward Agent ic Retrieval-Augmented Generation for Enterprise Retail Workflows over Heterogeneous Data Sources
链接: https://arxiv.org/abs/2610.25096
作者: Jeonghwan Choi,Jibin Hwang,Hayoon Park,Hyeonjae Cheon,Hwanjun Song
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: Preprint. Under review
Abstract:Enterprise retail workflows require agents to chain evidence across heterogeneous sources, from structured product and order records to unstructured inquiries and reviews, yet existing agentic RAG benchmarks rarely capture this challenge. To enable controlled evaluation of such workflows, we introduce RetailBridge, a compositional retail benchmark built from persona-driven simulations. Its questions are constructed by linking entities and intermediate answers across heterogeneous sources, requiring agents to identify the relevant sources, carry evidence between them, and produce a grounded answer. To systematically evaluate LLMs on these retail workflows, we develop RetailRAG, a retail-aware framework equipped with auto-schema briefing, source-routed plan chaining, and diagnostic plan revision. RetailRAG consistently outperforms existing agentic RAG baselines across all evaluation axes, serving as a strong reference system for assessing LLM agents on retail workflows. On RetailBridge, evaluating ten LLM backbones with RetailRAG reveals multi-source chaining as the dominant bottleneck across providers and model scales, suggesting that reliable retail agents require explicit cross-source orchestration rather than scaling alone. Code and data are publicly available at \urlthis https URL.
[IR-29] Structure-Aligned History Learning with Subspace-Structured Semantic IDs for Generative Recommendation
链接: https://arxiv.org/abs/2610.25093
作者: Kaiwen Wei,Kejun He,Jie Zhang,Xiaomian Kang,Yuming Yang,Aijia Zhou,Xie Zhou,Yining Wang,Nayu Liu,Li Jin,Zhenyang Li,Jiang Zhong,He Bai,Jingxuan Wei,Junnan Zhu
类目: Information Retrieval (cs.IR)
备注: 34 pages, 10 figures, including references and appendices
Abstract:Generative recommendation constructs semantic IDs for catalog items and predicts the next ID from user interaction histories. However, existing methods face limitations in both stages. First, conventional quantizers may produce imbalanced code assignments and lack coherent intermediate semantic units between individual tokens and complete items, limiting history reconstruction. Second, recommendation models remain dominated by next-item prediction, treating historical interactions mainly as conditioning context and providing limited supervision for recovering persistent preferences across the historical path, particularly when relevant preference signals are distributed over distant interactions. To address these limitations, Structure-Aligned History Learning (SAHL) reconnects the two stages by using the structure created during quantization directly for history reconstruction. Its Semantic Residual Balanced Quantization (SRBQ) forms a balanced token-subspace-item hierarchy, and SAHL reconstructs history at these same granularities. Predictive entropy prioritizes uncertain transitions, while curriculum optimization progressively shifts training to inference-aligned next-item prediction. On four Amazon Review categories, SAHL improves NDCG@10 by 18.6%-30.0% over RPG and 2.7%-6.0% over strong method MHL. Controlled analyses verify balanced code use, the benefit of following quantizer-defined groups during reconstruction, and dependence on user history. The relevant code is available at this https URL.
[IR-30] Learning to Program Retrievals: Composing Search Primitives via Reinforcement Learning
链接: https://arxiv.org/abs/2610.24854
作者: Kyuyoung Kim,Zilin Xiao,Zhiyuan Wang,Avinash Atreya,Ke Yang,Jinwoo Shin,Kecheng Hao
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Retrieval-augmented language models increasingly use multi-step search to solve knowledge-intensive tasks, yet existing systems typically expose retrieval through fixed pipelines or collections of atomic tools. We introduce Retrieval Program Learning (RPL), a programmable search framework that allows a policy to compose search primitives into executable programs with local state and control flow. This enables adaptive search that can retrieve broadly, process intermediate results, and selectively construct model context without tying retrieval breadth to context consumption. However, the resulting flexibility creates a difficult exploration problem: useful programs may require several coordinated decisions, while reinforcement learning often supplies only sparse feedback. To address this challenge, we propose program mutation, a grammar-constrained and execution-verified procedure that generates local variants of suboptimal programs sampled during RL, converts them into training trajectories, and distills them back into the policy. We evaluate RPL on open-domain question answering and corpus-level reasoning tasks, comparing it with a range of retrieval and reasoning baselines. RPL consistently improves open-domain QA performance across model sizes and in- and out-of-domain benchmarks, with particularly large gains on multi-hop tasks. Notably, a 4B model trained with RPL outperforms all 8B model baselines. On corpus-level reasoning, RPL achieves the highest answer F1 and document-coverage F1 for both model sizes while substantially reducing context consumption and retrieval turns. Our results show that programmable search expands the retrieval strategies that agents can learn, while program mutation provides an effective mechanism for exploring useful program structures beyond direct on-policy sampling.
[IR-31] VisR-RAG -Fin: Benchmarking Multimodal RAG over Visually Rich Financial Documents SIGIR
链接: https://arxiv.org/abs/2610.24822
作者: Omar El Bachyr,Laura Maria Bernardy,Saad Ezzini,Jacques Klein,Tegawendé F. Bissyandé,Charles Mont’e,Anas Zilali,Ulrick Ble,Anne Goujon
类目: Information Retrieval (cs.IR)
备注: Accepted at the main research track of ACM SIGIR-AP 2026, Singapore, December 13-15, 2026
Abstract:Enterprise documents such as financial reports are increasingly processed by automated systems, yet they are primarily designed for human reading and exhibit complex visual structure, combining dense text with tables, charts, diagrams, and rich layouts. These characteristics pose significant challenges for Retrieval-Augmented Generation (RAG) systems, which must retrieve relevant pages and leverage visual and structural information during answer generation. Despite recent advances in multimodal retrieval and vision-language models, there remains a limited understanding of how design choices across retrieval and generation components affect end-to-end RAG performance on visually rich documents. In this paper, we present a comprehensive empirical study of multimodal RAG for page-level Question Answering (QA) over financial documents. We introduce VisR-RAG-Fin (Visually Rich RAG for Finance), an open-source benchmark constructed through a general pipeline for visually rich documents that preserves original layout and visual content. Using this benchmark, we systematically analyze retrieval and generation behavior under text-only, multimodal, and hybrid configurations, providing a foundation for rigorous evaluation of RAG systems on complex enterprise document collections. Overall, this work provides actionable insights for practitioners designing and evaluating multimodal RAG pipelines over complex, visually rich enterprise documents.
[IR-32] MemRevise: A Self-Revising Memory for Conversational Search
链接: https://arxiv.org/abs/2610.24672
作者: Mohammed Ali,Abdelrahman Abdallah,Adam Jatowt
类目: Information Retrieval (cs.IR)
备注:
Abstract:In conversational search, users correct misun-derstandings, revise their requests and return to earlier topics, so part of the history no longer reflects what they want. Existing benchmarks do not isolate such turns. Current search methods either re-read the history or keep a record of every turn, and none of them marks earlier content as dropped. We introduce CONVREVISE, a benchmark of 4,429 test questions in six tiers of returns and revisions over TopiOCQA, CORAL and OR-QuAC. We also propose MEMREVISE, which combines a revisable memory graph, a 3B writer distilled from teacher memories, and a search guided by the memory. The writer reads each turn once and adds a node that records its topic, its facts and the operations it performs, such as adding, abandoning or reactivating a topic. Topic status follows from these operations, so a dropped topic keeps its facts and can be restored. With a frozen retriever, MEMREVISE reaches a mean MRR@10 of 0.701, against 0.589 for the best competitor, ConvSearch-R1-FT, and 0.617 for a writer distilled the same way without the memory, so the memory and the steps built on it add 0.084. MEMREVISE is significantly better on Topi- OCQA and OR-QuAC, statistically tied with prompted 32B rewriters on CORAL, and never significantly behind the strongest competitor of any tier.
[IR-33] RAG it: Turning Document Preprocessing into an Experimental Variable for RAG
链接: https://arxiv.org/abs/2610.24659
作者: Omar El Bachyr,Saad Ezzini,Tegawendé F. Bissyandé,Jacques Klein
类目: Information Retrieval (cs.IR)
备注: 4 pages, 2 figures, 4 tables. Accepted at the Tool Demonstration Track of the 33rd Asia-Pacific Software Engineering Conference (APSEC 2026)
Abstract:Document preprocessing choices can substantially affect retrieval in Retrieval-Augmented Generation (RAG) pipelines, yet they are often selected heuristically and are difficult to compare under consistent conditions. We present RAGit, an open-source experimentation workbench for the systematic and reproducible evaluation of document preprocessing choices through downstream retrieval. RAGit represents parsing, chunking, indexing, retrieval, and evaluation as configurable experimental stages, enabling controlled comparisons while the remaining downstream settings are kept fixed. It preserves provenance from retrieved chunks to source pages and maps heterogeneous retrieval units to a common page-level ranking for evaluation. The current implementation supports multiple parsers and chunking strategies, dense and late-interaction retrieval, reusable intermediate artifacts, and multi-configuration benchmarking. We demonstrate RAGit through controlled experiments comparing alternative preprocessing configurations under a fixed retrieval setup, illustrating how preprocessing can be treated as an explicit experimental variable. RAGit and its documentation are publicly available on GitHub (this https URL). A demonstration video is available at this https URL.
[IR-34] DuoLM: Fast and Effective Personalized Recommendation with Language Models
链接: https://arxiv.org/abs/2610.24481
作者: Jiang Wu,Yutong Song,Hongbo Wang,Guanghe Cao,Jeff Liang
类目: Information Retrieval (cs.IR)
备注: 19 pages. Jiang Wu and Yutong Song contributed equally
Abstract:Language models can make personalized recommendations from rich item content, user history, and feedback. However, evaluating many candidates with full-text inputs incurs substantial online computation and limits serving throughput. We introduce DuoLM, a framework for fast and effective personalized recommendation that integrates a compact recommendation model with an efficient serving runtime. An encoder LLM learns compact item representations for offline reuse, while a personalized ranker LLM combines them with user history and feedback for recommendation. We train both models end to end with recommendation supervision and teacher distillation. The runtime further accelerates inference through virtual token IDs for embedding access, shared history computation, and CUDA Graph replay. Benchmark results show that our framework achieves recommendation quality comparable to that of the expert teacher model. On a single H100, it delivers approximately 26,400 candidate scores per second, about 60\times the throughput of the teacher served with vLLM and automatic prefix caching.
[IR-35] Optimal Fusion Weights Form Intervals Not Points: A Latency-Aware Study of Query-Adaptive WRRF
链接: https://arxiv.org/abs/2610.24341
作者: Iris Ma,Zahra Shekarchi,Stuart Mcilroy,Mahtab Farrokh,Fattane Zarrinkalam,Joel Stremmel
类目: Information Retrieval (cs.IR)
备注:
Abstract:Reciprocal Rank Fusion (RRF) is widely used to combine sparse and dense retrievers in hybrid information retrieval, but a critical limitation is that it assigns a fixed, uniform weight to each retriever regardless of the query. Query-aware methods treat the per-query optimal weight as a single point, an assumption whose empirical validity has not been examined. We present an empirical study of eight dynamic fusion weight prediction strategies across four datasets and four sparse–dense retriever pairs. When we compare these approaches to static fusion and a per-query oracle that bounds achievable performance, several findings emerge. First, the per-query optimal fusion weight is typically an interval rather than a point. Queries can have multiple optimal intervals, and these can take on a variety of shapes. Second, 74 of 128 dynamic fusion methods tested were significantly better than standard RRF after false discovery rate correction. Third, dynamic fusion methods exhibit differing quality and latency characteristics that should be considered when choosing an appropriate fusion method. To address this last finding, we present a tiered decision framework that maps practitioner constraints, dataset properties, label availability, hardware, and latency budget to an appropriate method class. We release the code for our experiments at this https URL.
[IR-36] Candidate-dependent residual routing trades top-of-list accuracy for recall in sequential recommendation
链接: https://arxiv.org/abs/2610.24200
作者: Han Chen,Yingrui Li
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 33 pages, 3 figures. Includes 18 pages of Supplemental Information with 18 supplementary tables
Abstract:Candidate-dependent routing lets a sequential recommender weight a user’s past behaviors differently for each item it scores. We test whether this flexibility helps by adding a small residual routing module to a self-attentive baseline, in two versions that differ only in what sets the routing weights: the user’s history representation or the candidate item’s embedding. Across five paired seeds on the REES46 cosmetics-shop dataset, with matched training for all three models, candidate-dependent routing leaves the primary metric, normalized discounted cumulative gain at cutoff ten (NDCG@10), essentially unchanged (mean relative change +0.003%). Secondary and exploratory analyses show that this flat average hides two effects that cancel. The candidate-dependent model places more relevant items in its top ten (recall +1.81%) but ranks the first one lower (mean reciprocal rank -1.09%), and most of its new hits land at positions six to ten. Its gains come from queries with at least one target among the user’s last 50 events, while queries whose targets all lie outside that window get worse. In this form, candidate-dependent routing trades accuracy at the top of the list for recall further down.
[IR-37] Cache2LoRA: Understanding and Improving Context Internalization
链接: https://arxiv.org/abs/2610.24107
作者: Yi Fang,Gaurav Srivastava,Zhenyu Bi,Sirui Ding,Xuan Wang
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Context-to-LoRA methods encode documents as low-rank parameter updates, allowing language models to answer without retaining the source text in their decoding context. How reliably do these updates preserve facts and support their use under different instructions? Controlled diagnosis of existing context-to-LoRA methods separates fact discrimination, target retention, and response adaptation. Competing sources influence answers, while target-only construction and memory-state optimization improve recovery at fixed update size. Recovering facts also does not ensure that answers follow response instructions. We introduce Cache2LoRA, which uses each query to aggregate evidence from a reusable frozen document cache before generating LoRA updates. Trainable encoding operates only on the query and memory-token suffix, and multi-style supervision teaches both evidence preservation and answer expression. Across 17 tasks, Cache2LoRA improves on existing context-to-LoRA methods and achieves macro F1 of 0.663, approaching full-context in-context learning at 0.671. For 32K documents and 256-token outputs at batch size 16, its decoding throughput is 5.9 \times that of full-context ICL. At matched memory size, generated LoRA preserves answer accuracy more reliably than the learned-KV variant under long, diverse intervening text and combines more effectively with retained document KV, yielding higher hybrid F1 and EM.
[IR-38] Location Location Generation: Where You Ask Matters for Generative Engine Optimization
链接: https://arxiv.org/abs/2610.24081
作者: Mahe Chen,Nick Koudas
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL); Machine Learning (cs.LG); Social and Information Networks (cs.SI)
备注: 4 pages
Abstract:Generative search engines answer local and commercial questions with a written answer and a few cited sources. A growing practice of generative engine optimization (GEO) monitors which sources and brands these answers show, typically from a single place. In this paper, we present a controlled audit of how the user’s location changes the sources that generative engines cite. We ask 300 questions of ChatGPT with 19 declared locations, and of ChatGPT and Gemini from five network locations, each twice to measure run-to-run noise (31,200 answers). The engines take location from different channels. ChatGPT follows the location an application declares, down to the city: it writes the place into its own searches, and its search backend localizes results even when the queries name no place. It ignores the network. Gemini ignores the declared location and follows the country of the network, changing most of its sources even for questions that do not depend on place. The businesses the answers recommend change with the sources. Thus, visibility in generative search is specific to the user’s location, and audits must control both channels.
[IR-39] When Does Evidence-Graph Structure Predict Answer Correctness? Label-Free Abstention for Structured Retrieval
链接: https://arxiv.org/abs/2610.24030
作者: Anjan Goswami
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 34 pages, 5 figures, 37 tables. Under review at ACM Transactions on Information Systems
Abstract:Retrieval systems return scored answers, but a high score does not show that an answer should be trusted. The proposed training-free abstention method applies the boundary cut to the selected evidence in an evidence graph. The signal is the weighted cut between selected and unselected candidates and requires no correctness labels or confidence-model training. On MemoryAgentBench (MAB) FactConsolidation multi-hop, with unresolved queries ranked last for every selector, it raises area under the precision-coverage curve from 0.736 for the best score baseline to 0.785 (+0.049, 95% CI [+0.008, +0.092]); under the native evaluation, where score baselines keep their unresolved values, the gap is +0.108. A zero-cut subset contains 71 of 400 queries at 100% observed precision. In this pipeline, the subset corresponds to structural closure; controlled analyses attribute much of the remaining separation to resolution status, with little benefit from continuous edge weights. Across external benchmarks, usefulness varies with graph construction, edge semantics, and relation; typed structure alone does not ensure useful confidence. A supervised gradient-boosted selector achieves 0.814 versus the boundary cut’s 0.785 when the retrieval stack matches. These findings provide empirical guidance for using available evidence graphs for lightweight, auditable post-retrieval abstention without additional model inference.
人机交互
[HC-0] AI Misconception and Adoption Among Indonesian K-12 Teachers
链接: https://arxiv.org/abs/2610.26758
作者: Alham Fikri Aji,Afifa Amriani,Rendi Chevi,Ayu Purwarianti,Derry Wijaya
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: Accepted at AIED 2026
Abstract:Indonesia’s fragmented geography and uneven infrastructure pose persistent challenges for adopting new educational technologies. While national initiatives aim to mainstream AI in K-12 schools by 2025, policy readiness does not automatically mean teachers are ready for the classroom. This study surveys Indonesian K-12 teachers to examine: (1) their understanding of AI, (2) how they and their students use AI, and (3) their views on the acceptability of AI in learning activities. Through semi-automated thematic analysis, we identify common misconceptions and use cases. Our findings reveal a notable mismatch: while teachers find AI use in education acceptable and support it for the sake of efficiency, they remain fully aware of the risks to learning depth and academic honesty. We also find that this high level of acceptance is currently coupled with a poor understanding of how AI actually works.
[HC-1] Comparing Computer Use Behavior Across Multiple Levels of Abstraction
链接: https://arxiv.org/abs/2610.26703
作者: Ananya Gubbi Mohanbabu,Rosiana Natalie,Anhong Guo,Amy Pavel
类目: Human-Computer Interaction (cs.HC)
备注: 21 pages, 4 figures, 12 tables
Abstract:Interaction traces record how people and agents use computers, and comparing them can reveal usability barriers, differences in expertise, and human–agent similarity. Yet traces are recorded as low-level, interface-specific actions, making comparisons difficult across tasks, interfaces, and input modalities. Existing approaches either compare these actions directly, reduce traces to aggregate outcomes, or summarize them at higher-level goals, obscuring where behavior differs. We present MAP (Multi-level Action Parsing), a hierarchical semantic representation and annotation pipeline that maps traces into a common vocabulary across linked levels of abstraction. MAP compares traces by sequence, recurring patterns, and effort, localizing differences by type and abstraction level. A technical evaluation across three computer use datasets shows high agreement with manual annotations and broad coverage of behavior. Applying MAP shows that sighted and blind users increasingly converge at higher levels despite navigation differences, while agents omit corrective, help-seeking, and recovery behaviors present in human traces.
[HC-2] apID: Rapid touch interaction in virtual reality using wearable sensing
链接: https://arxiv.org/abs/2610.26598
作者: Manuel Meier,Paul Streli,Andreas Fender,Christian Holz
类目: Human-Computer Interaction (cs.HC); Signal Processing (eess.SP)
备注: accepted to IEEE VR 2021
Abstract:Current Virtual Reality systems typically use cameras to capture user input from controllers or free-hand mid-air interaction. In this paper, we argue that this is a key impediment to productivity scenarios in VR, which require continued interaction over prolonged periods of time—a requirement that controller or free-hand input in mid-air does not satisfy. To address this challenge, we bring rapid touch interaction on surfaces to Virtual Reality—the input modality that users have grown used to on phones and tablets for continued use. We present TapID, a wrist-based inertial sensing system that complements headset-tracked hand poses to trigger input in VR. TapID embeds a pair of inertial sensors in a flexible strap, one at either side of the wrist; from the combination of registered signals, TapID reliably detects surface touch events and, more importantly, identifies the finger used for touch. We evaluated TapID in a series of userstudies on event-detection accuracy (F1 = 0.997) and hand-agnostic finger-identification accuracy (within-user: F1 = 0.93; across users: F1 = 0.91 after 10 refinement taps and F1 = 0.87 without refinement) in a seated table scenario. We conclude with a series of applications that complement hand tracking with touch input and that are uniquely enabled by TapID, including UI control, rapid keyboard typing and piano playing, as well as surface gestures.
[HC-3] Small Averag e Error Wide Individual Error: General-Purpose LLM s Estimating Energy Intake from Meal Photos in a Feeding Trial
链接: https://arxiv.org/abs/2610.26538
作者: Xiaomeng Li
类目: Human-Computer Interaction (cs.HC); Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
备注: 20 pages, 1 figure, 12 tables. Code, prompts, pre-registration and per-meal outputs: this https URL
Abstract:Photo-based estimation of food energy could ease dietary assessment, and general-purpose multimodal large language models (LLMs) make such estimates easy to obtain. Most evaluations use one photo taken before eating and compare with energy served, while dietary assessment needs energy eaten. We evaluated six models from two vendors (Claude Haiku, Sonnet and Opus 5.5; GPT-6 luna, 6.1 sol and 6 astra) on 802 meals from ACETADA, a controlled feeding trial with weighed intake, with the before photo only, before and after photos in separate calls (subtraction), or both photos in one request. Three results ran against intuition. Subtraction under-estimated intake for five of six models, because the models still report hundreds of kilocalories for a nearly empty tray. Scaling the before-only estimate by the average share eaten (0.87) matched much of the gain from the after photo; beyond that constant, the second photo lowered meal-level error by only 2.2 to 3.4 points, mainly in meals with food left. With one photo, two models had a near-zero intake bias only because they under-estimated the served energy and ignored uneaten food. In ablations, adding the after photo improved accuracy under both prompt formats for five models, while prompt effects varied across models. Over 148 feeding days, three models kept the 90% confidence interval of their mean error within a pre-set 10% margin, but no model reached the pre-set 30% limits of agreement. Protein was under-estimated by every model, most in protein-dense meals; repeated calls changed error by at most 1.4 points; a high reasoning effort helped only the smallest model (4.2 points) and changed the others by less than 1 point. A low average bias is not evidence of accurate individual intake estimation; such estimates should support, not replace, assessment by a dietitian.
[HC-4] Promptile: Recovering and Reusing Prompting Strategies from LLM Interaction Histories
链接: https://arxiv.org/abs/2610.26485
作者: Qingyang Wan,Chang Liu,Zhoutong Ye,Chengwen Zhang,Yanru Qian,Zhike Wang,Jie Cai,Chun Yu,Yuanchun Shi
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:As people use large language models (LLMs) over time, they develop their own ways of prompting to achieve desired results. These useful experiences often remain scattered and unorganized across conversations. Yet existing personalization approaches primarily capture user preferences or explicitly defined workflows. Through a formative study, we found that users develop prompting strategies that range from implicit instructions to explicit templates from these experiences. We further find that interaction behaviors provide evidence for these strategies, which can be decomposed and recombined for reuse across tasks. Building on these findings, we present Promptile, a framework that identifies and validates users’ prompting strategies from interaction histories. Promptile organizes these strategies into reusable components that can be retrieved, combined, and adapted to users’ current tasks. Our evaluation shows that Promptile accurately recovers and reuses these strategies, better aligning model outputs with users’ expectations while reducing repetitive prompting effort.
[HC-5] Comparing Experiences of Speech-Driven Game Modification in Solo Play Paired Play and Paired Watching Using the Fighting Game PATCH FIGHTER
链接: https://arxiv.org/abs/2610.26447
作者: Maya Grace Torii,Takahito Murakami,Yoichi Ochiai
类目: Human-Computer Interaction (cs.HC)
备注: 17 pages, 5 figures, 4 tables
Abstract:Players often have ideas about a game while playing it, but turning them into changes usually means stepping out of play to edit it. We built PATCH FIGHTER, a two-player fighting game defined by 91 types of adjustable parameters. A large language model (LLM) reads players’ conversations during play and rewrites these parameters between games, so players can change the game just by talking. We conducted a within-subjects experiment with 27 pairs (54 participants) comparing three conditions, solo play, paired play, and paired watching. Participants rated enjoyment after each game and answered a questionnaire on speech and changes after each condition. Paired play was the most enjoyable condition. In paired play, participants also felt more able to say what they thought about the game than in solo play. In interviews, participants valued talking with the partner and checking changes through their own play. Paired watching was less preferred because they could not try the changes themselves. As our design cannot separate the enjoyment of editing from that of play, we outline a design that measures them separately and the evaluation of the resulting games as future work.
[HC-6] Exploring Visual Access and Proactive AI Tutoring for Handwritten Mathematics
链接: https://arxiv.org/abs/2610.26411
作者: Nachiket Kotalwar,Conrad Borchers,Shubham Gandhi,Partha Sarathi Purkayastha,David Yaron,Carolyn P. Rosé
类目: Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 24 pages, 7 figures, 13 tables
Abstract:Handwritten work captures mathematical reasoning that learners may not express in typed or voice-dictated messages. We built a multimodal AI tutoring platform in which a phone camera gives the tutor a live view of learners’ work on paper. In a randomized online study, 117 adults solved calculus problems with Text-Reactive, Visual-Reactive, or Visual-Proactive assistance. Analysis of 2,140 tutor messages showed that visual access shifted assistance toward checking ongoing work, while proactivity enhanced error identification and directive hints. Learners valued feedback grounded in their writing; those who received proactive help rated it as supportive, and about a third wanted it more often. These findings inform when tutors should intervene and what they should address. We release the platform, prompts, and instruments, together with 255,984 timestamped images of handwritten work linked to tutoring conversations, intervention decisions, and learning outcomes, to support multimodal tutoring systems that preserve the affordances of handwritten work while providing adaptive feedback.
[HC-7] From Tasks to Turns: Measuring Human Labor Contributions in Human-Agent Joint Production
链接: https://arxiv.org/abs/2610.26365
作者: Likai Wang,Dezhao Su,Jing Hu,Yunyun Wan,Jingdi Lei,Xinhai Li
类目: Human-Computer Interaction (cs.HC)
备注: 18 pages, 6 figures, 19 tables, including appendices
Abstract:In agent workflows, a brief human instruction can alter subsequent search, reasoning, and tool execution, yet final task outcomes rarely reveal the contribution of an individual instruction to human-agent joint production. We propose a Human Turn-level framework for measuring human contributions, using task quality and token usage to characterize the collaborative production outcomes associated with each human instruction. We compare continuations from adjacent checkpoints under matched models, budgets, and termination conditions, and assess stability through repeated measurements. Across four archived and simulated interaction settings, we conduct 29,124 comparisons between predicted and observed directions of quality and computational change at both Human Turn and session levels. On HumanBench, we obtain 506 valid comparisons across 85 real-user scenarios. Results show that quality and computational changes associated with human guidance depend on task conditions. Expert-style and ordinary-style guidance differ in information provision, correction, and repetition. Even when task quality after continuation is identical, remaining computational demands can differ substantially across checkpoints. Our framework advances the measurement of human labor contributions from tasks and sessions to individual turns, providing a repeatable, fine-grained foundation for studying human contributions and their allocation in human-agent joint production.
[HC-8] Remote Expert Evaluation of Tabletop Avatar Gaze Control in Asymmetric Multi-Scale MR–VR Collaboration
链接: https://arxiv.org/abs/2610.26360
作者: Carlos Mosquera,Neven Elsayed,Ernst Kruijff,Joseph Newman,Valeriya Lehrbaum,Eduardo Veas
类目: Human-Computer Interaction (cs.HC)
备注: Accepted for ACM Symposium on Virtual Reality Software and Technology (VRST) 2026
Abstract:This paper proposes an integrated setup for a multi-scale asymmetric collaboration scenario in the context of a remote-expert use case, focusing on the expert side. In this system, the remote expert (with an exocentric view in Mixed Reality (MR)) uses a Tangible User Interface (TUI) to precisely position and rotate their avatar within a scaled tabletop World-in-Miniature (WIM). Inside the WIM, the expert’s avatar and the operator share the same scale and egocentric view, while the expert retains a global tabletop overview of the plant to support collaboration. Interviews with industry experts during the participatory design phase revealed the need for an egocentric view inside the plant on the remote-expert side to enable full collaboration. We designed and studied three interaction-visualisation combinations to integrate and control the expert avatar’s gaze: VR (fully immersive with direct head control), Head (floating screen with direct head control), and Touch (on-avatar screen with touchpad control). We evaluated these methods with 29 participants across two fault-detection tasks: Same Machine, where both collaborators inspect the same machine, and Twin Machine, where they inspect identical machines in separate locations. The VR condition, which combines immersion, direct head control, and a full-scale view, yielded the strongest overall performance and user experience, while the tabletop avatar setup was well received for collaboration. Our findings inform the design of gaze-control techniques for asymmetric MR-VR collaboration and reveal challenges and opportunities for future work. Comments: Accepted for ACM Symposium on Virtual Reality Software and Technology (VRST) 2026 Subjects: Human-Computer Interaction (cs.HC) Cite as: arXiv:2610.26360 [cs.HC] (or arXiv:2610.26360v1 [cs.HC] for this version) https://doi.org/10.48550/arXiv.2610.26360 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Related DOI: https://doi.org/10.1145/3822517.3848661 Focus to learn more DOI(s) linking to related resources
[HC-9] SplatXR: A Toolkit for Multi-Layered Extended Reality via On-Headset 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.26338
作者: Cyrus Vachha,Mohamed Kari,Parastoo Abtahi
类目: Human-Computer Interaction (cs.HC)
备注: UIST 2026, Project page: this https URL
Abstract:Recent advances in Gaussian splatting, 3D foundation models, and video-passthrough mixed reality offer unprecedented fidelity for capturing and blending physical and virtual worlds. However, established XR methods rely on polygon-based pipelines that treat physical and digital environments as separate domains, making seamless integration across recent novel representations difficult. Because splats lack explicit surface geometry, they impede standard graphics techniques necessary for cross-reality interactions, such as shadow or collision handling. Furthermore, while 3D foundation models can reconstruct physical objects on-the-fly into digital splats, current XR systems fall short of leveraging this capability for user experiences that transition seamlessly between physicality and virtuality. To address this, we organize XR as a multi-layered reality of interacting primitives, letting pixels and objects move across physical and digital boundaries. We contribute SplatXR, a Unity toolkit that operationalizes this structure as building blocks for interactive, immersive apps leveraging 3D Gaussian splatting: a standalone renderer, splat-to-mesh reconstruction, splat physics, splat blending, and offloaded splat object reconstruction via on-headset capture. SplatXR unites vision and graphics techniques to make splat-based XR development broadly available, and we validate it through demonstrative applications and a small developer study.
[HC-10] Bringing physics-based controller policies to virtual reality
链接: https://arxiv.org/abs/2610.26278
作者: Joan Llobera,Pierre Nagorny,Balint Hodossy,Valérie Juillard,Caecilia Charbonnier
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Nowadays, it is easy to set up any element of a VR experience so users can interact with it physically, except for interactive VR characters. Can we extend this to animated VR characters? In this work, we studied whether physics-based character animation (PBCA) controllers can produce high-quality movement in VR-compatible environments (Unity3D, WebXR). We evaluated the quality of the results, as well as the losses at each stage of the processing pipeline: data ingest, training and inference. For data ingest, we computed accuracy metrics related to the retargeting method used. For training, we computed accuracy metrics related to different training policies and different training environments. We developed new JAX implementations of state-of-the-art PBCA controllers (DeepMimic, AMP and ADD) which can be trained in MuJoCo-MJX in minutes, 50 times faster than in a traditional CPU-based training environment. The benefit of these particular implementations is that the resulting control policies can be readily transferred to Unity3D and support cross-platform compilation, including Android, something critical for use in VR production. For inference, we tested these controller policies in Unity3D and WebXR. Our results suggest that DeepMimic and ADD trained in MuJoCo-MJX generate controller policies with better animation fidelity than other methods. To simplify the comparison between controller policies, and different reference animations, we implemented an animation fidelity index (afi). Its results are analogous to other more common metrics, but afi is more intuitive to interpret and thus simplifies the comparison of training methods across the successive stages of the processing pipeline. Subjects: Human-Computer Interaction (cs.HC) Cite as: arXiv:2610.26278 [cs.HC] (or arXiv:2610.26278v1 [cs.HC] for this version) https://doi.org/10.48550/arXiv.2610.26278 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[HC-11] Context-Aware Topological Distillation Graph Network for Asymmetric EEG-Eye Movement Emotion Recognition
链接: https://arxiv.org/abs/2610.26084
作者: Peijun Shi,Xiujin Zhu,Zhenguo Lao,Chee-Onn Chow,Wei Ru Wong,Zhen Liu,Xiaobing Jiang
类目: Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注:
Abstract:EEG-eye movement emotion recognition can support mental-health monitoring and adaptive interfaces, but most fusion models treat the two signals as parallel features, classify each window alone and degrade when eye data are lost. We propose the Context-Aware Topological Distillation Graph Network (CTD-GNet), in which EEG features form the nodes of an electrode graph, and eye movements and preceding windows modulate its edges, nodes and readout, the edges through electrode-specific prompts. Bidirectional distillation of the graph, node relations and predictions couples this multimodal teacher with a single-window EEG student, improving the teacher and aligning the two graphs 35-fold. In leave-one-subject-out tests on SEED and SEED-IV against 17 baselines, CTD-GNet has the highest mean accuracy, 84.91% and 76.17%, significantly above 9 and 6 of 13 eye-based baselines after Holm correction, the smallest cross-subject spread among multimodal models and the highest worst-subject accuracy. It is the most accurate model at every rate of missing eye data on SEED-IV, keeps 66.1% on SEED without eye data, more than any EEG-only model, and stays 9.3 and 11.9 points above all single-window fusion models under eye-movement noise of one standard deviation. On film clips unseen in training, it leads on SEED with 77.1%.
[HC-12] Cling: Supporting Cross-Page Coordination with Human-Directed In-Situ and Small-Step Automation
链接: https://arxiv.org/abs/2610.25879
作者: Zhicheng Huang,Yining Cao,Haijun Xia
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Web-based information work spans independently developed pages, with goals and coordination needs often emerging through browsing. Manual coordination requires repeated low-level actions, yet explicitly specifying operations for automation can interrupt engagement with information that shapes users’ next steps. Our approach constructs reusable, schema-based page representations and identifies coordination actions through correspondences across schemas. We instantiate this approach in Cling, which surfaces coordination opportunities in situ for users to select, configure, and execute incrementally while directing the broader task. A technical evaluation demonstrates accurate identification of useful opportunities across diverse domains. A 12-participant study found that Cling reduced task durations by more than half compared with both manual interaction and an end-to-end AI agent, while requiring 61% fewer clicks and 80% fewer tab switches than manual interaction. Its small-step automation supported control and verification as actions unfolded, while accommodating intent as it emerged and evolved through information encountered on the page.
[HC-13] When Care Expands and Control Declines: Understanding Digital Privacy in End-of-Life Care
链接: https://arxiv.org/abs/2610.25773
作者: Yibo Meng,Bingyi Liu,Lionel Z. Wang,Zhuoran Lu
类目: Computers and Society (cs.CY); Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC)
备注: Accepted by IEEE SP 27, Lionel Z. Wang and Zhuoran Lu are the corresponding authors
Abstract:Terminally ill patients generate extensive digital health data through hospital portals, physiological monitoring devices, consumer health applications, and end-of-life care platforms. Yet their ability to control this data is systematically undermined by the structural conditions of terminal care. Through semi-structured in-depth interviews with 21 terminally ill patients, 11 family caregivers, and 8 physicians recruited from tertiary hospitals, this study examines the digital privacy circumstances and subjective attitudes of patients in end-of-life contexts. Thematic analysis reveals five structural dimensions of digital privacy challenge: continuous bodily digitization without meaningful informed consent; cognitive decline driving involuntary proxy control of accounts; inferential data generation exceeding patients’ epistemic boundaries; dissolution of individual privacy boundaries amid relational network restructuring; and the absence of institutional governance over post-mortem data. Patient privacy attitudes span three profiles: fundamental absence of privacy awareness, sustained active management under severe tool constraints, and perception of problems followed by resigned abandonment. Interviews with caregivers and physicians reveal benevolent substitution in the name of care needs, expansion of access beyond the initial task, and unresolved boundaries under absent institutional guidance. These findings reveal a compounding privacy vulnerability unique to end-of-life contexts and call for privacy-by-design principles, granular permission mechanisms, and institutional governance frameworks tailored to the needs of dying patients.
[HC-14] Voice-Based AI Feedback in Higher Education: A Field Study on Quality and Actionability
链接: https://arxiv.org/abs/2610.25560
作者: Thilo Tamme(Technical University of Munich),Leander Schilling(Technical University of Munich),Anton Hantel(Massachusetts Institute of Technology)
类目: Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 9 pages. Research proposal accepted, following double-blind peer review, for presentation at the 46th Annual Conference of the Strategic Management Society (SMS), Berlin, October 17-20, 2026
Abstract:Student feedback is a critical input for continuous improvement in higher education, yet traditional end-of-semester written surveys suffer from low response rates, superficial answers, satisficing, and limited actionability. Voice-based conversational AI agents can conduct structured interviews at scale, asking open-ended questions and probing follow-ups, and may capture richer qualitative feedback than written instruments. This research proposal asks what characteristics, depth, and strategic value feedback collected through voice-based conversational AI agents has in higher education, and whether such agents elicit detailed, honest, and actionable feedback that can support instructional improvement and institutional learning. Drawing on survey-mode-effects research, the Computers As Social Actors (CASA) paradigm, and dynamic capabilities theory, we develop three hypotheses on elaboration, social presence and disclosure, and actionability. A completed technical pilot at the Technical University of Munich established the feasibility of telephone-based AI interviews; the main field study is planned for the “Advanced Leadership Communication” course at MIT Sloan with approximately 60 students, using a convergent mixed-methods design and a four-dimensional coding framework for actionability (specificity, thematic richness, constructive orientation, contextual depth). The study extends CASA theory to evaluative educational settings, characterizes voice-based AI as a distinct feedback modality, and links micro-level feedback quality to institutional sensing capabilities.
[HC-15] AI-Mediated Employee Voice and Performance: Disentangling Informational and Affective Pathways
链接: https://arxiv.org/abs/2610.25548
作者: Thilo Tamme(Technical University of Munich),Andrej Levin(Technical University of Munich)
类目: Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 10 pages, 1 figure. Research proposal accepted, following double-blind peer review, for presentation at the 46th Annual Conference of the Strategic Management Society (SMS), Berlin, October 17-20, 2026
Abstract:Employee voice – the discretionary expression of ideas, concerns, and suggestions intended to improve organizational functioning – creates value inside firms, yet organizations increasingly replace or augment human-led listening with AI-enabled systems that collect, analyze, and route employee input at scale. Whether algorithmic listening replicates the value-creation mechanisms of human voice, and whether it alters the pathways through which voice translates into performance, remains an open question. This research proposal theorizes that traditional voice bundles two distinct functions, knowledge extraction and relational signaling, and that AI mediation may decouple them: strengthening the informational pathway (implemented improvement measures) while weakening the affective pathway (the experience of being heard). We propose a randomized field intervention at a global strategy consulting firm, covering approximately 2,000 client-facing consultants, that varies the form of listening across three conditions – no structured voice, human-led survey-based voice, and AI-enabled conversational voice – and links weekly team survey data to longitudinal performance ratings. A dual-mediation model is tested in which voice affects performance through (1) implemented process improvements and (2) affective well-being. The study extends voice theory to knowledge-intensive professional services, disentangles the micro-mechanisms of voice, and provides causal evidence on whether AI can serve as an effective listener.
[HC-16] When Interpretations Persist: Downstream Influence in Gaze-Aware Assistants
链接: https://arxiv.org/abs/2610.25512
作者: Mingcong Ding,Chenyan Zhang,Zhuojiang Cai,Dongyang Guo,Babette Bühler,Enkelejda Kasneci
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: 25 pages, including references and appendix
Abstract:AI assistants may carry forward not only observations but also interpretations of what a person is doing or may do next across time. Once retained, these interpretations become context for later predictions and system-triggered outputs. We study whether this representation step changes later behavior in gaze-aware streaming assistance. Holding retrieved events, their order, and the target input fixed, we find that Free-form Interpretation amplifies historical influence in both directions: correct histories produce more helpful changes and wrong histories more harmful changes. Unsupported and broader interpretations about activity or future behavior are associated with greater harmful influence. Evidence-Bounded Interpretation preserves distinctions between observation, inference, uncertainty, and temporal scope, reducing unsupported and overcertain content and lowering wrong-source harmful influence while retaining most helpful influence. The same effect structure appears with a second model. These findings identify persistent interpretation as a consequential design choice for history-aware human-AI systems.
[HC-17] Simulating Infant Senses with Age-Tuned Networks
链接: https://arxiv.org/abs/2610.25377
作者: Somesh Mohapatra
类目: Machine Learning (cs.LG); Human-Computer Interaction (cs.HC); Neurons and Cognition (q-bio.NC)
备注: 21 pages, 7 figures, 2 tables. Code and website: this https URL and this https URL
Abstract:What a newborn sees, hears and feels, and how this changes month by month, has been measured sense by sense, with different methods and in different units. We bring these measurements together for the first two years of life and ask whether small neural networks tuned to a given age can turn them into a quick, inspectable simulation of sensory perception. Each sense’s immaturity is written as a fixed, age-indexed filter on the input, and a small prediction network learns behind it over the first year. The filters alone reproduce the order in which the senses mature, by construction. The networks add what filters cannot: in vision they extract far less than their input allows at early stages (at the newborn stage a toy’s position cannot be read out, R^2 = 0.00 , against 0.57 for a network trained directly on the same input), and a critical period and perceptual narrowing emerge without being built in; in hearing they track what their input allows. The same filters drive the Baby Senses Lab, a free website that shows a user’s own images and sounds as a baby of a chosen age might sense them. It is a research and teaching tool, not a diagnostic test.
[HC-18] UnexpectedParrots: Iterating Behavioral Experiment Design with In-Silico Simulations
链接: https://arxiv.org/abs/2610.25343
作者: Sanchaita Hazra,Bodhisattwa Prasad Majumder,Pao Siangliulue,Peter Clark,Amy X. Zhang,Joseph Chang
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Running behavioral experiments is expensive, slow, and prone to design flaws, often detectable after collecting data. We introduce UnexpectedParrots to iteratively refine behavioral study designs before a researcher recruits costly human participants. UnexpectedParrots relies on reasoning traces in addition to responses from LLM participants to uncover diagnostic critiques of structural errors and missed opportunities in existing study designs. When evaluated against a human-pilot or a no-simulation design-only baseline, UnexpectedParrots adds most value, producing unique, high-quality critiques. In a moderated user study with 12 proficient researchers, we find UnexpectedParrots useful, time-saving, and thought-provoking. Critiques provided by UnexpectedParrots often converge on issues researchers identify only during independent human pilots or additional formative interviews, even when the fidelity of the simulated populations remains questionable. Our users deem UnexpectedParrots useful not only as a tool to conduct early pilot studies and identify errors but also as an ideation tool to expand one’s research horizon.
[HC-19] Its Complicated: Factors Influencing the Effectiveness of LLM -Generated Feedback in a Live Programming Course
链接: https://arxiv.org/abs/2610.25244
作者: Tsvetomila Mihaylova,Arto Hellas,Bita Akram,Narges Norouzi,Peter Brusilovsky,Juho Leinonen
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:We report on three iterations of an LLM-generated feedback system deployed in a live web development course, spanning three underlying models and three feedback formats. We ask how feedback presence and its characteristics (length, complexity, format) relate to student performance, whether prior feedback exposure is associated with final-project performance, and how performance changes after first exposure. Feedback was associated with more favorable immediate outcomes in several comparisons, but the pattern varied by outcome and period. In pooled models, longer feedback was associated with lower next-submission correctness, more remaining attempts, and longer resubmission time. Prior same-part feedback was associated with lower final-project solve probability. Student-level comparisons found differences on final-project steps begun before exposure, while within-student changes across steps were uncertain. These comparisons do not identify the causal effect of feedback. Changes on standard exercises were more favorable among exposed students, but the groups’ different course-stage splits limit causal interpretation. Bounded message length and student choice of format are directions for experimental evaluation. Our final-project analysis captures follow-up performance rather than directly measuring learning or understanding.
[HC-20] Leverag ing Mental Models for Human-Centered Explanations: A Systematic Design Approach Illustrated by Explainable Fake Review Identification
链接: https://arxiv.org/abs/2610.24903
作者: Sidra Naveed(1),Md Shajalal(1),Gunnar Stevens(1 and 2),Alexander Boden(2) ((1) University of Siegen, Germany, (2) Hochschule Bonn-Rhein-Sieg (H-BRS), Germany)
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: 38 pages, 10 figures, 8 tables. Keywords: Explainable Artificial Intelligence (XAI); Human-centered AI; Mental models; Fake review detection; User-centered explanation design; Natural language explanations; Cognitive heuristics
Abstract:Explainable AI (XAI) often prioritizes model fidelity over user intelligibility, yielding explanations that satisfy technical transparency yet fail to support human sense-making. We propose a human-centered approach, MIND-X (Mental-model INtegrated Design for eXplainability), which aligns explanations with users’ mental models, commonsense concepts, and judgment heuristics. Using fake review detection as a case study, we contribute: (1) a qualitative study (n=10) that elicits how people assess review credibility and produces an actionable, design-guiding taxonomy of perceptual cues; (2) a set of design principles that link those cues to explanation strategies and prompt LLMs to generate adaptive, language-based explanations; and (3) a quantitative evaluation (n=166) showing that these explanations offer distinct advantages over token-level transparency methods widely used in explainable text classification in terms of understandability, satisfaction, perceived trust, persuasiveness, and decision-making support. Rather than claiming general superiority, our evaluation illustrates how grounding explanations in users’ everyday reasoning practices can produce more intelligible and context-sensitive accounts. Our findings advance human-centered explanation design beyond a mere “gloss of transparency,” demonstrating that a mental-model-aligned design process can improve intelligibility and support accountability across user-centered explanation tasks.
[HC-21] Predicting Individual Differences in Lipreading From Eye-Tracking and Pupillometry During Silent Visual Speech
链接: https://arxiv.org/abs/2610.24898
作者: Puneet Kumar,Satyarth Malik,Gao Zheng,Guoying Zhao,Xiaobai Li
类目: Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注: Submitted to the Journal of Speech, Language, and Hearing Research. 5 figures, 2 tables
Abstract:Lipreading ability varies widely, including among people who are hard of hearing. Although mouth-directed attention is linked to speechreading outcomes, it remains unclear how much lipreading skill can be inferred from viewing behavior and which signals carry this information. We examine whether individual differences in lipreading ability are reflected in viewing behavior during silent speech and quantify the contributions of gaze dynamics, pupil-linked signals, and spatial viewing allocation. We introduce the Lipreading Perception Study (LiPS) dataset and a tracker-independent preprocessing pipeline that synchronizes SR Research EyeLink gaze and pupil recordings with the stimuli. Twenty-four adults (age 32 \pm 10 years) completed a four-alternative forced-choice silent lipreading task. An extreme-groups protocol yielded 18 participants labeled as stronger or weaker lipreaders. Each participant-trial included four aligned views: lip-region images, frame-level gaze, pupil summaries, and spatial viewing proportions. A four-branch neural network with cross-modal attention predicted labels using participant-disjoint splits and systematic ablations. The full model reached accuracy 0.680 \pm 0.040 across five folds, outperforming naive fusion baselines in accuracy. The pupil and viewing-summary branch was most informative: removing it caused the largest degradation (0.618 accuracy, 0.623 AUC), while it alone achieved 0.675 accuracy. The lip-image stream contributed little to participant-level discrimination. Individual differences in lipreading are reflected in pupil-linked and frame-level gaze signals rather than in how much time a viewer spends looking at the mouth. Synchronized gaze and pupillary measures may provide objective markers of visual speech perception, warranting investigation in larger, clinically diverse populations. LiPS provides a resource for extending this work.
[HC-22] Collaborative Hand Drawing to Conceptualize Usable Privacy and Security Research Problems: Students Perspectives (Extended Version)
链接: https://arxiv.org/abs/2610.24748
作者: Eman Alashwali,Joud Alraimi,Maryam Almaghrabi
类目: Human-Computer Interaction (cs.HC); Cryptography and Security (cs.CR); Computers and Society (cs.CY)
备注: This document is the authors’ extended preprint for a short paper in the proceedings of Koli Calling’26, ACM (DOI: https://doi.org/10.1145/3856208.3856277 )
Abstract:In this paper, we explore students’ perspectives on incorporating hand drawing to conceptualize usable privacy and security research problems. We surveyed 64 female students who participated in an in-class collaborative hand drawing activity to conceptualize usable privacy and security research problems as part of a group-based student research project on usable privacy and security in an information and computer security course for undergraduate students. Our survey responses suggest that students perceived the drawing exercise positively. That is, they found it helpful in several aspects, enjoyable, and they recommend it to other students taking the course. The students also said that they would like to have such an exercise in other courses, such as Principles of Human-Computer Interaction and Fundamentals of Data Networks. Only three students mentioned downsides of the exercise, namely doubts about their drawing skills. Our results provide useful insights for educators seeking to enhance the student learning experience.
[HC-23] Say It or Grab It? Conversational Control of UI Adaptations in Mixed Reality
链接: https://arxiv.org/abs/2610.24640
作者: Sina Elahimanesh,Joao Marcelo Evangelista Belo,Isidora Jeknic,Alexander Koller,Anna Maria Feit
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Mixed reality interfaces must continuously adapt to a user’s changing task and environment, yet adaptation policies are fixed in advance, either preset by the system or configured by the user. Natural language offers a way to state changing needs mid-task, but whether users can express interface behavior this way, and what it gains over direct manipulation, is unknown. We first elicited how users verbalize adaptation needs (N=19). We then built ConverseMR, which translates a spoken request into adaptation objectives and constraints and satisfies them continuously as the user’s context changes. In a within-subjects study (N=24), conversational and direct control proved complementary: direct manipulation for precision, ConverseMR for reach and expressive breadth, with their combination rated highest. Spoken commands were rarely final and users frequently refined and replaced them upon experiencing the result, characteristic of, what we call, user-led adaptation, where users communicate adaption needs and the system identifies a solution.
[HC-24] From Neglect to Resignation: Understanding the Evolution of Mobile Privacy
链接: https://arxiv.org/abs/2610.24389
作者: Yibo Meng,Ziqi Liu,Lionel Z. Wang,Shuning Zhang,Xiaolan Ding
类目: Human-Computer Interaction (cs.HC)
备注: Accepted by IMWUT 2026
Abstract:This decade-long longitudinal qualitative study follows 31 smartphone users in China across four interview waves (2014-2023) to examine how mobile privacy understanding evolves alongside shifting computing infrastructures. Findings reveal three stages shaped by distinct interactional mechanisms: early functional acceptance of permission-based data access, growing vigilance triggered by behavioral tracking feedback loops, and eventual resignation as AI-driven inference made ecosystem-level profiling feel inescapable. Rather than a stable attitude, digital resignation emerges through cumulative infrastructure exposure, as each new system layer retroactively reinterpreted earlier permissions and exhausted users’ capacity for self-management. The study challenges permission-centric design assumptions and proposes inference visibility, context isolation, and automated privacy agency as system-level responses.
[HC-25] GazeAgent : VLM-Guided Implicit Gaze Calibration from Egocentric Multimodal Behavior for AI Glasses
链接: https://arxiv.org/abs/2610.24364
作者: Yuhong Zhang,Khanh An Nguyen,Xinyu Tian,KungYi Chang,Abraham Padua,Yuchen Xu,Ying Choon Wu,Gert Cauwenberghs,Tzyy-Ping Jung
类目: Human-Computer Interaction (cs.HC); Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 14 figures, 3 tables
Abstract:AI glasses typically require explicit, user-specific eye-tracker calibration, and accuracy can drift when the device shifts during use. We present GazeAgent, a VLM-guided framework that continuously and adaptively improves gaze-calibration accuracy from naturally occurring behavior without asking users to interrupt their activities. GazeAgent identifies precise visual attention (PVA) events from multimodal streams of inertial measurement unit (IMU), gaze, and wearer-hand signals, validates their task relevance, and localizes functional targets as pseudo-calibration targets. In a 10-participant scripted study, the system detected all 820 predefined PVA opportunities and, under device disturbance, reduced median gaze error by 44.1% compared with explicit calibration. In a five-participant case study, autonomous calibration produced the lowest workload and interruption ratings, and all participants preferred it for routine use. By deriving task-critical targets from natural multimodal behavior, GazeAgent demonstrates the potential for continuous, adaptive, and low-burden gaze interaction on egocentric AI glasses.
[HC-26] Indirect Influences of Artificial Intelligence Advice on Complex Sepsis Treatment Decisions
链接: https://arxiv.org/abs/2610.24313
作者: Venkatesh Sivaraman,Joel Levin,Eric P. Mason,Victor Talisa,Suchi Saria,Jean Feng,Andrew J. King,Jeremy M. Kahn,Adam Perer
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Quantitative Methods (q-bio.QM)
备注:
Abstract:Sepsis is a major public health concern, and many sepsis patients die due to ineffective hemodynamic management. Artificial intelligence could support hemodynamic decisions, particularly in complex cases where conflicting information and comorbidities drive uncertainty and practice variation. However, these systems’ real-world value depends on not just accuracy, but how they influence the quality of clinician decisions. We first retrospectively evaluated an AI system’s recommendations in 6,928 sepsis patients, finding that they improved upon the original providers’ decisions overall but had significantly worse performance in complex cases. We then conducted a pre-registered experiment with 228 US critical care physicians to test their use of this imperfect advice in six complex treatment decisions, using one of three advice types designed to vary reasoning autonomy. Participants given simple model recommendations overcame poor advice, choosing optimal treatments in 56.3% of decisions compared to 31.8% if they had always followed the recommendations. They also made significantly better decisions than those receiving a less prescriptive advice type (42.9%). Physicians selectively adapted and improved weak advice, likely because it prompted them to reconsider their assessment. Our findings suggest that improving sepsis care decisions with AI may require accounting for these indirect influences resulting from human-AI collaborative reasoning.
[HC-27] Desirability Confirmation: Relational Sycophancy in AI Intimate Roleplay
链接: https://arxiv.org/abs/2610.24265
作者: Huiqian Lai,Runze Ding
类目: Computers and Society (cs.CY); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注: Accepted for publication in the Proceedings of the 60th Hawaii International Conference on System Sciences (HICSS-60), 2027
Abstract:Research on LLM sycophancy has largely focused on propositional agreement, in which models echo users’ beliefs, preferences, or judgments. Adult intimate roleplay presents a different case because what the user seeks may be preservation of a preferred relational position within a generated scene. Using a controlled prompt-based audit of 2,016 outputs from two open-weight LLMs, we identify four recurrent relational moves: legitimating desire, presenting the user as wanted, preserving non-rejection while communicating limits, and keeping intimacy available within fantasy or roleplay frames. Bringing the sycophancy literature into dialogue with sexual script theory, we conceptualize this configuration as desirability confirmation, an output-level pattern that captures how user-pleasing can operate through relational positioning. The findings extend sycophancy research and intimate-AI evaluation beyond binary compliance or refusal by showing how outputs can differ in the ways they communicate boundaries, preserve relational acceptance, and mark intimacy as fictional.
[HC-28] Evaluating Lightweight Contact Feedback in Dual-Operator Collaborative Humanoid Teleoperation
链接: https://arxiv.org/abs/2610.24257
作者: Austin Graves,Chuizheng Kong,Yunho Cho,Yuanshao Yang,Advaith Nidumukkala,Shreyas Kousik,Mohsen Moghaddam
类目: Robotics (cs.RO); Human-Computer Interaction (cs.HC)
备注: 8 pages, 4 figures, 2 tables. Project website: this https URL
Abstract:Dual-operator humanoid teleoperation enables collaborative manipulation through two separate robots, but operators lack direct awareness of robot-object contact that is normally available during physical interaction. To address this, we present eXtended Reality Teleoperation (XRT), an extended reality teleoperation framework that returns a robot’s fingertip contact through lightweight vibrotactile and/or auditory feedback, and a synchronized dual-humanoid testbed built on it for two operators. Sixteen dyads performed a collaborative crate lift-and-flip under vision only and the three feedback conditions in a counterbalanced within-subject design with verified feedback-channel delivery. Three results follow. First, operators rated our finger-localized vibrotactile display as more effective than our open-ear sonification: all 32 ranked it above the sonification, 18 ranked it most effective, and its cues were rated far easier to perceive and interpret than the auditory cues. Second, completion time, failures, workload, discomfort, and partner awareness showed no detected condition differences, with wide completion-time intervals and smaller meaningful effects unresolved. Third, performance changed substantially with practice, especially on the harder flip-over task: completion time improved across blocks, and attempt-to-attempt variation within a dyad (SD x1.51) was larger than the largest condition contrast (x1.16). Exploratory analyses detected no difference in immediate grip or motion responses between feedback conditions. For novice collaborative teleoperation, our fingertip vibrotactile cues were rated more effective and easier to interpret than our auditory cues, while their effect on task performance remains unresolved.
[HC-29] Usability User Experience and Acceptability of the Trail Making Test in Virtual Reality in Younger and Older Adults
链接: https://arxiv.org/abs/2610.24237
作者: Panagiotis Kourtesis,Dante O’Connor,Sarah E. MacPherson
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: 20 pages, 3 figures, 4 tables, supplementary material are included
Abstract:Immersive virtual reality (VR) adaptations of neuropsychological tests change how a test operates. Before their scores are interpreted, the individuals they are intended to assess should find them usable, acceptable and tolerable, and for older adults this evidence is usually missing. This study evaluated the experience of the Trail Making Test in Virtual Reality (TMT-VR), a head-gaze adaptation of the Trail Making Test that requires no hand-held controller, in 47 younger (20-38 years) and 44 older adults (64-81 years). After the task, participants rated usability, pragmatic and hedonic user experience, acceptability, and cybersickness. The study had three aims: to compare the two age groups on these ratings; to examine whether the ratings were associated with age group, gaming and technology experience, and whether cybersickness was associated with motion sickness susceptibility; and to examine whether the ratings were related to TMT-VR completion and error-adjusted times. Both groups rated the TMT-VR as highly usable, acceptable and pleasant to use. Older adults gave slightly lower usability and pragmatic ratings, but neither the age-group differences nor the associations with age group or previous experience were statistically significant after correction. Cybersickness was low, with 38 of the 91 participants reporting no symptoms; it did not differ between the age groups and was not associated with motion sickness susceptibility. No correlation between the ratings and TMT-VR performance survived correction. The TMT-VR with head-gaze selection is therefore usable, acceptable and well tolerated by older as well as younger adults. Because the ratings were favourable across the whole range of TMT-VR scores, slower or less accurate performance is unlikely to reflect difficulty with the interface.
[HC-30] rinketry: Materializing Elements to Anchor Traceable Design Histories
链接: https://arxiv.org/abs/2610.24221
作者: Wen-Fan Wang,Yi-Ting Chiu,You-Yi Hsieh,Bing-Yu Chen,Max Kreminski
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Generative tools enable visual designers to rapidly explore diverse design directions—but as flawed generated outputs proliferate, designers may lose track of promising ideas and the context of their development. We introduce trinkets: user-preserved design elements that function as reusable materials, retrieval queries, and navigational anchors for tracing how ideas emerge, transform, and recur throughout design history. We implement this concept in Trinketry, a visual exploration canvas. In a within-subjects study with 12 concept designers, Trinketry supported focused element-level exploration and greater perceived control than a simple branching image generation history baseline. A follow-up study with six returning designers examined delayed design history sensemaking, continuation of work, and reconstruction of others’ processes. Compared with whole-image histories, Trinketry supported more precise reconstruction of process structure, turning points, and goals, while helping designers recover overlooked materials and continue exploration. However, opportunities remain to augment trace representations with explicit statements of design rationale.
[HC-31] Landmark-Free and Scale-Blind Multi-View Registration of Preoperative Liver Models to Intraoperative Surfaces
链接: https://arxiv.org/abs/2610.24218
作者: Jiaming Feng,Gerardo Loza Galindo,Shahid Farid,Sharib Ali
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 10 pages
Abstract:Augmented-reality guidance in laparoscopic liver resection requires aligning preoperative 3D liver models from CT/MRI with intraoperative surfaces. Monocular reconstruction in most cases cannot guarantee scale, albeit rigid-motion registration requires scale as an input. Our proposed framework ``SimCompose’’ estimates the full similarity transformation \Sim , including scale, without relying on landmarks and applies the same transformation across multiple views with known relative geometry. However, reliable detection of registration failures remains challenging in surgical settings, where ground-truth alignment is generally unavailable. For this, we propose a geometry-based selector that evaluates candidate transformations without requiring ground truth. Large disagreement among these transformations is used to flag potentially unreliable registrations. Finally, we propose to use a non-rigid refinement based on a low-dimensional deformation model from Neo-Hookean simulations in the Simulation Open Framework Architecture (SOFA), without additional network training. SimCompose is trained exclusively on synthetic partial views. Its encoder processes point clouds normalised to a unit root-mean-square radius, without access to metric scale, and filters each view before multi-view fusion. Evaluation on unseen synthetic data (300 single-view pairs per run; 100 three-view registrations) shows that multi-view integration and candidate selection increase the success rate under stringent \Sim accuracy criteria from 38.4% for a single-view, single-method configuration to 77%. Existing learning-based baselines do not satisfy these criteria even when provided with the exact scale. On five in-vivo patients with 92 keyframes, the rigid registration stage reduces surface error more than three-fold relative to learning-based baselines with external scale information.
[HC-32] You Already Have Me–Why Do You Still Use Blued? How Gay Dating Apps Shape Intimate Relationships Among Chinese Gay Men
链接: https://arxiv.org/abs/2610.24215
作者: Yibo Meng,Lyumanshan Ye,Bingyi Liu,Linghao Li,Huidi Lu,Xiaolan Ding
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Gay dating applications such as Blued are widely used for meeting partners, yet their role does not end once an intimate relationship is formed. When these platforms persist within an established relationship, they transform from matchmaking tools into a form of relational infrastructure that both partners must continuously manage. Building on existing CSCW and HCI research that has primarily examined how social media triggers jealousy, surveillance, or trust issues in intimate relationships, this study investigates how the persistent presence of gay dating applications shapes the dynamics of intimate relationships among Chinese gay men. Through semi-structured interviews with 23 self-identified gay men aged 19 to 55, we identify a recurring cyclical process: platform signals-such as online status, profile updates, distance indicators, and usage records-trigger suspicion and conflict; partners invest negotiation labor through intent explanation, visibility control, rule formation, or silence; and these negotiations eventually sediment into temporary governance settlements that reduce trigger frequency but remain conditionally stable. We identify four types of governance settlements: infrastructure removal, transparency governance, silence agreements, and layered/open agreements. Each settlement addresses the governance problems of platform presence, visibility distribution, and loyalty boundaries differently, yet none achieves permanent stability due to platforms’ reversible entry mechanisms and continuous visibility cues. Our findings extend existing CSCW research on digital technology and intimate relationships by revealing a more complex process than previously documented: platform signals not only trigger emotional responses but prompt partners to engage in sustained relational negotiation through explanation, visibility management, rule formation, or silence.
[HC-33] Navigating Deception on Gay Dating Apps: A Qualitative Study of Users Experiences Verification Strategies and Community Knowledge in China
链接: https://arxiv.org/abs/2610.24207
作者: Yibo Meng,Lyumanshan Ye,Bingyi Liu,Huidi Lu,Xiaolan Ding
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Gay dating applications have become essential platforms for sexual minority men to seek relationships, intimacy, and community, yet they also expose users to a range of deceptive interactions that remain underexplored in HCI and CSCW research. This study examines how gay male users in China experience, identify, and respond to deception on dating applications. Through semi-structured interviews with 22 participants who used platforms such as Blued, Aloha, Fanka, and Soul, we identify a typology of deceptive practices that users encounter, including relational misrepresentation for sexual access, sustained emotional deception, financial fraud, marketing behavior disguised as ordinary users, and false identity construction. We further analyze the verification and coping strategies that users develop through long-term platform use, such as judging relational intent through interactional pace, identifying financial risk signals, recognizing commercial promotion accounts, and progressively verifying identity authenticity. Finally, we show that risk recognition is not merely an individual capacity but a collaborative practice that gradually emerges through the circulation of experience within the community, the summarization of recurrent deceptive tactics, and the learning of shared rules. Our findings contribute to CSCW scholarship by demonstrating how users of a marginalized community collectively construct practical knowledge about deception, and by revealing that trust assessment on dating platforms operates as a layered, probabilistic, and socially distributed process rather than a binary judgment.
[HC-34] Over reliant and underprepared: First responder resilience to disruption of operational technology
链接: https://arxiv.org/abs/2610.24186
作者: James Connor,Courtney Crooks,Ryan Shandler
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: 33 pages, 1 figure, 5 tables
Abstract:The Emergency Services Sector (ESS) is responsible for taking life-saving action in the face of extreme crises. To enhance its response capabilities, the sector has spent the last two decades investing in new technology to improve their operational efficiency. This has created significant interdependencies and a culture among first responders of technical overreliance, raising questions about what would occur if those tools were suddenly disrupted. Several works have discussed the potential for terrorism to intentionally target ESS technology as part of a larger kinetic attack. Despite this, the resilience of first responder agencies to technological disruption and their preparedness for such an attack has largely remained unexplored. To bridge this gap, we implemented a novel scenario-based interview methodology involving senior officials from emergency service agencies across the United States. These interviews explored how each agency prepared for a disruption of operational technology and what factors might influence different levels of preparation between them. Participants were asked to describe their responses to a hypothetical scenario modeled after historical cyber-attacks against ESS infrastructure. The results indicate that while ESS agencies are prepared for short-term technological disruptions, they are likely to struggle with long-term or hybrid incidents. The data further revealed that agencies often lack dedicated expertise in cybersecurity, technology, and emergency management that could otherwise elevate resilience to interruptions of operational technology. These findings emphasize the need for investment in redundant analog systems, increased training for frontline personnel, and a focused effort against loss of specialized technical expertise among ESS agencies.
[HC-35] Modeling Group Socialization in Online Communities
链接: https://arxiv.org/abs/2610.24140
作者: Vinay Koshy,Iris Zhong,Karrie Karahalios,Benjamin Mako Hill,Amy X. Zhang
类目: ocial and Information Networks (cs.SI); Human-Computer Interaction (cs.HC)
备注:
Abstract:Newcomers to online communities must learn a group’s norms to participate successfully. This process is known as socialization, and has been the subject of considerable study in the online communities literature. Foundational work on group socialization in social psychology, however, suggests that socialization is actually a bidirectional process: individuals and communities adapt to each other’s needs. In this work we develop a formal quantitative model of socialization, and present the first large scale empirical validation of the bidirectional socialization hypothesis using Reddit data. More specifically, we focus on linguistic socialization, and model users’ comments in online communities as a mixture of in-group language and language from ``background’’ communities they previously participated in. On a dataset of 112 subreddits (7.8 million tokens) find that our bidirectional model makes predictions that better generalize across users and time than two unidirectional socialization models. Furthermore, we conduct a qualitative analysis in which we demonstrate that our model is able to extract interpretable socialization patterns from the data that align with the existing literature. Specifically our model is able to learn that users conform more to group norms in topically niche communities with stricter content moderation rules, and that long time users conform more than newcomers. We emphasize that the claims made in this paper are explicitly falsifiable – we are staking out a clear claim about what the socialization process looks like, and encourage future work to challenge our model.
[HC-36] From Prefabricated Content to Situated Generation: A Four-Wave Longitudinal Study of AI-Mediated Museum Surfaces and Spaces
链接: https://arxiv.org/abs/2610.24076
作者: Yibo Meng,Xiujun Chen,Bingyi Liu,Xiaolan Ding
类目: Human-Computer Interaction (cs.HC)
备注: ACM ISS 2026 PACMHCI full paper; revised manuscript; 5 figures; 3 tables; accepted paper
Abstract:Generative AI changes a basic condition of interactive surfaces and spaces (ISS): visitor-facing content can be produced during interaction rather than fully authored before it. We report a longitudinal qualitative study with 23 Chinese science museum practitioners, interviewed annually from 2023 to 2026 (retention 23 to 19) and analyzed using reflexive thematic analysis. Participants’ accounts moved from anticipated epistemic and operational risks, through backstage experimentation and visitor-facing trials, toward situated governance and renewed questions about the purpose of museum interaction. Following these changes over time shows how generative AI reconfigured the design problems of museum ISS: generated content appeared under institutional authority; inference delay became publicly and bodily experienced; concurrent touch, gesture, and speech created ambiguous multi-user intent; and post-generation editing could leave curators selecting within an AI-defined possibility space. We conceptualize the resulting shift as a movement from prefabricated display to situated generation and formulate four design propositions concerning provenance at the point of display, spatially appropriate fallback, multi-intent negotiation, and curatorial control before generation. These propositions are grounded in practitioner accounts rather than evaluations of implemented solutions.
[HC-37] Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation
链接: https://arxiv.org/abs/2610.24010
作者: Renye Yan,Jialun Zhong,Jikang Cheng,Yi Gao,Yanzeng Li,Ling Liang
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注: Human-AI Co-Creation, Reinforcement Learning
Abstract:Generative AI is becoming an important medium for artistic creation, yet existing interactions often assume that creators can specify intentions in advance and accurately communicate them through text prompts. Artistic intentions, however, may be vague, latent, or recognizable only after seeing concrete visual possibilities. We propose a human-AI co-creation paradigm beyond explicit intent expression. By expanding model exploration, we expose artists to diverse visual possibilities, allowing them to identify previously unarticulated preferences through observation, comparison, and selection. We then integrate artists’ curatorial judgments into reinforcement learning, assigning different learning strengths based on preferences over generation trajectories, so recognized directions are more strongly reinforced and the model gradually adapts to individual aesthetics. This creates a continuous “explore-identify-shape-re-explore” loop in which artistic intent is progressively discovered, externalized, and shaped. Through three user studies, we examine latent-preference discovery, personalized preference alignment, and the externalization of vague, difficult-to-express creative intentions.
[HC-38] When Better Opportunities Backfire: Service Worker Responses to Parallel AI Deployment in Customer Acquisition
链接: https://arxiv.org/abs/2610.25603
作者: Xinyu Fu,Shulong Luo,Narayan Ramasubbu,Peng Shen,Xincheng Wang,Yucheng Wang
类目: General Economics (econ.GN); Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET); Human-Computer Interaction (cs.HC)
备注:
Abstract:As AI voice agents get deployed in high-volume contact centers, firms can route low-intent leads to AI and allocate high-intent leads to human workers. Drawing on sequential search theory, we posit that AI-induced task allocation reshapes work, enriching opportunities facing human workers but also raising their reservation value for continuing with the lead in hand, making them less willing to persist. We study the effects of an AI voice-agent deployment at a Chinese online health-insurance brokerage using more than a half million customer outreach records surrounding the rollout. Adopting a regression-discontinuity-in-time design, we trace how worker behavior changes across the workday. Results show that the enrichment of human-facing opportunities concentrates early in the workday, raising the probability that a call is a high-intent lead by about 23 percentage points on the first three calls and fading to zero later. However, customer acquisition by human workers declines precisely where the enrichment is greatest, falling about 5 percentage points on the first three calls and flattening over the day. Consistent with the theorized reservation value mechanism, workers reallocate their earliest leads about 8 percentage points more often and persist less with them, shortening those calls by about 13%. A calibrated decomposition analysis further shows that the modest gain from improved lead ordering is outweighed by a much larger negative behavioral response. By reshaping the quality and within-workday distribution of opportunities, parallel AI deployment alters how workers pursue tasks, highlighting opportunity distribution as an important service operations state variable beyond workload and congestion.
[HC-39] Is User Calibration Still Necessary for Continuous Myoelectric Control? A Closed-Loop Comparison of Calibration-Free Controllers
链接: https://arxiv.org/abs/2610.24802
作者: Amir Reza Hariri,Kevin B. Englehart,Erik Scheme
类目: ignal Processing (eess.SP); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注: Submitted to IEEE Transactions on Neural Systems and Rehabilitation Engineering
Abstract:The need for user-specific calibration remains an often-cited barrier to widespread adoption of myoelectric control. Cross-user models trained on large multi-user datasets can now recognize gestures from unseen users without calibration, but they have been evaluated almost exclusively offline, leaving open whether calibration is still needed in closed-loop control. This study compared six calibration-free controllers, trained on 306 users of the EMG-EPN-612 dataset and used without any calibration data from the test participant, against three calibrated controllers in a closed-loop Fitts’ law target-acquisition task with sixteen participants. The calibrated controllers achieved higher offline balanced accuracy (90.6% against 73.1% for the best controller of each paradigm), but completion rates did not differ significantly between paradigms, and the highest completion rate was achieved by a calibration-free controller. Calibrated controllers moved faster on successful acquisitions, partly because of their individually fitted speed mappings, though not significantly after correction for multiple comparisons. However, every calibration-free controller produced more direct and stable trajectories than every calibrated controller, even after adjusting for cursor speed. Offline accuracy did not predict completion rate across controllers or within participants, and across paradigms higher offline accuracy was associated with poorer trajectory quality in 15 of 16 participants. Without any calibration data from the end user, a model trained across a large population matched calibrated control in completion rate and exceeded it in trajectory quality. The offline accuracy gained through calibration should therefore not be taken as evidence of a closed-loop benefit.
[HC-40] RACE: Governing AI-Assisted Research with Principles from Computer Science
链接: https://arxiv.org/abs/2610.24035
作者: Amirhossein Zohrehvand,Aldona Kapačinskaitė
类目: General Economics (econ.GN); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 27 pages, 7 tables
Abstract:Artificial intelligence (AI) makes it easier than ever to produce research outputs. The pressing question is how researchers should organize their work with this technology. The question is not new: software engineering and security research have long studied how teams of people and machines keep control of a shared body of work with many contributors. We distil principles from the practices these literatures report and transpose them to AI-assisted research. The principles form TRACE: Track records, Review against sources, Accept independently, Control transitions, and Evaluate corrections. Together, the five principles give researchers a common basis for remaining in control while integrating AI into their workflows. By bounding units of work and attending to how each is shaped, they help researchers pace AI-assisted work to human attention rather than to the speed at which AI produces content. We also introduce TRACE-OS, an open source operating system for AI-assisted research projects. It puts the principles into practice through shared instructions, tools, and records that fit researchers’ existing workflows. The shared standard can contribute to building trust, which allows teams to delegate and realize the productivity that AI promises. More broadly, we aim to open a conversation about the governing principles through which the research community can benefit from AI.
计算机视觉
[CV-0] xt-to-Image Generative Modeling Without Language Pretraining
链接: https://arxiv.org/abs/2610.26863
作者: Keshigeyan Chandrasegaran,Yue Zhao,Armin W. Thomas,Ehsan Adeli,Jiajun Wu,Juan Carlos Niebles,Li Fei-Fei,Michael Poli
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages; Project website: this https URL
Abstract:Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2 \times less pretraining compute. Further, Halcyon-1.6B outperforms JiT-T2I-XXL/8++ (35.3 vs. 41.0 gFD) with 275 \times less pretraining compute. Controlled experiments show that text-only pretraining does not improve T2I generation. Halcyon also learns image captioning by replacing caption dropout with I2T supervision, maintaining competitive T2I generation quality under the same pretraining budget. We show that strong T2I systems can be trained from scratch without internet-scale language pretraining. Code and models are available at this https URL
[CV-1] VideoMaMa: Temporally Consistent Video Matting via Preserve-and-Refine Token NEURIPS2026
链接: https://arxiv.org/abs/2610.26857
作者: Sangbeom Lim,Seoung Wug Oh,Heeji Yoon,Seungryong Kim,Joon-Young Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: NeurIPS 2026. Project page: this https URL Code: this https URL
Abstract:Diffusion-based video matting has recently emerged as a leading approach, leveraging video diffusion priors from internet-scale pretraining to achieve strong zero-shot generalization from limited synthetic data. However, the heavy compute of these diffusion backbones limits inference to a short chunk of just 6 to 24 frames, so even ordinary videos span multiple chunks. Because each chunk is matted independently from coarse mask guidance, existing methods fail to maintain consistency across chunk boundaries, producing flickering edges and discontinuous transparency. We address this with a mask-guided diffusion-matting framework built on two ideas. The RGB and mask conditioning are represented as separate token streams that interact through full spatio-temporal self-attention, letting the model adaptively combine fine RGB texture with coarse mask layout rather than committing to a fixed channel-wise fusion. On top of this, we propose a novel pair of learnable embeddings, the preserve token and the refine token, that act as a per-frame conditioning interface and enable a sliding-window inference scheme in which already-generated mattes are propagated across chunk boundaries for local temporal consistency, complemented by a globally shared reference anchor at every chunk to prevent error accumulation in long sequences without any additional training. Across standard video matting benchmarks and in-the-wild long footage, our framework outperforms prior diffusion-based methods in both fine-detail accuracy and temporal consistency, and uniquely maintains consistency across chunk boundaries on long videos.
[CV-2] Reinforcement Learning with Visually Attributable On-Policy Self-Distillation
链接: https://arxiv.org/abs/2610.26856
作者: Shiao Xie,Siyu Chen,Jianwei Lv,Xiandong Li,Yujin Wang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Multimodal reinforcement learning with verifiable rewards (RLVR) provides reliable task-level supervision, yet offers limited guidance for localizing errors in visual reasoning. On-policy self-distillation (OPSD) complements this supervision with dense token-level corrections from a privileged teacher equipped with an evidence-centered visual view. However, the resulting corrections may conflate visual evidence with non-visual shifts induced by linguistic priors, response style, or the privileged view itself. We formalize this phenomenon as privileged-correction contamination: a token-level correction is contaminated when its update direction cannot be consistently attributed to the privileged visual evidence. Directly distilling contaminated corrections may therefore reinforce erroneous token-level updates. To address this issue, we propose reinforcement learning with visually attributable on-policy self-distillation, an evidence-consistent paradigm that augments task-level RLVR supervision with privileged token-level corrections grounded in visual evidence. We instantiate this paradigm with Visual-Attribution Gating (VAG), which uses the verifier-derived advantage as the optimization anchor and validates each privileged correction against the contrast between evidence-present and evidence-degraded teacher views. By admitting only corrections directionally supported by this visual contrast, VAG preserves visually grounded guidance while suppressing inconsistent residuals. Experiments across six visual reasoning benchmarks show that VAG consistently outperforms GRPO and OPSD methods, establishing new state-of-the-art average performance.
[CV-3] Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
链接: https://arxiv.org/abs/2610.26855
作者: Tianqi Liu,Hanfei Zhao,Qifeng Liang,Hui En Pang,Jiarui Xu,Sayang Mu,Penghao Qiu,YingSheng Zhang,Jun Wang,Ziang Cao,Mingze Sun,Yukang Cao,Fangzhou Hong,Zhaoxi Chen,Ziwei Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Project page: this https URL ; Code: this https URL
Abstract:World models enable embodied agents to anticipate action consequences, yet jointly modeling locomotion and manipulation across diverse embodiments remains challenging due to heterogeneous action spaces and limited multimodal supervision. We present Ropedia W0, the first egocentric loco-manipulation world foundation model spanning humans and robots with diverse embodiments. We curate Xp-Data, comprising 10k hours of egocentric human data and 3k hours of robot data with unified multimodal annotations and systematic quality control. A unified kinematic representation encodes locomotion through camera poses and manipulation through projected skeletons, with robot skeletons combining embodiment-specific forward kinematics with extensive manual keypoint annotation and correction. We also introduce Xp-Teleop, a long-horizon loco-manipulation benchmark based on high-quality dexterous humanoid teleoperation data. Evaluations demonstrate state-of-the-art performance across five cross-embodiment benchmarks (DreamDojo, \Psi_0 , \omega -HOME, Xp-Teleop, and AgiBot) and on held-out human egocentric video, from zero-shot inference to few-shot and full-data post-training. Scaling human pre-training data improves performance on both human and robot evaluations. After distillation, our model supports real-time, minute-long rollouts, enabling live VR interaction. Applications in policy evaluation, model-based planning, and world-action modeling further demonstrate the model’s utility for embodied decision-making.
[CV-4] InfiniDream: Open-Ended Photorealistic 3D Worlds Through Procedural Generation
链接: https://arxiv.org/abs/2610.26853
作者: Yi-Chuan Huang,Hao-Jen Chien,Zheng-Hui Huang,Yu-Lun Liu,Kaipeng Zhang,Zhixiang Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Procedural generation plays a key role in synthetic data creation and game development, offering greater controllability, editability, and diversity than static mesh representations. However, VLM-based code generation often produces relatively simple 3D assets, while sophisticated photorealistic procedural systems require extensive manual development. Existing VLM-guided procedural methods remain largely constrained by predefined factories and APIs, limiting their ability to create novel asset types. We introduce InfiniDream, a VLM-driven framework for open-ended procedural 3D world generation. First, we propose Symbolically Constrained Asset Generation, which leverages existing procedural algorithms to create novel, photorealistic assets while preserving editability and seed-based diversity. Second, we develop an end-to-end text-to-world pipeline that combines semantic scene planning with Deterministic Frustum-Guided Grid Placement to construct large-scale 3D worlds with precise object placement, procedural population, and animation. Extensive evaluations on 100 object and 20 scene generation tasks demonstrate improvements across four perceptual metrics. Diversity analysis, ablation studies, and human preference evaluations further validate the effectiveness of our framework. Project page: this https URL
[CV-5] Manga2Animate: An Agent ic Framework for Faithful Manga Animation with Script and Storyboard Planning
链接: https://arxiv.org/abs/2610.26851
作者: Sheng-Wei Zhang,Hsi-An Chen,Yan-Ting Liou,Chih-Hsuan Chen,Shih-Chen Tseng,Yu-Lun Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Adapting existing manga into coherent animation that stays faithful to the source is a valuable yet challenging task. Existing methods either access the manga mainly through text, losing its characters and art style, or feed manga pages directly into video generation models, which misinterpret the manga’s structure and produce incorrect storylines. In this paper, we propose Manga2Animate, a multi-agent framework that grounds every stage of generation in the source manga. We decompose the task into two intermediate representations, an animation script and a visual storyboard. Each representation is produced by a creator agent and reviewed against the source panels by a paired reviewer agent. We tailor representation-specific selection and dual-level feedback to progressively refine both before synthesizing the final animation. Extensive experiments and a user study demonstrate that Manga2Animate produces animations more faithful to the source manga and more consistent across shots than state-of-the-art approaches. Project page: this https URL
[CV-6] GenOlympiad: Benchmarking Frontier Image Models at Their Capability Boundaries
链接: https://arxiv.org/abs/2610.26850
作者: Xi Hu,Martin Bell,Liang Zheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 30 pages, 16 figures, 6 tables
Abstract:Text-to-image models have made tremendous progress and can produce high-quality images. However, a question which would be asked in any other discipline is where the frontier models fall short. We introduce GenOlympiad, a benchmark which provides the most challenging text prompts from nine capability axes for stress-testing frontier text-to-image models. The assessed capabilities include generating large quantities, performing deep reasoning, ensuring hard physical consistency, following complex rules and so on. The prompts are designed such that prompt alignment is visually verifiable (e.g., counting) instead of relying on subjectiveness (e.g., style). Based on this, we develop an automated evaluation framework which decomposes the prompt into easy and hard questions, and for most hard questions, draws help lines to ground evidence for accurate grading. We evaluate 20+ strong text-to-image models, including Nano Banana Pro, GPT-Image-2.5, Seedream-5.0 and so on. We report that even the best-performing T2I model (Nano Banana Pro) only achieves a strict score of 23.9 out of 100 and that coding models such as GPT-6 Astra have higher prompt alignment scores (but not as visually pleasing). We also derive insights in model evolution, alignment with Elo scores, and effectiveness of our grounding-based evaluation protocol. Code is available at this https URL and the project page is at this https URL
[CV-7] Where is Physical RSI capability from? An Observational Scaling Law of LLM -as-policy
链接: https://arxiv.org/abs/2610.26848
作者: Zekai Li,Jiayi Yu,Jinlu Zhang,Luyang Niu,Zekun Qi,Jun Zheng,Xuchuan Chen,Dairu Liu,Yintianrun Zhang,He Wang,Li Yi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Tech Report
Abstract:In this paper, we systematically investigate the mechanisms, scaling behavior, and generalization of physical Recursive self-improvement. We introduce PhysRSI-eval to quantify physical RSI through a five-axis metric: First-Attempt Performance, Efficiency, Success, Budget and Improvement from failure. Different VLMs are evaluated on 18 manipulation tasks under a fixed interaction budget. The analysis of experimental results motivates the Physical Barrel Hypothesis: different capabilities may constrain different parts of improvement. The five-axis metrics also distinguish systems with close public benchmark averages and show that high scores need not come with early or efficient gains. We study observational scaling from two perspectives: predicting the relative performance of VLM agents from their public capabilities, and forecasting later progress from early physical feedback. Together, these analyses ask how quickly and how far an agent can improve. Moreover, we conduct workspace interventions and then analyze what improvement leaves behind. Removing history reduces late-round success, frozen workspaces remain useful on unseen layouts, and three cross-task cases achieve success earlier. The results suggest that scene-specific adaptation and reusable experience can coexist.
[CV-8] SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration
链接: https://arxiv.org/abs/2610.26837
作者: Javad Rajabi,Hakki Can Karaimer,David B. Lindell,Babak Taati,Michael S. Brown
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 34 pages, 19 figures
Abstract:Diffusion Transformers (DiTs) achieve high-fidelity image and video generation, but they suffer from substantial inference costs due to repeated transformer evaluations across many denoising steps. Feature caching reduces this cost by reusing features across denoising steps. Recent forecasting methods extend this idea by extrapolating features to skipped steps instead of reusing them. However, existing forecasting methods extrapolate the features they predict uniformly, even though the spatial-frequency components of these features change most at different stages of denoising. They also overlook that denoising progresses nonuniformly, so equal step intervals can correspond to very different amounts of change. These mismatches amplify forecast errors as more steps are skipped, limiting the achievable acceleration. We propose SpecKCast, a training-free method that forecasts the model output in the frequency domain. SpecKCast applies a 2D discrete cosine transform (DCT) to the output and tracks each coefficient with a lightweight Kalman filter along a band-specific time coordinate derived from the noise schedule. The filter tracks each coefficient’s value and rate of change, and the transformer is evaluated only when the filter’s forecast uncertainty exceeds a threshold. Experiments on FLUX.1-dev, Qwen-Image, Wan2.1-14B, and HunyuanVideo show that SpecKCast achieves a better trade-off between acceleration and fidelity than existing caching and forecasting methods, reaching up to 6x speedup with less than 0.05 GiB of additional memory.
[CV-9] Open-Form Geometric Style Transfer
链接: https://arxiv.org/abs/2610.26820
作者: Adar Elad,Gal Metzer,Omer Dahari,Min Lu,Daniel Cohen-Or
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:A geometric style learned from one family of shapes is typically applied to new shapes from the same family. We consider a more challenging setting: learning the geometric design language of a stylized Latin alphabet and applying it to new forms, from Greek glyphs to entirely different structures such as human poses. We introduce Open-Form Geometric Style Transfer, where the forms to which a learned geometric style can be applied are not restricted to the shape family in which that style was observed. This setting is deceptively challenging: the difficulty is not merely learning a style, but learning the style without learning the shape family in which it was observed. We represent target forms by structural sketches and disentangle structural adherence from style realization using a Geometry LoRA and a style-specific LoRA. Critically, open-form transfer requires the Geometry LoRA to be universal. To achieve this, we devise a geometry-enrichment strategy that constructs a large and purposefully diverse collection of synthetic structural sketches spanning a broad range of unfamiliar geometries. Our experiments and ablations demonstrate the importance of both disentanglement and this universal geometry prior, enabling geometric styles learned from glyphs, pictograms, and other shape families to be transferred to substantially different, unseen forms.
[CV-10] Real-Time Radar Semantic and Panoptic Scene Understanding
链接: https://arxiv.org/abs/2610.26816
作者: Santosh Parit,Julia Hindel,Rohit Mohan,Georgi Pramatarov,Paul Newman,Daniele De Martini,Abhinav Valada
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robust scene understanding under varying environmental conditions is essential for safe autonomous driving. While cameras and LiDAR degrade in adverse weather and poor lighting, Frequency-Modulated Continuous-Wave (FMCW) spinning radar provides dependable 360° perception at extended ranges with minimal compute cost. However, real-time FMCW radar scene understanding is challenging due to complex signal returns, limited temporal resolution, and scarce annotated datasets. We introduce StreamRPS, a novel framework for streaming FMCW radar panoptic segmentation, comprising an automated label-generation pipeline via LiDAR scene completion and a real-time streaming network for semantic and panoptic segmentation. Our encoder-decoder streaming architecture comprises a radar-specific hierarchical Transformer encoder and task-specific decoders for semantic or panoptic prediction. Our encoder introduces a stem to capture range-dependent signatures, anisotropic patch embeddings to preserve angular resolution, and axial polar attention to separately model the periodic azimuth and bounded range axes. We further introduce azimuth roll rotation augmentation to simulate ego-vehicle rotation during training. Comprehensive experiments on the SAX and BOREAS datasets validate both our auto-labeling methodology and streaming segmentation network against established baselines. We outperform the strongest baseline by at least 1.4pp while processing sectors at over 400 FPS in a 4-sector streaming setting. We additionally provide manual annotations for 100 SAX radar scans, enabling independent evaluation of the entire framework.
[CV-11] PICA-WAM: Visual Perturbation Invariance via Cross-Appearance Self-Alignment in World Action Models
链接: https://arxiv.org/abs/2610.26813
作者: Yude Zou,Yifei Yao,Guanbo Jiao,Jianming Ma,Guanjie Zheng,Xing Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:World Action Models (WAMs) jointly predict future video and generate actions, yet remain sensitive to task-irrelevant visual perturbations, i.e., changes in visual appearance that should not affect robot behavior. We identify a gap in conventional WAM training: video prediction encourages representations to preserve appearance information, while action generation does not explicitly enforce invariance to task-irrelevant appearance. Data augmentation exposes WAMs to such perturbations but does not require perturbed and nominal observations to produce consistent action representations, and alignment with external pretrained features does not guarantee the desired invariance. To address this gap, we propose PICA-WAM, a self-supervised training framework that casts perturbation invariance as an auxiliary cross-appearance self-alignment task for WAMs. An exponential-moving-average (EMA) teacher processes the nominal observation to provide a deeper-layer target, while the online WAM processes a perturbed observation and aligns its shallower-layer action representation to this target through a lightweight projection head. Across four strong WAMs evaluated on RoboTwin 2.0, LIBERO-Plus, EBench, and a real robot, PICA-WAM consistently improves robustness under visual perturbations, preserves nominal performance in simulation, and raises the real robot success rate. PICA-WAM further reduces the sensitivity of video features to task-irrelevant appearance and its propagation into actions, as shown by paired layer-wise diagnostics. Code: this https URL
[CV-12] State World Models: Execute the Rules Generate the Pixels
链接: https://arxiv.org/abs/2610.26811
作者: Chan Yan Ki,Yanran Zhang,Wenzhao Zheng,Xiaofeng Wang,Zheng Zhu,Lei Chen,Jie Zhou,Jiwen Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing world models can generate action-conditioned video from a single image. Many focus on camera movement, allowing users to explore the generated scene. However, navigation alone does not ensure that an action follows persistent rules when its outcome depends on state that is invisible in the current frame. We present Stateful Harness, a framework that separates world evolution from video rendering to support stateful and rule-governed interaction. It maintains an evolving world state under specified rules, renders each resulting event with a frozen video model, and uses the generated clip to update the world state for the next turn. At each turn, the harness resolves the action selected by the user and compiles the resulting event into an instruction, while a preview clip from an editable 3D scene provides spatial and camera guidance. To evaluate rule compliance, we build WorldState-Bench, a benchmark of 50 counterfactual twin pairs that share an initial frame and action but differ in hidden state and required outcome. Under rule-only prompting, the best of the 11 evaluated video backends produces the required endings for both cases in 32% of pairs. For MiniMax-H3, specifying the required endings raises this rate from 32% to 64%, while the full harness reaches 92% with the same frozen model. These results support separating rule execution from video rendering for world interactions with hidden state.
[CV-13] What Language Adds and at What Cost: Four Controlled Probes of Pathology Vision-Language Models
链接: https://arxiv.org/abs/2610.26805
作者: Wenhao Zhang,Zhongliang Zhou,Yi Ping,Zirui Wang,Sheng Li
类目: Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
备注:
Abstract:Vision–language models (VLMs) promise to make computational pathology queryable and able to draft reports, but how much language adds beyond the vision backbone is unclear, because almost every system is evaluated by its developers on test sets of their own choosing. We grade published evidence by how it was produced and test four claims with controlled probes that reuse released checkpoints and public data, varying one input at a time. Report text adds prognostic signal: adding a frozen embedding of the stage-redacted report to a slide embedding raises the pooled C-index by 0.019 for progression-free interval and by 0.016 for overall survival across 1,633 TCGA patients, and restoring the stage statement adds nothing. Prompts add no measurable accuracy at inference: with features and pooling fixed, zero-shot classification is statistically indistinguishable from a supervised linear probe for lung and renal cancer subtyping. Slide-level question answering is barely grounded: SlideChat answers 74.82% of SlideBench-VQA(TCGA) items correctly with the slide and 74.41% without it, whereas its language backbone alone reaches 43.52%. Paraphrasing leaves accuracy unchanged but flips 8–23% of answers. Self-reported gains from agentic navigation and few-shot language priors cannot yet be attributed to language. Language thus adds real but narrow value, and we propose minimum controls that would expose these failure modes in future evaluations.
[CV-14] Mise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation
链接: https://arxiv.org/abs/2610.26801
作者: Yingshuang Zou,Xiaoyu Shi,Yawen Luo,Ziang Li,Qinghe Wang,Mengfei Li,Chengfeng Zhao,Xintao Wang,Pengfei Wan,Kun Gai,Yuan Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent advances in text-to-video generation have enabled high-fidelity single-shot synthesis, yet practical storytelling creation requires multi-shot videos with both strong inter-shot consistency and fine-grained controllability. Existing multi-shot video generation approaches achieve strong consistency but still fall short in controllability. In this paper, we propose Mise-en-Video (MeV), a highly controllable multi-shot video generation framework that allows video creators to conduct staging the multi-shot in the 3D space and then generate the corresponding multi-shot video. To achieve this, our framework combines a hierarchical attention mechanism for multi-level narrative consistency with 3D-aware layout guidance for precise, per-shot spatial control. To support training and evaluation, we develop standardized annotation pipelines and construct a 160K-video dataset with hierarchical multi-shot texts and oriented 3D bounding boxes. Extensive experiments demonstrate that MeV consistently outperforms strong baselines in video quality, inter-shot consistency, and subject-level controllability.
[CV-15] V-JEPA Hand: Simple Accurate and Fast 3D Hand and Camera Trajectory Recovery from Egocentric Videos
链接: https://arxiv.org/abs/2610.26780
作者: Ruicheng Li,Qixiu Li,Yu Deng,Sicheng Xu,Jiaolong Yang,Baining Guo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Egocentric video supports diverse applications, including robot learning, augmented reality, and human activity understanding. Recovering metric 3D hand trajectories from such video requires estimating both hand motion and camera egomotion to express the hands in a consistent reference frame. Many existing pipelines recover hand and camera motion in separate stages, combining hand detection, hand reconstruction, camera tracking, and trajectory refinement. We present V-JEPA Hand, a simple feed-forward video model that jointly reconstructs bilateral hand motion and camera motion from egocentric clips. It uses a self-supervised V-JEPA 2.1 encoder to extract spatio-temporal features from full-frame video and a query-based predictor to decode these features into per-frame hand poses, hand presence, and camera motion. We use the predicted camera motion to express the hand trajectories in a common world frame. The model accepts known focal lengths or predicts them when unavailable. It reconstructs each clip in a single forward pass. We train both components end to end in a single stage on a mixture of diverse pseudo-labelled videos and high-precision 3D annotations. A single model achieves the lowest coverage-aware mean per-joint position error (MPJPE) among comparable baselines on ARCTIC, HOT3D, and EgoStandard. It also yields lower world-frame trajectory errors than the compared SLAM-based pipelines. Ablations support the benefits of video pretraining, temporal context, and mixed supervision. V-JEPA Hand achieves a forward-pass throughput of 93.4 FPS on a single RTX A6000. Project page: this https URL.
[CV-16] Beyond Omni: A Unified MLLM for Embodied Sensory Understanding
链接: https://arxiv.org/abs/2610.26751
作者: Shiyun Xiong,Dongming Wu,Peiwen Sun,Bokang Yang,Jiaxu Wang,Xiaolou Sun,Mingxing Wen,Xiangyu Yue
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent embodied systems increasingly build on multimodal large language models (MLLMs) for their pretrained understanding and reasoning capabilities, but their perception of the physical world remains largely limited to basic modalities such as vision and audio. However, broadening MLLMs to more diverse sensory modalities from the physical environment is challenging: their representations are often weakly connected to the MLLM and vary substantially in structure and statistics, while their interpretation still requires shared high-level understanding and reasoning. We therefore propose Unisense, a unified embodied MLLM framework that supports diverse physical sensory modalities within a shared multimodal architecture. Unisense follows a two-stage training process: (i) aligning sensory representations with language to establish a language-accessible interface; and (ii) tuning modality-specific experts within the shared LLM to provide specialized processing for different sensory modalities. To support broad sensory learning, we curate Unisense-1M, a large-scale embodied sensory training corpus comprising 1M samples across 8 sensory modalities. We further evaluate Unisense on 12 benchmarks spanning depth, point cloud, IMU, tactile, thermal, radar, motion, and smell understanding. Across these heterogeneous sensing tasks, Unisense achieves strong performance and outperforms both generalist and specialized models on most benchmarks, demonstrating the potential of pretrained MLLMs as unified models for embodied sensory understanding.
[CV-17] CheckDistill: Adaptive Checklist Distillation for Efficient Distributional Reward Modeling in Image Editing
链接: https://arxiv.org/abs/2610.26740
作者: Yaqi Cai,Zuan Gao,Mengting Chen,Wei Li,Bo Zheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 31 pages
Abstract:Reinforcement learning for instruction-guided image editing requires reward models that provide reliable, accurate, and efficient feedback. Direct scalar prediction compresses multidimensional editing preferences, whereas generative evaluators can expose their judgment through explicit reasoning. However, broad scoring criteria do not specify all instance-dependent editing and preservation constraints, and generating long reasoning trajectories is costly for online optimization. We introduce CheckDistill, a pointwise reward modeling framework that internalizes instruction-adaptive checklists. We decompose each editing condition into atomic, independently verifiable requirements and construct four-dimensional scoring labels together with structured verification evidence. After supervised score initialization, a privileged teacher conditions on checklists, item-level judgments, and dimension-level rationales to supervise the student’s own scoring trajectories. A reliability-weighted on-policy self-distillation objective adjusts supervision at score tokens according to agreement between the teacher’s expected scores and reference labels, while retaining candidate probability weights in a truncated token support. The deployed student directly predicts four score distributions without checklist inputs or explicit reasoning and aggregates their expectations into a scalar reward. CheckDistill achieves aggregate accuracies of 72.30%, 56.67% and 82.90% on MMRB2, MER-Bench and EditReward-Bench, respectively, exceeding the compared open-source baselines on all three metrics.
[CV-18] CAVA: Complementary-Label-Aware Data Valuation
链接: https://arxiv.org/abs/2610.26734
作者: Tan-Ha Mai,Hsuan-Tien Lin,Vu Nguyen
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 39 pages, 18 figures
Abstract:Complementary-label learning (CLL) trains a classifier from labels that name a class an instance does not belong to. Such labels are cheap to collect from crowdworkers or vision-language models (VLMs), making them attractive for large-scale annotation. However, both sources can mistakenly assign the true class as the complementary label. Such annotation errors are particularly harmful in CLL because the model is explicitly trained to suppress the correct answer. Optimal transport (OT) data valuation identifies harmful samples by comparing training data with a clean validation set, but transferring it directly to CLL creates a semantic mismatch: label c means not class c in the training but class c in the validation. We show that this makes the OT label groups noisy mixtures whose behavior depends on the noise rate. The transferred score is informative only at sufficiently high noise, becomes uninformative at the crossover rate 1/M for M classes, and reverses below it, where practical CLL annotation typically operates. We propose CAVA, which introduces a complementary label cost and a label margin that isolates the contribution of each complementary label to the transport potential. The margin keeps a consistent direction across noise rates, detects diagonal noise with a parameter-free threshold of closed-form precision, and allows detected samples to be recycled as ordinary-label supervision. On synthetic and human-annotated benchmarks, CAVA separates diagonal noise almost perfectly where directly transferred OT valuation performs worse than random. With only a fraction of the clean validation data, it also improves every CLL objective we evaluate, showing that OT-based valuation becomes reliable once the semantics of complementary supervision are modeled explicitly.
[CV-19] PlanarCityLoc: Breaking Silhouette Ambiguity in LoD Aerial Localization via 3D Planes WACV2027
链接: https://arxiv.org/abs/2610.26729
作者: Oussema Dhaouadi,Shao Jie Hu-Chen,Jakob Gollreiter,Samir Jusufi,Nikolas Rieger,Caroline Wendlandt,Christian Wender,Jingfeng Mao,Guangming Wang,Zuria Bauer,Marc Pollefeys,Daniel Cremers,Olaf Wysocki
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV 2027 - Project page: this https URL
Abstract:Aerial visual localization is crucial for autonomous drone navigation in urban environments where GNSS signals are often degraded. Lightweight textureless Level-of-Detail (LoD) city models provide a geometric map alternative to textured assets. Existing silhouette-based methods rely on image-space alignment, which can leave different camera poses indistinguishable and favor incorrect poses when segmentation errors distort the building outlines. We present PlanarCityLoc, which moves beyond silhouette overlap by leveraging the planar geometry of building facades and roofs. Following geometry-aware initialization, we match image-side planar patches to LoD model planes and use the resulting 3D-to-2D correspondences to refine pose with Perspective-n-Point (PnP). The framework uses off-the-shelf geometry and segmentation models without dataset-specific training or fine-tuning. Experiments on UAVD4L-LoDv2 and Swiss-EPFLv2 achieve state-of-the-art localization recall among the evaluated off-the-shelf LoD-based methods. By grounding image-space alignment in 3D planes, PlanarCityLoc enables more accurate aerial localization with lightweight, textureless city maps.
[CV-20] A Flexible ROI Analysis Tool for Spatially Consistent Vegetation Index Extraction from Turfgrass Imagery
链接: https://arxiv.org/abs/2610.26719
作者: Kibon Ku,Thai Thanh Tuan,Jinhyun Ahn
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages, 4 figures, 2 tables
Abstract:Time-series multispectral imaging is widely used to assess turfgrass condition; however, extracting plot-level vegetation index (VI) values remains labor-intensive when camera positions vary between acquisitions and automated plot tracking is unreliable. We developed an open-source software tool comprising (i) a graphical user interface (GUI) for the interactive placement of orientation-adjustable regions of interest (ROIs) with N -way subdivision to match plot geometry and (ii) a batch-processing script that computes plot-level mean VI values from single-band floating-point rasters and exports measurements and ROI coordinates in JSON format for reuse. Saved coordinates can be reapplied to measure NDVI, NDRE, and SAVI from identical pixel regions within a single time point without re-delineating ROIs. In a demonstration using rail-mounted multispectral turfgrass imagery (18 images), plot-level mean values were extracted in a single session, with coordinates reused directly for NDRE and SAVI rasters. The system reads non-georeferenced pixel-coordinate rasters natively, does not require GIS expertise, and supports reproducible VI quantification where static sensor deployment is impractical.
[CV-21] PVBench: Quantifying Physical Plausibility in Video Generation with Object-Level Physical Measurements
链接: https://arxiv.org/abs/2610.26711
作者: Kaiwei Chen,Moran Li,Zhenye Gan,Yabiao Wang,Chengjie Wang,Kexin Zhang,Yong Liu,Jiangning Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent video generation models produce visually compelling outputs, yet their physical plausibility remains poorly measured. Physical plausibility asks whether generated objects obey geometry, kinematics, dynamics, and conservation laws. Existing benchmarks rely on coarse visual-language model judgments or target embodied world models rather than video generation itself. We introduce PVBench, a physical video benchmark that quantitatively measures the plausibility of generated videos across 46 dimensions organized into seven domains, spanning: i) instruction and semantic fidelity, ii) behavioral and causal consistency, iii) perceptual and temporal quality, iv) geometric and spatial consistency, v) material properties and field states, vi) kinematic and spatiotemporal persistence, vii) and mechanical plausibility with conservation laws. PVBench comprises 600 carefully crafted prompts, of which 30% involve physical embodied agents such as robot arms, humanoids, quadrupeds, and multi-agent systems. For challenging physically tractable dimensions, it pairs each with specialized vision models that derive verifiable, physically grounded measurements, such as object-level 6D pose, metric scale, surface normals, kinematics, and numerical conservation-law residuals. It thus avoids relying solely on subjective scoring, while supporting both text-to-video and image-to-video generation. Finally, we validate the agreement between automatic metrics and human judgments through human evaluation. Meanwhile, we evaluate mainstream open-source and closed-source models and present several insights. By releasing the benchmark, prompts, and calibration protocol, we aim to foster physically grounded video generation.
[CV-22] Foundation Model Driven Optical Flow NEURIPS2026
链接: https://arxiv.org/abs/2610.26690
作者: Praroop Chanda,Suryansh Kumar
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 19-pages, 5 figures, 16 tables. Accepted for publication at 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Workshop: RoCo-Spring: The Robust Correspondence Challenge
Abstract:We investigate how frozen pretrained visual and depth representations support dense optical flow when coupled with a supervised matching network. Building on TTSFlow, this paper studies Dense Prediction Transformer (DPT) feature decoding and one coarse-to-fine refinement stage. Our method fuses features from frozen DINOv2 and Depth Anything V2 models and estimates optical flow through a trainable transformer-based matching pipeline. After optical-flow training on FlyingChairs and FlyingThings3D, it achieves 2.37 endpoint error (EPE) on the Sintel training split, Final pass, compared with 2.52 for the FlowSeek (T) configuration reported here. Our RoCo-Spring 2026 entry, this http URL, obtains a robustness-balanced score of 1.1038, Spring EPE of 1.411, and RobustSpring proxy EPE of 4.739, with a lower balanced score than seven of the twelve listed reference baselines. These results characterize the complete pipeline, which reuses frozen foundation backbones and performs one refinement stage. Code is available at: \urlthis https URL
[CV-23] BrushOff: Stroke-Based Re-rendering as a Watermark Removal Attack on Painting
链接: https://arxiv.org/abs/2610.26682
作者: Hao Chen,Junnan Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR)
备注: 8 pages, 4 figures, 2 tables
Abstract:Invisible watermarks are used to mark who owns a painting. We show that a simple attack removes the most robust one we test. The attack redraws the watermarked painting with brush strokes, using a pretrained stroke-based renderer. We call this attack BrushOff. It needs no training, no knowledge of the watermark, and no generative model. We test four watermarks on 500 paintings from five styles. VINE, a recent watermark built to survive image editing, survives all 14 conventional distortions we test, with at least 97.8% detection. Repainting lowers its detection to 1.8%, while the output keeps a no-reference quality score close to that of the clean paintings. Diffusion regeneration can also remove VINE, but at the same removal level it moves the image twice as far from the painting in DreamSim distance, and it changes what the painting shows. Because the renderer draws a countable number of strokes, we can control the attack strength step by step. Finally, we test what kind of signal survives. A simple probe that stores bits in the relative brightness of neighboring regions survives repainting, while a probe that uses absolute brightness does not. The same probe fails under diffusion regeneration, while VINE fails under repainting. Robustness to these two attacks is therefore a separate property, and benchmarks should test both.
[CV-24] From Short Clips to Real-Time Streams: Identity-Aware Supervision and Context-Efficient Generation for Video Face Swapping
链接: https://arxiv.org/abs/2610.26681
作者: Zekai Luo,Yuwei Liang,Ruiqi Ma,Haoyu Huang,Xuehao Wang,Huanyi Zheng,Mingyu Liu,Linhao Zhong,Chenchen Jing,Yuling Xi,Hao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project webpage: this https URL
Abstract:Video face swapping transfers the reference identity to a target video while preserving its other attributes and motion. Despite recent progress in reference-guided flow-based methods, jointly achieving faithful identity transfer, high fidelity, and real-time generation over long video streams remains challenging. To address this challenge, we rethink the task-specific structure of reference-guided video face swapping. First, face swapping changes only identity while preserving most visual content, yet flow-matching supervision is weakly sensitive to these changes. We therefore develop Joint Rec-ID Loss, which balances explicit identity supervision and latent reconstruction according to the noise level. Second, the target video already provides non-edited attributes and temporal dynamics, enabling generation with fewer spatial tokens and shorter temporal contexts through Face-RoI Generation and Short-Context Rollout Training. By integrating these designs, we propose StreamingSwap, a reference-guided flow-matching framework for real-time streaming video face swapping. We also introduce HumanFaceVid, comprising 1,000 hours of cinematic training data and evaluation sets of 200 short and 200 long videos. Experiments demonstrate that our method achieves competitive identity transfer and visual quality at 20 FPS on a single NVIDIA RTX 4090 GPU, with stable generation over videos up to 10 minutes long. Project webpage: this https URL
[CV-25] CaptionFuse: Training-Free Composed Product Retrieval Via VLM Intent Captioning and Weighted Embedding Fusion
链接: https://arxiv.org/abs/2610.26661
作者: Zhao Gao,Mina Metias,Mohamed Noordeen Alaudeen,Bassel Saleh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Composed Image Retrieval (CIR) retrieves products matching a reference image modified by natural language—a core capability for eCommerce search. Existing training-free methods fuse CLIP embeddings via arithmetic, but we identify a fundamental limitation: user text modifications are attribute-sparse—a user who types make it red with shorter sleeves'' specifies what to change but not what to keep (the floral pattern, the V-neck, the fit-and-flare silhouette). We introduce language-space fusion: a Vision-Language Model (VLM) jointly interprets the reference image and modification to generate a visually grounded product-title caption---e.g., Red floral fit-and-flare dress with short sleeves and V-neck’'—that fills in the missing attributes from the image. Our method, CaptionFuse, requires no task-specific training and is model-agnostic. A controlled experiment confirms that attribute content— not linguistic style—drives the gains: imperative and descriptive captions with identical attributes achieve equal R@10, while both outperform raw modifications by +6.5 points.
[CV-26] Dual-Scale Condition-Shift Editing for Robust Real-World Image Demoiréing
链接: https://arxiv.org/abs/2610.26644
作者: Shiyang Zhou,Haijin Zeng,Congpei Qiu,Zhengcen Li,Ping Chen,Xiangming Wang,Zheng Li,Yongyong Chen,Jingyong Su
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robust demoiréing of screen-recaptured images remains challenging: moiré varies in frequency, color, and spatial support across display-camera pairs and capture conditions. Dedicated restoration networks trained on finite paired data struggle with this diversity, especially when low-frequency color banding and high-frequency stripes co-occur. We present DemoiréEdit, a content-preserving editing framework that adapts the natural-image prior of a pretrained flow-matching editor to demoiréing. First, an endpoint-supervised training objective supplements velocity matching with pixel, perceptual, and frequency-domain restoration losses. Second, a dual-scale condition-shift strategy performs global correction at low resolution, then refines the image at full resolution by shifting the denoising condition from the first-stage output to the original input for detail recovery. We also introduce DailyMoiré130, a benchmark of challenging moiré in everyday screen photography, comprising 130 smartphone captures across diverse displays with low-frequency color banding, fine high-frequency stripes, and mixed-frequency interference. On the challenging DailyMoiré130 benchmark, our method achieves the lowest residual moiré area and receives 71.3% overall-quality preference in a blind Best-of-7 study with 31 participants. We also evaluate on paired FHDMi/UHDM, reducing residual moiré confidence by 47.1%/9.3% and increasing MUSIQ by 4.3%/2.0% over the best baseline for each metric.
[CV-27] EgoXperience: Learning World-Grounded Egocentric Representations from Human Experience
链接: https://arxiv.org/abs/2610.26625
作者: Xinhao Ji,Hui En Pang,Hao Li,Kairui Hu,Jiner Wang,Mingze Sun,Ziang Cao,Yalun Dai,Thang Luu,Qin Ma,Hongyi Chen,Yukang Cao,Fangzhou Hong,Zhaoxi Chen,Ziwei Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reconstructing human activity from egocentric video requires more than recovering individual 3D components: the observer’s motion, surrounding geometry, and hand-object interactions must fit together in a consistent world. Achieving this coherence requires both joint reconstruction and aligned supervision, yet existing methods and datasets offer limited support for either. We present EgoXperience, a full-stack paradigm that addresses these needs through data collection, automatic annotation, unified modeling, and benchmarking. Central to EgoXperience is an automatic annotation engine, EgoX-Engine, that couples geometrically verified object tracking with SLAM-based world alignment, yielding consistent supervision for the camera, scene, hands, and objects. This yields EgoX-Data-6M, a 50-hour dataset of task-driven manipulation and locomotion within and across indoor and outdoor environments, with second-level task captions. We carefully select one hour across these activity categories as EgoX-Bench to evaluate joint reconstruction. To translate this supervision into reconstruction, we introduce EgoX, trained with EgoX-Data-6M and complementary annotations from additional public datasets. Given monocular egocentric video and first-frame object masks, our model achieves joint metric reconstruction of the camera, scene, hands, and objects in a shared world coordinate frame with a single forward pass. Extensive experiments demonstrate state-of-the-art performance across scene estimation and world-space hand-object pose recovery. Against the strongest evaluated baselines, EgoX reduces camera, hand, and depth errors by 68.9%, 65.6%, and 42.9%, and improves object pose scores by 28.0%.
[CV-28] Scaling Vision Foundation Models for Toxicologic Pathology
链接: https://arxiv.org/abs/2610.26586
作者: Jessica Loo,Theodoros Pissas,Byunghak Kang,Vincent A. Ella,Xiaoyan Zhang,Kevin Sweeney,Walter Georgescu,Prateek Katiyar,Julie Boisclair,Holger Hoefling
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注:
Abstract:Here, we present TOX-FM, a family of non-clinical PFMs trained on large-scale diverse histopathology datasets of over one million whole slide images consisting of over 60 organs across six animal species from over 14,000 toxicology studies across six decades. We introduce a comprehensive benchmark of 18 non-clinical downstream tasks, incorporate an established benchmark of 9 clinical downstream tasks, and evaluate TOX-FM against state-of-the-art clinical PFMs across both non-clinical and clinical domains. Notably, we explore open questions surrounding domain-specific training and cross-domain generalization, offering insights into the opportunities and challenges of applying PFMs in the non-clinical domain.
[CV-29] Drive2Gauss: FeedForward 4D Gaussian Reconstruction from Multimodal Generative Driving Videos
链接: https://arxiv.org/abs/2610.26585
作者: Tong Zhao,Zhiyuan Han,Lin Chen,Bowen Xie,Yunxi Qiao,Bohan Li,Weiqing Xiao,Gen Li,Shu Han,Xuyang Dai,Cheng Bi,Hao Zhao,Chaojian Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generating reusable driving environments requires both controllable future observations and an explicit representation that supports rendering from different viewpoints, yet existing methods still recover explicit scenes from generated RGB videos through optimization for each sequence. To overcome these limitations, we present Drive2Gauss, a feedforward framework that converts jointly generated RGB, metric depth, and dynamic flow into dynamic 4D Gaussian scenes. A Multimodal Video Generator predicts the three modalities from reference multiview images and future control signals. Furthermore, we introduce a Gaussian Query Decoder that uses depth to initialize sparse 3D queries and flow to associate the queries across time, and integrates appearance information through spatiotemporal feature aggregation and iterative query refinement. The reconstruction stage operates in a single feedforward pass without optimization for each scene. Extensive results on nuScenes demonstrate that our approach attains an FID of 6.38 and an FVD of 12.43 for future RGB generation, a PSNR of 29.91 dB when reconstructing its generated observations, and supports both direct rendering under lateral camera shifts and downstream video fixing. Code and more visualizations are available at https://zttt04. this http URL.
[CV-30] DenseHap: Dense Hand Haptic Estimation from Monocular Images
链接: https://arxiv.org/abs/2610.26583
作者: Junhan Xue,Rolandos Alexandros Potamias
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:As dexterous robotic manipulation advances, scalable annotations of human hand-object interactions are becoming increasingly important. Furthermore, previous hand contact estimation methods suffered from issues such as dorsal-side misclassifications and insufficient precision, while lacking a unified framework for integrating multiple hand to environment interaction modalities, including distance fields, contact thresholds, and pressure. To address that, we present DenseHap, a unified end-to-end approach to estimate hand distance field recognition, comprehensive hand contact detection, hand pressure analysis, and hand reconstruction from a single monocular image. By incorporating three distinctly defined heads, along with a pretrained 3D reconstructed skeleton and an enhanced dataset, DenseHap outperforms previous state of the art models on four evaluation datasets, improving the F1 score by 8.5 percentage points (a 16% relative improvement) on untrained datasets and by 3 percentage points (5% relative improvement) on in-distribution datasets. Furthermore, DenseHap has effectively improved short-range hand estimate and pressure evaluation. Code and models will be publicly available upon the paper’s acceptance.
[CV-31] PrismWAM: Decoupling State Representations in World Action Models
链接: https://arxiv.org/abs/2610.26564
作者: Tianyu Xin,Wanqi Zhong,Bowen Hu,Zhangchi Guo,Lijie Wang,Yiming Wang,Haoran Liao,Heyang Yu,Xiaoyu Tian,Zhiding Yu,Yiming Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:World Action Models (WAMs) typically represent the world through latent embeddings from pretrained video models, which entangle heterogeneous aspects of the world, from how it looks (e.g., texture and color) to how it is physically configured (e.g., occupancy and pose). This raises an underexplored question: \textbfwhat should constitute the state of a WAM? We argue that the world state should separate how the world appears from how the world evolves under action: the former captures semantic context, while the latter is governed by physical state. To this end, we propose \textbfPrismWAM , which decouples physics from semantics while jointly modeling state and action. PrismWAM factorizes an embodied task into a \mathrmSemantic\ Planning\rightarrow\mathrmPhysical\ Rollout\rightarrow\mathrmExecutable\ Action pipeline: given only the initial RGB observation, semantic planning learns to predict the final goal image, while physical rollout predicts the fine-grained evolution toward this goal in depth space; executable actions are then generated from the depth rollout. Without any embodied action pretraining or additional data augmentation, PrismWAM achieves strong manipulation performance in both simulation and the real world. Notably, PrismWAM even surpasses a broad range of models with large-scale embodied action pretraining (e.g., OpenWAM- \alpha ). More broadly, we reframe World Action Models from modeling how the world looks to how it responds to action, paving the way for action-centric world models grounded in occupancy, flow, tactile sensing, proprioception, and beyond.
[CV-32] MRI Reconstruction Risk Estimation from Inference Dynamics
链接: https://arxiv.org/abs/2610.26558
作者: Tamir Shor,Alex Bronstein
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Iterative MRI reconstructors produce intermediate estimates before returning a final image. We study these inference trajectories as a model-internal representation of realized reconstruction risk, measured by log-NMSE. Model-native states are projected to magnitude images, summarized by a fixed 16-dimensional descriptor, and mapped to risk using a source-trained ridge readout. Across three reconstruction backbones, trajectory features are strongly predictive of reconstruction risk on held-out source data. Under zero-shot target shift, their behavior is backbone-dependent, exhibiting both preserved positive association and readout reversal. Conditional analyses show that trajectory information can remain associated with error beyond an initialization-only predictor. A prefix analysis further shows that useful trajectory information can emerge before reconstruction completes. Repeated-mask and low-shot calibration studies indicate that trajectories primarily encode persistent case difficulty, while their orientation and calibration can change under domain shift.
[CV-33] Holi-World: Building Holistic World Intelligence from Human-Ego Experience
链接: https://arxiv.org/abs/2610.26549
作者: Hao Li,Xiyuan Zheng,Yuanyuan Gao,Yalun Dai,Xinhao Ji,Xuan-Li Chen,Jing Zhao,Fangzhou Hong,Zhaoxi Chen,Ziwei Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: this https URL
Abstract:Human egocentric experience offers a rich foundation for learning how the world is organized and how it changes through interaction. Realizing this potential at scale requires turning growing volumes of recordings into learning resources that connect scene structure, object semantics, and interaction histories across viewpoints and time, while reducing dependence on costly manual annotation. We introduce HOLI-WORLD, an integrated ecosystem connecting experience capture, world representation, automated data construction, model learning, and independent evaluation. To sup- port learning at scale, HOLIWORLD-Foundry combines complementary static and dynamic perception to automatically construct world records that connect persistent scene structure and evolving object states with interaction histories and supporting observations. HOLIWORLD-Data organizes these records from approximately 100 self-collected scenes into grounded learning resources, including approximately 80K QA pairs used to train HOLIWORLD-VL-4B. Independent human annotations of heldout recordings form HOLIWORLD-Bench, supporting evaluation of both annotation quality and learned spatial and temporal understanding. Experiments on ScanNet, ScanNet++, Aria Digital Twin, and HOLIWORLD-Bench demonstrate strong performance in static perception and dynamic tracking, while training on HOLIWORLD-Data substantially improves HOLIWORLD-VL-4B’s spatial and temporal reasoning on HOLI- WORLD-Bench and EgoDynamic4D. HOLIWORLD provides a practical foundation for expanding learning resources from the continuous flow of human experience and advancing machine understanding of the dynamic world.
[CV-34] Interpretable Face Contrast Modeling for Smartphone Portrait Rendering WACV2027
链接: https://arxiv.org/abs/2610.26517
作者: Yujin Cho,Benoît Pochon,Jean-Michel Morel,Gabriele Facciolo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027. 37 pages, 37 figures, including references and supplementary material
Abstract:The quality of smartphone portrait photography has become increasingly important to both manufacturers and users. Recent devices use complex computational pipelines to improve facial rendering under varied lighting. These pipelines vary widely across manufacturers, so the same portrait can look noticeably different across smartphones. These differences affect multiple aspects of facial appearance, making perceptual preference difficult to characterize with a single measure. To address this, we propose an interpretable framework that decomposes face contrast into three components: overall facial tonal separation, eye-region contrast, and micro-contrast of fine skin details. We formulate dedicated metrics for each component and validate them via pairwise human judgments, demonstrating that our approach correlates more closely with human contrast perception than traditional contrast measures. To support this analysis, we introduce a realistic mannequin-based dataset designed for systematic facial contrast evaluation, comprising 508 full-resolution captures from 29 smartphones under diverse lighting, with region-specific masks and extensive pairwise perceptual annotations. As a downstream analysis, we show that combining the three component-wise metrics provides a meaningful basis for modeling balanced facial contrast preference in a controlled setting, and we present a preliminary evaluation on a smaller real-portrait dataset indicating that insights derived from the mannequin study can extend to real images. Our work provides a controlled first step toward defining and evaluating face contrast in portrait imaging.
[CV-35] PDNet: Exploring Pruning and Distillation for Real-Time Worker Activity Recognition in Industrial Assembly
链接: https://arxiv.org/abs/2610.26491
作者: Malek Baba,Thinesh Thiyakesan Ponbagavathi,Alina Roitberg
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published by SciTePress. ISBN 978-989-758-804-4
Abstract:In this work, we address real-time fine-grained human activity recognition (HAR) in industrial assembly and introduce PDNet, a new framework that jointly exploits structured pruning and knowledge distillation for 3D convolutional networks. We start by training a 3D-CNN backbone in a supervised manner. Then, PDNet applies channel-wise structured pruning to obtain a more compact network. In the third step, our method is optimized using knowledge distillation from the large unpruned teacher to recover accuracy lost due to pruning. On the challenging MECCANO dataset with an X3D-M backbone, the pruned model without distillation reaches a Top-1 accuracy of 25.54 %, while PDNet raises this to 39.46 %, with only 6.90 MB model size and 3.78 GFLOPs. These results show that distillation within PDNet is highly effective in restoring pruned model performance and highlight its high potential for resource-constrained HAR in industrial assembly.
[CV-36] 3D Diffusion Models for Medical Imaging
链接: https://arxiv.org/abs/2610.26475
作者: Frank-Peter Schilling,Daniel Barco,Philipp Denzel,Marc Stadelmann
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, accepted for publication in Informatics in Medicine Unlocked
Abstract:Generative deep learning methods such as Variational Autoencoders, Generative Adversarial Networks or Diffusion Models have achieved remarkable results in image synthesis. In particular, Diffusion Models have demonstrated superior image fidelity and sample diversity compared to other approaches on standard benchmarks, while also offering more flexible training dynamics as evidenced both in natural image domains and in medical applications. In medical imaging including Computed Tomography, Magnetic Resonance Imaging or Positron Emission Tomography, such models have various important use-cases including data augmentation, domain transfer, image reconstruction, inpainting, image augmentation and artifact correction. Further, generated images can be conditioned on patient characteristics, medical reports, segmentation masks or other images. Specific challenges arise due to the 3D volumetric nature of medical images, which translate to substantial demands in terms of memory, compute, and training data. This article provides a comparative overview of 3D Diffusion Models which have been developed for medical images, with a focus on current performance, limitations, and an outlook for future work.
[CV-37] YUTELA: A Scalable Edge Cloud AIoT System for Multi-Site Video Anomaly Detection with Scene Specific Adaptation
链接: https://arxiv.org/abs/2610.26468
作者: Shanle Yao,Babak Rahimi Ardabili,Amir Kimiyaie,Hamed Tabkhi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Building upon our previous Ancilia framework, this work presents \emphYUTELA, an edge–server–cloud AIoT system for long-term, multi-site video anomaly detection. YUTELA replaces workstation-based on-site processing with NVIDIA Jetson Thor edge devices and redesigns the multi-camera pipeline through camera batching and redistribution, shared memory, reduced inter-stage data transfer, task integration, and TensorRT-accelerated inference. It maintains a separate anomaly model per camera, retrained on a centralized GPU server once enough recent pseudo-normal data are collected, with updated weights returned to the edge. With eight cameras, TensorRT increases throughput from 14.17 to 29.94 FPS under normal density, and the optimized edge node exceeds the throughput of the workstation-based Ancilia system in seven of nine workload settings, reaching up to 2.03 times its throughput at extreme density. This gain comes at the cost of higher latency under lighter loads with four or eight streams. YUTELA was then operated continuously for 170 hours at four public sites using 31 existing CCTV cameras. The deployment shows that runtime load is driven by scene activity and input frame rate rather than by camera count, and that retraining frequency varies by about three times across sites. The 101 automatically triggered training events occupied only about 6.6% of a single GPU, indicating that one global node can support adaptation for tens of sites. Mean end-to-end alert latency ranged from 6.8 to 15.4~s depending on site traffic and anomaly type. These results show that YUTELA enables practical multi-camera edge processing with low-cost, camera-specific adaptation across heterogeneous sites.
[CV-38] When Guidance Becomes a Guard: Early Interception of Unsafe Inputs in Text-to-Image Generation
链接: https://arxiv.org/abs/2610.26452
作者: Yanchun Li,Yujin Wu,Gangkai Shen,Ming Yang,Li Zeng,Xianlong Wang,Hangtao Zhang,Shujuan Tian
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-to-image diffusion models can produce unsafe content, while output filtering intervenes only after synthesis. We investigate whether native generation signals enable early detection of unsafe and adversarial inputs. We find that classifier-free guidance exposes strong safety signals in conditional responses, accessible through compact statistics and linear readouts. Based on this finding, we propose Conditional Semantic Trajectory Interception (CSTI), which aggregates activation contrasts between conditional and unconditional responses at early checkpoints for calibrated risk detection. CSTI fits concept-specific detectors using canonical concept and safe examples, without attack samples, intermediate image decoding, or generator modification. Across five generators, four adversarial attack collections, and two general NSFW datasets, CSTI achieves the strongest overall performance among six baselines at a matched 5% FPR. Early interception reduces generation latency by over 60% on intercepted paths. These results demonstrate the effectiveness of native conditional responses for early safety interception.
[CV-39] Long-form Video-to-Text Summarization using Recursive Language Models
链接: https://arxiv.org/abs/2610.26431
作者: Alessia Intini,Lorenzo Vaiani,Enrico Chen,Luca Cagliero
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Paper currently under review
Abstract:While VideoLLMs have made significant progress in understanding short video benchmarks, they often struggle to summarize long-form ones. This is mainly due to performance degradation caused by long-context prompts, the inherent difficulty of preserving temporal continuity and a lack of semantically rich concepts in event descriptions. To tackle these issues, this paper introduces the first recursive Video Language Summarization Model (RVLSM) designed for video processing and tailored to long-form video-to-text summarization. RVLSM recursively decomposes videos into semantically coherent segments and delegates chunk-level reasoning to VideoLLM sub-models and external tools. It employs (1) Semantic Temporal Decomposition (STD), which preserves temporal coherence and continuity through a scene-aware chunking strategy that dynamically partitions videos according to semantic boundaries, and (2) Adaptive Multimodal Tool Calling (AMTC), which ensures semantic richness of the summary via a routing mechanism that selectively invokes auxiliary modules such as action recognition, optical character recognition, automatic speech recognition, and object detection. Experiments on the VideoXum and MixTyVSum benchmarks show that RVLSM consistently improves over zero-shot VideoLLM baselines across multiple model settings and performance metrics, and is competitive with a state-of-the-art fine-tuned model for video-to-text summarization. Importantly, the named entities in the RVLSM’s summaries align more closely with those detected in the ground-truth summaries than those detected by baseline and state-of-the-art methods, confirming the saliency of the event descriptions it generates.
[CV-40] Policy-Adaptive Credit Assignment for Spatial Reasoning with Verifiable Rewards
链接: https://arxiv.org/abs/2610.26410
作者: Hailiang Yang,Shengchao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 8 figures, 11 tables
Abstract:Image-based spatial reasoning requires perception, decomposition, and inference, but the stage that limits accuracy can change during training. Answer-level rewards neither localize failures to individual stages nor provide discriminative group-relative feedback when all rollouts receive the same reward. Process supervision provides intermediate feedback, but fixed weights cannot adapt reward allocation to these changing bottlenecks. To address this gap, we propose CALIPER, an adaptive process-reward framework supported by SpatialTrace, a corpus of stage-decomposed spatial reasoning traces. CALIPER jointly estimates the associations between stage-wise reference coverage and answer correctness through within-group ridge regression, using these associations as utility proxies to gradually update reward weights. The resulting process reward supervises all rollout groups, including uniformly failing ones. Because utility estimates can become unreliable when answer-discriminative groups are scarce, a closed-loop harness monitors estimator reliability and validation performance to guide checkpoint rollback, rollout diversification, and early stopping. Extensive experiments across eight benchmarks and three backbones show that CALIPER achieves the highest mean under both one-shot and harness training, with consistent gains over fixed allocation across RL optimizers. Further analyses confirm the benefits of adaptive reward allocation and harness control, with favorable scaling in model/data size.
[CV-41] One Fact Many Places: Redundant Evidence in Long Visual Documents Breaks Single-Gold Evaluation of Retrieval Citation and Attribution
链接: https://arxiv.org/abs/2610.26401
作者: Xiaokang Yuan,Binhui Yao,Luwei Yang,Ruimin Chen,Zeng Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 5 figures; includes supplementary material. Code, data and editable figure sources provided as ancillary files
Abstract:Document understanding is increasingly judged on where an answer comes from, scored against annotated gold locations as if each fact lived in one place. In long visual documents, a fact often lives in several places: a chart value is restated in the text, a key figure reappears in the summary. We ask how often such copies occur in standard benchmarks and what they do to evaluation. We annotate 420 questions from MMLongBench-Doc and MMDocRAG with the locations that independently establish each required fact, checked by two independent reviewers. Copies are common and often cross-modal: 19.5% of MMLongBench-Doc test questions have a strict copy outside the gold, and in 28% of MMDocRAG questions a non-gold candidate alone supplies a required fact. Single-gold scoring then fails in three predictable ways. A retriever that finds a copy is scored as a miss: page recall@1 of text and visual retrievers is underestimated by 13-24 points on questions with a cross-page copy. A model that cites a copy is scored as citing wrongly: crediting equivalent evidence reverses 11% of pairwise orderings among 65 competitive MMDocRAG systems. Masking the gold region leaves the fact available elsewhere: with a copy in the input, 41-56% of correct answers across four models survive it yet fail once every annotated instance is masked. Set-level masking with automatically found copies closes 78% of the gap to masking every annotated instance at matched granularity (three models); an unverified key-string rule lowers survival at least as far but masks three times the area. We release the annotations and redundancy-aware metrics.
[CV-42] From Legacy to Contemporary: Scalable Color Modernization for Long Video
链接: https://arxiv.org/abs/2610.26391
作者: Dongyun Kim,Jiwoo Choi,Jaeseong Lee,Seon Joo Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Preprint
Abstract:Bringing legacy video back to contemporary screens does not necessarily make it look contemporary. Earlier imaging pipelines often leave color casts, muted chroma, weak con- trast, and tonal imbalance, while correcting these character- istics still relies heavily on expert colorists and scales poorly to long, multi-shot videos. We formulate long video color modernization, a task that updates the chromatic and tonal appearance of legacy color footage while preserving scene content and coherence across frames and shots. We introduce a scalable, training-free framework that jointly modernizes grids of temporally adjacent keyframes using a pretrained text-guided image editing model, producing coherent con- temporary references without paired supervision. Compact shot-specific color transforms then propagate these adjust- ments throughout each shot, preserving spatial structure and temporal stability. We conduct a multidimensional evaluation on diverse legacy videos spanning five decades. We combine a human evaluation of perceived modernness with quantita- tive measures of content fidelity, temporal stability, percep- tual quality, and efficiency, and introduce Source-Weighted Inter-shot Grade Consistency (SW-IGC) to assess grade co- herence across shots. On our evaluation set, our frame- work achieves the highest perceived modernness while re- maining competitive across complementary measures, open- ing a path toward the systematic study of long video color modernization. Code is available at this https URL this http URL
[CV-43] HGASurv: Hierarchical Granularity-Aware Integration for Multimodal Survival Prediction
链接: https://arxiv.org/abs/2610.26357
作者: Peike Yu,Wuchao Liu,Jihong Guan,Shuigeng Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Preprint
Abstract:Objective: Multimodal survival prediction jointly analyzes histopathological whole-slide images (WSIs), genomic profiles, and clinical variables to stratify patients by risk. However, existing approaches often inadequately account for granularity differences across modalities, particularly in spatial organization and feature dimensionality. To address these challenges, we propose HGASurv, which introduces a hierarchical WSI encoding module and a clinical-guided graph propagation module for granularity-aware multimodal integration. Methods: The hierarchical WSI encoding module maps WSI patch features onto a spatial grid using their original coordinates and applies parallel convolutions to construct representations at patch, local-neighborhood, and regional scales. Genomic signature embeddings then query these scale-specific WSI representations through cross-modal co-attention, producing signature-conditioned morphological features. The clinical-guided graph propagation module uses clinical variables to define patient-similarity edges and propagates fused image-genomic node embeddings with graph attention, integrating clinical information through topology rather than feature concatenation. Results: Across three TCGA cohorts, HGASurv achieved the strongest overall C-index performance, with ablation studies supporting the complementary contributions of the hierarchical WSI encoding and clinical-guided graph propagation modules. Conclusion and Significance: By introducing hierarchical WSI encoding and clinical-guided graph propagation, HGASurv unifies multi-scale pathology representation with topology-based clinical integration in a hierarchical granularity-aware framework. This design shows that matching representation and integration mechanisms to modality-specific granularity improves multimodal survival prediction.
[CV-44] riScore: Post-Hoc Out-of-Distribution Detection with Energy Boundary Probes and Transform Consistency
链接: https://arxiv.org/abs/2610.26323
作者: Ege Gursoy,Nicolas Mansard
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Post-hoc out-of-distribution detection usually relies on a single signal extracted from a frozen classifier, such as confidence, energy, features, or gradients. This dependence becomes fragile under classifier-level adversarial shift, where the shift can preserve the detector’s chosen cue while changing the model’s decision behavior. We investigate whether this failure mode can be mitigated by combining complementary detection criteria instead of selecting a single score. We introduce TriScore, a post-hoc OOD detector that combines three ID-calibrated signals from the same frozen backbone: an energy prior, a boundary-fragility score, and a transform-consistency score. These signals are standardized on ID validation data and fused by an input-dependent ID-residual gate, requiring no OOD labels, auxiliary model, training, or stored feature bank. Across four ImageNet-scale backbones, TriScore achieves the best mean AUROC among 15 post-hoc baselines under a non-adaptive adversarial-shift protocol, while remaining competitive on standard OOD benchmarks. Ablations show that the gains come from combining complementary cues rather than optimizing a single detector signal. These results support multi-criterion post-hoc detection as a practical route toward more robust OOD detection under classifier-level shift. The source code is publicly available at this http URL.
[CV-45] MCDA: Multi-View Graph Contrastive Learning with Dynamic Topology Augmentation for Functional MRI Analysis
链接: https://arxiv.org/abs/2610.26322
作者: Hao Zhang,Xiaoyun Liu,Shuo Huang,Yue Ma,Liang Cao,Mingxia Liu,Li Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 14 pages, 8 figures
Abstract:Resting-state fMRI (rs-fMRI) is a vital non-invasive tool for diagnosing brain disorders and uncovering pathological mechanisms. However, existing graph contrastive learning (GCL) methods often underexploit augmentation histories across training epochs and the relationships among different augmented views. In this paper, we introduce a topology-guided Multi-view graph Contrastive learning framework with Dynamic graph topology Augmentation (MCDA) for fMRI analysis. Unlike traditional GCL that uses fixed noise or edge-dropping, MCDA employs a Dynamic Augmentation Update (DAU) module that iteratively refines the augmented topology. Specifically, we incorporate self-aware and cross-sample augmentation strategies to simultaneously capture individual-specific functional signatures and inter-subject heterogeneity. We then design a topology-guided multi-view graph contrastive learning scheme that explores interrelationships not only between the original network and its augmented views, but also across different augmented views to maximize structural information preservation. We characterize the score-level optimization directions and establish a lower bound for the contrastive objective. The framework combines classification and contrastive losses for brain disorder diagnosis and examines discriminative functional connectivity patterns as potential biomarkers. Experiments on two public rs-fMRI datasets comprising 1,236 subjects demonstrate that MCDA outperforms the state-of-the-art methods. The code has been made publicly available at: \urlthis https URL.
[CV-46] CARNet: End-to-End Coupled Association-Recognition Network for Fine-Grained Multi-Drone Collaborative Detection
链接: https://arxiv.org/abs/2610.26312
作者: Zhao Zhang,Ruigang Fu,Yu Zhang,Jiajun Xiang,Zixuan Xiao,Ping Zhong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multi-drone collaborative detection aims to enhance target recognition by leveraging complementary observations from overlapping aerial viewpoints. Existing approaches typically follow a two-stage pipeline that first establishes cross-view correspondences and then performs fusion-based recognition. However, this decomposition suffers from inconsistent optimization objectives, redundant pairwise modeling, and irreversible error propagation, where incorrect associations inevitably undermine subsequent recognition performance. To overcome these limitations, we propose the Coupled Association-Recognition Network (CARNet), a unified framework that jointly optimizes cross-view target association and recognition fusion through a shared target-pair representation. To improve association reliability in densely populated scenes with visually similar targets, CARNet explicitly incorporates spatial cues into pairwise relation modeling. Moreover, we introduce two complementary causal-inspired intervention strategies: the first enhances cross-view association by emphasizing appearance-confusing pairs, while the second bolsters the robustness of fusion-based recognition by learning stable multi-view representations under incomplete visual evidence. In addition, we present MDMT-CD, a fine-grained extension of the MDMT dataset that enriches the original vehicle annotations with detailed category labels, providing the first benchmark specifically designed for multi-drone collaborative vehicle detection. Extensive experiments demonstrate that CARNet consistently outperforms existing state-of-the-art methods. The MDMT-CD dataset and source code will be available at this https URL.
[CV-47] From Cause to Effect: Generating Intermediate Event Chains in Videos EMNLP2026
链接: https://arxiv.org/abs/2610.26275
作者: Ridouane Ghermi,Vicky Kalogeiton,Ivan Laptev
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at EMNLP 2026. Project page: this https URL
Abstract:Video-Language Models (VLMs) excel at visual recognition but struggle with causal reasoning. Current evaluations are flawed: multiple-choice QA allows shortcuts, and open-ended QA lacks robust automatic evaluation. We introduce Causal Chain Generation (CCG), a novel task where models have to generate the sequence of intermediate events connecting a cause and an effect. This structured setup demands genuine reasoning and enables reliable automatic scoring. We accompany this task with two complementary metrics: SODA-Embedding, which measures the semantic matching of a generated chain against a reference, and Causal-QA, which probes the chain’s causal coherence by asking whether it contains the specific causal facts needed to reconstruct the narrative. We provide two test sets: SF20K-CC, a manually annotated diagnostic on 20 short films, and CMD-CC, a movie clip benchmark derived from audio descriptions. Our analysis reveals that state-of-the-art VLMs struggle with narrative causality understanding, frequently hallucinating or omitting key events. To address this, we fine-tune VLMs with GRPO using a composite reward that combines a format reward, SODA-Embedding, and Causal-QA accuracy, demonstrating substantial improvements over strong baselines.
[CV-48] IM: Temporal Decoupling with Iterative Mutual-Refinement Model for Longitudinal Radiology Report Generation CVPR2026
链接: https://arxiv.org/abs/2610.26271
作者: Yiheng Dong,Yi Lin,Shilong Huang,Xiyan Yang,Xin Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by CVPR 2026
Abstract:Automatic radiology report generation (RRG) aims to translate medical images into diagnostic text, reducing radiologists’ workload and standardizing clinical documentation. Nonetheless, existing approaches mainly focus on single-timepoint analysis and fail to capture temporal disease evolution across longitudinal examinations. While recent longitudinal RRG (LRRG) approaches incorporate historical data, they often combine images from different time points within a single representation space, leading to blurred semantics and inconsistent temporal reasoning. In this work, we propose a Temporal Decoupling with Iterative Mutual-Refinement Model (TIM), a two-stage framework that explicitly decouples spatial pathology from temporal progression and iteratively refines reports through mutual feedback. Stage I performs temporal-decoupled representation learning, separating temporal evolution patterns from disease-specific features and generating radiology reports for both prior and current studies. Stage II introduces a mutual report refinement mechanism that identifies diagnostic inconsistencies within prior reports and iteratively rectifies both prior and current reports through error-sensitive feedback. Experiments on the Longitudinal-MIMIC dataset demonstrate that TIM surpasses existing single-image and longitudinal baselines, achieving new state-of-the-art performance across both language and clinical metrics. Code is available at this https URL.
[CV-49] Dynamics-Aware Sparse Attention for Efficient Autoregressive Video Diffusion NEURIPS2026
链接: https://arxiv.org/abs/2610.26262
作者: Yuanyu He,Zhuokun Chen,Yefei He,Zhiwei Tang,Jiasheng Tang,Yinghao Yu,Jianfei Cai,Bohan Zhuang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: NeurIPS 2026, Project page: this https URL
Abstract:Autoregressive Diffusion Transformers have emerged as a powerful paradigm for long video generation. However, they are often constrained by prohibitive computational overhead and substantial memory footprints. This inefficiency arises from dense computation mechanisms that allocate uniform processing resources across all spatiotemporal regions, failing to leverage the inherent temporal redundancy of video where motion is typically confined to sparse areas. In this paper, we propose Dynamics-Aware Sparse Attention (DASA), a unified framework that shifts from dense computation to sparse, motion-driven generation to eliminate both storage and computational redundancies. Inspired by video codec principles, our approach explicitly decouples video content into dynamic and static regions. To mitigate memory bottlenecks, we selectively compress the historical KV cache by retaining only dynamic features within the generated chunk. Furthermore, to reduce computational overhead, we employ a region-level computation allocation strategy based on localized dynamics. Additionally, a dynamics-guided distillation method is introduced to enhance performance via a lightweight distillation training process. Experiments on VBench demonstrate that our approach accelerates autoregressive video generation by 2.16x, while preserving high visual fidelity.
[CV-50] XGLASS: X-Ray Gaze and Language Assisted Slice Sequencing for CT Volumes WACV2027
链接: https://arxiv.org/abs/2610.26255
作者: Michael Glöckler,Timo Ropinski
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027. 24 pages, including references and supplementary material. Code and checkpoint: this https URL
Abstract:Vision-language models (VLMs) achieve remarkable performance on 2D medical images, such as X-rays or pathology images, while 3D VLMs that directly process CT/MRI volumes lag behind due to limited 3D training data and the predominance of 2D foundation architectures. To leverage powerful 2D VLMs for analyzing medical volumes, which contain hundreds of slices, it is therefore necessary to focus only on those few slices, which are diagnostically relevant. To enable such a prioritization, we propose XGLASS, a method that transfers medical expertise from 2D X-rays to 3D CT for task-aware slice prioritization. XGLASS uses disease-conditioned importance maps derived from language-aligned radiologist gaze on X-rays and spatially aligns them with coronal CT projections, bringing gaze priors and CT content into a shared 2D representation. The fused gaze-prior/CT projection is then used to predict a disease-specific ordering of axial slices, prioritizing diagnostically relevant slices. XGLASS significantly improves budgeted retrieval over 2D and 3D-native baselines, achieving a top-10 hit rate up to 75.9% for the best-performing disease. In practice, this means that after inspecting only the top 10 slices (4.17% of a 240-slice scan), XGLASS identifies a diagnostically relevant slice in more than 75% of cases. For downstream slice-wise finding classification, an XGLASS-selected 30-slice candidate set reduces model calls by 87.5% while improving F1 by 11.3 percentage points over all-slice inference, 4.2 points over the strongest 2D selector, and remains competitive with 3D-native selectors while requiring up to 130.8 times less time and 10.9 times less GPU memory per finding, supporting its use as a lightweight preselector for volumetric CT analysis. The code and pretrained checkpoint are available on this https URL.
[CV-51] RECAP: Reciprocal Sparse Attention and Hierarchical Caching for Efficient Video Diffusion
链接: https://arxiv.org/abs/2610.26244
作者: Tiange Xu,Kai Liu,Linghe Kong,Yulun Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Code is available at this https URL
Abstract:Diffusion transformers generate high-quality videos but remain computationally expensive. Sparse attention and temporal caching reduce computation within and across denoising steps, respectively, offering a natural composition. However, predictions produced with sparse attention become sources for temporal reuse, coupling their errors along the sampling trajectory. We observe that improving the accuracy of these sources recovers fidelity without changing the reuse schedule, and that sparsification errors remain directionally aligned across nearby steps, suggesting that correction information can be shared rather than recomputed. These observations motivate RECAP, a training-free method that couples sparse attention with hierarchical caching through reciprocal correction, adaptively coordinated by online error feedback. Cached dense refinement corrects omitted interactions in current sparse estimates, while fresh sparse statistics track input changes and regulate the aging correction. The resulting predictions then renew the sources for whole-step prediction reuse, allowing prediction history to advance between dense refreshes. We evaluate RECAP on multiple video diffusion models against sparse-attention methods, caching methods, and their direct compositions. On HunyuanVideo at 720p 5-second generation, RECAP achieves a 3.05x speedup and outperforms all tested direct compositions in both fidelity and inference speed.
[CV-52] SAR2Net: Learning Spatially Anchored Representations for Retrieval-Guided Cross-Stain Alignment CVPR2026
链接: https://arxiv.org/abs/2610.26217
作者: Tianle Shen,Fang Yan,Xiaofan Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, CVPR 2026
Abstract:Achieving spatial alignment between whole-slide images(WSIs) across stains is essential for integrating morphology and molecular information, yet it remains challenging due to extreme resolution, tissue fragmentation, and large nonlinear deformations. Conventional registration pipelines depend on global pre-alignment and spatial consistency, which often collapse under such distortions. We present SAR2Net, a framework that learns spatially anchored representations and reformulates cross-stain alignment as a region-level feature retrieval problem. Instead of estimating explicit transformations, SAR2Net learns pointwise representations encoding the relative spatial relationships to tissue landmarks. Given landmarks and arbitrary coordinates, it predicts spatially anchored features that serve as deformation-invariant descriptors of tissue topology. A twostage retrieval framework then establishes correspondences between slides, even when global alignment is infeasible. Experiments on biopsy-oriented HE-IHC datasets show that SAR2Net achieves robust region-level alignment under severe tissue distortions, outperforming previous methods. Code is available at this https URL.
[CV-53] RoAR-VAE: Improving Latent Diffusability through Robust Prediction
链接: https://arxiv.org/abs/2610.26216
作者: Yang Yue,Fangyun Wei,Yitong Wang,Jinjing Zhao,Henry Zheng,Dong Chen,Gao Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code is available at this https URL
Abstract:An effective visual tokenizer must preserve image fidelity while producing latents that downstream diffusion models can learn efficiently. Prior work improves latent diffusability through semantic alignment or joint tokenizer–diffusion training. Revisiting REPA-E, a representative approach, we find that removing its diffusion objective and dedicated generative blocks while retaining semantic prediction from corrupted latents yields comparable downstream generation performance under a matched setting. This observation motivates a general principle of robust prediction: generation-relevant information should remain recoverable from corrupted latents. We instantiate this principle with two complementary objectives. RobustAlign predicts pretrained semantic features from corrupted latents to preserve high-level structure, while RobustRecon reconstructs the clean image to retain fine-grained spatial and appearance information. Together, they form Robust Alignment and Reconstruction VAE (RoAR-VAE). Our analysis suggests that robust prediction improves diffusability by making useful information more accessible in high-variance latent directions, which diffusion models tend to learn earlier. Empirically, RoAR-VAE improves average generation benchmark performance by 2.19 points over the strongest baseline tokenizer under the same training budget, while improving small-text reconstruction accuracy by 16.93 percentage points over FLUX.1-VAE. These results demonstrate that RoAR-VAE achieves a favorable balance between reconstruction fidelity and downstream generative learnability.
[CV-54] ExMR: Extremity Motion Refinement for Monocular Human Motion Capture
链接: https://arxiv.org/abs/2610.26213
作者: Tom Wehrbein,Bodo Rosenhahn
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Extended version of the 3DV 2026 paper “Improving 3D Foot Motion Reconstruction in Markerless Monocular Human Motion Capture”. 17 pages main paper + 4 pages supplementary material. Project page and video: this https URL
Abstract:Detailed hand and foot articulation is important for animation and motion analysis, yet remains challenging to recover from monocular video. Recent expressive video methods infer hand articulation from shared whole-body representations. Our conference work suggests that this underexploits local evidence: the Foot Motion Refinement (FootMR) module refines ankle rotations from foot keypoints, improving reconstruction even when the initial estimator receives the same keypoints. We extend this principle to hands with the Hand Motion Refinement (HandMR) module. Because 2D finger keypoints are often missing or unreliable, HandMR combines image features of a pretrained hand reconstruction model with initial arm and hand rotations to refine elbows, wrists, and fingers. We integrate FootMR and HandMR into ExMR (Extremity Motion Refinement), a system for reconstructing 3D body, hand, and foot motion from monocular video. An initial temporal estimator predicts whole-body motion, after which both refiners revisit local evidence to predict residual joint rotations. Experiments on ARCTIC, MOYO, RICH, and MOOF show that ExMR outperforms state-of-the-art methods in hand and foot reconstruction, reducing hand per-vertex error and ankle joint angle error on MOYO by 19% and 18%, respectively, relative to the strongest competitor. Code will be made available for research purposes. Project page: this https URL
[CV-55] Focused Omni-Modal GUI Agents
链接: https://arxiv.org/abs/2610.26203
作者: Fuyao Zhang,Wenwen He,Jiaming Zhang,Yurong Hao,Xuan Wang,Guowei Guan,Che Wang,Longtao Huang,Wei Yang Bryan Lim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Graphical User Interface (GUI) agents increasingly act over streaming omni-modal observations spanning screenshots, audio, video, interaction history, and device states. However, exposing this complete context uniformly at every step ignores that relevance is transient and sparse: the signal determining one action can become clutter at the next. Because executed actions feed back into the context, these errors compound into Sequential Distraction, driving the agent into irreversible trajectory drift. We formulate the underlying requirement as Focused Perception: a representation that is step-adaptive, action-sufficient, and non-intrusive. We realize this in FocusGUI, a lightweight adapter that dynamically prioritizes multimodal channels against a step-specific decision reference. Instead of uniformly processing all inputs, it compresses the most relevant evidence into a compact latent and subtly injects it into the frozen decoder’s embeddings, guiding the model’s attention without altering its native parameter space. FocusGUI trains from action supervision alone, requiring neither modality-relevance labels nor backbone fine-tuning. Across four omni-modal backbones, it raises overall Type Match by +33.2 and Exact Match by 3.0 times over the frozen backbone on average, and Type Match by +30.1 over the strongest existing enhancement method.
[CV-56] GeoEdit: Adversarial De-Evidencing for Silent Geoprivacy Protection via Image Editing
链接: https://arxiv.org/abs/2610.26194
作者: Yu Yan,Shengjia Cheng,Fei Ma,Zhe Wang,Peng Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent efforts have demonstrated the strong capabilities of Multimodal Large Reasoning Models (MLRMs) in geographic reasoning. By jointly analyzing subtle visual cues, these models can infer the locations of photos, thus enabling doxing even by non-expert adversaries. Existing defense methods struggle to balance the shareability of photos with effective protection, and often fail to sufficiently disrupt the implicit geo-evidence chains embedded in scene semantics. To this end, we propose GeoEdit, an adversarial de-evidencing framework for silent geoprivacy protection via image editing. Specifically, we apply minimal edits to high-risk evidence anchors in the image, thereby disrupting the underlying spatial reasoning chains while preserving overall scene semantics. We identify these anchors by combining structured knowledge with adversarially discovered geographic cues. To further navigate the large editing action space, we adopt a learnable policy to iteratively refine editing decisions, distilling successful trajectories into transferable natural-language insights for consistent and efficient editing. Evaluated on LifePhotoBench, GeoEdit achieves strong protection against 6 mainstream MLRMs, outperforming visual perturbation baselines by over 0.59 in image fidelity.
[CV-57] Wan-Alpha: Animating a Single RGBA Asset into Controllable Transparent Video Effects
链接: https://arxiv.org/abs/2610.26189
作者: Haotian Dong,Xin Wang,Wenjing Wang,Chen Li,Jing Lyu,Yan Tina Luximon,Ping Li,Wanchao Su,Ke Xu,Di Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages, 13 figures
Abstract:RGBA video generation is essential for creating compositable visual assets in filmmaking, game development, and digital content production. However, generating high-quality RGB appearance alongside temporally coherent transparency remains a significant challenge. Existing text-to-RGBA video methods often entangle RGB and alpha distributions and offer limited control over the precise appearance of a target asset. In addition, incorporating image conditioning typically suffers from the emergence-disappearance effects, which fail existing first-frame or first-and-last-frame conditions due to their fully transparent boundary frames. To address these issues, we present Wan-Alpha++, a unified framework that brings controllable RGBA video generation to both text- and image-to-video diffusion backbones through a shared RGBA latent space, extending our conference version Wan-Alpha from the text-conditioned setting to image-conditioned generation with an interior reference frame. Our method first learns shiftable RGB-alpha distributions while preserving the pretrained RGB prior. Specifically, we introduce a transparency-aware bidirectional diffusion objective to produce distinguishable RGBA latent representations, and a transparency-guided mean sampler that shifts the noise distribution to improve transparency generation and enable spatial controllability. Building upon this RGBA generation prior, we introduce temporal phase reparameterization for image-conditioned effect generation. By cyclically transforming a zero-positive-zero alpha pattern into a positive-zero-positive pattern, it relocates visible content to both temporal boundaries, facilitating effective appearance conditioning with limited RGBA data.
[CV-58] HP-PnP-Flow: Hybrid Time Scheduling and Projection for Plug-and-Play Image Restoration via Flow Matching
链接: https://arxiv.org/abs/2610.26183
作者: Yong Li,Xiaojie Wang,Xiaoyan Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Image restoration is a fundamental inverse problem that seeks to recover high-quality images from degraded observations. Plug-and-Play (PnP) methods provide a flexible framework for integrating pretrained denoisers into iterative reconstruction, whereas flow matching offers an efficient generative modeling approach based on continuous-time ODE dynamics. Very recently, Martin et al.\ (ICLR, 2025) developed a PnP-Flow method, which integrates the PnP framework with flow matching through a time-dependent denoiser constructed from a pretrained flow matching this http URL incorporating a different time schedule and an additional pixel-space projection operator, this paper proposes a novel efficient PnP-Flow algorithm (HP-PnP-Flow) that enjoys rigorous convergence guarantees under mild assumptions, without imposing restrictive conditions on iterates required by the aforementioned work. Numerical experiments on CelebA and AFHQ-Cat demonstrate that the newly proposed HP-PnP-Flow algorithm significantly outperforms PnP-Flow on several restoration tasks such as denoising, deblurring, super-resolution, random inpainting, and box inpainting. Notably, our algorithm achieves comparable or better reconstruction quality with substantially fewer iterations, reducing the iteration count from 500 to 100 for deblurring and super-resolution on AFHQ-Cat.
[CV-59] Learning Goal-Conditioned Subgoals through Frozen Hierarchical World Models
链接: https://arxiv.org/abs/2610.26179
作者: Xiang Kaibo,Liu Qingfeng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Hierarchical world models plan over long horizons by passing predicted states from a coarse model to a lower-level controller. Choosing these states can be expensive: the upper-level planner repeatedly differentiates through candidate macro-action trajectories. We replace this search with a lightweight goal-conditioned network trained through the frozen high-level predictor. The network supplies a conditioning vector, while the predictor generates the next latent waypoint. Training uses hindsight goals and state-prediction losses on offline trajectories, without search-solution targets or macro-action labels. Inference retains an autoregressive high-level rollout, but requires no high-level candidate optimization. We evaluate this approach on the shared-latent HWM checkpoints distributed with H-JEPA, using three backbone seeds and three head initializations per backbone. With a common evaluation protocol, mean success is 70.4% on Cube, 48.7% on Push-T, 100.0% on FourRoom, and 71.1% on Visual AntMaze. The method improves over the corresponding HWM search baseline on three environments and has the same mean on Push-T. Dedicated measurements on an NVIDIA L40S give 1.28-4.39x end-to-end speedups for batched evaluation. Ablations identify state-adaptive conditioning as a consistent contributor, while predictor-mediated subgoals and unrestricted conditions have environment-dependent benefits.
[CV-60] riTD: Tri-Partite Trajectory-Distribution Distillation for Real-Time Autoregressive Video Generation
链接: https://arxiv.org/abs/2610.26165
作者: Jia Li,Xurui Peng,Haowei Zhu,Tianyu Zhao,Jiexi Wang,Youwei Zheng,Yuxi Ren,Xinglong Wu,Xing Wang,Hayden Kwok-Hay So
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 21 pages, 8 figures
Abstract:Autoregressive (AR) video diffusion has emerged as a leading paradigm for efficient video generation, yet its practical deployment remains bottlenecked by the trade-off between generation quality and inference efficiency. We attribute this bottleneck to three error sources in AR rollouts: the base model error, the train-inference distribution mismatch, and the upstream error propagation. Building on this decomposition, we derive closed-form upper bounds on the per-step and cumulative errors, providing a principled basis for jointly optimizing generation quality and inference efficiency. Guided by this analysis, we propose TriTD, a Tri-Partite Trajectory-Distribution co-optimized distillation framework with three complementary modules. First, for ODE initialization, we adopt a Next-Node Trajectory Distillation (NTD) objective, which shortens the supervision horizon and admits a lower regression-error floor. Second, we introduce Chunk-Graded Noising (CGN), a monotonically non-decreasing per-chunk noising schedule that helps narrow the expected train-inference noise mismatch and removes redundant KV-cache updates, yielding substantial speedups at no quality cost. Third, we propose Orthogonal-Parallel DMD (OP-DMD), a distribution-matching loss that retains the original DMD supervision while introducing an additional regression constraint in the orthogonal subspace to suppress off-axis drift, improving both training stability and final generation quality. Extensive experiments show that TriTD consistently surpasses existing baselines in visual quality while reaching a real-time throughput of 26.8 FPS on a single H100 GPU, delivering a balanced quality-efficiency solution for interactive streaming video generation.
[CV-61] Optimal Multi-Kernel Combination for Online Multi-Label Classification
链接: https://arxiv.org/abs/2610.26123
作者: Tingting Zhai,Wei Liu,Renjian Yu
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: This manuscript has been submitted to Pattern Recognition for consideration and is presently under review
Abstract:Multi-label classification is particularly challenging in streaming scenarios, where data arrives sequentially. Online kernel methods offer a promising solution. However, prevailing methods rely on a single preselected kernel, making them prone to suboptimal choices since the global data distribution is unknown in advance in online settings. Although the recent method, Online Kernel Selection for Online Multi-label Classification (OKS-OMC), enables data-driven kernel selection, it fails to fully harness the complementary information from multiple kernels and encounters the curse of kernelization. To address these limitations, this paper introduces a novel online multi-kernel combination method for large-scale multi-label classification. First, an efficient and robust online single-kernel multi-label method is designed. Second, learning an optimal convex combination of multiple multi-label kernel classifiers is formalized as a structural risk minimization problem. An improved alternating optimization procedure is developed, in which the combination coefficients are updated by directly minimizing the objective function, while each kernel classifier is updated by minimizing a surrogate upper bound. This procedure decouples the learning of kernel classifiers from combination coefficients, allowing independent classifier updates via our single-kernel method and accelerating convergence to high-quality solutions. Our multi-kernel method is proven to asymptotically converge to the optimal fixed convex combination of the best fixed single-kernel models chosen in hindsight. Ablation studies confirm that our method’s components outperform their counterparts in OKS-OMC. Comprehensive evaluations across 11 datasets and seven metrics demonstrate our method’s overall superiority over existing online multi-label methods.
[CV-62] PnP-Transformer: A Neural Pose Solver for 6D Object Pose Estimation ECCV2026
链接: https://arxiv.org/abs/2610.26096
作者: Emirkan Burak Yilmaz,Yakup Genc
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to ECCV 2026 R6D Workshop
Abstract:We present PnP-Transformer, a neural pose solver designed as an iterative-free, differentiable alternative to classical PnP-RANSAC algorithms. PnP-Transformer regresses the 6D object pose by processing predicted semi-dense 2D-3D correspondence fields alongside confidence scores, while mapping geometric inputs into a continuous latent space. To evaluate its practical utility, we integrate PnP-Transformer into a zero-shot 6D pose estimation pipeline, where the correspondence module and pose solver are trained separately. PnP-Transformer achieves a 41.2% mean AR score with low latency (0.46 s/obj) on seven core BOP benchmark datasets, surpassing all prior baselines (such as RayPose with 39.1% AR) but one (Co-op with 58.4% AR). When the pose solver is replaced with EPnP-RANSAC (similar to Co-op) while keeping the correspondence module unchanged, the RANSAC alternative outperforms PnP-Transformer on six of the seven datasets by reaching 52.2% mean AR at a cost of an additional 60 ms per object, revealing an accuracy-latency trade-off. Code and pretrained models are available: this https URL
[CV-63] How Robust Are Prognostic Radiomic Features? Comparing Selection Techniques on Pooled Glioblastoma Cohorts
链接: https://arxiv.org/abs/2610.26095
作者: Yannick Sodonougbo,Marlene Schnieders,Alexia Rizoudis,Santiago Cepeda,Frank Kramer,Dominik Müller
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Quantitative Methods (q-bio.QM)
备注: 7 pages, 1 figure
Abstract:Radiomic survival studies in glioblastoma report heterogeneous prognostic features, but how much of this stems from the methodology itself is rarely examined. We analysed the association of 463 radiomic, anatomical and clinical features from pre-treatment MRI with overall survival in 705 IDH-wildtype glioblastoma patients from three public cohorts, using three techniques: a log-rank test after median split, univariate Cox regression (both Benjamini-Hochberg corrected) and an XGBoost regressor interpreted with SHAP. The log-rank test selected 99 features, Cox regression 72 and SHAP its top 20; five were selected by all three: age, a peritumoural T2 texture feature, the number of connected peritumoural mask components and two image-level intensity maxima. The two survival tests shared 51 features, about half of the SHAP top 20 were also selected by either test, and where selections overlapped, effect directions agreed. Many selections were PyRadiomics diagnostic descriptors that are not restricted to tumour voxels. Analytical pipeline choices, including ranking method, pre-filtering and descriptor handling, can substantially influence which features appear prognostic.
[CV-64] GeoNaviDiff: Transferring Disentangled Semantic Controls to Diffusion Features via Local Riemannian Navigation
链接: https://arxiv.org/abs/2610.26091
作者: Tao Sun,Huanting Guo,Baao Xie,Zhou Fang,Qiuyu Chen,Hao Zheng,Wenjun Zeng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Disentangled real-image manipulation requires factor-specific semantic control while preserving image fidelity and non-target content. StyleSpace S provides interpretable controls but is limited by GAN inversion, whereas diffusion models offer stronger image realization without an explicit factorized control basis. We introduce GeoNaviDiff, an unsupervised cross-space framework that transfers disentangled StyleSpace interventions to local Riemannian trajectories in diffusion h -space. GeoNaviDiff constructs reconstruction-decoupled supervision by extracting a spherical Log-map tangent from a matched StyleGAN reconstruction-edit pair and parallel-transporting it to the real-image diffusion state. A shared, factor-conditioned Navigator predicts this tangent at inference, while norm-preserving geodesic realization and source-anchored rollout stabilize the reverse process without edited references or factor-specific models. Experiments across multiple datasets show that GeoNaviDiff outperforms the compared methods in fidelity, factor selectivity, edit locality, and trajectory smoothness.
[CV-65] RGB–Event Object Detection Across Operating Frequencies via Reliability-Supervised Routing and Asymmetric-Resolution Fusion
链接: https://arxiv.org/abs/2610.26090
作者: Peijun Shi,Jiannan Zhao,Xiujin Zhu,Zhenguo Lao,Chee-Onn Chow,Wei Ru Wong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Autonomous vehicles must detect objects faster than a frame camera delivers images. Event cameras record brightness changes with microsecond resolution, so RGB–event detectors can decide between frames, at an operating frequency, the rate of decisions, far above the frame rate. As this rate rises, each decision receives a sparser event window, and the reliability of the events relative to the frame varies over the scene and over time. Existing detectors weight the sensors per sample and encode the frame at the reduced event resolution. Reliability-supervised routing assigns a fusion weight to every location and supervises it with targets from the losses of two single-modality classifiers used only in training, moving loss-based supervision of fusion weights from whole samples to individual locations. Asymmetric-resolution fusion lets the frame carry the decision where the events fail: it encodes the frame at native resolution and pools every feature level onto the event grid, reusing all pretrained weights. On the DSEC-Det dataset, routing halves the accuracy loss from 20 to 180 Hz, from 6.3 to 3.1 mAP, and lowers it in every seed. The resulting detector, ERA-Det, sets a new state of the art at 63.2 mAP, 3.7 mAP above FlexEvent, leads at every operating frequency, and at 180 Hz is more accurate than FlexEvent at 20 Hz. On two DAVIS346 datasets, the router lowers the frame weight at night and in low light and raises it when the events vanish. ERA-Det upgrades released FlexEvent weights in 5,000 iterations on one GPU.
[CV-66] DeOverlay: A Benchmark and Source-Anchored Flow for Video Overlay Removal
链接: https://arxiv.org/abs/2610.26081
作者: Deshui Miao,Yuhang Ma,Shaoyong Jia,Wenzhao Gao,Yue Zhang,Shaohui Jiao,Xin Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video overlay removal aims to recover scenes obscured by subtitles, titles, and logos while preserving surrounding content. Existing methods use temporal propagation or source-conditioned generation, but can leave residual overlays and distort scene structure when useful source information is obscured. We propose DeOverlay, a source-anchored flow model that reconstructs the clean video from the source latent with fixed source guidance. Specifically, we train a shared velocity network along paired source-to-clean paths using region-weighted flow matching, enabling successive cor- rections from intermediate states. To select local source cues, we develop visibility-normalized temporal aggregation that excludes masked positions and averages the remaining observations. Global source context and edit-region embeddings complement these cues through additive condi- tioning, retaining a single video-token sequence. To evaluate scene recovery across diverse overlay conditions, we introduce DVTE-Bench, comprising 1,631 paired videos across eight text-overlay categories, with OCR-screened clean targets, frame-aligned pixel masks, and a common evaluation protocol. Experiments on the complete release show that DeOverlay outperforms four baselines on all five reconstruction and perceptual metrics, reducing Mask LPIPS and Mask VFID by 58.3% and 70.3%, respectively, relative to the strongest evaluated baseline. Additional qualitative examples demonstrate the removal of graphic logos and stickers. We will make the complete benchmark, evaluation code, and test outputs availabl. Project page:this https URL
[CV-67] oolkit for Facial Deidentification
链接: https://arxiv.org/abs/2610.26079
作者: Esteban Leiva-Montenegro,Matthieu Pillot,Manfred Gonzalez-Hernandez,Vitomir Struc,Peter Peer,Blaž Meden
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 3 pages, 2 figures, 2 tables. This work was presented as a poster at the 6th IEEE International Conference on BioInspired Processing (BIP 2024), poster session, Liberia, Costa Rica, December 4-6, 2024
Abstract:This work presents a novel open-source toolkit designed for the evaluation of face deidentification techniques with available evaluation models. The framework provides a complete evaluation pipeline, including preprocessing, deidentification, and performance assessment, based on data utility, image quality, and identity protection metrics. It supports a variety of datasets, enabling researchers to easily benchmark new deidentification techniques. Results show the effectiveness of some techniques on maintaining privacy while preserving image utility. The toolkit is aimed at unifying future research by providing a standard evaluation framework for privacy-enhancing technologies in facial biometrics. Implementation is available at this http URL.
[CV-68] DCR-TSF: Dual-Level Compositional Reasoning with Triadic Spherical Fusion for Training-Free Zero-Shot Composed Image Retrieval ACCV2026
链接: https://arxiv.org/abs/2610.26078
作者: Gwang-Ho Na,Ho-Joong Kim,Seong-Whan Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to ACCV 2026. 14 pages, 6 figures, 6 tables
Abstract:Zero-shot composed image retrieval (ZS-CIR) aims to retrieve a target image that reflects an intended textual modification while maintaining semantic consistency with a reference image. Existing training-free ZS-CIR methods typically rely on fine-grained generated target descriptions, which often overemphasize reference-specific visual details and lead to appearance-biased retrieval. Moreover, since these descriptions are generated through multimodal large language models (MLLMs), the direct influence of the original image-text query can be weakened in the final retrieval representation. In this paper, we propose DCR-TSF, a training-free ZS-CIR framework that combines Dual-level Compositional Reasoning (DCR) and Triadic Spherical Fusion (TSF). DCR generates fine- and coarse-grained descriptions from the reference image and modification text, capturing complementary target-side semantics across different granularities. To preserve the original image-text query information, TSF constructs a composed-query anchor by spherically interpolating the reference image feature and modification text feature, and fuses it with the fine- and coarse-grained representations using a weighted Karcher mean on the unit hypersphere. The resulting final query representation balances detailed visual evidence, high-level semantic abstraction, and direct preservation of the original composed query. Extensive experiments on CIRCO, CIRR, and Fashion-IQ demonstrate that our method achieves state-of-the-art performance among training-free methods.
[CV-69] Not All Similarity Is Leakage: Distribution-Aware Identity Leakage Assessment in Synthetic Face Datasets
链接: https://arxiv.org/abs/2610.26077
作者: Paweł Borsukiewicz,Daniele Lunghi,Wendkûuni C. Ouédraogo,Fadi Boutros,Jacques Klein,Tegawendé F. Bissyandé
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Synthetic data provides the means for more ethical and less biased facial recognition. Nonetheless, the limitations of generative models and the conditioning process they rely on pose a risk of identity leakage, resulting in synthetic samples that are overly similar to those seen in a generator’s training set, mitigating privacy benefits. In this work, we show that previously proposed approaches are affected by two key limitations: high classifier dependence and a tendency to identify leakage with respect to real datasets that were not used in training. To address these shortcomings, we propose a novel metric called Scaled Overlap (SO), which adapts a non-parametric data-copying test for facial recognition and extends it to the identity level. By leveraging a multi-model setup that compensates for errors in individual classifiers, it provides insights into the true strength of identity leakage. Our analysis identifies leaking datasets (CemiFace, SFace2, IDNet) with the most acute leakage introduced by generative model conditioning in CemiFace, demonstrates that real datasets do not contribute equally to generator training, and shows that the majority of synthetic images are drawn from multiple identities rather than relying on a single identity. Ultimately, our study provides new insights into the quality of synthetic datasets and gives more reliable techniques to help the community ensure the trustworthiness of synthetic data.
[CV-70] Risk Decomposition for Hallucination-Aware Video Sampling
链接: https://arxiv.org/abs/2610.26069
作者: Chenyan Zheng,Fan Wu,Ruibo Li,Jiaxing Li,Minghao Yang,Bang Zhang,Guosheng Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Despite achieving high visual fidelity, video models still suffer from physical hallucinations, limiting their reliability as world simulators. Existing methods leverage model-internal or external signals to improve physical plausibility, but their roles and limitations require a unified explanation. We propose a two-dimensional physical-hallucination risk framework that defines low-density sampling risk and high-density drifting risk. The first risk captures generations assigned low likelihood by the video model, for example, a ball suddenly teleporting despite the model favoring continuous motion, while the second risk captures physical inconsistencies in high-likelihood generations, for example, a ball bouncing progressively higher without additional energy. Our analysis explains how model-internal signals steer sampling toward higher-density regions of the learned distribution to address low-density sampling risk. However, higher model likelihood does not guarantee physical plausibility, motivating external guidance to address high-density drifting risk. Based on this framework, we develop Hallucination-Aware Sampler (HA-Sampler), which applies external physical guidance within latent refinement to generate physically plausible videos, and HA-Scorer, which derives a likelihood-based signal from the generator itself for evaluating generated videos using continuous normalizing flow (CNF), with coarse ODE integration and finite-difference divergence estimates. Experiments on Physics-IQ and VideoPhy benchmarks support our theoretical analysis and demonstrate that our methods outperform the compared sampling and scoring SOTA baselines.
[CV-71] PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning
链接: https://arxiv.org/abs/2610.26067
作者: Yuanhang Lei,Baofei Liu,Xingxuan Li,Tao Cheng,Zixuan Wang,Boming Zhao,Yujun Shen,Siyuan Huang,Ruizhen Hu,Peter Yichen Chen,Hujun Bao,Zhaopeng Cui
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: Project Page: this https URL
Abstract:Generalizing real-time physics-based animation across diverse 3D shapes and spatial discretizations remains challenging. In this paper, we introduce PhysSkin++, a physics-informed framework that learns continuous neural skinning fields from static geometries to construct compact deformation subspaces. Specifically, we propose a Neural Skinning Fields Autoencoder that predicts shape-dependent deformation subspaces without per-object network training. To learn these subspaces without simulation trajectories, we develop a Physics-Informed Self-Supervised Learning strategy that combines skinning-field normalization with conflict-aware gradient correction, stabilizing joint optimization under physical and orthogonality constraints. The learned representation supports efficient reduced dynamics with user control and contact interactions, together with direct animation binding to meshes, point clouds, and 3D Gaussians. For prescribed rig animation, we further introduce a self-supervised complementarity objective to learn complementary elastic responses without enforcing complementarity constraints at runtime. Experiments demonstrate cross-shape generalization of the learned skinning fields, real-time reduced simulation, and applications to controllable, collision-aware, and rig-complementary animation.
[CV-72] Learning the Dose Manifold: A Latent Diffusion Framework for Volumetric Dose Prediction in Automated Radiotherapy Treatment Planning
链接: https://arxiv.org/abs/2610.26064
作者: J. Boyer-Chammard,R. Vauclin,M.C. Biston,P. Maury,M. Gau,S. Stathakis,V. Grégoire,N. Paragios,E. Deutsch,N. Komodakis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 7 figures plus graphical abstract
Abstract:Automated treatment planning (ATP) aims to reduce the unwarranted inter-planner variability of manual planning, while improving treatment quality and efficiency. Most ATP pipelines rely on the prediction of the achievable three-dimensional dose distribution for a given anatomy and the subsequent optimization of machine parameters to reproduce it. Diffusion-based methods are largely slice-wise due to the high dimensionality of radiotherapy data, and assume a one-to-one mapping between anatomy and optimal dose despite the multi-objective nature of treatment planning. They are evaluated on their ability to match reference doses and rarely on their integration in ATP frameworks. We present LaDose, a latent diffusion framework for fully volumetric dose prediction. An autoencoder learned on dose distributions, trained so that perturbed latents decode to plausible doses, compresses dose maps 128-fold and enables 3D diffusion on a single GPU. Prescription values are represented as exact scalar vectors that bypass the image encoder, avoiding resampling and latent compression. Multiscale classifier-free guidance allows for the modulation of clinical trade-offs at inference, without retraining or additional labeled data. On 900 clinically delivered head-and-neck volumetric modulated arc therapy plans from five centers, LaDose reached a median absolute error of 1.18 Gy (1.4–1.8% of the prescription) and was the only one of six frameworks free of systematic bias with respect to the reference distribution. Guidance scales acted predictably on target coverage and organ-at-risk sparing. Integrated with direct aperture optimization, the pipeline generated plans in 15.2 minutes; in a blinded review, senior clinicians rated 14 of 15 plans acceptable without editing.
[CV-73] Procedural Exposure Bias in Skill-Augmented Language Model Agents
链接: https://arxiv.org/abs/2610.26062
作者: Qing Yang,Yuqi Xiong,Pengcheng Huang,Zhipeng Xu,Zhenghao Liu,Zulong Chen,Yu Gu,Ge Yu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 26 pages, 12 figures, 8 tables
Abstract:External skills provide language-model agents with reusable procedural knowledge, yet how models use such knowledge after it is supplied remains poorly understood. We identify Procedural Exposure Bias (PEB), the tendency of language models to increasingly favor an action as its associated procedural pattern is repeatedly exposed in a skill, even when the interaction state and action relevance remain unchanged. We find that PEB strongly depends on how procedural knowledge is represented: presenting the same content visually substantially attenuates exposure-induced shifts in action preference relative to textual presentation. Further analysis shows that this attenuation primarily arises from visual encoding rather than structured layout. Although repeated procedural signals remain recoverable from Visual representations, they exert a weaker influence on downstream action prediction, distinguishing signal accessibility from its use as action evidence. This effect also extends to executable action selection, where increasing exposure more strongly favors the corresponding action under Text than Visual. Layer-wise analyses show that exposure-dependent action evidence becomes more pronounced toward the model output, with stronger evidence under Text, while activation interventions reveal architecture-specific roles of internal components in transmitting or counteracting the preference shift. Complete-agent evaluations across ALFWorld, ScienceWorld, and WebShop further show that Visual skills achieve higher macro-average task success than Text skills for all four evaluated models. These findings establish skill representation as an important factor governing how procedural knowledge is translated into agent decisions. All code is available at this https URL.
[CV-74] AVTime: Reinforcing Dense Audio-Visual Captions with Bidirectional Alignment
链接: https://arxiv.org/abs/2610.26046
作者: Mingyang Wu,Kaituo Feng,Bohao Li,Kaixiong Gong,Zihao Yin,Xiangyu Yue
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Dense temporal audio-visual captioning is important for multimodal video understanding and generation. However, prior work is constrained by three main limitations: (1) the lack of a dataset combining high temporal resolution, longer videos, and bidirectional alignment between time and events; (2) the reliance of most existing RL methods on checklists; and (3) the lack of a benchmark that bidirectionally evaluates event semantic correctness and temporal accuracy. To address these challenges, we propose: (1) AVTime-50K, a high-quality 50K-scale dataset with temporally precise, semantically accurate captions and longer videos; (2) AVTime, a state-of-the-art model optimized with a novel Bidirectional Temporal-Semantic Alignment (BiTSA) reward that imposes explicit bidirectional constraints between time and events; and (3) AVTime-Bench, a specialized benchmark that introduces a novel bidirectional evaluation of temporal accuracy and semantic correctness. Our code, models, and datasets are available at this https URL.
[CV-75] After the Federation: Benchmarking Knowledge Portability to Heterogeneous Newcomers
链接: https://arxiv.org/abs/2610.26026
作者: Xinyuan Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages
Abstract:Once historical clients become unavailable, a federation must serve newcomers through the artifact it leaves behind. How much of its knowledge remains useful in an architecture chosen afterward? We introduce FEDNBO, a seven-dataset benchmark that separates knowledge supply, policy identification, and private-model compilation under disjoint data roles. Controlled studies expose inflated oracle gains, architecture-dependent transfer reversals, and a gap between teacher accuracy and private-model utility. Ordinary public-response distillation provides a competitive reference without public task labels or newcomer samples: gains over strengthened local training are 23.52 and 16.62 percentage points on 300 CelebA identities and 300 FedEMNIST writers, respectively. In a complete-pipeline FedEMNIST comparison, this reference reaches 97.13% accuracy with 12.4 seconds of compilation per architecture, versus 97.03% and 67.5 seconds for a HeteroFedTTA reconstruction at the same epoch budget. Equal-count public-input interventions extend support sensitivity to trainable visual backbones. Together, these results establish an auditable benchmark and a strong reference for evaluating the utility, cost and reliability of post-federation knowledge transfer.
[CV-76] amedFlow: Trajectory-Aware Distillationfor Teacher-Assisted Diffusion Sampling
链接: https://arxiv.org/abs/2610.26025
作者: Shuaiqi Cheng,Jiahao Zhang,Boyuan Gu,Siyu You,Yuxi Chen,Haiying Xu,Zexin Lu,Huiyu Zhou,Xuming Hu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Diffusion Transformers (DiTs) deliver high-fidelity image and video generation, but inference requires many sequential model evaluations. Two acceleration strategies are increasingly important: model compression makes each evaluation cheaper, while adaptive inference avoids selected teacher computation through reuse, forecasting, or speculative execution. Their combination, however, is brittle: directly applying the evaluated adaptive methods to a compressed student can degrade quality, an observed failure mode we call the \textbfCombination Gap. We link this gap to a mismatch between pointwise distillation and short-horizon execution: a student trained at teacher states need not remain reliable when its own predictions are chained. We propose \textbfTamedFlow, which trains a compressed DiT for this deployed role through cross-architecture feature alignment, a near-identity Velocity Alignment Head, and randomized rollout supervision. At inference, a lightweight Learnable Trajectory Estimator routes completed drafts: accepted drafts bypass the teacher, while rejected drafts roll back and receive one teacher rescue step. Distilling DiT-XL/2 into DiT-B/2, TamedFlow reaches FID 2.58 at 2.25\times measured speedup over DDIM-50 on ImageNet 256\times256 . Ablations support trajectory supervision, and routing controls favor LTE over random decisions at nearly matched latency. Evaluations on ImageNet 512\times512 , MS-COCO, and Latte/FaceForensics extend the empirical scope.
[CV-77] FrameGRPO: Incentivizing Video Reasoning via Multi-Frame Rollouts and Temporal Credit
链接: https://arxiv.org/abs/2610.26022
作者: Jiaxuan Sun,Biao Wang,Shuai Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning with verifiable rewards, particularly Group Relative Policy Optimization (GRPO), has improved video reasoning in multimodal large language models. However, standard video GRPO varies only linguistic trajectories while conditioning every rollout in a group on the same sampled frames, which can produce homogeneous rewards and weaken group-relative learning when the shared visual evidence is insufficient or distracting. We propose \textbfFrameGRPO, a simple and effective reinforcement-learning framework that requires neither auxiliary networks nor changes to the model architecture. Specifically, FrameGRPO combines two lightweight components: (1) \textbfMulti-Frame Rollouts, which use only uniform temporal subsampling of a single maximal-frame decode to construct aligned views that diversify visual evidence and reasoning trajectories; and (2) \textbfGated Temporal Efficiency Credit, which selectively reinforces a correct lower-ratio trajectory when all rollouts in a fixed full-ratio reference subset fail, without imposing an unconditional preference for fewer frames. Extensive experiments across diverse video understanding and reasoning benchmarks demonstrate robust overall improvements over standard GRPO, highlighting temporal-view diversification as an effective mechanism for exploration and credit assignment in video MLLMs.
[CV-78] Curvature as a Baseline: Monocular Pose Measurement of Cylindrical Objects with Dual Colour Stripe Markers
链接: https://arxiv.org/abs/2610.26003
作者: Jaesung Hong,Dong Won Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Thin cylindrical objects offer little space for planar markers. We propose a dual-colour-stripe marker that uses the known curvature of an object to improve pose measurement with a single camera. The boundary between two colours provides a long centre line, while dots and printed edges supply features at different depths. We analyse how this depth separation improves the estimation of rotation about the cylinder axis, or roll. For a camera-facing marker at 15 cm, the Cramér-Rao lower bound on roll standard deviation is 2.00°for a flat marker, 0.38°for a single curved marker and 0.21°for a wrap-around marker. Monte Carlo roll RMS errors are within about 10% of the corresponding standard-deviation bounds. A first-order relation connects the sagitta of the curved marker to the roll sensitivity provided by a stereo baseline. The analytical and simulation results explain how curvature improves monocular measurement and provide a basis for designing marker geometry.
[CV-79] On the Benefits of Uncertainty Estimation for Species Classification: A Case Study on Optimizing Expert re-Annotation
链接: https://arxiv.org/abs/2610.25977
作者: Kamenan N’Gadi(LORIA,BISCUIT),Martin Laviale(LIEC,LTSER-ZAM),C{é}dric Pradalier,J{é}r{é}my Fix(LORIA,BISCUIT)
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Automated identification of diatoms from microscopic images is a cornerstone of large-scale freshwater biomonitoring, yet the high diversity of species and strong class imbalance pose significant challenges for reliable classification. In this study, we develop a ResNet-18 baseline for a 759-class diatom dataset and investigate how calibrated confidence estimates support selective classification to reduce expert workload without sacrificing taxonomic integrity. While the initial model exhibits overconfidence with an Expected Calibration Error (ECE) of 0.117, post-hoc temperature scaling (T = 2.12) dramatically improves reliability, lowering the ECE to this http URL leveraging these calibrated probabilities within a triage framework, we demonstrate that a Selective Macro-F1 score of 99.01% can be maintained at 66.2% coverage (rejection threshold \tau = 0.934). Qualitative analysis reveals that the system effectively functions as an intelligent filter, consistently deferring rare taxa-such as AAMB and AASI-and ambiguous morphologies for expert review. Our findings, supported by the torch-uncertainty framework, highlight the potential of simple calibration techniques to make automated diatom identification practical and robust, prioritizing the preservation of rare species metrics essential for ecological assessment. To support reproducibility, our code, environment specifications, and calibration results are available anonymously at https://anonymous.4open. science/r/greenprwksp-B53D/README.
[CV-80] HORIA: 4D Human and Object Reconstruction from a Monocular Video with Interaction-Aware Re-Rendering
链接: https://arxiv.org/abs/2610.25960
作者: Hyeongjin Nam,Taeryung Lee,Daniel Sungho Jung,Kyoung Mu Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages including the supplementary material
Abstract:Joint 4D reconstruction of a human and object from a monocular video is essential for understanding how people interact with their environment. While recent methods have made noticeable progress, capturing such human-object interactions remains difficult due to two main challenges. First, a video typically contains composite semantic context spanning multiple interaction phases, making it difficult to track the temporal changes of the interaction in the reconstruction. Second, under human-object occlusion, a single-view observation contains incomplete visual information about the interaction, resulting in geometrically inconsistent reconstructions. To address these challenges, we introduce HORIA, built upon two core designs. First, we decompose the entire input video into temporally grounded and semantically distinct interaction phases with motion-aware captioning, obtaining a specific text description for each phase. Second, conditioned on these text descriptions of the interaction, we propose an interaction-aware re-rendering that compensates for the limited single-view observation. Specifically, we render the initially reconstructed 4D human and object from novel viewpoints and refine the renderings with a text-to-video diffusion prior, which provides auxiliary novel-view supervision beyond single-view cues. As a result, our framework produces temporally coherent and accurate 4D reconstructions under diverse interaction scenarios, achieving state-of-the-art performance.
[CV-81] OmniGround: Strengthening Omni-Modal Temporal Grounding via Attention-Consistency Training
链接: https://arxiv.org/abs/2610.25929
作者: Zixuan Chen,Depeng Wang,Hao Lin,Ke Xu,Ya Guo,Huijia Zhu,Tanfeng Sun,Xinghao Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Omni-modal large language models can jointly understand video, audio, and text, yet they remain weak at temporal grounding, a capability essential for real-world audio-visual applications. To understand this limitation, we look beyond final predictions and examine the model’s internal temporal evidence. We identify a temporal generation gap: the internal attention of the model localizes the target moment more accurately than the timestamps it generates. To close this gap, we propose TACG (Temporal Attention-Consistency Grounding), a post-training method that treats internal attention as a reliable temporal pivot and aligns it with both the model prediction and the ground-truth annotation through attention-consistency training. We further build a 580K-sample training dataset by reconstructing temporal QA from dense audio-visual annotations and organizing general omni-modal instruction data. Using TACG and this training dataset, we train OmniGround, a general-purpose omni-modal model with stronger temporal grounding capability, without injecting any additional temporal cues. OmniGround reaches state-of-the-art omni-modal temporal grounding, including 37.5 mIoU on LongVALE and 35.7 mIoU on UnAV-100, while staying competitive on both general audio-visual understanding and vision-only grounding.
[CV-82] CondenseVLA: Learnable History Condensation for Efficient Multi-Frame VLA NEURIPS NEURIPS2026
链接: https://arxiv.org/abs/2610.25927
作者: Feiyang Hong,Yujie Wei,Bo Zhao,Xiu Su,Hongxun Yao,Shuo Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Accepted at NeurIPS 2026. 39 pages, including supplementary material and the NeurIPS paper checklist. Project page: this https URL
Abstract:Robot manipulation often depends on short-term cues - such as brief contacts, temporary occlusions, and intermediate object states - that are ambiguous in a single frame and can only be resolved by integrating observations over time. However, injecting historical information solely by enlarging the history window is computationally costly and frequently yields limited gains due to substantial redundancy and task-agnostic, inflexible sampling of observations. We propose CondenseVLA, a multi-frame VLA policy with a learnable decoder-side history condensation module for using short-term visual context in action decoding. Given a sequence of cached per-frame features from the VLM backbone, CondenseVLA distills observation history into a fixed set of slot tokens using learnable queries and multi-head cross-attention. Simple auxiliary regularization penalizes overlap among slot attention maps and decorrelates slot embeddings, encouraging complementary attention and reducing slot redundancy. Across the evaluated tasks, CondenseVLA improves Simpler-WidowX success from 58.7% to 69.8% (+11.1 percentage points), achieves 97.2% on LIBERO, and reaches 52.8% on real-world tasks. It achieves 21.6% success on memory-dependent RMBench tasks. These results support selective condensation of bounded recent visual evidence for action decoding, while harder multi-observation tasks expose the limits of a finite history window. Project page: this https URL
[CV-83] MIRA: A Multimodal Illustrated Deep Research Agent for Controllable Reports
链接: https://arxiv.org/abs/2610.25926
作者: Hongrui Wu,Kaiwen Tuo,Congcong Wang,Gen Li,Shuai Dong,Xinlei Yu,Yuqi Xu,Haowen Hou,Zelin Peng,Jiaqi Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 33 pages, 8 figures, 9 tables
Abstract:Deep research agents are increasingly capable of searching the open web and synthesizing evidence, yet they remain poorly suited to controllable report generation in multimodal deep research. This task requires agents to handle interleaved textual and visual inputs, gather multimodal evidence through multi-step web search, and deliver comprehensive illustrated reports under user-specified requirements. Existing systems face two limitations. During reasoning, they often underuse images discovered along the search trajectory; during generation, they tend to produce fixed-form answers with limited control over report granularity, multimodal composition, and evidence grounding. We introduce MIRA, a Multimodal Illustrated Deep Research Agent for controllable reports. MIRA integrates textual and visual evidence throughout research and writing. At its core, MIRA-Harness translates user requirements into generation contracts and connects four stages: evidence collection, report submission, feedback-driven revision, and final delivery. Building on this harness, MIRA-Data constructs high-quality multi-hop multimodal research trajectories, and we post-train foundation models to operate within the workflow. Using Qwen3.6-35B-A3B, MIRA achieves state-of-the-art performance among open agents on 7 representative, established multimodal-search and deep-research-agent benchmarks, while producing user-aligned illustrated reports with traceable textual and visual evidence.
[CV-84] SAPO: Self-Supervised Attention Policy Optimization for Improving Temporal Video Grounding in Vision-Language Models
链接: https://arxiv.org/abs/2610.25923
作者: Zixuan Chen,Depeng Wang,Wei Chen,Ke Xu,Ya Guo,Huijia Zhu,Tanfeng Sun,Xinghao Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video vision-language models (VLMs) have strong semantic understanding, yet remain unreliable at precise temporal grounding. Existing gains from explicitly aligning video frames with absolute time suggest that VLMs may already encode accurate temporal locations internally but lose precision when expressing them as timestamps. Our analysis across four VLMs and two grounding datasets confirms this hypothesis: attention intervals outperform predicted intervals on approximately 66% of valid examples. We term this discrepancy the temporal generation gap. Because precise temporal annotations are expensive, this finding motivates closing the gap to improve temporal grounding using only unlabeled video-query pairs. We propose Self-Supervised Attention Policy Optimization (SAPO), which uses internal temporal attention as self-supervision for Group Relative Policy Optimization and applies a ground-truth-free criterion to select high-value samples. SAPO consistently improves temporal grounding across multiple datasets and models spanning weak to strong initial grounding ability. Relative mIoU gains reach 33.7% on Qwen2.5-VL-3B and 5.4% on the strong Time-R1-7B. The improvements transfer across datasets while preserving general video understanding.
[CV-85] LEAP: Accelerating ODE-based Diffusion Sampling with Optimal Linear Leaps
链接: https://arxiv.org/abs/2610.25918
作者: Haowei Zheng,Zhenyu Zhou,Jiawei Chen,Can Wang,Defang Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Diffusion models primarily rely on general-purpose ODE-based numerical methods for training-free accelerated sampling, but performance degrades with coarse time discretization. In this paper, we propose optimal linear leaps (LEAP), a specialized method grounded in theoretical and empirical investigations of sampling trajectory geometry. Built upon low-dimensionality and cross-trajectory consistency, LEAP formulates a supervised least-squares optimization problem within the local trajectory subspace, achieving a precise approximation of future states without model evaluations. Extensive experiments demonstrate that LEAP outperforms previous methods across diverse modalities, including text-to-image (Stable Diffusion 3.5, FLUX.1-dev) and image-to-3D (TRELLIS.2) generation. Furthermore, LEAP can be seamlessly integrated into existing solver-based methods, offering a plug-and-play enhancement for modern diffusion models. Code is available at this https URL.
[CV-86] Multi-scale Fusion of Fast Vision Mamba and Foundation Model Features for Patch-Level Gleason Grading in Computational Pathology MICCAI2026
链接: https://arxiv.org/abs/2610.25904
作者: Zihe Zhang,Weimiao Yu,Piumi Don Simonge
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 4 figures and 4 tables (including supplementary), Accepted to COMPAYL workshop MICCAI 2026, Comments are welcome!
Abstract:Automated Gleason grading of prostate whole slide images (WSIs) enables faster and more consistent cancer risk stratification. However, it remains a challenging task due to the subtle morphological differences between grades, the severe class imbalance arising from rare patterns, and the diagnostic features that span multiple spatial scales. Pathology foundation models (FMs) extract strong patch representations but often operate at a single-scale, which limits spatial context and disproportionately affects classes that depend on broader context. The challenge of fusing multiple scales and complementary models, with minimal data and computational resources, remains unresolved. We propose a multi-scale classifier built on state-space models. Our method combines two complementary feature sources: a frozen foundation model, and a lightweight FastVim, pre-trained on ImageNet and distilled from the foundation model, that reads the raw image crops. At each scale, a Mamba model encodes the features from both sources with linear-time spatial context, and a jointly trained head fuses the encoded representations into one prediction. On the PANDA dataset, our method reaches macro F1 of 0.894 and Gleason 5 F1 of 0.802 under 5-fold cross-validation, a 12% and 21% relative improvement over the best single-scale model. A controlled study over 5 foundation models, 5 base classifiers and 3 scales confirms that the gain holds across encoders, with larger improvements observed at smaller scales.
[CV-87] Out-of-Distribution Detection in Continual Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.25901
作者: Nimeshika Udayangani,Flora D. Salim,Christopher Leckie,Sarah Erfani
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the 40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:Out-of-distribution detection in continual learning (OOD-CL) is highly challenging, as it requires (i) learning a sequence of tasks without catastrophic forgetting, and (ii) identifying unknown samples (OOD) without access to known (in-distribution (ID)) data from past tasks, where standard confidence-based OOD scores tend to degrade. Existing methods largely rely on a rehearsal buffer of past data or assume closed-world settings, limiting their scalability and reliability in real-world scenarios. In this paper, we propose a novel OOD detection framework for rehearsal-free continual learning, inspired by Kolmogorov-Arnold Networks (KAN), which mitigate catastrophic forgetting via their inherent local neuroplasticity. To better leverage KAN in OOD-CL, we introduce a task-adaptive classifier architecture, TA-KAN, which mitigates recency bias and enhances ID–OOD separability through task-specific control of activation locality. Furthermore, we propose a geometry-guided OOD score that complements TA-KAN classifier confidence without relying on past ID data. Our method significantly improves OOD detection while maintaining strong ID classification performance, achieving state-of-the-art results in OOD-CL.
[CV-88] Rethinking Sparse-View 3D Gaussian Splatting: A Few-Shot Learning Perspective on Overfitting Prevention SIGGRAPH
链接: https://arxiv.org/abs/2610.25887
作者: Lejia Ye,Bin Wu,Jingjing Gao,Yiyang Wu,Xiaohu Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 5 pages, 5 figures. First-place solution in the sparse-view track of the SIGGRAPH Asia 2025 Volumetric Video Challenge
Abstract:Sparse-view novel view synthesis (NVS) is essential for AR/VR and real-time communication applications, yet remains challenging when reconstructing 3D scenes from limited viewpoints. While 3D Gaussian Splatting (3DGS) excels in dense-view scenarios, it suffers from overfitting and geometry degradation under sparse inputs. Current approaches often tackle these problems from a local perspective, addressing either data quality or training regularization individually, lacking a unified and interpretable framework for systematic improvement. We introduce a new perspective that reframes sparse-view 3DGS from a few-shot learning viewpoint, treating each Gaussian primitive as a neuron and leveraging deep learning regularization techniques. Under this unified framework, our approach generates more effective supervision from scarce observations through two strategies: (1) data-related supervision enrichment via optical flow-guided pseudo-view synthesis and joint camera-scene optimization; and (2) training-time regularization via motion-aware dropout and early stopping, which suppress overfitting while preserving dynamic-scene stability. Extensive experiments on the SIGGRAPH Asia 2025 volumetric video challenge sparse-view dataset demonstrate significant improvements over state-of-the-art optimization-based and feed-forward baselines. Our system won first place in the sparse-view track of the SIGGRAPH Asia 2025 challenge. Comments: 5 pages, 5 figures. First-place solution in the sparse-view track of the SIGGRAPH Asia 2025 Volumetric Video Challenge Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.25887 [cs.CV] (or arXiv:2610.25887v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.25887 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-89] Vision-Language-Delta-Action: Learning Compact Deltas as Actionable Transition Interfaces for Robotic Manipulation
链接: https://arxiv.org/abs/2610.25874
作者: Lingsen Zhang,Wei Li,Jie He,Henghua Zhang,Kai Mo,Shiyong Li,Yanpeng Wang,Rui Shao
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 37 pages, including appendix. Code framework: this https URL
Abstract:Robotic manipulation is not merely observation-to-action generation, but the realization of task-specified state changes in the physical world. Existing vision-language-action (VLA) methods directly learn observation-to-action mappings, imposing a substantial learning burden due to the gap between semantic understanding and action execution. Although world-action models (WAMs) improve control through future prediction, modeling complete future states wastes capacity on static appearance and action-irrelevant changes. We propose Vision-Language-Delta-Action (VLDA), a paradigm that reformulates full future prediction as compact actionable transition modeling. Its core is a compact Delta variable that captures latent world-state changes, serving as an interface between vision-language understanding and action generation. VLDA learns Delta through a unified objective composed of the Local Transition Filter (LT-Filter) and Global Transition Anchor (GT-Anchor). LT-Filter captures task-relevant manipulation-centric changes while suppressing static reconstruction biases. GT-Anchor provides coarse future-level regularization to preserve the overall direction of state changes. Delta-Driven Action Refinement (DDAR) further grounds Delta in the action manifold and injects it as a structured seed to guide action denoising. We implement VLDA within LoongForge, an open-source training framework for large-scale foundation and embodied models, using its modular and distributed infrastructure for unified two-stage optimization. Experiments show that VLDA improves long-horizon consistency, robustness, and generalization, validating actionable Delta as an alternative to direct action prediction and full future-state modeling.
[CV-90] ActionRoPE: action as coordinates for spatially consistent control in planar-camera game world models
链接: https://arxiv.org/abs/2610.25861
作者: Zhiyang Deng,Yuzhe Wu,Zijun Lin,Yeying Jin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Game world models have made rapid progress toward real-time interactive generation. Yet action control is still predominantly implemented as feature conditioning, leaving the model to learn from training data how each action should move the visual world. This formulation is poorly matched to a broad class of planar-camera worlds, such as top-down and isometric games. In these worlds, player motion induces a geometrically well-defined planar camera displacement. We observe that the key to spatially consistent control in such worlds is to represent the effect of each action directly in world coordinates. To this end, we introduce ActionRoPE, a parameter-free action interface that converts action-driven camera displacement into world-coordinate rotary positions. As a result, tokens corresponding to the same world location retain the same positional address across time. These world coordinates further expose a spatial asymmetry in generation: regions covered by the initial observation are geometrically known, while newly revealed regions remain to be synthesized. We therefore propose Coordinate-Conditioned Noise Scheduling (CNS), which assigns token-wise denoising schedules according to this known-new structure. It allows observed regions to stabilize earlier, while new content is generated with reference to them. Compared with four representative action-conditioning mechanisms, our approach follows the commanded motion more faithfully and reduces the revisit drift by 66% over the strongest baseline. The same interface transfers across video backbones from 1.3B to 33B parameters and to out-of-distribution scenes. Project Page: this https URL
[CV-91] Auditing Semantic and Relational Reliability in Open-Vocabulary Human-Object Interaction Evaluation WACV2027
链接: https://arxiv.org/abs/2610.25849
作者: S M Asif Hossain,Ruksat Khan Shayoni,Md Kishor Morol
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Under review at WACV 2027
Abstract:Open-vocabulary human-object interaction detectors describe interactions in free text, and this creates a problem for evaluation. Category scoring rejects valid paraphrases, while semantic similarity scoring can reward predictions that name the wrong action or reverse the roles of the participants. We present a diagnostic protocol that holds every prediction fixed and scores it under five views, namely official category evaluation, an identity control, lexical normalization, graded semantic credit, and strict propositional equivalence. Comparing these views separates the effect of the matching engine from the effect of semantic credit. The protocol pairs controlled transformations that preserve or break meaning with a blinded audit of natural predictions, and we derive consistency checks that link mean credit, invalid acceptance, and transformation counts. Applied to six systems on HICO-DET, the views disagree on which system leads. GEN-VLKT leads under official evaluation, SL-HOI leads under SHOE, and GEN-VLKT leads again under strict equivalence. Semantic gains range from 1.6 to 20.3 mAP, and AgentHOI keeps less than half of its gain once equivalence is enforced. We recommend that semantic scores be reported together with relational validity and paired uncertainty, so that recovered credit is not read as better interaction understanding.
[CV-92] SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
链接: https://arxiv.org/abs/2610.25844
作者: Yanjin He,Cuifeng Shen,Wanzhou Lei,Maohua Li,Tao Lan,Kan Liu,Hanlin Tang,Qing Qu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 38 pages, including appendices
Abstract:Training high-quality diffusion models is computationally expensive, and inference-time guidance often adds further sampling cost. In this work, we introduce Swift, a simple yet effective self-guided training method that uses a frozen weak denoiser to improve a stronger model without adding inference overhead. For each noisy input, Swift extrapolates the clean regression target away from the weak denoiser’s prediction, amplifying components that remain underfit. The weak guide can be an early training checkpoint, requires no backpropagation during main-model training, and is discarded afterward. The resulting model retains the original sampler and supports both conditional and unconditional generation without label dropout or an unconditional branch. We explain this mechanism through a spectral analysis of finite-time population gradient flow in an RKHS. With an early-stopped weak guide, we prove that a nonempty range of extrapolation strengths yields strictly lower posterior-mean error than matched standard training, despite shifting the regression target away from the posterior mean. Experiments on CIFAR-10 and ImageNet-256 demonstrate improvements across model scales and representations. In one ImageNet-256 configuration, Swift surpasses the baseline’s best FID using roughly half the total training compute, including guide preparation and training overhead. On RAEv2, it achieves FID 1.15 without inference-time guidance and 0.99 with guidance. These results establish target extrapolation as an effective approach to weak-to-strong diffusion training with efficient single-model inference.
[CV-93] Hihaze: One-Step Diffusion for High-Resolution Real-World Image Dehazing
链接: https://arxiv.org/abs/2610.25832
作者: Zheng Chen,Yuchen Huang,Xiaoyang Liu,Jue Gong,Linghe Kong,Yulun Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code is available at: this https URL
Abstract:Diffusion models have recently achieved encouraging results in real-world image dehazing. However, restoring high-resolution images remains challenging due to spatially non-uniform haze and inconsistent predictions across local patches. To address these issues, we propose Hihaze, a one-step diffusion framework for high-resolution real-world dehazing. Hihaze adopts two haze-aware modulation modules: attentive skip compensation (ASC) and a spatial degradation adapter (SDA). ASC selectively transfers reliable encoder information to the variational autoencoder (VAE) decoder, preserving structural details while suppressing haze-degraded features. SDA extracts multi-scale spatial degradation representations and injects them into the U-Net to guide restoration according to local haze variations. Furthermore, we design a training-free global-guided patch fusion (GPF) method. It uses a global prediction to align patch appearance and downweight unreliable predictions, followed by structure-guided detail compensation. Notably, this training-free design also makes GPF a plug-and-play module for other dehazing models. Experiments across four real-world benchmarks show that Hihaze surpasses state-of-the-art approaches in both visual quality and inference speed. Specifically, it delivers PSNR improvements of 3.62~dB on Dense-Haze and 3.69~dB on NH-HAZE compared with the strongest baselines. Code: this https URL.
[CV-94] Do We Need Learned Targets? Classical Gradients as Supervision for 2D Gaussian Initialization
链接: https://arxiv.org/abs/2610.25809
作者: Kyungjune Baek
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, 8 tables, preprint
Abstract:Learning-based Gaussian image initializers can require costly supervision obtained by per-image optimization. We investigate whether the Pseudo-PPM targets used by Instant GaussianImage can be replaced by classical image gradients. Our framework uses normalized Sobel maps in two tracks. A training-free analytic initializer combines probability boosting, parallel grid-based sampling, and geometric processing, producing initial parameters in 0.32 seconds on average. A neural initializer is trained on DIV2K with online Sobel supervision, which removes offline target optimization and allows training to scale to 10,000 LSDIR images without additional preprocessing. Both tracks achieve competitive long-budget reconstruction on Kodak and DIV2K. Under domain shift, the DIV2K-trained neural initializer surpasses Instant-GI on Manga109 from 500 fitting iterations onward (by 1.40 dB at 5K iterations), and training on LSDIR increases this margin to 1.85 dB. On Kvasir-SEG, the training-free analytic initializer exceeds Instant-GI at every reported budget. A controlled pipeline-time comparison yields an estimated 4.9\times reduction for DIV2K. These results establish inexpensive analytic targets as a practical alternative to optimization-derived supervision for scalable Gaussian initialization.
[CV-95] UniJev: Separating Accuracy from Calibration in Trimodal System One Decision Models
链接: https://arxiv.org/abs/2610.25784
作者: Jianing Wang,Youwen Jian,Derong Wang,Jinyang Yu,Qingyi Si,Chuanyu Qin,Yanghua Xiao,Hongcheng Guo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 21pages, 8figures
Abstract:System One models produce probabilistic judgments that software can consume directly, making reliable confidence as important as answer accuracy. Yet joint training across text, images, and video remains underexplored, and the contributions of supervised fine-tuning (SFT) and reinforcement learning for calibrated decisions (RLCD) are difficult to separate. We introduce UniJev, a trimodal framework that scores candidate answers with a shared head under joint supervised training across text, images, and video. Supervised decision models improve accuracy over their backbone readouts across all three modalities, while the existing RLCD algorithms change accuracy little and have inconsistent effects on visual calibration across recipes. These findings motivate a separate supervised objective for confidence calibration. We introduce Cora (Confidence–Outcome Reliability Alignment), which groups current predictions by confidence and optimizes each prediction’s confidence toward the empirical accuracy of its bin. At both 2B and 9B, Cora lowers image and video expected calibration error (ECE) while preserving accuracy, without a reinforcement stage. We release our code at this https URL.
[CV-96] See the Future Classify the Present: Stochastic Latent World Modeling for Hyperspectral Images
链接: https://arxiv.org/abs/2610.25781
作者: Muhammad Ahmad,Gemine Vivone,Ronny Hänsch,Ahmed Sohaib,Manuel Mazzara,Danfeng Hong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 11 figures
Abstract:Hyperspectral image classification (HSIC) relies on static spatial-spectral representations learned from labeled samples, limiting its ability to model structured dependencies in the latent representation space. To address this limitation, this work proposes Stochastic Latent World Modeling (LWM), a trajectory-based representation-learning framework that enriches classification through stochastic latent-state prediction. Given a hyperspectral patch, LWM tokenizes and encodes spectral and spatial information separately using dedicated transformer branches, then fuses their outputs into a unified representation. This representation is further organized into an environment state that captures center, neighborhood, spectral, and global contextual information. A stochastic latent transition module then predicts multiple future latent states through Gaussian parameterization and reparameterization, with recurrent refinement used to model dependencies across successive states. The resulting latent trajectory is adaptively aggregated by a rollout planner to construct a planned representation, which is integrated with the base representation for final classification. By extending static spatial-spectral encoding with stochastic multi-step latent-state modeling, LWM enables the classifier to exploit structured dependencies in the representation space and provides richer features for HSIC. Across five benchmark datasets, LWM achieves overall accuracies ranging from 89.21% to 99.52%, demonstrating consistent classification performance under limited-label settings.
[CV-97] VHA: VQA-based Hallucination Assessment for Generative Image Super-Resolution
链接: https://arxiv.org/abs/2610.25770
作者: Donghun Ryou,Inju Ha,Jaeyoo Park,Bohyung Han
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Generative image super-resolution produces visually realistic high-resolution images, but may also introduce hallucinations by adding unsupported content or altering existing details. Because existing super-resolution metrics fail to measure whether generated details remain grounded in the reference image, we present a hallucination assessment technique based on bidirectional visual question answering (VQA), which we term VQA-based Hallucination Assessment (VHA). The proposed approach uses a vision-language model (VLM) to determine whether the semantic content and low-level visual evidence in a super-resolved image are supported by its ground-truth counterpart. To this end, it constructs question–answer pairs from each image and evaluates them on both images. Discrepancies between the resulting scores in both directions reveal missing or altered ground-truth information and unsupported additions. However, obtaining these scores requires computationally expensive VLM inference. We therefore develop VHA+, a learned, lightweight model that directly predicts VHA scores from ground-truth and super-resolved image pairs. We construct a hallucination benchmark, VHA-bench, through a human study and show that both VHA and VHA+ correlate highly with human judgments of hallucination. The code, benchmark, and model weights are available at this https URL
[CV-98] Functional Cross-layer Experts Routing for One-step Diffusion Distillation
链接: https://arxiv.org/abs/2610.25742
作者: Zile Huang,Ser-nam Lim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:One-step distillation compresses diffusion trajectories into a single forward evaluation, yet how internal capacity should be structured within that remaining pass remains largely unexplored. In this setting, identical parameters must simultaneously preserve general generative priors and resolve prompt-specific details without iterative refinement. We formalize this challenge as functional allocation. Our analysis reveals that decoupling representations is beneficial only when expert interference outweighs shared transfer, favoring an always-active shared backbone augmented with conditionally routed residuals over standard isolated MoE architectures. We propose DeMoE, which pairs a persistent shared expert with conditionally routed expert slices across image FFNs, trained under a one-step distribution-matching objective with executed-route supervision. On one-step FLUX.2-4B generation, DeMoE achieves a GenEval score of 0.8220 while activating only 80.5% of the image-FFN width, substantially outperforming dense one-step baselines, matching the 50-step teacher overall, and surpassing it on complex composition and counting. Across COCO-10K, DeMoE establishes state-of-the-art fidelity and diversity among single-step models, demonstrating that structured internal computation reallocation provides a superior compression paradigm for diffusion models. Code is available at this https URL.
[CV-99] hink Densely Act Sparsely: Latent Expert Cognitive Chains for Vision-Language-Action Autonomous Driving NEURIPS2026
链接: https://arxiv.org/abs/2610.25734
作者: Jie Wang,Guang Li,Zhijian Huang,Jinlong Li,Chenxu Dang,Hangjun Ye,Yahong Han,Long Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: NeurIPS 2026
Abstract:Autonomous driving fundamentally demands dense understanding of visual, semantic, and geometric information, yet existing Vision-Language-Action (VLA) models are typically trained with sparse supervision such as language instructions or trajectory signals. This direct mapping from dense visual observations to sparse representations forces the model to compress or even discard critical scene information essential for safe driving. We argue that this limitation does not stem from insufficient model capacity, but from the lack of explicit mechanisms that encourage the formation of dense world representations in the latent space. In this spirit, we propose a novel paradigm for driving intelligence, Think Densely, Act Sparsely, and instantiate it with an efficient and interpretable framework, LECDrive (Latent Expert Cognitive Chains for Vision-Language-Action autonomous Driving). The core idea is to mimic the hierarchical perception process of humans by embedding a progressive chain of latent expert cognition within the VLA model. Concretely, we inject a compact set of task-specific latent tokens as carriers, enabling the model to internalize dense knowledge from foundational vision models, including visual representations (DINOv3), semantic structures (SAM3), and spatial geometry (DepthAnything3), while maintaining efficient inference. This interdependent cognitive chain follows a curriculum learning principle, where supervision is progressively structured from representation to semantics and geometry. Extensive experiments demonstrate that LECDrive consistently improves the performance of base VLMs across multiple autonomous driving benchmarks, covering key tasks such as object perception, state prediction, and trajectory planning. These findings suggest that constructing dense world models through structured latent expert chains is a promising direction for VLA-based autonomous driving.
[CV-100] LEVI: Benchmarking Visual Identity Fidelity with Verifiable Low-Entropy Constructs
链接: https://arxiv.org/abs/2610.25716
作者: Xiao Ma,Zhonghao Zhan,Hamed Haddadi,Qi Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 24 pages, including supplementary material
Abstract:Modern generative models can produce visually plausible images while violating exact identity constraints. Evaluating such failures requires distinguishing violations of the requested identity from permissible variation in appearance. We introduce Low-Entropy Visual Identity (LEVI), which characterizes visual identities determined by compact facts and explicit rules. The low-entropy requirement concerns identity uncertainty, while appearance may remain diverse. We formulate identity fidelity in terms of the probability that a conditional image distribution assigns to the identities permitted by the prompt. Visual Identity Spaces (VIS) formalize this distinction by specifying admissible identity values and which different realizations express the same identity. Building on this formulation, we develop Verifiable Visual Grammar (VVG), a declarative language whose typed primitives and compositional constraints describe identity-bearing structures. Using this grammar, we build a benchmark for image generation. It assesses each output against the task’s identity requirements, drawing evidence from the final pixels. The curated benchmark contains 1,375 tasks spanning 120 base identity spaces across six domains. Among ten evaluated model configurations, the highest whole-task pass rate is 58.69%. These results reveal a persistent gap between visually plausible images and faithful realization of their specified identities. Project page (forthcoming): this https URL.
[CV-101] Multi-View Structural Representations for Physically-Consistent TeX Generation
链接: https://arxiv.org/abs/2610.25695
作者: Daoheng Li,Jianjun Qian*,Mengkai Yan,Jian Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Infrared and visible images provide complementary cues for low-light image generation, and physics-based methods can better exploit this complementarity. However, existing physics-based methods still rely heavily on paired data. To address these limitations, we propose Multi-View Routing Mixture-of-Experts (MVR-MoE), a framework for physically consistent TeX generation that learns multi-view structural representations. MVR-MoE mines intrinsic structural cues from infrared and visible images to organize them into multi-view representations, while the proposed multi-view hyper connections expand the feature space and preserve both shared evidence and modality-specific details. A hierarchical routing mechanism further regulates feature utilization at both the structural-view level and the expert level, enabling adaptive exploitation of heterogeneous cues. The predicted temperature, emissivity, and texture maps are constrained by the Stefan-Boltzmann law to ensure physically consistent. Extensive experiments on challenging datasets demonstrate that MVR-MoE achieves superior robustness and discriminability while remaining lightweight.
[CV-102] BitFlat: Error-Localized Bitstream Design for Robust Diffusion Image Compression
链接: https://arxiv.org/abs/2610.25682
作者: Xinfeng Deng,Li Zhou,Haitao Chen,Shengjie Wu,Lingjin Kong,Nan Li,Haitao Zhao,Dongtang Ma
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Zero-shot diffusion image compression based on reverse channel coding achieves high perceptual quality at low bitrates without retraining, yet its bitstream remains vulnerable to bit flips. Prior work has shown that a joint combinatorial representation can amplify a single bit flip into substitutions of multiple atoms and that indexing each atom independently confines this error propagation. However, independent indexing reduces the number of atoms that fit within the bit budget and gives the decoder no indication of which atoms are corrupted. To address these issues, we propose BitFlat, a training-free error-localizing bitstream framework with two components. Blockwise Locality Alignment (BLA) decomposes the globally coupled atom selection at each diffusion step into independent selection units, each describing one signed atom from a dedicated codebook with a fixed-width field, so that any bit error affects only the unit in which it occurs. Reliability-Aware Diffusion Control (RADC) uses the bits saved by BLA to impose structured valid-codeword constraints within each field and converts the decoding state of each field into a binary reliability gate, which turns detected corrupted atoms into erasures at known positions and excludes them from noise synthesis. BitFlat requires neither target residuals nor additional network forward passes and leaves the diffusion model unchanged. Experiments on Kodak, DIV2K, and CLIC2020 show that BitFlat outperforms conventional, learned, and diffusion-based codecs at bit error rates from 10^-4 to 10^-2 and, even with 26.7% fewer bits, surpasses Robust Turbo-DDCM.
[CV-103] RxGS: Relightable Gaussian Stylization via BRDF Decomposition
链接: https://arxiv.org/abs/2610.25681
作者: Wanrong Hong,Liu Jiang,Mukesh Prasad,Jun Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reference-based 3D scene stylization transfers the appearance of a reference image onto a reconstructed 3D scene while preserving scene geometry. Recent methods perform this transfer by optimizing the appearance representation of neural scene models including Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). These methods optimize an RGB appearance representation, in which a single per-point colour combines the effects of material reflectance, environment illumination, together with the shading it produces, and the viewing direction. When stylization modifies this combined colour, the illumination effects present during stylization are baked into the stored appearance, so the stylized scene cannot be rendered under different illumination. To address this limitation, we present RxGS, a relightable 3D scene stylization framework built on a Bidirectional Reflectance Distribution Function (BRDF)-aware 3D Gaussian representation. Rather than modifying a per-point RGB colour, RxGS optimizes only the diffuse albedo field recovered by inverse rendering, while leaving roughness, metallicity, surface normals, and the estimated environment lighting unchanged. As a result, the stylized scene keeps all the components a physically-based renderer needs other than the edited albedo, and can therefore still be relit. RxGS optimizes the albedo field using a nearest-neighbour feature matching loss together with a spatial smoothness prior over Gaussian centres. We evaluate the method using a Geometry Consistency Cost (GCC) based on optical-flow warping error. RxGS achieves the lowest GCC on all evaluated NeRF-Synthetic scenes and remains competitive on real-captured scenes, while additionally supporting relightable stylization, which existing reference-based 3D scene stylization methods do not provide.
[CV-104] Frechet Forcing: Training Long-Horizon Streaming World Models from Short Clips
链接: https://arxiv.org/abs/2610.25666
作者: Yian Zhao,Yu Yan,Hongcan Guo,Ruochong Zheng,Jian Zhang,Jie Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Streaming video world models generate the future chunk by chunk, conditioned on their own past and a stream of actions. Trained on short clips, they drift once a rollout exceeds the training horizon, as prediction errors re-enter the context and compound. A common remedy is to train the model on its own rollouts via distribution matching distillation (DMD). For action-conditioned world models, the teacher becomes a bottleneck: an external video model is agnostic to the actions, whereas the model’s own pretrained network drifts in the same manner as the student. We observe that a long rollout need not match a teacher, provided it remains a realistic continuation of its own scene that follows the commanded actions. Realism and scene fidelity can be measured against real-data statistics, which remain valid at any horizon, while action following is preserved by the model’s conditional generation loss. We propose Fréchet Forcing, which fine-tunes a world model together with its deployed multi-step sampler under three constraints. Generative anchoring applies the flow-matching loss after a self-generated history ending on real data. Context fidelity, a Fréchet distance (FD) on latent statistics of the free rollout centred on each clip’s own ground truth, keeps every rollout faithful to its scene. Reality matching, an FD to real frames in a frozen perceptual space, keeps the generated frames realistic. Because all three targets are derived from short real clips without comparing a rollout against a long recorded future, training requires no teacher, no online critic and no long videos. Fine-tuning MiniWorld-1B, pretrained on 253-frame clips, for at most 36 GPU-hours on 381-frame clips yields rollouts that remain stable beyond 4,000 frames and continue to follow their actions in both robot manipulation and camera-controlled exploration.
[CV-105] SpaRV: Sparse Radiance Voxels for Scalable Geometrically Consistent LiDAR-Visual Mapping and Rendering
链接: https://arxiv.org/abs/2610.25665
作者: Jianheng Liu,Yunfei Wan,Fangming Cheng,Xiangyu Hong,Zuhao Zou,Haotian Li,Longji Yin,Chunran Zheng,Jiarong Lin,Fu Zhang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robotic simulation, navigation, and inspection need photorealistic and geometrically accurate maps. Radiance fields achieve photorealism, but building them from robot data raises three challenges. A robot’s trajectory serves its task, not reconstruction, so images from few viewpoints leave the geometry under-constrained. Residual errors in camera poses and intrinsics blur the map. The map’s GPU memory grows with scene size, which limits its level of detail in large scenes. We present \textbfSpaRV, a \textbfSparse \textbfRadiance \textbfVoxel map rendered by differentiable ray marching, which lets irregular LiDAR scans supervise it alongside images. LiDAR depth and occupancy anchor the under-constrained geometry, and normal and smoothness regularizers keep it smooth between LiDAR points. To correct camera errors, SpaRV alternates map training with photometric optimization of the camera poses and focal lengths. To bound GPU memory in training and rendering, SpaRV holds one submap at a time at full resolution and the rest at coarser octree levels. Experiments show that SpaRV keeps the geometry consistent away from the captured trajectory and sharpens textures by optimizing the camera parameters. On one 24,GB GPU, SpaRV reconstructs a university campus captured in 63,582 images and simulates other cameras and LiDARs. Our code will be released at \urlthis https URL.
[CV-106] PoSTCache: From Forecasting to Postdictive Smoothing for Revisable Diffusion Caching
链接: https://arxiv.org/abs/2610.25664
作者: Liwei Lan,Hui Chen,Jungong Han,Guiguang Ding
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 37 pages, 13 figures
Abstract:Diffusion and flow matching models achieve impressive visual generation quality but remain computationally expensive due to iterative sampling. Caching-based acceleration provides a practical training-free solution by reusing or predicting intermediate features. However, existing methods largely enforce a forward-only information flow: approximations produced for skipped model evaluations are immediately committed to the sampling trajectory and cannot benefit from feedback generated by later full-computation steps. We propose \textbfPoSTCache (\emphPostdictive Smoothing Trajectory Cache), a training-free and plug-and-play framework that reformulates diffusion caching as revisable trajectory inference. PoSTCache first uses an existing caching strategy to construct forward proposals for skipped steps. At a later full-computation step, the discrepancy between the refreshed prediction and its forward proposal yields a correction signal, which PoSTCache propagates backward to refine and reintegrate the preceding predictions. Viewing full-computation steps as sources of corrective feedback also changes how they should be selected. We therefore introduce the Posterior Variance Reduction (PVR) strategy, which estimates cross-timestep dependencies from offline denoising trajectories and selects anchors according to their expected reduction in surrogate trajectory uncertainty rather than local approximation error alone. Experiments on representative image and video generation models show that PoSTCache consistently improves the quality–efficiency trade-off over representative caching methods.
[CV-107] FluidGen: Physics-Aware Fluid Video Generation via Motion-Guided Post-Training
链接: https://arxiv.org/abs/2610.25660
作者: Yang Zheng,Zizhang Li,Boyang Deng,Jan Ackermann,Lei Shu,Yinghao Xu,Jiajun Wu,Gordon Wetzstein,Leonidas Guibas
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: Project page: this https URL
Abstract:Video generation models produce visually compelling motion but systematically fail on fluid phenomena, including smoke, fire, and explosion debris, where outputs lack coherent vortical structure and collapse to bland dynamics that violate the physics of real fluids. Classical solvers achieve high physical fidelity but require detailed 3D scenes and carefully tuned parameters, making them impractical for in-the-wild imagery and video generation. We address this gap by injecting a data-driven fluid prior into a pre-trained video model, trained on a mixture of high-fidelity Houdini simulation data with paired velocity-field ground truth and curated in-the-wild fluid video data for appearance diversity. The core of our approach is a dual stream architecture that predicts per-frame velocity fields alongside RGB, supervised by simulation ground truth. This auxiliary signal regularizes generation toward physically consistent motion and exposes an interpretable physical quantity that we further leverage as a reward for velocity-aware direct preference optimization. Our model produces substantially more realistic fluid dynamics sequences than strong video baselines and optimization-based methods. Project page: this https URL
[CV-108] InteractAgent : Agent ic Human Motion Interaction with Memory-Augmented Reflection
链接: https://arxiv.org/abs/2610.25649
作者: Tao Zhang,Senhe Zhang,Zeyu Zhang,Yujia Zhang,Dong Gong
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Long-horizon human-scene interaction requires coordinated spatial reasoning, physical feasibility, and the ability to recover from compounding execution errors. Text-driven motion models are largely scene-agnostic and lack closed-loop correction, while geometry-conditioned approaches depend on paired scene-motion training or explicit 3D preprocessing that limits generalization to novel layouts and open-vocabulary goals. We present InteractAgent, an agentic multimodal LLM system for long-horizon human-scene interaction that plans, executes, observes, and reflects in a closed loop, using a single MLLM as both planner and evaluator and a frozen MotionLLM as the executor. InteractAgent integrates Online Observation for step-wise recovery decisions (continue/redo/replan) and Memory-Augmented Reflection for multi-round plan refinement with adaptive stopping. On UniHSI, InteractAgent improves task success and reduces penetration compared with training-free baselines. Code: this https URL.
[CV-109] Negation without Confusion: Decomposing Bias and Recomposing Style in Text Embeddings for Negation-aware Image Generation
链接: https://arxiv.org/abs/2610.25645
作者: Sooyeon Park,Hemraj Singh,Kyunghyun Min,Youngjae Yu,Sung-Bae Cho
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Negation-aware image generation remains challenging for text-to-image diffusion models, which often struggle with negation due to implicit linguistic biases that are deeply entangled within text representations. As a result, bias-aligned semantics frequently dominate over intended negative constraints. To overcome this limitation, we propose a simple training-free method of Decompose and Recompose text embedding (Dero), which constructs a semantic-preserving subspace and a bias subspace using auxiliary prompts and performs semantic disentanglement based on bidirectional SVD. By decomposing entangled representations and selectively recomposing disentangled semantic components, Dero realigns text representations toward intended negation semantics while restoring residual stylistic features to preserve semantic fidelity and image quality. As a model-agnostic inference-time framework operating solely on text representations, Dero can be seamlessly applied to diverse diffusion architectures, including both U-Net and DiT based models. Extensive experiments on both object- and concept-level negation demonstrate consistent improvements in negation-aware image generation, achieving stronger text-image alignment, better semantic fidelity, and higher human preference scores.
[CV-110] Visual Memory and Textual Alignment for Cross-Modal Few-Shot Segmentation
链接: https://arxiv.org/abs/2610.25641
作者: Tianqi Wei,Zhi Chen,Xin Yu,Scott Chapman,Zi Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Few-shot segmentation (FSS) aims to segment target objects with limited annotated examples. Recent studies have begun to explore large foundation models to improve generalization for this task. In particular, SAM 3 provides strong visual representations and supports text-guided segmentation. However, its performance remains limited in specialized domains, where semantic prompts and visual representations are often poorly aligned. To address this challenge, we propose ViTA-SAM, a framework built upon SAM 3 that integrates visual memory and textual features through cross-modal alignment. Specifically, we construct a visual memory module that encodes support features together with their corresponding masks, providing support-aware visual guidance for query prediction. In parallel, we leverage the text encoder of SAM 3 to extract semantic features as complementary priors. To bridge the gap between semantic and visual representations, we introduce a cross-modal alignment module that conditions textual features on memory-enhanced visual representations, yielding more consistent and visually grounded predictions. In addition, we propose a pseudo-label refinement strategy that updates the memory with predicted query masks, enabling query-adaptive representation learning. Extensive experiments on multiple benchmarks show that ViTA-SAM achieves the best overall performance among the compared methods, outperforming the strongest competing method by over 5% in average mIoU.
[CV-111] ACorD: Adaptive Correspondence Discovery for Medical Image Registration
链接: https://arxiv.org/abs/2610.25638
作者: Arghavan Rezvani,Kun Han,Xiaohui Xie
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Medical image registration is a fundamental task for aligning anatomical structures across images. Exhaustively searching for point-wise correspondences is prohibitively expensive in both computation and memory, particularly for high-resolution 3D medical images. Most recent learning-based methods directly predict correspondences in a coarse-to-fine manner, but often propagate intermediate estimates across scales without explicitly weighting correspondence evidence, which can allow coarse-scale errors to persist during refinement. To address these challenges, we propose ACorD, an adaptive correspondence discovery framework designed to mitigate error propagation with an attention-guided process. ACorD first predicts multiple potential correspondence candidates through Correlation-guided Cross-scale Offset Prediction (CCOP), which combines correlation maps as spatial matching priors with cross-scale features as global deformation context. Then, Multi-Head Deformable Aggregation (MHDA) aggregates these candidate correspondences using attention, producing a learned, non-uniform weighting over competing matches rather than treating them equally. The multi-head design further captures complementary deformation patterns, ensuring both global alignment and local precision. By integrating adaptive offset prediction with selective, attention-based aggregation of correspondence evidence, ACorD enables accurate deformation refinement. Extensive experiments on four public medical registration benchmarks demonstrate that adaptive correspondence discovery within a multi-scale refinement framework improves registration robustness across diverse anatomical structures.
[CV-112] NVD-SC: Diagnosing Scan Context Failures in Forests with Terrain-Normalized Vertical Distributions
链接: https://arxiv.org/abs/2610.25637
作者: Jineon Kim,Jhonghyun An
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 8 pages, 5 figures
Abstract:Long-term LiDAR place recognition in forests is difficult because repetitive vegetation and uneven terrain create severe structural aliasing. We diagnose Scan Context (SC) by separating candidate retrieval from fine matching: on the standardized TreeLoc++ Wild-Places protocol, a TNVD support key places a valid location in the top-25 set for approximately 99% of queries, while conventional SC matching on those candidates reaches only 61.84% mean R@1. With an identical raw-SC candidate set, terrain normalization improves max-height cells by only 1.21 points but vertical-distribution cells by 19.72 points. These observations motivate Terrain-Normalized Vertical Distribution Scan Context (TNVD-SC), which preserves the SC ring-sector topology while encoding terrain-relative vertical probability distributions. Under the standardized TreeLoc++ protocol, TNVD-SC reaches 91.83% mean R@1 and attains the highest R@1 mean and R@1 stability among the compared methods, while TreeLoc++ remains stronger in mean MR, MF1, and AUC. The unchanged method reaches 85.14% mean R@1 on the four-session CVTNet-style NCLT evaluation. A lossless count-coded implementation with exact hierarchical Jensen-Shannon search reduces descriptor payload by 3.86x and compute latency by 23.8% without changing the audited retrieval outputs.
[CV-113] Clean Forcing: Drift-Resistant Autoregressive Video Diffusion with a Frozen Base
链接: https://arxiv.org/abs/2610.25634
作者: Wenqing Wang,Joonghyuk Shin,Jonathan Tremblay,Chan Hee Song,Yun Fu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Project page: this https URL
Abstract:Autoregressive (AR) video diffusion models generate streaming video of arbitrary length by conditioning each new chunk of frames on those already generated. However, they suffer from exposure bias, where a model trained on clean context must continue from its own imperfect outputs, and errors compound until quality collapses within seconds (drift). Existing methods either require cluster-scale base retraining, apply training-free inference corrections with inconsistent results, or train frozen-base correctors for sampling and cache errors. We instead find that drift is largely deterministic and therefore learnable, as holding the noisy state fixed while varying only the history reveals a counterfactual velocity gap that is ~95% systematic across noise seeds. Based on this observation, we introduce Clean Forcing, which trains a 5.9M-parameter LoRA corrector on the frozen base model to regress this gap, with a closed-loop stage that adds DAgger-style aggregation and a drift-contraction objective to expose the corrector to its own rollouts. Because the clean histories can be the base model’s own single-shot generations, training needs no real videos, and the merged corrector adds no inference cost. In 50 s text-to-video generation, Clean Forcing reduces \Delta-drift from +11.3 to +1.65 without real video and to -0.16 with 40 real clips, improving on the best published result of +3.57 with over 150x less data. Extensive experiments demonstrate that Clean Forcing outperforms prior methods in aesthetic quality and data efficiency with reduced drift for long-horizon video generation.
[CV-114] Confidence-Guided Pseudo-Label Refinement for Unsupervised Cross-Domain Person Re-Identification
链接: https://arxiv.org/abs/2610.25633
作者: Sundas Iqbal,Danish Ali,Qing Tian,Yuhui Zheng,Jianping Gou,Weihua Ou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Unsupervised cross-domain person re-identification remains challenging because domain shifts generate noisy pseudo labels that are reinforced during iterative adaptation. Existing clustering-based methods often rely on limited reliability cues or global confidence criteria and may impose hard identity supervision on ambiguous target samples. To address these limitations, we propose CGPLR, a confidence-guided pseudo-label refinement framework for unsupervised domain-adaptive person re-identification. CGPLR estimates sample-level reliability by jointly modeling centroid compatibility, local neighborhood consistency, separation from the strongest competing cluster and temporal assignment stability. The resulting confidence distribution is used to derive cluster-adaptive upper and lower thresholds for a three-way selective decision. Reliable assignments are preserved, recoverable samples are reassigned only when an alternative cluster receives sufficient geometric and neighborhood support and highly ambiguous samples are deferred from hard identity supervision. The learning objective adapts to these decisions through confidence-weighted supervision for preserved samples, cautious corrected-label weighting for reassigned samples and label-free representation consistency for deferred samples. CGPLR achieves 76.4% Rank-1 accuracy and 47.6% mean average precision (mAP) on Market-1501 to MSMT17, and 95.1% Rank-1 accuracy and 87.7% mAP on MSMT17 to Market-1501, demonstrating effective reliability-aware selective adaptation. The code is available at this https URL.
[CV-115] Where Do the Gaussians Go? Foreground Attraction for Camera-Based Weakly Supervised 3D Occupancy
链接: https://arxiv.org/abs/2610.25631
作者: Jeagon Song,Jhonghyun An
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 7 pages, 3 figures
Abstract:Camera-based 3D semantic occupancy models can be trained without 3D labels by rendering their predictions into the images and supervising them with 2D pseudo labels. In sparse Gaussian representations trained this way, Gaussian positions are learned only through pixel-wise rendering losses, which we hypothesize give small foreground objects little weight. In trained GaussianFlowOcc models, the Gaussian density on foreground voxels is on average only 1.62 times that on background voxels. We propose a training-only foreground attraction loss that back-projects foreground pseudo-label pixels with pseudo depth and pulls nearby Gaussians toward these 3D points, favoring Gaussians that already predict the target class; the inference network is unchanged. On Occ3D-nuScenes, the loss raises this concentration to 2.40 over three seeds. Over three seeds, the mean foreground IoU increases from 6.29 to 6.96 and is higher for each seed, whereas the change in mean mIoU (13.53 to 13.82) is within the seed-to-seed variation of the baseline. Single-run ablations further indicate that pseudo-label targets outperform targets placed at ground-truth voxels and that acting on the Gaussian positions alone is not sufficient, and a single-seed test shows that the model degrades less when fewer Gaussians are kept at inference.
[CV-116] Recurrent Vision Transformers for Event-Based Spacecraft Pose Estimation Under Slow Relative Motions
链接: https://arxiv.org/abs/2610.25622
作者: Sofia McLeod,Mohsi Jawaid,Carl Seubert,Tat-Jun Chin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-based object pose estimation is vital for conducting spacecraft rendezvous and proximity operations (RPO). Recent research has demonstrated the robustness of event cameras against the harsh lighting conditions that often occur in orbit, thanks to the extremely high dynamic range of event sensors. However, event-based spacecraft pose estimation (EBSPE) suffers when the relative motion between the chaser and target is slow, since the event rate drops significantly, leading to data with low signal-to-noise ratio. Unfortunately, for safety reasons, real RPO trajectories often involve slow relative motions (e.g., down to 1 mm/s), and operating under low illumination exacerbates the problem. Addressing slow relative motions is thus crucial to raise the practicality of EBSPE. To this end, we propose using recurrent vision transformers (RVT) for EBSPE. The key idea is to exploit recurrent memory to capture longer-term information across sparse event streams, and to learn the temporal attention span from full trajectories rather than fixing the time window. We developed an RVT-based heatmap regressor to produce 2D-3D correspondences for pose estimation, which was validated on real event data collected under slow relative motions representative of RPO trajectories. Results indicate the superior pose estimation accuracy and stability of our RVT-based method, relative to previous per-event-frame-based or fixed-time-window EBSPE methods.
[CV-117] CheXFuse: Guided Differential Joint Attention Fusion for Robust Multimodal Chest X-Ray Classification in Data-Scarce Medical Settings ICDM
链接: https://arxiv.org/abs/2610.25617
作者: Muntaqim Ahmed Raju,Priyanka Siddappa,Md Shifat Haider Al Amin,Ruizhe Ma,Md Ripon Ahammed
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: IEEE ICDM DMBIH 2026
Abstract:A central challenge in multimodal learning is how to fuse modalities that differ substantially in predictive strength. When one modality provides a much stronger representation than another, indiscriminate cross-modal interaction can introduce weak, redundant, or unstable signals, limiting or even degrading performance relative to the strongest unimodal representation. To address this problem, we propose CheXFuse, a multimodal fusion framework based on Guided Differential Joint Attention (GuDJA). CheXFuse combines image and text token representations within a joint multimodal sequence, where learned token guidance emphasizes informative representations and differential attention reduces less useful attention responses. We evaluate CheXFuse on the Indiana University (IU) Chest X-ray dataset, using binary (normal vs. abnormal) classification as the downstream task. Experiments are conducted under both original-report and masked-report settings, with the latter serving as a controlled stress test under reduced textual informativeness. CheXFuse with GuDJA achieves the strongest mean overall performance across the evaluated baselines and ablations, reaching 95.73% accuracy, 95.79% F1-score, 96.54% recall, and 95.28% sensitivity at 95% specificity (Sens@95Spec) with original reports, and 95.58%, 95.60%, 95.60%, and 94.34%, respectively, under masked reports. These results suggest that GuDJA provides a promising fusion strategy under asymmetric or partially reduced modality informativeness.
[CV-118] ViMoLa: Discrete Motion Tokens for Language-Guided Video Generation
链接: https://arxiv.org/abs/2610.25595
作者: Bishoy Galoaa,Sarah Ostadabbas
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Language-guided video generation often captures scene content while missing the requested motion. We present ViMoLa (Video Motion Language), which predicts a motion prior from language and injects it into a frozen image-to-video backbone. A spatial vector-quantized autoencoder compresses dense trajectories into 64 tokens from a 512-entry codebook; a 28M-parameter transformer predicts the tokens from text; and zero-gated cross-attention adapters condition every frozen Wan~2.2 DiT block through forward hooks. Trained on 462K videos from Something-Something~v2 and Kinetics-400, ViMoLa raises directional accuracy from 0.287 to 0.442 (chance 0.250), with similar measured generative quality. Across-prompt shuffled codes achieve 0.283, supporting the importance of prompt-matched token content. A low-rank adapter reaches 0.428 at 1.45% additional forward-pass FLOPs, and reusing the tokenizer and language model with newly trained CogVideoX adapters raises accuracy from 0.301 to 0.398. Control remains imperfect: oracle codes reach 0.531, compound-motion gains are smaller, and the adapter does not explicitly preserve token-to-patch assignments.
[CV-119] Finite Interaction Lens: Structured Precision Sharing for Vision-Language Models
链接: https://arxiv.org/abs/2610.25577
作者: Nozomu Fujisawa
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 32 pages
Abstract:Identifying which visual interventions interact in a vision-language model requires comparing their joint and separate effects. The resulting four-output difference can be small enough for rounding error to obscure it. We introduce the Finite Interaction Lens (FIL), a mixed-precision method for evaluating finite intervention families. At each selected linear projection, FIL decomposes each candidate’s input into a native state, two single-intervention components, and a full-dimensional mixed residual. It projects the single-intervention components at higher precision and reuses the same numerical contributions in the joint candidates, while retaining every mixed residual and nonlinear trajectory. We derive a local cancellation identity and a conditional bound on error propagation through a smooth nonlinearity. In one MLP of Qwen3-VL-4B-Instruct, FIL reduces final interaction MSE by 94.4% on 16 synthetic families and by 99.85% and 99.35% in two separate four-image COCO studies, relative to a control that has the same nominal BF16-product budget and higher-precision single interventions but projects each joint candidate independently. One image accounts for most of the absolute gain in each COCO study. Higher-accuracy multi-product BF16 arithmetic on all candidates gives lower aggregate error at similar total query times. Single-product BF16 already recovers the reference choices in both candidate-selection tasks we tested. On these development inputs, FIL improves interaction fidelity at a fixed nominal precision budget. Its practical value depends on a task that needs the extra fidelity.
[CV-120] DiffGround: Semantic Difference-Guided Zero-Shot Video Question Grounding ACCV2026
链接: https://arxiv.org/abs/2610.25570
作者: Cheng-Han Lee,Amir Salimi,Hsiang-Wei Huang,Che-Chun Su,Min Sun,Cheng-Hao Kuo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: This paper is accepted by ACCV 2026. We plan to publicly release the source code at this https URL
Abstract:Video Question Grounding (VideoQG) requires both accurate question answering and precise temporal localization. Existing supervised methods rely on costly temporal annotations, while existing zero-shot approaches are primarily designed for Video Question Answering (VQA) and are not optimized for temporal grounding. We propose DiffGround, an action-driven zero-shot framework built upon frozen Multimodal Large Language Models (MLLMs). Our framework first predicts an answer-relevant action cue that transforms the input question into an explicit semantic target for grounding. Action Cue converts a reasoningoriented question into an action-centric retrieval query. Guided by the action cue, a Vision-Language Model (VLM) efficiently selects informative keyframes for reasoning. Finally, we introduce Semantic Diff Refinement, a training-free temporal refinement method that exploits semantic differences between neighboring SigLIP2 embeddings to recover complete action segments from sparse predictions. Extensive experiments on NExT-GQA and STAR show that DiffGround consistently outperforms existing zero-shot and supervised methods, establishing a new state of the art for Video Question Grounding.
[CV-121] ricking Synthetic Image Detectors with Amortized Generative Attacks
链接: https://arxiv.org/abs/2610.25554
作者: Mohamed Awad,Khaled Elhady,Yasser Rohaim,Samuele Poppi,Nils Lukas
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Rapid advances in image synthesis have increased the need for reliable synthetic-image detectors, yet their robustness to transferable attacks remains incomplete and understudied. Existing evaluations generally examine whether an attack constructed against one detector transfers to another, but rarely isolate how its effectiveness changes when the attacked images originate from different generators. We conduct a systematic evaluation of three representative attacks: two iterative perturbation-based methods, MIG and MuMoDIG, and the amortized generative attack Tricker. Across four detectors and five generators spanning diverse architectures, we introduce a two-axis evaluation protocol that independently varies the surrogate detector used to construct or train the attack and the generator supplying its input images. We assess attack effectiveness, cross-detector transferability, perceptual quality, runtime, and interaction with common post-processing operations. We revisit Tricker and show that its data and training requirements can be substantially reduced, enabling broader and more efficient evaluation. Our results demonstrate that transferability is not a detector-only property: the input-generator distribution strongly influences both evasion effectiveness and attacked-image quality. MIG and MuMoDIG exhibit pronounced dependence on the detector–generator pair, whereas Tricker provides more uniform white-box evasion and image quality. We additionally observe that post-processing maintains the asymmetric transferability topology. These findings expose a neglected dimension of attack evaluation and motivate transfer-aware protocols that jointly account for detector and generator variation. We will release our code, evaluation data, and trained attack models with a responsible-use statement.
[CV-122] StreamAhead: Real-Time Streaming Video Question Answering via Adaptive Precomputation
链接: https://arxiv.org/abs/2610.25551
作者: Haoyang Fan,Dhruv Parikh,Rajgopal Kannan,Viktor Prasanna
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 5 figures, 6 tables
Abstract:Real-time streaming video question answering with vision-language models must answer queries that arrive at any time with low latency while sustaining the video-processing rate required by the application. Existing systems construct key-value (KV) caches for the entire video as it arrives, although a query typically uses only a few of the preceding frames as evidence. This highlights a fundamental trade-off: performing more computation as video arrives reduces query-time computation but can increase processing lag, whereas deferring computation can increase the video-processing rate but requires additional query-time computation for retrieved content. We present StreamAhead, an algorithm-system co-design that exploits this trade-off through adaptive precomputation: dynamically controlling how much computation to complete for each video segment before a query arrives. StreamAhead encodes video segments independently and lets each remain at a different representation level, from inexpensive visual features to costly KV caches, completing KV caches only for the retrieved segments when a query arrives. StreamAhead profiles offline the cost of advancing a segment to each representation level. Its runtime uses these profiles to schedule segment computation concurrently with query execution, pipelining it across heterogeneous CPU and GPU resources while maintaining the required video-processing rate. StreamAhead achieves accuracy competitive with state-of-the-art methods with up to 10.4x lower time-to-first-token (TTFT), and up to 5.9x lower TTFT than state-of-the-art streaming systems. It sustains 30 FPS with P95 TTFT under 0.5 s, and up to 57 FPS with P95 TTFT under 2 s. To our knowledge, StreamAhead is the first real-time streaming video QA system with competitive accuracy, setting a new Pareto frontier between video-processing rate and TTFT. Comments: 12 pages, 5 figures, 6 tables Subjects: Distributed, Parallel, and Cluster Computing (cs.DC); Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.25551 [cs.DC] (or arXiv:2610.25551v1 [cs.DC] for this version) https://doi.org/10.48550/arXiv.2610.25551 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-123] Residential Construction Image Dataset Using CSI Divisions for Deep Learning-based Object Detection
链接: https://arxiv.org/abs/2610.25535
作者: Fahad Iqbal,Shayan Mirzabeigi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Accurate recognition of construction site elements is essential for the application of deep learning (DL) in the construction industry. Despite rapid advances in computer vision, the adoption of DL in construction has been limited by the lack of structured, publicly available datasets aligned with established industry standards. To address this gap, we introduce the Residential Construction Dataset Using CSI Divisions (ResCD-CSI) image dataset, a manually labeled image benchmark organized according to the divisional structure defined by the Construction Specifications Institute (CSI). The dataset comprises high-resolution images collected from active residential construction sites across the United States, capturing diverse building components, materials, and equipment under real-world conditions. In total, 16,365 objects spanning 14 major CSI divisions were precisely annotated using bounding boxes, ensuring consistency and reliability for object detection tasks. By aligning with the CSI divisional framework, ResCD-CSI establishes a standardized resource that bridges construction practice with computer vision research and supports the development of Artificial Intelligence-driven solutions for residential construction monitoring.
[CV-124] Modaliff: A Unified Token Diffusion Model for Any-to-Any Generation
链接: https://arxiv.org/abs/2610.25533
作者: Duc Kieu,Kien Do,Minh Khoa Le,Tung Kieu,Dang Nguyen,Thin Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Any-to-any generation seeks a single model that translates flexibly among text, image, and audio, yet existing approaches either require fully aligned multimodal tuples or a shared representation that suppresses modality-specific structure. We introduce Modaliff, a novel token-based diffusion model that learns from widely available bimodal pairs while preserving each modality in its native latent sequence. Modaliff assembles conditions and targets into a unified token sequence, with per-token attributes to disambiguate token identities and role-based positional embedding to preserve each modality’s internal structure across sequence layouts. This design enables a single denoiser to support diverse generation tasks without task-specific architectural changes or a shared cross-modal latent space. Beyond one-to-one generation, Modaliff supports multiple conditions either by combining single-condition predictions at inference or by lightweight finetuning on captions that jointly describe multiple inputs. Experiments across text, image, and audio show that Modaliff achieves the best fidelity among any-to-any models on most generation tasks while using substantially fewer parameters and training data.
[CV-125] ReDT: Resolution-Decoupled Visual Geometry Transformer
链接: https://arxiv.org/abs/2610.25517
作者: Abteen Arab,Guile Wu,Dongfeng Bai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical report
Abstract:We introduce ReDT, a feed-forward visual geometry transformer designed to scale multi-view reconstruction to high-definition (HD) inputs. Existing visual geometry transformers are typically trained at fixed, standard definition (SD) resolutions and exhibit degraded reconstruction quality and poor computational scaling to high input resolutions. While recent efficient architectures reduce the complexity of inter-frame attention, expensive intra-frame and encoder attention remain largely unaddressed, making HD reconstruction impractical with existing architectures. ReDT addresses both limitations with a resolution-aware architecture designed to scale to high-resolution inputs. The key idea of ReDT is to first map each view to a fixed-size set of basis tokens, compute correspondences in this compact constant-size basis space, and then project the outputs back to the original resolution for dense prediction. By decoupling the internal representation from input resolution during attention, ReDT substantially reduces the computational cost of high-resolution 3D reconstruction and improves model performance at all resolutions. Extensive experiments on multiple datasets spanning SD to FHD (2K) show that ReDT outperforms state-of-the-art visual geometry transformers in both reconstruction quality and computational efficiency. For instance, on the ETH3D (2K) dataset, ReDT achieves a 16\times speedup, a 34% reduction in peak GPU memory usage, and a 33% reduction in chamfer distance over VGGT- \Omega . \hrefthis https URLProject Website.
[CV-126] GroundedVDR: Grounding Video-to-Web Evidence Dependencies in Video Deep Research
链接: https://arxiv.org/abs/2610.25493
作者: Xiyuan Liang,Yu Wang,Zuhao Yang,Ziting Wang,Duo Li,Sudong Wang,Zhongyu Yang,Xiaoqin Zhang,Ling Shao,Shijian Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video large multimodal models have improved at identifying cross-frame visual evidence and conducting open-web retrieval, as reflected in recent video deep research (VDR) benchmarks. However, existing VDR benchmarks primarily evaluate tool-use trajectories and final responses without verifying whether a model actually uses the temporal video information or the retrieved web evidence. To fill this gap, we introduce \method, an evidence-grounded benchmark with three features. First, it combines human review with restricted-access probing across multiple models to exclude questions answerable from parametric knowledge or a single information source. Second, each question carries structured video-to-web evidence annotations that link the relevant video segments to their supporting web evidence. Third, its evaluation protocol separately scores video grounding, web evidence grounding, and final-answer correctness. Across eight multimodal agent systems, the best reaches 76.0% answer accuracy, yet none recovers the complete required web evidence for more than 8% of questions. Our benchmark is publicly available at \urlthis https URL.
[CV-127] SlimFormer: Token Mixing Where It Matters for Efficient 3D Medical Image Segmentation
链接: https://arxiv.org/abs/2610.25488
作者: Yang Hong,Lei Zhang,Zhihong Liu,Jianqing Mo,Xujiong Ye
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:To determine where computation is needed in volumetric segmentation networks, we analyze block-level redundancy and stage-wise token-mixer contribution. Centered Kernel Alignment reveals high input-output representational similarity within blocks of representative convolutional, hybrid, and Transformer-style segmentation networks. Similarity-guided pruning reduces their parameter counts by 39-75% while decreasing mean Dice by less than 0.4 percentage points. To examine token mixers within blocks, we introduce Anatomical Relation Distortion, which quantifies how bypassing a mixer distorts feature relations between labeled anatomical positions. Applied to an unpruned candidate architecture, Centered Kernel Alignment indicates high representational redundancy in intermediate encoder mixers, whereas Anatomical Relation Distortion shows that they contribute little and the bottleneck mixer most. These findings motivate SlimFormer++, which uses Selective Mixing blocks with windowed Mamba mixing in the first encoder stage and only residual channel MLPs in deeper stages. At the bottleneck, Augmented Mixing injects similarity-weighted sparse long-range context into the axial taps of a depthwise convolution at cost linear in voxel count. Evaluated against 12 baselines across eight public CT, MRI, and ultrasound datasets (1,809 cases), SlimFormer++ achieves the highest mean Dice (87.84%) and lowest mean HD95 (6.50 mm) with the fewest parameters (5.98 million). These results support stage-aware token mixing as a design strategy for improving the accuracy-efficiency trade-off in volumetric segmentation. The source code is available at this https URL
[CV-128] PCXR-Copilot: Device-aware pediatric chest radiographic analysis with segmentation suppression and edema-related cues
链接: https://arxiv.org/abs/2610.25486
作者: Xueyang Li,Alec Girten,Peng Zeng,Mingze Jiang,Shaoyang Huang,Gelei Xu,Danny Z. Chen,Yiyu Shi,John N. Kheir
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Patients admitted to pediatric ICU and cardiac ICU often require intensive perioperative support and frequent chest radiography to monitor support devices, extravascular lung water, and body-wall edema. Radiographic interpretation and analysis is often difficult due to superimposed anatomic and iatrogenic structures that obscure clinically relevant signals. While deep learning can provide decision support, translation to pediatric radiographs is still limited by high device burden and scarcity of pediatric datasets with clinically grounded device annotations. To address this gap, we first introduce PediaDeviceSeg-1K, a curated dataset of 1,000 anterior-posterior radiographs of pediatric CICU patients with clinically grounded, fine-grained annotations of common ICU devices. To our best knowledge, this is the first publicly available pediatric CICU CXR resource with clinically grounded device annotations for device-focused benchmarking. Building on PediaDeviceSeg-1K, we further present PCXR-Copilot, a device-aware pipeline that outputs lung and device masks, optional device-suppressed views, and edema-related visualization maps and scalar indices. The full pipeline has been integrated and runs in real time within the TWIST platform at Boston Children’s Hospital, demonstrating feasibility for real-world clinical deployment.
[CV-129] PRIMA: Pose Refined Inductive Motion Adapter for Zero Shot Social Bodily Behavior Recognition ACM-MM2026
链接: https://arxiv.org/abs/2610.25482
作者: Mingcong Ding,Babette Bühler,Tim Fütterer,Ulrich Trautwein,Enkelejda Kasneci
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 10 pages, 5 figures. Accepted at ACM Multimedia 2026 (ACM MM 2026)
Abstract:Recognizing fine-grained bodily behaviors, such as self-grooming, postural shifts, and backchanneling, is a useful intermediate step for computational social signal analysis. But the behaviors that matter vary across social contexts, and retraining a model for each new taxonomy is expensive. Vision-language models (VLMs) can adapt to new categories through text prompts, yet they lack sensitivity to body structure and often confuse subtle postural distinctions. We propose PRIMA (Pose Refined Inductive Motion Adapter), a lightweight module that injects a body-structure prior into frozen VLMs using human pose as a training-time supervisory signal, requiring no pose input at inference. PRIMA separates a model-specific projection head from a shared core that encodes transferable body-configuration knowledge; adapting to a new VLM requires retraining only the head. We train PRIMA on a micro-action dataset and evaluate entirely through zero-shot prompting. The adapter improves recognition on its training domain, transfers to additional VLM backbones at different scales with the core frozen, and generalizes to two unseen target datasets from social psychology and daily activity recognition. In a supporting downstream study, the improved behavior predictions also yield higher correlations with human ratings of rapport and dominance in a collaboration context. Within the scope of the tested models and indoor datasets, the results indicate that pose-guided training can produce body-structure priors that generalize across unseen behavior vocabularies and can be partially reused across VLM backbones while preserving RGB-only inference.
[CV-130] Multimodal-Conditioned Latent Diffusion Models for Spatiotemporal Brain MRI Forecasting WACV2027
链接: https://arxiv.org/abs/2610.25481
作者: Antonio Scardace,Silvia Ingala,Sebastiano Battiato,Daniele Ravì
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to WACV 2027
Abstract:Spatiotemporal forecasting of 3D anatomical changes is a critical generative vision challenge for modeling Alzheimer’s Disease (AD) progression. While brain MRI forecasting from single-modality data has advanced, other key complementary multimodal biomarkers remain underutilized. In this work, we propose the Multimodal-Conditioned Latent Diffusion Model (MC-LDM), a novel framework for patient-specific brain MRI synthesis. MC-LDM leverages multimodal biomarkers—MRI, tau-PET, Clinical Dementia Rating (CDR), and APOE4—during training, but relies only on baseline MRI, APOE4, and patient metadata at inference, reflecting realistic clinical availability. Our contributions are threefold: (1) a Latent Diffusion Model (LDM) fine-tuned via Low-Rank Adaptation (LoRA) for data-efficient learning; (2) a multimodal conditioning that integrates molecular and clinical signals to model heterogeneous disease trajectories; and (3) auxiliary predictors estimating unobserved multimodal biomarkers from MRI features and patient covariates to condition the diffusion model, removing the need for these assessments at inference. Trained on 1,355 visits from 852 subjects in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), MC-LDM reduces average regional volumetric MAE by 18% over the second-best baseline on an independent test set of 664 visits. To our knowledge, MC-LDM represents the first diffusion-based framework to leverage multimodal priors for MRI forecasting. This conceptual advancement enables patient-specific trajectory generation that captures aspects of AD heterogeneity not observable from structural anatomy alone, bridging the gap between rich multimodal supervision and the strict constraints of routine clinical deployment. Code is available at: this https URL.
[CV-131] st-Time Scaling of Text-to-Image Flow Matching Models with Noise-Rejuvenated Sequential Monte Carlo
链接: https://arxiv.org/abs/2610.25479
作者: Wenyi Lian,Zheng Zhao,Thomas B. Sch{ö}n
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Test-time scaling of diffusion models with sequential Monte Carlo (SMC) has become a promising strategy for sampling under user-specified rewards, such as human preference scores. However, the standard SMC framework typically relies on stochastic proposals and therefore does not directly transfer to flow matching models, in which inference is governed by deterministic ODEs. In this work, we introduce noise-rejuvenated SMC, an inference-time scaling method tailored for text-to-image flow matching models. Our key idea is to perturb each resampled particle by changing its source noise and constructing a new intermediate state using its current endpoint estimate. This moves the particle to another state that remains representable under the flow-matching interpolation form, enabling particle exploration without converting the sampler to an SDE. We further introduce a look-ahead mechanism to refine resampled particles before weighting. Experiments on Stable Diffusion 3, FLUX.1-dev, and Z-Image demonstrate consistent improvements in reward-guided generation, while ablation studies validate the contributions of noise rejuvenation and look-ahead refinement. Code is available at this https URL.
[CV-132] BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100 Simulation Tasks
链接: https://arxiv.org/abs/2610.25459
作者: Yong Wang,Kecheng Zheng,Xing Zhu,Chao Xu,Yujun Shen,Shengze Cai
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Bimanual manipulation with dexterous hands has attracted growing attention in research on robotic action models. However, few studies have developed benchmarks specifically tailored for dexterous manipulation, leaving the dexterous manipulation capabilities of existing models insufficiently characterized. In this paper, we present BiDex-100, a simulation benchmark for bimanual dexterous manipulation that combines broad task coverage, multi-dimensional evaluation, and high-DoF hand embodiments. Concretely, the benchmark comprises 106 tasks organized along five capability axes: generalization, instruction following, limited demonstrations, long horizon, and memory. Each task is paired with an out-of-domain variant featuring controlled visual and spatial changes, allowing systematic assessment of robustness. In addition, the benchmark supports two embodiments and provides 8,080 success-validated teleoperated demonstrations, totaling 6.25 million frames. We evaluate seven recent Vision-Language-Action models under a unified protocol, establishing baselines for capability-specific performance, robustness, and data efficiency. The website is available at this https URL.
[CV-133] UBD-Bench: Benchmarking and Analyzing Unimodal Branch Degradation in Multimodal Perception for Autonomous Driving
链接: https://arxiv.org/abs/2610.25448
作者: Ang Liu,Xiang Gu,Ruikai Li,Qiyue Wang,Yizhe Fan,Hao Shan,Ziyang Yan,Han Jiang,Zhenning Li,Xiaoshuai Hao,Yilong Ren,Zhiyong Cui,Haiyang Yu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 5 figures
Abstract:Multimodal perception improves autonomous-driving performance by combining complementary sensor information. However, stronger multimodal predictions do not necessarily imply stronger representations in each modality-specific branch. Such branch-level weaknesses can remain hidden in conventional output-level evaluation. We introduce UBD-Bench, a systematic benchmark for studying unimodal branch degradation (UBD) in multimodal perception. UBD-Bench uses a Paired Frozen-Branch Probe to compare the task-relevant representation quality of the same modality after independent unimodal training and multimodal joint training under matched decoding conditions. Across four autonomous-driving perception tasks and 10 models, we find that Camera branches consistently degrade after multimodal joint training, whereas LiDAR branches exhibit weaker and more model-dependent changes, revealing a pronounced modality asymmetry. Hierarchical probing further shows that Camera-branch degradation extends from the visual backbone through the Camera-to-BEV pathway and is most severe at the pre-fusion Camera BEV representation. Guided by this diagnosis, we introduce unimodal-to-multimodal distillation at the degraded representation. This intervention consistently improves both Camera-branch Probe performance and full-model performance across all four tasks. These results show that multimodal gains can coexist with degraded unimodal representations, highlighting branch-level representation quality as an important dimension for evaluating and improving multimodal perception.
[CV-134] A New Hope for PRNU on Modern Smartphones? Evidence That Device-Specific Fingerprints Survive Computational Photography
链接: https://arxiv.org/abs/2610.25438
作者: Benjamin Loison(1 and 2),Quentin Bammey(3),Thibaud Ehret(1),Rafael Grompone von Gioi(2),Pablo Musé(2 and 4),Marina Gardella(2) ((1) Agence Ministérielle pour l’IA de Défense, France, (2) ENS Paris-Saclay, Université Paris-Saclay, CNRS, Centre Borelli, Gif-sur-Yvette, France, (3) Télécom Paris, Institut Polytechnique de Paris, LTCI, Palaiseau, France, (4) IIE, Facultad de Ingeniería, Universidad de la República, Uruguay)
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注:
Abstract:Photo Response Non-Uniformity (PRNU) is widely regarded as the most reliable marker for source camera identification. Recent studies, however, report high correlations between images and fingerprints from different modern smartphones, which calls into question the discriminative power of PRNU on these devices. Using 30 recent smartphones under controlled acquisition conditions (e.g., no digital zoom), we investigate whether the fingerprint itself is at fault or the way it is compared. We find that a component shared by all fingerprints of the same model inflates the standard signed peak-to-correlation energy (sPCE). To counter this, we propose \Delta sPCE: the sPCE between a query image and a candidate camera’s fingerprint, minus the mean sPCE between that image and the fingerprints of the remaining candidates, in the spirit of T-norm score normalization. \Delta sPCE improves attribution accuracy, suggesting that the device-specific PRNU signal is still present. Although the limited dataset size and the dependence of \Delta sPCE on the candidate camera set currently restrict operational applicability, these results suggest that PRNU-based attribution of modern smartphones remains feasible when fingerprints are compared appropriately.
[CV-135] Drag onMoCap: Monocular Video-based Motion Capture for Diverse Drag on Topologies SIGGRAPH
链接: https://arxiv.org/abs/2610.25437
作者: Zhongfei Qing,Xiaoxiang Zhao,Ling-Hao Chen,Zhongang Cai,Wenjia Wang,Yifan Wu,Qianying Liu,Wei Zeng,Bo Hu,Songnan Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Siggraph Asia 2026
Abstract:This work explores monocular motion capture for dragon-type characters, which currently relies on manual keyframing due to the lack of high-quality datasets and parametric models. The primary obstacle lies in the extreme topological diversity across dragon rigs, where varying joint counts and naming conventions render traditional fixed-template regressors inapplicable. To address this, we introduce DragonMoCap, a novel framework that treats the target rig as a dynamic input rather than a fixed prior. By deliberately scoping the framework to the most prevalent production topologies, we ensure high-fidelity results across diverse morphologies. The core of our approach is the design of language-grounded joint query tokens, which establish a dual-property representation: ensuring intra-rig discriminability while maintaining cross-rig similarity. We further propose a non-parametric query-token retrieval mechanism to enable zero-shot adaptation to unseen rigs without retraining. Extensive evaluations demonstrate that DragonMoCap consistently outperforms HMR-style baselines across diverse dragon topologies. The practical utility of our framework is further evidenced by its integration into professional animation workflows, highlighting its potential for real-time industrial applications.
[CV-136] Preserving the Manifold: Frozen Expert Alignment for Label-Efficient SAR Scene Understanding ECCV2026
链接: https://arxiv.org/abs/2610.25429
作者: Eli A. Underwood,I. Taylor McKechnie,Leslie M. Collins
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages. A shorter version of this work was presented at the ECCV 2026 Workshop TerraBytes II: Towards Global Datasets and Models for Earth Observation (non-archival)
Abstract:Cross-modal alignment of synthetic aperture radar (SAR) imagery to natural language is constrained by two distinct failures. The first is modality mismatch: text encoders pre-trained on optical imagery, such as CLIP, cannot be efficiently fine-tuned on SAR, because active microwave backscatter is physically incompatible with passive optical reflectance and the available SAR caption datasets are orders of magnitude too small to bridge this gap without substantially degrading the learned semantic manifold. The second is physical forgetting: SAR encoders pre-trained on large-scale self-supervised objectives encode radar physics that are progressively erased when the encoder is fine-tuned for text alignment on limited caption data. A single design principle addresses both failures: keep both encoders frozen and learn only a constrained, structure-preserving bridge. Our Frozen Expert framework aligns a frozen CROMA SAR encoder with a frozen CLIP text encoder via a lightweight constrained residual bridge. We formally define physical forgetting via Procrustes disparity and show our bridge preserves SAR geometry significantly better than fine-tuning (disparity: 0.210 vs. 0.351). We validate generality across multiple SAR backbones and text encoders. On zero-shot SAR scene classification, our approach outperforms fine-tuned baselines using up to 8x less training data: DFC2020 (+10.3% R@1, +15.5% mAP), EuroSAT-SAR (+15.7% R@1, +12.2% mAP), BigEarthNet-MM (+14.7% R@1). These results establish frozen manifold alignment as a principled, data-efficient alternative to fine-tuning in high-disparity cross-modal settings.
[CV-137] From Label Priors to Task Evidence: Efficient Long-Video Frame Selection via Bayesian GRPO
链接: https://arxiv.org/abs/2610.25418
作者: Shuochen Chang,Bingjie Gao,Qingyang Liu,Qianli Ma,Yibo Miao,Haonan Zhao,Zhaohe Liao,Xiaofeng Zhang,Jiangtong Li,Li Niu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal Large Language Models (MLLMs) face significant scalability challenges in long-video understanding due to quadratic attention complexity and context window constraints. Existing retrieval-based approaches typically rely on supervised learning from static keyframe annotations. However, these label priors often fail to align with the actual information required by downstream MLLMs for complex reasoning, creating a misalignment between the training target and inference needs. To bridge this gap, we propose a probabilistic framework that reformulates frame selection as latent-variable inference. We first initialize a lightweight policy using supervised learning to capture general visual saliency. Then, we treat a frozen MLLM as a task environment and use its answer likelihood as dense task evidence to refine the policy via Bayesian Group Relative Policy Optimization (GRPO). Within this formulation, the negative log-likelihood (NLL) over ground-truth answer tokens provides a principled dense reward, while a prior-matching KL term preserves semantic consistency with the elicited label prior. This process transitions the selection principle from static label priors to task evidence. Extensive experiments on Video-MME, MLVU, and LongVideoBench demonstrate that our selector achieves competitive accuracy with MLLM-based scorers while retaining the inference efficiency of lightweight methods. The selector is trained once with a frozen MLLM environment and transfers directly to multiple downstream MLLMs without retraining. Code is available at \urlthis http URL.
[CV-138] UniState: Unifying Memory and Intervention for Controllable World Simulation
链接: https://arxiv.org/abs/2610.25414
作者: Jaeseok Jeong,Kyungmook Choi,Sohyun Chung,Youngsik Yun,Youngjung Uh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:World models offer increasingly rich environments for exploration, yet intervening in a world while observing the consequences from freely chosen viewpoints remains challenging. We tackle these challenges by introducing UniState, a framework for controllable world simulation built on a unified world-state representation. This representation unifies spatial memory and user interventions. Users specify object motion directly within spatial memory, while independently choosing the camera trajectory from which the resulting world state is observed. This unification makes the relationship between the scene and the requested changes explicit, removing ambiguity between object motion and viewpoint change from the control specification. Crucially, specifying an intervention does not require prescribing every resulting movement: the video model can synthesize additional motion within controlled objects and plausible interactions with their surroundings. We develop training and sampling strategies that help generated videos follow the specified object motion and camera trajectory while preserving this capacity to generate responses beyond the prescribed transformations. Experiments demonstrate more accurate joint object and camera control than the existing methods, alongside qualitative evidence that intervention-induced scene changes persist across rollouts.
[CV-139] MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion
链接: https://arxiv.org/abs/2610.25404
作者: Houyuan Chen,Fanzhou Wang,Xiangyu Fan,Zesong Qiu,Yukun Wei,Songlin Yang,Zhitao Yang,Lvmin Zhang,Pei Yang,Hao Zhao,Lei Yang,Anyi Rao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Multi-turn image editing aims to modify an image according to a sequence of instructions, enabling more effective interactive visual content creation for users. However, existing methods often exhibit noticeable drift after only a few editing turns, manifesting in two aspects: 1) unstable visual quality, with accumulating artifacts and degraded details; and 2) inconsistent image content, with unintended changes to preserved regions and prior edits. To address these challenges, we propose MT Forcing, a novel training paradigm for multi-turn image editing with unified multimodal models (UMMs) based on autoregressive diffusion. Specifically, we propose Single-Turn Distribution Matching Distillation (Single-Turn DMD), which supervises student self-rollouts with a teacher that applies cumulative edits to the source image in a single turn, providing visual-quality guidance without recursive error accumulation. Additionally, we introduce Teacher-Forcing Guidance, which supervises history-dependent transitions on reference trajectories during warmup and as a regularizer during distillation, encouraging cross-turn content consistency beyond what per-turn distribution matching alone can ensure. Extensive experiments demonstrate that our approach significantly improves the multi-turn image editing capabilities of UMMs, yielding more stable visual quality and stronger cross-turn consistency.
[CV-140] RAHF-Net: Range-Aware Heterogeneous Fusion for LiDAR-Assisted RGB Semantic Segmentation
链接: https://arxiv.org/abs/2610.25400
作者: R.D. Lin,Pengcheng Weng,Ge Wang,Han Ding,Fei Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:This paper studies LiDAR-assisted RGB semantic segmentation, where LiDAR point-cloud features improve dense pixel-wise predictions in the RGB image plane rather than 3D point labels or detections. Existing 2D fusion methods often project point clouds into shallow depth/range maps and apply spatially uniform fusion, discarding rich 3D structure and ignoring the range-dependent reliability of outdoor LiDAR. We propose RAHF-Net, a Range-Aware Heterogeneous Fusion Network. Its Cylindrical-to-Perspective (C2P) module projects pretrained cylindrical voxel features into the image plane, preserving structured 3D semantic priors. A Range-Aware Heterogeneous Fusion (RAHF) module then uses a soft depth-derived range mask to perform LiDAR-guided geometric enhancement in reliable near-field regions and visual-context enhancement in sparse far-field regions. On KITTI-360, RAHF-Net achieves 65.76 mIoU, outperforming representative RGB-only and RGB-LiDAR baselines, including CMX by 1.45 points. Experiments on DELIVER further demonstrate improved robustness under adverse weather, illumination degradation, and sensor corruption.
[CV-141] PolarVis: A Dataset and Diffusion Framework for RGB-to-Polarization Perception in Adverse Driving Scenes
链接: https://arxiv.org/abs/2610.25396
作者: Ruilin Huang,Jue Wang,Zhen Tian,Xia Hua,F. Richard Yu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reflections, glare, transparent structures, and material-dependent appearance variations can introduce ambiguity into RGB-based visual perception in vehicle-relevant environments. Polarization provides complementary reflection- and material-sensitive information, but its acquisition requires dedicated sensing hardware. This paper presents RGB-P Wild, a real-world dataset containing 10,000 paired RGB and four-angle linear-polarization image sets, together with PolarVis, a diffusion-based framework for recovering polarization representations from a single RGB image. PolarVis combines hierarchical RGB feature extraction, diffusion-based feature refinement, four angle-specific decoders, and multi-scale attention recalibration to reconstruct polarization intensity images at 0^\circ , 45^\circ , 90^\circ , and 135^\circ . On RGB-P Wild, PolarVis achieves a normalized-domain MSE of 0.0465, a PSNR of 27.83 dB, and an SSIM of 0.8597. Compared with the released PolarAnything model under the adopted evaluation protocol, PolarVis obtains an AoLP error of 25.66^\circ and a DoLP MAE of 0.0404, compared with 46.03^\circ and 0.1084, respectively. Cross-dataset evaluation on vehicle-mounted imagery further examines transfer beyond the training domain. In downstream evaluation, a frozen PCDNet achieves AP50/AP50:95 of 86.3%/59.1% on 990 RGBP-Car test images using PolarVis-generated polarization, compared with 87.0%/59.8% using measured polarization. These results indicate that RGB-derived polarization retains useful polarimetric information for the evaluated perception tasks, while remaining distinct from calibrated physical polarization measurements.
[CV-142] A 3D-Driven Framework for Moving-Camera Perspective-to-360° Video Generation WACV2027
链接: https://arxiv.org/abs/2610.25395
作者: Michael Wakeham,Mrinmoy Kumar Das,Wenqi Wang,Ayush Tewari,Boqing Gong,Deepti Ghadiyaram
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. Project page: this https URL
Abstract:Turning a moving-camera perspective video into a 360° video requires more than just generating content outside the input frame: observations revealed over time must be preserved, unseen regions completed, and the resulting scene kept consistent as the camera viewpoint changes. Existing methods assign these different requirements to a single conditional video model, requiring them to be jointly learned from panoramic video data. We instead present a modular, 3D-driven framework that distributes these three responsibilities across specialized components, all connected through a shared scene representation. Input observations are reconstructed explicitly in 3D, while a pretrained panoramic image model completes unobserved content from a selected panoramic render. This generated content is lifted into the shared 3D scene, where it remains spatially anchored across camera viewpoints. Without requiring any additional panoramic-video training, our framework achieves state-of-the-art input-view fidelity, perceptual quality, and geometric consistency on a static-scene subset of the Argus evaluation set. These results support our decomposed, 3D-driven formulation as a promising alternative to direct video generation in this setting.
[CV-143] ATLAS: Attention-Guided Test-Time Latent Selection for Compositional Visual Autoregressive Generation
链接: https://arxiv.org/abs/2610.25390
作者: Hossein Shahabadi,Raya Namazi,Mahdieh Soleymani Baghshah
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages
Abstract:Visual autoregressive models produce convincing images but can misassign attributes and spatial relations. Searching multiple generation paths can improve composition, yet completing every candidate is costly. We introduce ATLAS, a training-free framework that uses the frozen generator’s internal attention to guide search before final images are available. Within-prompt diagnostics identify attention-derived scores that predict final compositional preferences on Infinity, with category-dependent transfer to Switti. ATLAS uses these on-the-fly signals to progressively concentrate computation on promising trajectories. It maintains multiple coarse alternatives, prunes candidates without decoding at intermediate scales, and reserves image-space evaluation for a final choice among partial trajectories. ATLAS scales effectively on both backbones, achieving higher mean scores on the evaluated T2I-CompBench and GenEval categories as more computation is devoted to search. At the largest tested budget, the targeted T2I-CompBench mean improves by over 10% relative to each backbone. Separate profiles show over 50% less generation time than completing and evaluating all candidates at equal initial counts, alongside lower peak allocated GPU memory. Relative time savings are larger at the higher tested candidate count. Together, these results support scalable compositional search: additional sampling expands exploration, while attention-guided selection limits continuation cost without parameter updates or gradient-based optimization.
[CV-144] Only Means Only: what a harmonization benchmark score averag es over
链接: https://arxiv.org/abs/2610.25387
作者: Imad Buljić,Delaida Muminović
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages including references, 3 figures in the paper and 1 in the supplement. The supplement is included as an ancillary file and published with the preprint record, DOI https://doi.org/10.5281/zenodo.23404530
Abstract:Image harmonization methods adjust the colour and lighting of an object pasted into a photograph so that it looks native to its new scene, and they are ranked by average error on one benchmark whose test images are real photographs with one object recoloured. One photograph can yield several such test images, and the benchmark promises that no photograph appears in both the training and the test split. We find that 1,399 photographs do, and the composites built on them are a quarter of the test set. Duplicate checks over file names or bytes miss this, because every composite is a different file. On that quarter the three methods we score end to end read two to five times lower error than on the rest, but most of the gap is composition rather than the sharing: the shared composites come almost entirely from one of the benchmark’s four source collections, and inside that collection the gap falls to about one and a half times. An official score is therefore an average over two populations, mixed in a proportion the benchmark does not report. We release a test list of 5,581 composites whose construction identities do not appear in training, with four confirmed near-duplicate pairs disclosed rather than removed, adoptable without retraining at the price of one rescore per checkpoint. We leave open whether the sharing itself changes what a model learns.
[CV-145] Painting Change Map as guidance for generating painting processes
链接: https://arxiv.org/abs/2610.25369
作者: Markus Pobitzer,Chang Liu,Niccolò Biondi,Nicu Sebe
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Digital art generation has undergone a massive paradigm shift, moving from static image synthesis to dynamic video generation that can mimic the human painting process. While existing models can synthesize visually stunning time-lapses of paintings being created or “unpainted,” they largely treat the creative process as a black box. Current conditioning mechanisms are either too rigid, relying on static geometric cues like Canny edges or depth maps, or too holistic, utilizing simple scalar progress variables that offer no localized control. Consequently, users lack the agency to steer how a painting evolves over time. In this work, we argue that true control over the video generation of artwork requires a joint understanding of two fundamental dimensions: where an artist acts, and when those actions take place. To achieve this, we introduce the Painting Change Map, a novel spatio-temporal control mechanism that explicitly tracks normalized structural changes across a 1,000-timestep continuum, a resolution sufficient to capture fine-grained stroke-level changes while remaining computationally tractable. By mapping localized visual differences into an encoded RGB space, our representation enables fine-grained guidance for both the forward painting and reverse unpainting processes. Extensive benchmarking demonstrates the efficacy of our approach. We outperform existing control baselines and achieve competitive or superior results compared to state-of-the-art painting methods. Furthermore, under oracle conditions, our method surpasses all baselines by a significant margin. Code is available at this https URL.
[CV-146] LateMatch: Breaking the Interaction Bottleneck via Cacheable Matching for Efficient Visual Localization
链接: https://arxiv.org/abs/2610.25358
作者: Mohammad Omama,Aeyan Ashraf,Sandeep Chinchali
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:Visual localization is critical for embodied agents, but deploying state-of-the-art deep feature matchers on edge devices is fundamentally limited by their massive computational cost. Standard deep matchers tightly interleave intra-image context building and inter-image matching at every layer, creating a severe interaction bottleneck. Localizing a single query against K database images forces the system to execute the full network K independent times with no capacity for caching. To break this bottleneck, we advocate for a paradigm shift toward cacheable image matching. We realize this paradigm with LateMatch, a versatile approach that restructures existing matchers by deferring inter-image interaction to their final layers. By isolating intra-image processing into offline-cacheable Context Blocks, only a small set of Interaction Blocks must be executed online. To further optimize LateMatch, we introduce E-SToRE (Entropy-guided Spatial Token Reduction and Expansion). E-SToRE leverages the decreasing Shannon entropy of the late Interaction Blocks to dynamically reduce token counts just prior to cross-image interaction. Evaluated across HPatches, ScanNet, MegaDepth, ETH-3D, and Aachen, LateMatch reduces effective computational cost by over 2x compared to standard deep matchers, while E-SToRE yields an additional 10% to 14% savings with negligible degradation in accuracy.
[CV-147] Local Evidence Global Failure: Diagnosing Speech-Vision Temporal Composition in Audio-Visual Language Models
链接: https://arxiv.org/abs/2610.25352
作者: Jaehong Myung,Seunghun Oh,Unsang Park
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Multimedia (cs.MM)
备注:
Abstract:Audio-visual large language models (AV-LLMs) can fail at full-video speech-vision decisions even when useful visual evidence is locally recoverable. We introduce Temporal Evidence Decomposition (TED), a training-free diagnostic framework that separates local visual verification from speech-relative aggregation. Automatic speech recognition supplies a transcript and utterance interval. TED queries audio-free windows with the transcript for Temporal Alignment and Forecasting, or answer candidates for Cross-Modal Temporal Binding, then aggregates evidence relative to that interval. Across four frozen AV-LLMs on SVHalluc, Temporal Forecasting accuracy rises from 29.3-33.7% under full-video questioning to 79.3-82.2% with TED. Query replacement reduces recovery; the observed anchor remains beneficial when temporal-region availability is held fixed. Physical-isolation and timing-shortcut controls test alternative explanations. The findings distinguish locally recoverable evidence from successful full-video decisions, motivating explicit separation of narration content, utterance timing, and time-indexed visual evidence.
[CV-148] Predicted Class Correction: Conformal Thresholds for Classes with No Calibration Data
链接: https://arxiv.org/abs/2610.25349
作者: One Octadion,Novanto Yudistira,Yuta Nakashima
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 23 pages including supplementary material, 4 figures
Abstract:Class-conditional conformal prediction calibrates a separate score threshold for every class and therefore requires labelled calibration examples of every class. In long-tailed deployments these are exactly what is missing: rare classes are left with few or no held-out examples, and existing methods are either undefined for them or fall back to the shared marginal threshold. We ask whether the per-class correction can instead be predicted from descriptors of the class that require no labels at all. We propose Predicted Class Correction (PCC), which regresses the offset between the per-class and the marginal conformal quantile on geometric descriptors of the classifier head, shrinks the prediction by a scalar selected on the labelled classes, and refits a single marginal offset. The descriptors depend only on the model’s weights, so a class with zero calibration examples still receives its own threshold, and since only one scalar is calibrated, split conformal’s marginal guarantee is retained. On ImageNet, PCC improves worst-class coverage on such classes by 0.058 [0.006, 0.110] at matched average set size, 42% of an oracle that sees the test labels. Of eight baselines on the same splits, six change the statistic by exactly zero and the best reaches 0.016 with an interval containing zero, so PCC is the only one distinguishable from no improvement; the gain persists with only 100 labelled classes out of 1000. A second finding concerns evaluation: worst-class coverage becomes unmeasurable before it becomes unachievable. Below roughly 35 test examples per class no improvement can be shown even for an oracle, so a null result there reflects the evaluation set, not the method. Restricting evaluation above this depth recovers a significant improvement on Pl@ntNet-300K, where the effect had appeared absent.
[CV-149] MCSplat: Multi-View Photometric and Geometric Consistent Feed-Forward Gaussian Splatting for Driving Scenes NEURIPS2026
链接: https://arxiv.org/abs/2610.25345
作者: Yifei Wang,Wei Tian,Wenhua Bao,Jia Zhou,Hanshi Wang,Fudong Ge,Jianhua Wu,Shuai Yuan,Zhipeng Zhang,Xingliang Liu,Xianming Zeng,Jianyun Xu,Bingzhao Gao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 24 pages, 14 figures, 6 tables; Accepted to NeurIPS 2026
Abstract:Driving-scene reconstruction via feed-forward Gaussian splatting from sparse multi-view images remains challenging due to two factors. First, cross-camera photometric inconsistency, caused by variations in exposure, ISP, and illumination, can be incorrectly absorbed into Gaussian geometry. Second, unreliable local geometric supervision arises from normal-based constraints near depth discontinuities. To address these issues, we present \textbfMCSplat, a multi-view consistent feed-forward Gaussian splatting framework for driving scenes. MCSplat constructs a canonical appearance space from depth-guided multi-camera correspondences to decouple geometry learning from camera-specific appearance. A multi-scale Bilateral Grid Transformer restores raw camera appearance from canonical renderings, enabling photometric supervision without contaminating Gaussian geometry. To improve local geometric reliability, we estimate Fisher-information-based uncertainty from photometric sensitivity and use it to adaptively weight surface-oriented self-supervision. Extensive experiments demonstrate improved reconstruction quality and generalization across driving datasets. Geometry-oriented evaluations further reveal enhanced cross-view depth consistency and local surface-normal agreement, while qualitative results show fewer geometric artifacts and more coherent structures across overlapping views.
[CV-150] CASE: Complementary Articulation-Aware Speech Encoder to Enhance Speech-Driven 3D Facial Animation
链接: https://arxiv.org/abs/2610.25342
作者: Jaehong Myung,Seungho Eum,Taegon Kim,Hyunjung Chung,Unsang Park
类目: Computer Vision and Pattern Recognition (cs.CV); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:General-purpose self-supervised speech encoders provide strong acoustic representations, but their pretraining objectives are not explicitly aligned with the visible articulatory dynamics required by speech-driven 3D facial animation. We propose CASE (Complementary Articulation-aware Speech Encoder), a complementary speech encoder adapted by contrastively aligning short speech clips with continuous 3D lip-mesh VQ-VAE latents. After adaptation, the geometry branch is removed, and the audio-only CASE representation is fused with the baseline SSL speech representation through residual cross-attention, while retaining the motion-decoder architecture and objective. Across four VOCASET baselines, CASE reduces mean lip vertex error by 2.8-6.0%, and mean full-face vertex error decreases in every system. Under the same fusion protocol, CASE yields lower mean LVE than the unadapted AV-SE stream on all four generators, with model-dependent margins. Because adaptation uses Korean speech-geometry pairs and evaluation uses English VOCASET, the result is consistent with cross-language downstream transfer under a shared English-pretrained audio-visual prior.
[CV-151] Edge Deployable Lightweight Framework for Real-Time Driver Safety System with Integrated Lane Departure and Traffic Sign Recognition
链接: https://arxiv.org/abs/2610.25340
作者: Hassaan Ali Qureshi,Danish Hussain,Ali Hassan,Aon Safdar,Anas Bin Aqeel,Muhammad Rafiq,Zohaib Riaz,Farhan Hussain
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Lane departure and traffic-sign negligence are leading contributors to road accidents, particularly in regions with non-standardized signage and limited annotated data. Existing Advanced Driver Assistance Systems (ADAS) solutions are often developed as disjointed submodules, are rarely validated for real-time operation, and struggle to balance accuracy with low-latency inference for embedded deployment. We address these gaps with a unified, edge-deployable ADAS framework that simultaneously performs lane-departure warning (LDW) and traffic-sign detection and recognition (TSD/TSR), and validate the proposed system on automotive-class devices. Our framework combines a lightweight, lane-segmentation-driven LDW, which includes a drivable-area fallback for degraded markings, with a two-stage TSD/TSR pipeline consisting of efficient sign localization followed by fine-grained classification, designed to preserve accuracy at embedded latency. To support region-specific deployment under data scarcity, we curate a 35-class local traffic-sign dataset and introduce a generalized augmentation and balancing profile, expanding the corpus from 359 images to 8,750 samples. On an NVIDIA Jetson Orin Nano, the integrated system sustains 15 FPS end to end (32 FPS for LDW and 18 FPS for TSD/TSR). The TSD/TSR branch achieves 97.4 mAP for detection and 95.6% validation accuracy (93.1% on a local real-world set) while maintaining an approximately 13.9 M parameter footprint. Relative to recent methods that target a single task or report results only on desktop-class GPUs, our approach delivers both LDW and TSD/TSR with real-time performance on edge hardware, providing a practical pathway to low-cost, field-ready ADAS in resource-constrained settings.
[CV-152] CoRe-WM: Conditioned Response-State World Modeling for Longitudinal Treatment-Response Prediction in Breast DCE-MRI
链接: https://arxiv.org/abs/2610.25333
作者: Lin Zhao,Bowen Chen,Qiaohui Gao,Shaowen Wan,Xiang Li,Wei Liu,Tianming Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Pathologic complete response (pCR) after neoadjuvant chemotherapy is a key prognostic indicator in aggressive breast cancer, yet existing prediction approaches either rely on pretreatment imaging alone, missing the patient’s evolving response, or optimize multi-visit classifiers under direct outcome supervision, coupling representation learning with the clinical endpoint. We formulate longitudinal pCR prediction as a world-modeling problem and propose Conditioned Response-state World Modeling (CoRe-WM), which learns a latent state-transition model that predicts the next imaging state from the observed sequence. The transition function is conditioned on patient clinical profile and assigned therapy through additive shift and feature-wise modulation, enabling patient- and treatment-specific dynamics. A Factorized Response State pathway further decouples treatment-response dynamics from imaging appearance by encoding the structural prior that response manifests primarily in lesion morphology. After pretraining, all parameters are frozen and a linear readout predicts pCR at each clinical decision point. We evaluate CoRe-WM on I-SPY2 under five-fold cross-validation and assess cross-cohort transfer on I-SPY1. CoRe-WM achieves the highest discrimination among all evaluated methods at every prospective decision point, and demonstrates robust cross-cohort transfer. Ablation studies identify patient-profile conditioning and the factorized response pathway as the principal contributors. These results establish world-model-based representation learning as a promising framework for outcome prediction in oncology and beyond.
[CV-153] MPIG: Multi-angle Polarization Image Generation from a Single Polarization Observation
链接: https://arxiv.org/abs/2610.25326
作者: Jue Wang,Ruilin Huang,Liquan Shen,Zhen Tian,Xia Hua
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Polarization imaging provides informative cues for material analysis, scene understanding, and visual perception. However, acquiring complete polarization observations typically requires specialized hardware and multi-angle measurements, thereby increasing system complexity and limiting practical deployment. In this work, we address the problem of predicting the missing 45^\circ , 90^\circ , and 135^\circ polarization views from a single 0^\circ observation by leveraging a learned scene prior. To this end, we propose MPIG (Multi-angle Polarization Image Generation), a Swin Transformer-based framework that couples cross-angle feature interaction with residual decoding to predict angle-dependent intensity variations while preserving shared scene structure. Specifically, measured Stokes differences and polarization-strength-aware supervision guide the recovery of polarization modulation, orientation, and magnitude, while low-rank factorization further reduces the decoding cost. We also construct HighPol, a large-scale real-world polarization dataset comprising 5,474 groups of high-resolution four-angle polarization images collected from diverse indoor and outdoor scenes. Experimental results demonstrate improved image and polarization accuracy, achieving 29.729 dB PSNR, 0.9505 SSIM, and 21.514^\circ AoLP MAngE while requiring 85.86 M parameters and 1.126 T FLOPs. Furthermore, downstream validation on glass and semantic-segmentation tasks suggests that the generated polarization images can preserve task-relevant information, demonstrating the effectiveness of MPIG and the practical utility of predicted polarization observations for downstream vision applications.
[CV-154] SENRI: Understanding the Space-Event Nexus in Real-Life Videos
链接: https://arxiv.org/abs/2610.25309
作者: Kehan Lan,Kaining Ying,Henghui Ding,Kaipeng Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Human memories are organized in a way that connects events with their corresponding spatial contexts. An AI assistant should also be able to memorize and recall the nexus of events and spaces. Thus we introduce SENRI, a task that studies this space–event connection in long, continuous real-life egocentric video. It contains six task types in two directions: Event \rightarrow Space, which recovers spatial knowledge from event cues, and Space \rightarrow Event, which retrieves events from regions and trajectories. To evaluate this, we build SENRI-Bench, containing 500 manually annotated multiple-choice questions over 10 uninterrupted self-recorded videos with around 25.3 hours of content. Also, we evaluate three typical long video understanding paradigms: direct perception, evidence seeking, and persistent memory. The task remains challenging across all three paradigms, highlighting substantial room for improvement. Then we show two representative failure modes of existing memory methods and address these evidence gaps with SENRI-Agent, a controller that uses Target-Guided Visual Check (TGVC) to propose local visual check tasks for parallel verifiers and integrate the reports into a working memory to help with reasoning. Built upon existing memory systems, SENRI-Agent improves the performance of existing baselines without modifying their stored memories. We hope SENRI will provide valuable insights into how real-life egocentric understanding should be evaluated, reveal the gaps in existing approaches, and offer a possible direction for improvement.
[CV-155] MotionShift: Environment-Compatible Motion Transfer via Planner-Evaluator Synergistic Adaptation
链接: https://arxiv.org/abs/2610.25290
作者: Yanchen Liu,Junyao Gao,Yanan Sun,Wensheng Chen,Guangming Lu,Wenjie Pei
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing motion transfer methods excel at exact, rigid motion replication from reference videos to target scenes, whereas they often fail to account for the compatibility between the transferred motion and the target scene environment in accordance with physical laws. In this work, we propose MotionShift, a novel motion transfer framework that achieves high-fidelity motion transfer while ensuring physical compatibility with the target scene. In particular, we design a planner–evaluator iterative motion adaptation mechanism, in which a Large Language Model (LLM) acts as the planner to generate motion trajectories by leveraging its strong semantic reasoning capabilities, while a Vision-Language Model (VLM) serves as the evaluator to rectify the motion based on its superior visual perception and scene understanding. Through this planner–evaluator synergistic reasoning process, the motion is adapted to the target scene in a training-free manner. The adapted motion is then used as a generative condition for an image-to-video (I2V) model to render videos in the target scene. Extensive experiments demonstrate that MotionShift achieves high-fidelity and physically compatible motion transfer across diverse and complex scenes. Moreover, we show that MotionShift can be readily integrated into existing trajectory-conditioned I2V models, endowing them with realistic motion transfer capabilities.
[CV-156] he Explainability Illusion: Must ViTs Exhibit Patch-Decoding Interpretability?
链接: https://arxiv.org/abs/2610.25289
作者: Ronen Schaffer,Hila Manor,Tomer Michaeli
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Code can be found in this https URL
Abstract:Transformers are widely assumed to exhibit token-decoding explainability. In particular, in vision transformer (ViT) classifiers, decoding individual patch tokens has become a common diagnostic tool due to its tendency to yield human-interpretable outputs. In light of the increasing popularity of this approach, an important question is whether token-decoding explainability is an inherent property of transformers that should be expected to hold across architectures, scales, and training objectives, or merely an artifact of some implicit bias. In this paper we systematically study token-decoding explainability across diverse ViT settings, spanning architectures, scales, and training objectives, using both controlled experiments and real-world models. We find that while some models indeed exhibit aligned and interpretable token decodings, others exhibit either inverted explainability maps or a complete lack of correlation with the image semantics. As we show, these tendencies toward different explainability behaviors are not intrinsic to the architecture and can vary arbitrarily with changes in the optimization landscape. We provide theoretical support for these observations by demonstrating a setting where a single fixed toy transformer architecture can provably implement the optimal classification rule in multiple ways, which differ in their patch-decoding explainability patterns. Our results show that token-decoding explainability is not an inherent property of trained transformers, but rather an outcome contingent on the problem setting and thus should not be a-priori assumed to hold. Code can be found in this https URL .
[CV-157] Vision as Proprioception: Estimating Humanoid Robot Pose from Video
链接: https://arxiv.org/abs/2610.25288
作者: Jiaxi Jiang,Matthias Heyrman,René Zurbrügg,Marco Hutter,Christian Holz
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Code: this https URL
Abstract:Estimating a humanoid robot’s internal state from video has a wide range of real-world applications but remains unexplored. During interactions with a robot, its onboard state information may be unavailable due to privacy, ownership, or communication constraints. Meanwhile, online and AI-generated robot videos provide rich sources of data but typically lack corresponding joint configurations, limiting their direct use in robot learning. To address these issues, we present HumanoidPose, a new framework that treats vision as proprioception to recover full-body joint configurations in world space from video alone. Specifically, we combine pretrained human-pose features with robot kinematics in a unified model that predicts each embodiment’s joint configurations. We retarget human motions to sixteen different humanoid embodiments, enabling scalable learning entirely from synthetic data. To improve real-world generalization, we further leverage unlabeled online videos through self-training to develop a 2D robot-pose detector. In our evaluation, HumanoidPose improves joint-angle and joint-position accuracy over the evaluated baselines and reduces joint-angle error by 18-32% relative to the best-performing baseline on each of the three real humanoid platforms. We further show improved global localization and generalization to unseen embodiments. HumanoidPose enables visual understanding of other robots during interaction and extraction of robot motion from online and generated videos.
[CV-158] Learning Lane-Structured Conditioning Fields for Online HD Mapping
链接: https://arxiv.org/abs/2610.25284
作者: Andrea Bertogalli,Anthony Dell’Eva,Marcello Ceresini,Federico Pirazzoli,Lorenzo Cipelli,Filippo D’Addeo,Nicola Marinello,Massimo Bertozzi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:High-definition maps provide critical structural information for downstream autonomous driving tasks. Recent methods increasingly rely on external priors to improve prediction accuracy and robustness. Yet, existing prior representations expose a trade-off: dense rasters are easy to integrate but ignore certain map geometric properties, while sparse vectors preserve geometric structure but require dedicated encoders and are difficult to plug into existing architectures. Moreover, external priors can be inaccurate, costly to acquire, or difficult to maintain, motivating a structural representation that can provide geometric guidance while remaining learnable and broadly applicable. We address this limitation with the Lane Field, a dense BEV-aligned geometric representation that encodes local lane structure at each spatial location and can be either predicted from onboard observations or constructed from external map priors. Building on this representation, we introduce a plug-and-play framework that enhances online HD mapping through two complementary mechanisms: BEV feature modulation, which injects lane-structured cues into latent BEV features, and deformable attention guidance, which steers query sampling toward the underlying lane geometry. We primarily evaluate our method on OpenLane-V2, showing that conditioning BEV features and guiding deformable attention with local lane geometry enables our framework to consistently improve existing methods and achieve state-of-the-art performance on lane segment perception. Project page: this https URL
[CV-159] Holo4D: Holistic 4D Reconstruction as Geometric Control for Video Diffusion NEURIPS2026
链接: https://arxiv.org/abs/2610.25276
作者: Yushi Lan,Zeren Jiang,Koichi Namekata,Xingang Pan,Chuanxia Zheng,Andrea Vedaldi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026. Project page: this https URL
Abstract:Synthesizing a monocular video from a new camera trajectory requires preserving the observed scene geometry and motion while completing regions that become newly visible. Feed-forward 4D reconstructors recover the structure of the observed scene, whereas video diffusion models provide a generative prior for completing unseen content. We introduce Holo4D, a video reshooting framework that uses reconstructed 3D trajectories as a holistic source-view 4D scaffold. We project these trajectories into the requested views and encode their geometry and source appearance to guide a pretrained video diffusion model. Together with the source video and target cameras, the scaffold grounds generation in reconstructed observations while leaving unsupported regions to the generative prior. Experiments demonstrate the value of trajectory-based conditioning for camera control and novel-view synthesis. The framework accommodates different 4D reconstructors without retraining the video generator, allowing it to benefit from future advances in source reconstruction. More details can be found in our project page.
[CV-160] GeoART: Geometry-Aware Refinement via Test-Time Scaling for Camera-Conditioned Video Diffusion
链接: https://arxiv.org/abs/2610.25273
作者: Liudi Yang,Ziyuan Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 11 figures
Abstract:Camera-conditioned video diffusion models often suffer from inaccurate camera control and geometric inconsisten- cies because camera constraints and estimated geometry are only weakly enforced during denoising. Existing reme- dies often require costly geometry-aware fine-tuning or ad- ditional 3D supervision. We introduce GeoART, a training- free self-refinement framework that improves camera con- trollability through geometry-aware test-time scaling. At selected denoising steps, GeoART generates multiple can- didates within the Predict-and-Perturb cycle and eval- uates them using a geometry prior, without directly modify- ing the denoising trajectory. A confidence gate suppresses unreliable geometric evidence, while rank-calibrated prob- abilistic selection combines score magnitude and relative rank to choose the next refinement state. Experiments demonstrate consistent improvements in camera controlla- bility, multi-view consistency, and visual quality over exist- ing sampling and self-refinement methods, without training or modifying the pretrained video diffusion model
[CV-161] PanoDent-KD: From Generalist Knowledge Distillation to Cross-Dataset Specialist Detection and Embedded Deployment in Panoramic Radiographs WACV2026
链接: https://arxiv.org/abs/2610.25256
作者: Muhammad Waqas Ashraf,Imad Ali Shah,Ali Hassan
类目: Computer Vision and Pattern Recognition (cs.CV); Performance (cs.PF)
备注: Accepted at WACV 2026
Abstract:Reliable panoramic-radiograph detection requires consistency across datasets with different class definitions and deployment software. We present PanoDent-KD, an auditable 1280-pixel YOLO26m pipeline linking a 12-class source generalist, feature-level knowledge distillation (KD), exact classifier-row transfer, teacher-free five-class specialization, and embedded deployment. A common training and evaluation pipeline compares three source-training arms across three seeds: a supervised baseline (B1), single-teacher KD (STKD), and multi-teacher KD (MTKD). Five shared classifier rows initialize nine OralXrays-9 specialists, while the backbone, FPN-PAN neck, and dual-detection-head topology are retained unchanged. On the 2,686-image test subset, all three arms achieve macro box F1 near 0.82. B1 achieves the highest three-seed arm-level mean macro box F1 (0.8239 \pm 0.0027) and mAP_50:95 (0.6130 \pm 0.0067). STKD seed 1, selected after development but before final testing, was used for deployment. Its Jetson Orin Nano Super FP16 TensorRT engine achieves 77.03-ms median and 80.86-ms P95 prediction-call latency, throughput of 12.90 images/s, and gross total-module energy of 1.39 J/image, with high numerical agreement between within-device PyTorch and TensorRT FP16 inference. These findings demonstrate the feasibility of an auditable, device-specific deployment workflow, but do not establish clinical validity or improved target-domain accuracy from KD.
[CV-162] Persistent Representational Misalignment Induces Vision-Language Model Failures
链接: https://arxiv.org/abs/2610.25252
作者: Yonatan Gideoni,Yoav Gelberg,Tim G. J. Rudner,Yarin Gal
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Vision-language models (VLMs) fail at some tasks that are simple for humans—but why? We hypothesize that many such failures stem from VLMs’ difficulties translating between vision and language representations, a misalignment the Platonic Representation Hypothesis predicts should disappear with scale. This paper is composed of two parts: in the first, we show that existing vision and language representations are linearly unalignable. This misalignment is an inherent property of the data and hence expected to persist regardless of either scale or the alignment’s complexity. In the second, we demonstrate how this misalignment both leads to a predictably induced generalization failure and correlates with existing failure modes, where more representationally aligned VLM backbones accordingly exhibit fewer failures. Overall, these results imply representations will not become aligned simply given sufficient scale, providing evidence against a strong version of the Platonic Representation Hypothesis.
[CV-163] Visual Scaffolding: Toward Native Multimodal Reasoning
链接: https://arxiv.org/abs/2610.25247
作者: Run Shao,Zhaoyang Zhang,Ziyu Li,Linrui Xu,Ji Qi,Xiu Su,Yichao Cao,Chao Tao,Haifeng Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The success of chain-of-thought (CoT) in complex reasoning can be attributed to a textual scaffolding effect: explicit intermediate results can be examined, reused, and composed to support subsequent reasoning, much as scaffolding supports a building under construction. However, mainstream multimodal LLMs lack an equivalent scaffold for visual distinctions that are difficult to express in concise verbal descriptions. We therefore ask: Can native visual representations become first-class citizens alongside language, providing a visual scaffold for reasoning? We define native multimodal reasoning as reasoning in which native visual representations and language jointly carry reusable intermediate states. As a concrete implementation, the visual scaffolding mechanism we proposed lets the MLLM autonomously initiate a retrieval action at any reasoning step. A lightweight retriever selects native visual features, which are then incorporated directly into the evolving CoT and remain available to subsequent steps. To evaluate this mechanism under controlled visual demands, we introduce NMR-Bench, comprising Burden Counting and Path Tracing. With 88M added parameters (approximately 2%) on Qwen3-VL-4B-Thinking, our model achieves 66.43% and 72.50% accuracy, respectively, surpassing the evaluated GPT-5.5 and Gemini 3.1 Pro baselines. Controlled interventions show that changing selected visual content affects task accuracy and can redirect subsequent visual selection even when intervening text is held fixed. These findings show that native visual states can function as intermediate reasoning material and point toward reusable scaffolds for additional modalities.
[CV-164] GeoV2V: Geometry-Grounded Video Diffusion Model for Driving Scene Generation ECCV2026
链接: https://arxiv.org/abs/2610.25245
作者: Yuchen Xi,Yuhan Guo,Chenwei Hou,Zixu Liu,Jin Fang,Jason Liu,Ruigang Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published at ECCV 2026. Author accepted manuscript. Project page: this https URL
Abstract:Novel view synthesis under trajectory changes is essential for autonomous driving simulation, yet existing methods struggle to generate consistent videos when extrapolating to unseen viewpoints such as lane shifts. Reconstruction-based approaches maintain geometric consistency but degrade in visual quality under large viewpoint changes, while reconstruct-then-restore methods often produce visually plausible frames yet suffer from geometric and temporal inconsistencies. We present GeoV2V, a geometry-grounded video-to-video diffusion framework that synthesizes temporally coherent driving videos along novel trajectories by integrating LiDAR and depth-derived geometric priors with full-video conditioning in the Wan 2.1 backbone. A key challenge in learning cross-lane transformations is the lack of synchronized multi-trajectory supervision in real-world datasets. To address this, we also introduce Para4D, a synthetic dataset capturing synchronized multi-lane driving videos within dynamic scenes, which provides explicit supervision for lane-shift generation and serves as a strong training prior for geometry-consistent view synthesis. Experiments on Waymo, nuScenes, and Para4D demonstrate state-of-the-art performance in visual quality, geometric consistency, and robustness to complex lighting dynamics under large trajectory shifts. Project page: this https URL
[CV-165] Evaluating Cloud and Self-Hosted OCR Systems for Arabic English and French Documents: Bounding Box Detection Text Recognition and Mathematical Expression Recognition
链接: https://arxiv.org/abs/2610.25235
作者: Jumana Alsubhi,Ahmed Alhusayni,Abdulrahman Gharawi,Israa Hamdine,Khadija Alansari,Alshaymaa Allahim,Lamees Alhumaid,Ahmad Shabana
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Choosing an OCR system for a multilingual archive requires balancing accuracy against deployment constraints. This study compares cloud and self-hosted systems on three tasks: bounding box detection on 48 Arabic, English, and French pages, text recognition on 315 pages across nine document types, and mathematical expression recognition on 83 English and Arabic samples. Detection was assessed by page-level visual grades, text by character error rate (CER) against a Gemini-generated reference transcription, and mathematics by scores against known answers. Main estimates are accompanied by 95% intervals and paired comparisons. Google Document AI and Surya produced Good line boxes on 47 of 48 (98%) and 46 of 48 (96%) pages. Datalab, Azure Document Intelligence, and Document AI had mean CERs of 0.145-0.152, with no difference detected in unadjusted paired comparisons. Surya had the lowest mean CER among self-hosted systems (0.199). No mathematical test had a statistically clear leader. Strict checks showed that near-perfect page scores could conceal a replaced Arabic variable or Eastern Arabic digits rewritten in Western form. Reliable evaluation therefore requires uncertainty estimates, structural checks, and notation-specific tests in addition to aggregate recognition scores.
[CV-166] Map2Vid: Embodied Navigation Video Generation with Human-Friendly Motion Control
链接: https://arxiv.org/abs/2610.25223
作者: Yulin Dai,Yuqian Fu,Tianwen Qian,Zhen Xing,Bojia Zi,Xiaoling Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Controllable video generation has recently achieved impressive visual quality, yet existing control paradigms remain poorly suited to embodied navigation. Text prompts are spatially ambiguous, while camera-trajectory control relies on low-level motion specifications that are unintuitive for human users and difficult to manipulate. In this work, we introduce Map2Vid, a novel task that reformulates embodied navigation video generation as a BEV-map-to-video synthesis problem, together with a large-scale benchmark for systematic evaluation. By using BEV maps, which provide a visual representation of global scene layout and intended navigation routes, Map2Vid enables human-friendly navigation video generation. To support this task, we construct a dedicated dataset with over 100K navigation sequences across 61 indoor scenes, each paired with egocentric videos, BEV maps, camera trajectories and language instructions. We further develop a unified framework that integrates multiple control signals, together with a curriculum training strategy that progressively weakens non-map inputs, enabling flexible inference under both full conditioning and map-only input. Extensive experiments across multiple foundation generation models demonstrate that controllable navigation video generation via BEV map is a viable and promising direction.
[CV-167] Image Dehazing as In-Context Tabular Inference
链接: https://arxiv.org/abs/2610.25142
作者: Le-Anh Tran
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, 8 tables
Abstract:Image dehazing is traditionally addressed through handcrafted physical priors or task-specific deep learning models, often requiring carefully designed architectures or extensive model training. Recent advances in tabular foundation models (TFMs), however, offer an alternative learning paradigm in which continuous relationships can be inferred directly from contextual examples without task-specific model training. Motivated by this capability, this paper investigates whether single-image dehazing can be effectively reformulated as an in-context tabular regression problem and introduces DehazeTab, a framework that estimates atmospheric transmission from structured image-patch representations. Specifically, overlapping patches extracted from hazy images are transformed into tabular samples using handcrafted haze-related descriptors, pretrained visual embeddings, or their hybrid combinations, while corresponding transmission patches are summarized as scalar regression targets. A pretrained TFM then performs in-context regression using labeled reference patches to predict transmission values for unseen images without any further fine-tuning. These predictions are spatially aggregated into a transmission map, refined through guided filtering, and incorporated into the atmospheric scattering model to recover haze-free images. Experiments on benchmark datasets demonstrate competitive restoration performance and reveal that physically motivated handcrafted features can outperform pretrained visual embeddings, while hybrid representations offer complementary benefits. Further analyses examine the effects of feature representations, patch selection, and contextual reference configurations on restoration quality. Beyond image dehazing, the findings highlight the potential of TFMs as an alternative paradigm for addressing low-level vision tasks through structured and in-context inference.
[CV-168] Reflection-Guided Dynamics Rectification for Text-to-Image Generation
链接: https://arxiv.org/abs/2610.25139
作者: Shuo Xu,Hanrui Ying,Debin Liu,Jianjun Li,Zhiyuan Ma
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: This paper is 20 pages long, including the appendices, and contains 10 figures. Interactive demos, code, and the dataset are available on the project page: this https URL
Abstract:Fine-grained rectification of existing generations under a limited additional budget remains a key problem in test-time optimization (TTO) for image generation. As for prior approaches, they could identify residual errors, but could not directly ensure that the model acts on the feedback during reasoning. Such errors stem from multiple sources, such as the initial noise or text prompt, and ultimately induce an execution gap of failing to handle potential deviations along the trajectory generating. However, classic models generally cannot remedy this gap. To narrow it, we propose Reflection-Guided Dynamics Rectification (RGDR), which turns reflections on completed outputs into corrective interventions on the process of sampling dynamics at the trajectory level. Specifically, RGDR could impact every sampling step by adapting to instantaneous intervention. Moreover, RGDR generates flexible candidates while adjudicating between them and the reference. As experiments show, rectification from RGDR starts from the existing base trajectory and reflection, and RGDR is substantially more effective than former trajectory exploration. Under matched budgets for additional image generation, RGDR’s gain in the cross-model average final checklist score over the reference is 1.51\times that of TIR prompt refinement and 1.81\times that of Dynamic CFG.
[CV-169] HullProbe: Measuring Interventional Part Reliance in Maritime Vessel Classifiers WACV2027
链接: https://arxiv.org/abs/2610.25137
作者: Faisal Mohammad,Mohamed Younis
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted to WACV 2027 (Evaluation Dataset Track). 20 pages including supplementary material, 12 figures, 15 tables. Code and data: this https URL
Abstract:Recognizing a vessel’s type from an image supports maritime surveillance and sanctions monitoring, but a classifier can reach high accuracy by leaning on the generic hull instead of the more distinctive superstructure, and is then fooled when a vessel is re-hulled or occluded. De-biasing methods claim to remove such reliance, yet the field certifies them with worst-group accuracy, a proxy that needs group labels and only infers reliance from accuracy. We measure the reliance directly. HullProbe swaps a registered anatomical part of a vessel for another instance’s and reads how much that part drives the decision, using a same-class swap to control for compositing artifacts. Because a surface vessel is a hull and a superstructure arranged along one axis, its parts form a 2x3 grid whose 2^6 = 64 swaps we evaluate exhaustively, and a single Walsh transform returns each part’s interventional influence and its interaction spectrum. On planted shortcuts, where the true reliance is known, the measure tracks worst-group accuracy (r = -0.94 over 45 configurations), and replacing its learned segmenter with human annotations changes the hull-share by less than the audit’s resolution. On the 26 vessel types of MARVEL, the classifier places about 46% of its corrected sensitivity on the generic hull yet follows the superstructure when the two conflict, and the group-label-free de-biasers available there do not detectably reduce the hull’s share. The findings hold across three architectures and a second maritime dataset, and the method extends to fine-grained vehicle recognition on Stanford Cars.
[CV-170] LC-TSRNet: A Land-Cover-Guided Two-Stage Network for Annual MODIS Surface Reflectance Time Series Reconstruction
链接: https://arxiv.org/abs/2610.25120
作者: Wenjuan Zhang,Xuesong Ma,Zhen Li,Meixiang Wang,Caimeng Yi,Jie Pan,Nan Su,Bing Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:MODIS surface reflectance (SR) products often suffer from data gaps due to cloud contamination and sensor malfunctions, limiting their applications in long-term and large-scale studies. Most existing methods rely on cloud-free observations from optical time series or introduce SAR time series as auxiliary information. However, due to the coarse spatial resolution of MODIS, cloud-free reference images are difficult to obtain, and SAR imagery is not well matched to the spatial resolution of MODIS, making it difficult to provide effective prior constraints. To address these issues, this study proposes LC-TSRNet for seamless reconstruction of annual MODIS SR products. In the first stage, an annual land cover map is introduced as prior knowledge, and CSPM is designed by leveraging the spectral similarity within identical land cover types, thereby providing prior spectral constraints for missing regions. In the second stage, to address large-area, persistent, and overlapping gaps in MODIS imagery, SMRM is designed to adaptively exploit and integrate multi-scale spatiotemporal contextual information. Meanwhile, to address the limited collaborative utilization of spatio-temporal-spectral information, 3DAM is designed to impose cross-dimensional joint constraints. Considering the lack of publicly available MODIS sample datasets, this study constructs the AIRMOD-TSR dataset. Extensive experimental results show that LC-TSRNet outperforms the comparison methods and achieves better reconstruction accuracy and spectral fidelity. Under the spatially adjacent scenario, LC-TSRNet improves PSNR by 0.56 dB and reduces SAM by 0.13°. The vegetation monitoring application further verifies the practical value of the reconstructed SR time series.
[CV-171] hink Visually Across Space and Time: Grounded Reasoning for Video Segmentation
链接: https://arxiv.org/abs/2610.25090
作者: Junwei Wen,Deshui Miao,Guangming Lu,Ming-Hsuan Yang,Xin Li,Wenjie Pei
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 21 pages
Abstract:Video Reasoning Segmentation (VRS) requires a model to identify an implicitly described target and predict its pixel-level mask trajectory, while resolving the spatial relations and temporal events that determine the answer. Text-only Chain-of-Thought (CoT) can explain a target choice without making the underlying visual references explicit, leaving a \emphvideo reference gap between linguistic inference and the evidence distributed across frames. We introduce \textbfTemporal Chain of Video Primitives (TCoVP), a spatiotemporally grounded form of Video CoT that organizes query-relevant observations into a chronological reasoning chain. Each primitive associates an entity reference with temporal support and timestamped spatial evidence, and the chain interleaves these references with language to guide reasoning about which entity is relevant, when the queried event holds, and where its evidence appears. The selected target primitives then provide box prompts and temporal support for video mask prediction. We construct TCoVP-SFT-15k and TCoVP-RL-13k to learn this representation through supervised fine-tuning and chain-aligned reinforcement learning. Extensive experiments across five VRS benchmarks and three model backbones demonstrate the strong performance of TCoVP, highlighting the effectiveness of organizing spatial and temporal evidence within the reasoning process. All models and code will be publicly released. Project page: \urlthis https URL
[CV-172] Sufficient Frame Subsets for Video Understanding
链接: https://arxiv.org/abs/2610.25084
作者: Lahiru Wijayasingha
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Understanding when during a video a model makes its critical observations is essential for interpreting video-understanding systems. However, existing attribution methods often spread evidence diffusely across near-duplicate frames. We introduce Sufficient Frame Subsets: compact sets of frames in which all other frames are masked, yet each subset independently preserves the model’s prediction. We evaluate our method on two complementary benchmarks, UCF101 and Something-Something v2, using both CNN- and transformer-based architectures. Our results show that the proposed algorithm identifies highly compact and effective frame subsets for explaining model decisions. We also find that models capable of relying on a greater number of sufficient frame subsets tend to perform better, possibly because they are more robust, consistent with prior research. Finally, the extracted frame subsets help us analyze model failure cases, reveal which aspects of a video the model has learned to emphasize, and identify opportunities for improving future training procedures.
[CV-173] MoCKEFA: Model-Conditioned Keyframe Extraction and Fair Attribution for Video Model Explanation
链接: https://arxiv.org/abs/2610.25075
作者: Lahiru Wijayasingha
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We show that a video can be represented with just a few keyframes without sacrificing a given model’s belief about the video. Our method is the first of its kind that extracts these keyframes conditioned on the model’s prediction, which is massively helpful for building explainable video models. We then fairly distribute the importance of the video to the model among the keyframes with game theoretically fair Shapley values. Through qualitative and quantitative experiments, we show that keyframe attribution is superior to the traditional frame attribution. Further, we conduct preliminary exploration on highlighting the important changes among the frames by looking at the differences between adjacent keyframes.
[CV-174] MoWE: Spatially Adaptive Mixture-of-Wavelet-Experts Pooling for Ultrasound Lesion Segmentation WACV2027
链接: https://arxiv.org/abs/2610.25074
作者: Reaz Hassan Joarder,Sidratul Tanzila Tasmi,Alif Arshad Bakshi,Md. Bakhtiar Hasan,Md Hasanul Kabir
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:We present Mixture-of-Wavelet-Experts (MoWE), a spatially adaptive routing mechanism that replaces max-pooling and strided convolutions in U-Net architectures for ultrasound lesion segmentation. Standard pooling operations suppress high-frequency spatial details, including potential boundary cues in speckled regions. Recent frequency-domain methods preserve some of these details, but a single wavelet family applies the same spectral bias across the image. MoWE instead uses a context-aware router to assign local regions to fixed wavelet families according to local texture and global feature context. The decoder reuses the encoder’s spatial routing maps during the inverse transform. Learnable channel-wise gates rescale the high-frequency subbands before reconstruction. We evaluate MoWE with U-Net on four ultrasound lesion benchmarks (BUSI, TN3K, UDIAT, BUS-UCLM) and one cardiac dataset (CAMUS), and with nnU-Net on BUSI. On BUSI, MoWE improves U-Net Dice from 72.59% to 74.55% and reduces HD95 from 47.01 to 43.49,px. With nnU-Net, MoWE improves Dice from 80.18% to 81.97% and reduces ASSD from 20.07 to 17.25,px. On U-Net, MoWE reduces trainable parameters by 6.4%.
[CV-175] Localization and Neural Splatting on the Edge: An FPGA Scene Memory for Robots
链接: https://arxiv.org/abs/2610.25068
作者: Que Haohua,Mingkai Liu,Handong Yao
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages
Abstract:Resource-constrained robots and autonomous vehicles need scene memories that localize a camera and render nearby views without a desktop GPU. Scene-coordinate regression (SCR) and 3D Gaussian splatting (3DGS) fit because they avoid an SfM stage, but ACEsplat-style pipelines are GPU-centered, and moving them to an edge FPGA is hard because SCR is pose-sensitive and 3DGS rendering is irregular and capacity-sensitive. We present LENS, an edge scene-memory acceleration system on the AMD KR260 that brings an ACEsplat-style workload to a low-power FPGA/ARM platform. Rather than proposing a new scene model, LENS co-designs the two dominant kernels with the board data path: an INT16 SCR accelerator that preserves pose accuracy, and a bit-exact fixed-point ordered 3DGS renderer that rewrites alpha transmittance as a log-sum-exp recurrence and removes provably useless streamed Gaussians. All LENS latency, power, and quality results are board-measured. The SCR engine preserves DSAC* localization accuracy at 25.42 ms, the on-board renderer reaches 34.11 dB on Cambridge against ACEsplat’s 33.20, novel-view synthesis on RealEstate10K reaches 27.84 dB against 26.11, and all ten Wayspots scenes reach the ACEsplat rendering target. LENS wins performance per watt over a Jetson Orin NX on every deployment task, by 1.9x to 5.7x at about a quarter of the power, and is faster in absolute latency on localization and Wayspots rendering. On-device training reaches ACEsplat-level quality but not the GPU’s time or energy, marking the remaining boundary for a future training accelerator.
[CV-176] PRET is a few-shot system for pan-cancer recognition without example training
链接: https://arxiv.org/abs/2610.25054
作者: Yi Li,Ziyu Ning,Tianqi Xiang,Qixiang Zhang,Zhihao Lin,Min Yi,Feiyan Feng,Baozhen Zeng,Xuexia Qian,Lu Sun,Jiace Qin,Ling Xiang,Chao Fan,Tian Qin,Qian Wang,Xiu-Wu Bian,Kun-Hsing Yu,Kang Zhang,Qingling Zhang,Xiaomeng Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Nature Cancer Accepted
Abstract:Pathological examination stands as the cornerstone in cancer diagnosis, impacting millions worldwide annually. With the shortage of pathologists globally, artificial intelligence (AI) has emerged rapidly to automate the diagnostics process. However, conventional AI models require substantial labeled data for each disease, posing huge challenges in scalability and practicality. Therefore, we introduce PRET (pan-cancer recognition without examples training), a few-shot system to achieve flexible, scalable, and effective cancer recognition across diverse organs, hospitals and tasks without training. Evaluated on 23 international benchmarks comprising 4,484 whole-slide images, our method outperforms existing approaches across 20 tasks, achieving over 97% area under the curve on 15 benchmarks with a maximum improvement of 36.76%. Notably, PRET delivers clinical-grade diagnostic performance in lymph node metastasis detection using only eight slide examples, outperforming 11 pathologists. By offering a flexible and cost-effective solution for pan-cancer recognition, PRET paves the way for accessible and equitable AI-based pathology systems, particularly benefiting minority populations and underserved regions.
[CV-177] IECLA: Interpretable Emotion Contribution Learning for Affective Image Editing ICLR2027
链接: https://arxiv.org/abs/2610.25015
作者: Yijie Guo,Dexiang Hong,Weidong Chen,Xinyan Liu,Zhendong Mao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 31 pages, 10 figures, 24 tables; under review at ICLR 2027
Abstract:Affective image editing (AIE) aims to evoke a target emotion while preserving the content and structure of a source image. Existing methods improve emotional expression or visual fidelity, but the target emotion alone does not reveal which source regions should be changed and which should remain intact, especially in complex scenes. We introduce Interpretable Emotion Contribution Learning for Affective Image Editing (IECLA), a framework that separates source-emotion attribution from target-aware editing. IECLA constructs signed object-level supervision through counterfactual interventions and trains a context-conditioned contribution predictor to localize evidence that supports or opposes the source emotion. At inference, an image-level evaluator estimates the source emotion directly from the input, while the predictor identifies editable and preservation evidence without repeated interventions. A target-aware editing agent then translates this evidence and the desired emotion into localized, content-aware operations for existing editors. Extensive evaluations show that IECLA predicts counterfactual contributions more faithfully on both in-domain and cross-dataset tests, while external VLM and independent human references support the quality of its localized evidence. Across six editing backbones, IECLA consistently strengthens target-emotion expression while preserving image content and structure, and user studies further show improved perceived target expression and naturalness.
[CV-178] FineSplat: Fine Patches Local Groups Detailed Splats
链接: https://arxiv.org/abs/2610.25011
作者: Hyebin Kim,Hwasik Jeong,Joonhyuk Park,Gyeongjin Kang,Jungwoo Kim,Younghyun Kim,Eunbyung Park
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Feed-forward 3D Gaussian Splatting (3DGS) promises instant scene reconstruction with high-quality novel-view synthesis. Yet, fine-grained geometry and appearance remain difficult to recover, motivating us to revisit how feed-forward models represent, aggregate, and refine detailed visual information. We introduce FineSplat, a feed-forward 3DGS framework built around fine-grained image patches. Since global multi-view attention becomes prohibitively expensive at such high token resolutions, we develop a dual-grouping strategy that combines two complementary forms of locality: view-wise grouping enables efficient interaction among neighboring views, while 3D spatial grouping reconnects geometrically related tokens that may never have interacted under view-wise grouping by lifting them into the 3D scene space. We further leverage reconstruction error as direct feedback, providing an explicit rendering-based signal to recover fine-grained structures and appearance, thereby overcoming the limitations of one-pass feed-forward prediction. Finally, a compact Gaussian generation scheme preserves the benefits of fine-grained features without unnecessarily increasing the number of primitives. Together, these designs enable FineSplat to recover richer geometry and appearance while remaining efficient and compact, achieving state-of-the-art results by a substantial margin, under the same or smaller Gaussian budgets than existing methods.
[CV-179] ReTrace: Visual Backtracking for Efficient Multimodal Large Language Models EMNLP2026
链接: https://arxiv.org/abs/2610.25005
作者: Guangzong Si,Dong Wang,Yifan Yu,Miao Liu,Panwang Pan,Wentao Zhu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to Findings of EMNLP 2026, forthcoming
Abstract:Multimodal Large Language Models (MLLMs) encode images into dense visual token sequences, creating substantial attention and KV-cache overhead. Existing visual token compression methods reduce this cost by pruning low-saliency tokens in shallow layers, but this practice couples temporary computational deactivation with irreversible information deletion. We show that this coupling breaks an intrinsic inference pattern in MLLMs. Through layer-wise attention analysis, we identify \textbfVisual Backtracking: visual pivots remain spatially stable, yet in middle layers they actively retrieve context from tokens that appeared unimportant earlier. Thus, early low-saliency tokens can be inactive for computation while still being useful for later reasoning. Motivated by this finding, we propose \textbfReTrace, a backtracking-aware compression framework that introduces a dormant token state. ReTrace caches non-pivot tokens outside expensive layer-wise computation, restores them during the backtracking window, and prunes only after their information has been absorbed. Experiments on Qwen2.5-VL and InternVL3 demonstrate nearly 90% reductions in FLOPs and KV cache, a 2.83 \times prefilling speedup, and strong performance retention across complex vision-language benchmarks.
[CV-180] UtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation Models NEURIPS2026
链接: https://arxiv.org/abs/2610.25001
作者: Yunbei Zhang,Janet Wang,Xi Xiao,Jihun Hamm,Xiao Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026. Code and project page: this https URL
Abstract:Weather transformers process dense reanalysis fields whose reliability varies across variables, locations, and weather regimes. Token merging can reduce computation, but feature similarity alone does not account for this heterogeneous input uncertainty. We introduce UtoMe, an observation-uncertainty-guided token merging framework that combines dynamics saliency with ensemble data assimilation spread as a relative input-reliability proxy. The resulting importance score protects selected tokens as anchors and calibrates bipartite matching between the remaining tokens. Three variants use external uncertainty at both training and inference, during training only, or neither stage. In ERA5 72-hour forecasting at two spatial resolutions, UtoMe improves Overall error over train-time ToMe at matched token budgets. At 1.40625^\circ and 50% token reduction, UtoMe-E lowers Overall error from 67.32 to 53.00 at the same 0.55x encoder FLOPs. Small reductions remain close to the full-token baseline; larger reductions incur measurable accuracy losses. Component ablations and spatial uncertainty controls support combining dynamics with aligned uncertainty when allocating a weather transformer’s token budget.
[CV-181] MultiAltitudeSR: Multi-Altitude Agricultural UAV Image Super-Resolution via Physical Ground Sampling Distance Conditioning
链接: https://arxiv.org/abs/2610.24992
作者: Xintong Jiang,Yu Tian,Yixue Liu,Valerio Hoyos-Villegas,Viacheslav Adamchuk,Shangpeng Sun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 12 figures
Abstract:Unmanned aerial vehicles (UAVs) enable high-throughput platforms for crop monitoring but face a fundamental trade-off between flight altitude and image spatial resolution. Many existing super-resolution (SR) methods for UAV imagery rely on oversimplified synthetic degradation models and are restricted to fixed integer upsampling scales, limiting their ability to handle real-world degradations and varying flight altitudes. This study proposes MultiAltitudeSR, a framework that conditions feature modulation and reconstruction on source and target ground sampling distances (GSDs). MultiAltitudeSR trained jointly on real UAV images at altitudes of 10 to 70 m above ground level supports all 21 source-to-target altitude pairs without pair-specific fine-tuning, covering scale factors ranging from 1.18 \times to 6.03 \times . The framework can be extended to additional source-target GSD combinations by incorporating corresponding image pairs into training. On the 20-to-10 m pair, the model achieved competitive perceptual performance compared with pair-specific baselines, with LPIPS of 0.2236, DISTS of 0.2431, and NIQE of 8.9020. Experiments using the same architecture demonstrated the value of acquiring real-world multi-altitude UAV image pairs for SR training, with real-pair training achieving better average reconstruction performance across the 21 altitude pairs than training with the tested synthetic degradation pipeline. Downstream zero-shot segmentation case studies illustrated task-dependent benefits of super-resolved imagery for crop monitoring, suggesting potential UAV flight-time savings through image acquisition at higher altitudes.
[CV-182] Visual Sufficiency Landscapes: Understanding Query-Conditioned Reliability in Visual Token Pruning
链接: https://arxiv.org/abs/2610.24968
作者: Zhenghan Li,Kexin Sun,Wei Zheng,Peng Hu,Xiao Wei,Anbo Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Different visual token pruning strategies retain different subsets while maintaining similar task performance. This motivates a combinatorial question: how many visual subsets can preserve model behavior, and how does their abundance depend on the query? We introduce visual sufficiency landscapes to characterize this abundance. At a fixed layer and budget, visual sufficiency probability is the fraction of uniformly sampled subsets that jointly preserve the dense model’s answer, reference-answer likelihood, and output distribution. Independent sampling replicates separate input variation from estimation noise. Across Qwen2.5-VL-7B, InternVL2-8B, and LLaVA-v1.5-7B, differences between inputs on GQA and TextVQA exceed replication error, and different questions about the same GQA image produce distinct landscapes. Sufficiency probability also improves failure prediction for full pruning methods. On LLaVA, adding this probability increases AUC by 0.1151 for FastV, 0.0683 for VisionZip, and 0.1401 for SparseVLM, while reducing log loss and Brier score. Several architecture-adapted configurations on the other two models reproduce these gains. Under controlled selection rules, predictors trained on the original images reduce log loss on new images by 9.0%, 13.0%, and 14.1%, respectively. These results establish successful-subset abundance as a measurable dimension of local prunability and connect it to the identification of high-risk pruning conditions.
[CV-183] Meta Inverse Prompting for Video Generative Models NEURIPS2026
链接: https://arxiv.org/abs/2610.24965
作者: Yerin Jung,Jongheon Jeong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026. Project page: this https URL Code: this https URL
Abstract:Recent advances in video generative models have enabled flexible conditional generation and editing pipelines, such as text-to-video, image-to-video, and reference-guided synthesis. However, these workflows are inherently one-directional, mapping conditions to videos without providing a mechanism to invert this process. In complex media production settings, the ability to recover controllable textual conditions from a given video is highly desirable, yet remains underexplored. While inverse prompting has been studied for image generation, directly extending these approaches to video is challenging due to the need to capture temporal dynamics beyond static appearance, as well as the substantial computational cost of per-video optimization. In this work, we propose the first practical framework for efficient video inverse prompting. Specifically, we formulate inverse prompting as a meta-learning problem and introduce Meta Video Inverse Prompting (MVIP), which learns a meta-prompt optimized over temporal dimensions to enable fast and effective prompt inference. Our approach amortizes the inversion process, eliminating the need for costly iterative optimization at test time. Experimental results demonstrate that MVIP recovers higher-quality inverse prompts with significantly improved reconstruction fidelity compared to image-based inversion and iterative captioning baselines, while achieving substantial gains in efficiency. Code and video results are available at this https URL.
[CV-184] LowNet: Looped Wavelet Neural Networks with 24 Learnable Parameters for Solving Poisson Equations
链接: https://arxiv.org/abs/2610.24941
作者: Yuanhao Gong
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV); Analysis of PDEs (math.AP); Numerical Analysis (math.NA)
备注:
Abstract:Poisson equations frequently appear in various fields, such as mathematics, physics and informatics. Solving Poisson equation, however, is challenging. Traditional methods are not data-specific, although have high certainty. They require a large number of iterations due to the CFL numerical stability. In contrast, deep learning methods are data-driven and more efficient, but require accurate training data and suffer from heavy computation. To tackle these issues, we propose a wavelet-inspired Poisson solver with 24 trainable scalars. The proposed network has several advantages. First, it is compact. It has four independent 5\times5 kernels, each represented by six parameters, leading to 24 scalars in total. Such compact format can be deployed on most devices. Second, thanks to the loop design, it achieves higher accuracy than previous methods. A residual loop reuses the wavelet block without increasing its parameters. Numerical experiments confirm that such loop indeed improves the solving accuracy. On 10000 images, one, two, and three loops of the wavelet block lead to PSNRs of 47.27 , 74.14 , and 100.74 dB, respectively. Third, the proposed method is computationally fast. Thanks to the wavelet transform, the proposed method has linear computation complexity, leading to high computation performance. In short, we propose a novel looped wavelet neural network with compactness, accuracy and efficiency for solving Poisson equations. Our method can be applied in various fields, such as computational fluid dynamics, image reconstruction, biomedical engineering, electromagnetic field simulation.
[CV-185] Multi-Speed Field Evaluation of a Semi-Autonomous Precision Sprayer for Weed Control in Vegetable Crops
链接: https://arxiv.org/abs/2610.24893
作者: Md Khairul Islam,Boyang Deng,Yuzhen Lu,Daniel Brainard,Ajit Srivastava
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 16 pages, 5 figures, 7 tables
Abstract:Machine vision-guided precision spraying is promising for sustainable vegetable production by enabling site-specific weed targeting, reducing herbicide application while protecting crop yield and quality. However, the effects of operating speed of sprayer systems on the integrated detection-tracking-actuation pipeline remain insufficiently characterized under field conditions. This study evaluated a semi-autonomous, tractor-based spraying system for targeted weed control in radish and quantified performance across three operating speeds. A five-season dataset comprising 10,599 images and 84,311 annotated instances across seven classes was curated for model development. A YOLOv11-medium model was integrated with ByteTrack tracking and a customized 12-nozzle sprayer on a semi-autonomous electric tractor. Field trials were conducted at 0.13, 0.27, and 0.45 m/s (i.e., 0.3, 0.6, and 1.0 mph) to characterize speed-dependent changes in weed detection, targeting, actuation, and crop damage. At the lowest speed of 0.13 m/s, the system achieved 94.6% mAP@50 and 93.2% recall for weed and crop detection. However, increasing platform speed substantially reduced system performance, with weed hit rate declining from 78.3% to 35.0% and crop contact rates increasing from 20.1% to 28.7%. Consequently, the system did not meet predefined commercial-viability thresholds. These results demonstrate that practical deployment requires improvements beyond weed detection alone, particularly coordinated improvements in motion-robust plant perception, spray-timing compensation, and crop-adjacent actuation precision. The identified performance bottlenecks provide specific engineering targets for advancing vision-guided precision spraying toward practical, high-speed vegetable weed management. The software programs developed in this study have been made publicly available to support future precision weeding research.
[CV-186] Goal Images Are Control Targets: Spatially-Grounded Goal-Image Synthesis for Robotic Manipulation WACV2027
链接: https://arxiv.org/abs/2610.24885
作者: Riddhi Chatterjee,Ritish Shrirao,Kushagra Mehrotra,Viswanath Gopalakrishnan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027. Project page: this https URL
Abstract:Goal-conditioned manipulation policies are steered by a goal image, a picture of the desired post-manipulation scene synthesized from the current observation and a language instruction. The prevailing assumption is that better image-editing quality yields better goals, so goal generators are built and compared using general image-editing metrics. We show this assumption is false. Across two independent goal-conditioned policies on CALVIN, the image editors that top aggregate editing-quality leaderboards (Gemini 3 Pro Image and GPT Image 1.5) produce worse control targets than our proposed far smaller finetuned open models, and the gap widens sharply for the multi-step (long-horizon) tasks robots actually face. A per-dimension analysis explains why. Downstream policy success is governed almost entirely by manipulation adherence (selecting the right object and the right final state; Spearman \rho \approx 0.9 with success), while photorealism and aesthetic fidelity, precisely where large proprietary editors excel, are uncorrelated with success ( \rho \approx 0 ). This dissociation holds on both CALVIN and the real-world-aligned SIMPLER-Bridge benchmark. We then present SGE-Goal, a spatially-grounded decomposed editing framework that predicts initial and final 2D edit-region masks before goal-image synthesis. SGE-Goal turns open base models (FLUX.1 Kontext, FLUX.2 klein) into the strongest goal generators we evaluate on CALVIN, and second only to Gemini 3 Pro Image on SIMPLER-Bridge. Base models fine-tuned with SGE-Goal also hold up under chaining far better than the best proprietary editor, sustaining roughly seven times the success rate at the longest (five-step) chains. We release code, datasets, checkpoints, and a human-aligned diagnostic evaluator. Our results argue that goal generators must be selected by downstream utility, not by editing-quality leaderboards.
[CV-187] SpriteStage: Long Horizon Animation Generation via Progressive Scene Programming and Replay
链接: https://arxiv.org/abs/2610.24869
作者: Zhaoqi Wang,Zijian Zhang,Jiamou Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Video generation models synthesize visual sequences from text and image inputs. Extending video generation to long animations requires actions, dialogue, camera coverage, and narrative developments to remain coordinated across scenes. Executable scene programs make these decisions available for inspection. Constructing an entire story before observing its execution can leave later scenes dependent on unrealized events. We present SpriteStage, a progressive scene programming framework that builds animation through iterative enactment and engine reorchestration. Iterative enactment generates each scene from an accepted world checkpoint and narrative snapshot, executes it on a shared spatial stage, and uses runtime and visual feedback to revise the relevant actions or story beats. The accepted outcome becomes the starting state for the next scene. Engine reorchestration maps the resulting draft to production assets and engine capabilities while retaining its events, geometry, and timing. Inspection of the assembled sequence and frozen replay carry the accepted performance into the final video. Three complete animations, ranging from approximately 9.9 to 14.3 minutes, demonstrate the framework’s effectiveness in producing extended narratives through scene programming and engine execution.
[CV-188] Arbitrary-Scale Anisotropic 3D MRI Super-Resolution with Gaussian Prototypes
链接: https://arxiv.org/abs/2610.24858
作者: Aditya Rastogi,Ramon Leiser,Prateek Singh,Martha Foltyn-Dumitru,Martin Bendszus,Fabian Bamberg,Klaus Maier-Hein,Kai Schlamp,Alexander Radbruch,Gianluca Brugnara,Philipp Vollmuth
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 26 pages, 6 figures, 15 tables (including appendix)
Abstract:Reducing MRI acquisition time improves patient comfort, reduces motion artefacts, and increases scanner throughput. Clinical protocols often shorten acquisitions by sacrificing spatial resolution, particularly along the slice axis. Super-resolution can recover higher-resolution images, but existing methods are typically specialised to one scale or acquisition geometry, limiting their use across heterogeneous clinical protocols. We present GSMR3D, a continuous volumetric model that reconstructs arbitrary isotropic and anisotropic resolutions with a single checkpoint per MRI contrast. It combines a learned dictionary of anisotropic 3D Gaussian prototypes with scale-conditioned frequency encoding and an exact measurement-consistency projection, keeping reconstructions consistent with the acquired low-resolution data. We train separate models on 3,731 contrast-enhanced T1-weighted and 30,254 T1-weighted brain MRI volumes from clinical and population-based cohorts. We evaluate GSMR3D against geometry-specific specialists for isotropic, in-plane, and through-plane super-resolution using reconstruction quality and downstream tasks: brain tumour segmentation, brain-metastasis detection, and brain parcellation. GSMR3D outperforms specialists at isotropic 2\times and achieves better LPIPS, DISTS, and FID across all geometries, although specialists retain higher PSNR in anisotropic settings. It more faithfully preserves glioma segmentations across all geometries, achieves the highest lesion sensitivity on an independent brain-metastasis cohort, and the highest brain-parcellation agreement for isotropic and in-plane reconstruction. Reconstruction takes about 6s per volume, adding little overhead relative to acquisition-time savings. GSMR3D offers a flexible approach to MRI scan-time reduction with one model per contrast across geometries while preserving information relevant to downstream tasks.
[CV-189] PCHMR: Empowering and Benchmarking Human Mesh Recovery in Privacy-Constrained Real-World Settings
链接: https://arxiv.org/abs/2610.24841
作者: Zeyu Cao,Qingxuan Wu,Zhiyang Dou,Rui Xu,Yuan Liu,Javier Fernandez-Marques,Nicholas D. Lane,Taku Komura,Wenping Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: TMLR 2026
Abstract:Fine-tuning directly on user-side data is an effective way to improve the performance of widely adopted data-driven human mesh recovery (HMR) models. However, doing so for existing HMR methods often assumes aggregating user-side data from real-world HMR-deployed devices to a central training server, posing a significant privacy risk as sensitive human images are transmitted. How can human mesh recovery be evaluated and improved in privacy-constrained real-world settings? We study this question through a benchmark in which state-of-the-art HMR models are trained with federated learning and a secure-aggregation variant, neither of which centralizes raw images, and evaluated across a wide range of client-scale and data-heterogeneity settings. We document that common HMR training pipelines are built around centralized data access and quantify how representative HMR backbones behave when that assumption is removed. Furthermore, to study the data bottleneck of privacy-constrained HMR training, we propose DePoser, a local annotation and fine-tuning pipeline enhanced with depth foundation models, with which collaboratively trained HMR models can be locally tailored to the end user’s distribution. On in-the-wild data, fine-tuning on DePoser annotations improves personalization accuracy, measured against DePoser-generated pseudo-ground truth. This benchmark aims to support future work on privacy-constrained HMR models and their real-world deployment and evaluation.
[CV-190] VTF900: A 900-Subject Visible-Thermal Face Recognition Dataset and Benchmark WACV
链接: https://arxiv.org/abs/2610.24832
作者: Jacob Abaare,Bianca Ianetta,Cedric Nimpa Fondje,Shuowen Hu,Benjamin S. Riggan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027
Abstract:Despite advancements in visible-thermal face recognition, where thermal probe (query) face images are matched against visible enrollment templates, existing benchmarks are currently limited by small gallery sizes (N \approx 100 subjects). This limitation inhibits the ability to accurately evaluate visible-thermal facial recognition at scale and to report significant recognition performance metrics, which have better operational relevance. Therefore, we introduce VTF900, a large-scale visible-thermal face recognition benchmark with 900 subjects and paired imagery that includes expression, pose, and occlusion variations. VTF900 is designed to support large-scale (i.e., large gallery) evaluations across verification (1:1 matching) and closed-set identification (1:N matching) for visible-thermal face recognition. Furthermore, VTF900 specifically addresses the challenging combination of cross-spectral and cross-dataset analysis. In this paper, after training models on the ARL-VTF dataset, we provide benchmark analyses across convolutional (baseline), vision transformers, and domain-adaptive neural network architectures to assess how these methods scale. At full gallery scale (N = 900), transformer models achieve strong discriminative performance under frontal conditions (e.g., ViT-B: 97.68% AUC, 51.10% TAR@1%FAR), while domain-adaptive methods such as DPIT-ResNet50 demonstrate greater robustness under off-pose conditions (94.68%AUC, 36.08% TAR@1%FAR). Our results show that high AUC at small scale does not translate to reliable performance at deployment scale, and that pose variation significantly amplifies performance degradation. VTF900 exposes these limitations, enabling more rigorous evaluation of cross-spectral face recognition systems.
[CV-191] SL-Ground: Sentence-Level Grounding for Fine-Grained Open-Vocabulary Object Detection
链接: https://arxiv.org/abs/2610.24814
作者: Yang-Syu Li,Cheng-Huan Yang,Chia-Yung Jui,Peggy Joy Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Grounding-based open-vocabulary detectors provide a widely used framework for text-conditioned localization, with a phrase-grounding formulation that aligns object queries to individual text tokens and recovers class scores through token-to-class aggregation. We observe that this token-level scoring design becomes a bottleneck for fine-grained open-vocabulary object detection, where the target category is defined by a complete compositional description rather than by isolated tokens. Based on this observation, we propose SL-Ground, a sentence-level grounding reformulation for grounding-based open-vocabulary detection. Instead of scoring queries against fragmented token evidence, SL-Ground encodes each category description as a complete sentence-level representation and directly scores object queries against these representations. To evaluate fine-grained recognition across label granularity, we construct a multi-granularity Stanford Cars benchmark spanning vehicle type, brand, and car model. At the finest car-model granularity, SL-Ground raises AP over the token-level variant from 14.5 to 70.6 under matched training candidate sampling and to 86.0 with full-vocabulary training, while better preserving open-vocabulary capability and avoiding repeated prompt chunking. Beyond Stanford Cars, SL-Ground also improves FG-OVD performance, achieving the best average AP among compared methods. These results show that sentence-level grounding provides a more suitable scoring unit for fine-grained OVOD and offers a practical path toward fine-grained grounding-based open-vocabulary detectors.
[CV-192] SwitchSplat: Learning Discrete Object States from Casually Captured Scenes SIGGRAPH
链接: https://arxiv.org/abs/2610.24812
作者: Yizheng Xie,Yalda Foroutan,George Shramko,Daniel Rebain,Shrisudhan Govindarajan,Andrea Tagliasacchi
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: SIGGRAPH Asia 2026 (Conference Papers). 19 pages, 15 figures, 8 tables. Project page: this https URL
Abstract:3D Gaussian Splatting has made high-fidelity 3D reconstruction fast and accessible, yet existing representations assume a static world. This assumption breaks down in casual captures, where a user naturally interacts with the environment – opening doors, toggling switches, or moving objects – introducing discrete state changes that no current framework can model. Time-conditioned dynamic methods such as 4DGS conflate when a state was observed with what state the scene is in, limiting them to replaying seen sequences rather than composing arbitrary object configurations. We present SwitchSplat, the first 3DGS representation to address this gap via a base+slot decomposition: a shared static base captures the invariant scene, while each slot independently holds the two binary states (on/off) of one interactable object. Because the assignment of Gaussians to slots is fully differentiable, the model discovers the geometry of each state directly from video, without manual annotation. At inference, the user can freely combine slot states and render the resulting scene from any viewpoint. We validate SwitchSplat on a new casual-capture dataset and show that it outperforms static and dynamic baselines in novel-view synthesis quality, while uniquely supporting rendering of state combinations never seen during capture. Project page: this https URL
[CV-193] Directionally Factorized Illumination with Spherical Harmonic Rotation ICRA
链接: https://arxiv.org/abs/2610.24811
作者: Yimeng Dou,Adam M. Saunders,Andre T. S. Hucke,Michael E. Kim,Adam W. Anderson,Eric R. Tkaczyk,Bennett A. Landman,Daniel Moyer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Submitted to ICRA
Abstract:Many real-world capture scenarios involve an active light source rigidly attached to the camera: camera flashes for dark-environment imaging, underwater vehicle lights, and endoscopic/laparoscopic medical procedures. As the camera moves, the co-mounted light moves with it, causing the illumination direction at each surface point to change across views. Neural rendering methods (NeRF, Gaussian Splatting) encode this pose-dependent illumination as surface coloration, producing appearance inconsistencies and making it harder to recover the underlying scene. We present \textbfDaFISH, which can be plugged into any novel-view synthesis (NVS) pipeline such as NeRF. It factorizes per-point emitted color into surface albedo and a spherical harmonic (SH)-represented illumination factor. During training, the SH illumination is evaluated directly, enabling stable joint optimization of albedo, lighting, and geometry. \textbfDaFISH provides analytical generalization of illumination to arbitrary camera poses without retraining. Building on the closed-form SH-to-irradiance conversion, the training pipeline is fully differentiable from SH evaluation through volume rendering. Averaged over the held-out frames of the five scenes, \textbfDaFISH attains the best mean PSNR (38.91,dB), SSIM (0.9613), and LPIPS (0.0582) among all compared methods. It improves over the strongest splatting baseline as well as the closest NeRF-based baseline, while recovering the scene consistently across views compared to prior methods.
[CV-194] UniAVC: Unified Audio-Video Creation via Progressive Training and Asymmetric Scheduling
链接: https://arxiv.org/abs/2610.24791
作者: Kien T. Pham,Yuhao Liu,Chong Hou Choi,Zhenqi He,Yanghao Wang,Chenyang Lei,Rui Liu,Qifeng Chen,Long Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: Page: this https URL Code: this https URL Note: Abstract is shortened to meet arXiv’s characters limitation
Abstract:Unified audio-video creation requires a single model capable of text-conditioned joint synthesis, bidirectional cross-modal generation, and instruction-guided editing across both modalities. Existing methods rely heavily on paired corpora, which are scarce for joint generation and even rarer for editing. Consequently, they are spent learning single-branch generative and editing capabilities that abundant unimodal data could readily teach, rather than on the cross-modal coupling. Furthermore, prevailing systems apply a single shared noise schedule to both modalities, neglecting two critical limitations. First, due to the significant discrepancy in information density, video requires prolonged high noise to organize spatial and motion structure, whereas the sparser audio signal can settle earlier during the synthesis process, the dynamics a uniform schedule might compromise. Second, lockstep denoising leaves asymmetric workflows, such as conditioning on one modality to synthesize or edit another, untrained. We present UniAVC, which addresses both obstacles. It first trains each branch on unimodal dataset to full generative and editing competence, and subsequently dedicates paired data to exploit cross-modal correspondence. Modality-decoupled schedulers meanwhile warp the noise schedule independently from a shared timestep, while subtasks with one clean modality directly supervise asymmetric configurations. These settings respectively enable intent-aware editing, which preserves one modality intact while enlisting it to constrain the other, and post-inference self-refinement, which regenerates only an unsatisfactory branch against an accepted one. UniAVC thus spans multiple creation tasks from a single set of weights without task-specific adapters.
[CV-195] Video-Language Entailment for Video Similarity Estimation EMNLP2026
链接: https://arxiv.org/abs/2610.24786
作者: Benedetta Liberatori,Lorenzo Vaquero,Elisa Ricci,Paolo Rota
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: EMNLP 2026 Findings
Abstract:Comparing videos to assess their semantic similarity remains a challenging task for current Large Multimodal Models (LMMs), which are primarily trained with single-video objectives. While most prior works have focused on computing global similarity, recent studies have introduced the more challenging problem of concept-based similarity, which measures how similar two videos are with respect to specific concepts (e.g., activity, location). However, no dedicated approaches have yet leveraged LMMs to explicitly address this task. In this paper, we aim to fill this gap introducing Video Caption Entailment for video similarity estimation (VICE): a dual-branch design that leverages LMMs for cross-video reasoning. Each branch jointly processes one video and the caption of the other, measuring their alignment for a specific concept via video-language entailment. The outputs from both branches are fused to produce a final similarity score, along with a textual explanation of the model reasoning. This approach leverages the model’s native single-video understanding while enabling fine-grained comparative reasoning between video and text. Experiments on ConViS-Bench and StepDiff show that VICE outperforms models that process both videos simultaneously, highlighting the promise of single-video reasoning for concept-based similarity estimation.
[CV-196] ATLAS: Learning to Adapt LVLM Attention for Reasoning Video Segmentation WACV2027
链接: https://arxiv.org/abs/2610.24783
作者: Hamidreza Dastmalchi,Ali Cheraghian,Aijun An,Abdelwahed Khamis,Lars Petersson
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV2027
Abstract:Reasoning Video Object Segmentation (RVOS) aims to localize and segment objects referred to by natural-language expressions throughout a video. Although large vision-language models (LVLMs) implicitly encode localization cues in their attention maps, these attentions are often noisy, diffuse, and scattered across layers, heads, and modalities, limiting their direct use for segmentation. Existing methods typically rely on handcrafted attention-selection heuristics or computationally expensive attention refinement. We propose ATLAS, an attention adaptation framework that learns to transform raw LVLM attentions into object-centric localization cues. ATLAS treats attentions from all layers, heads, and modalities as structured features, combining high-resolution frame-level attentions that capture spatial information with low-resolution video-level attentions that encode temporal context. A lightweight U-Net equipped with temporal attention and FiLM-based modulation fuses these complementary representations to generate coarse segmentation heatmaps, which are propagated across the video using SAM2. Extensive experiments on challenging referring and reasoning video segmentation benchmarks show that ATLAS outperforms existing frozen-backbone methods. Code and models are available at \urlthis https URL.
[CV-197] Design Anything Design at Scale
链接: https://arxiv.org/abs/2610.24771
作者: Xuchao Ren,Siyuan Feng,Jing Huang,Yanhua Yao,Jing Wang,Lingyu Alina Wang,Anan Chen,Hangjie Yuan,Tao Feng,Qiong Wu,Sai Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 47 pages, 25 figures, 16 tables. Project page: this https URL
Abstract:As multimodal resources are transformed into media outputs, agentic AI is expected to stay aligned with design intent over long horizons. Designers follow briefs specifying goals, constraints, and deliverables, with later work building on earlier decisions. Recent work reuses design experience across tasks, yet alignment among artifacts within one project remains challenging. Agents therefore need to preserve and apply this intent across domains and as projects grow. We introduce Design Anything, a graph-based harness that preserves design intent outside the model and binds deliverables to the workflow steps that produce them through an intent-execution graph. A task contract retains constraints and decisions, selected designs become explicit inputs to later generations, and a deterministic controller uses bindings and execution records to check output types and counts, identify outstanding deliverables, and guide subsequent turns. We also introduce DesignAnything-Bench, with 228 tasks across 19 design domains, including 38 project families at scales of 6, 14, and 24 artifacts. With the LLM held fixed, Design Anything satisfies 85.0% and 92.5% of the brief’s factual constraints with DeepSeek V4.1 Flash High and GPT-5.6 Sol High, respectively, exceeding the strongest same-model baselines by 15.4% and 14.4%. These gains reach 25.1% and 23.0% on 24-artifact tasks. Under each model, Design Anything leads or ties in constraint satisfaction in 18 of 19 domains. Professional designers also prefer its complete deliverable sets over those of same-model baselines in a blind evaluation.
[CV-198] CorrCLIPv2: Harnessing Unlabeled Images for Open-Vocabulary Semantic Segmentation
链接: https://arxiv.org/abs/2610.24757
作者: Dengke Zhang,Fagui Liu,Quan Tang,C. L. Philip Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 12 figures, 12 tables. Published in International Journal of Computer Vision
Abstract:Open-vocabulary semantic segmentation aims to classify pixels without being restricted to a fixed set of categories. Although Contrastive Language-Image Pre-training (CLIP) demonstrates remarkable potential for open-vocabulary recognition, its segmentation capability is impaired by noisy inter-class patch interactions. To this end, this work proposes a unified training-free object-centric framework, named CorrCLIPv2, from three complementary aspects of patch correlation, feature representation, and output regularization. Specifically, for patch correlation, we leverage class-agnostic mask generators to explicitly disentangle inter-class interactions. Moreover, we utilize spatially-coherent features from self-supervised models to compute attention scores to mitigate mask inaccuracies. Regarding feature representation, we devise a refinement mechanism to simultaneously strengthen spatial details and semantic representation. Within the output space, we unify predictions via voting in each mask to enhance spatial consistency. To mitigate contextual bias and enhance prediction accuracy of objects, we further extend the framework by incorporating a cross-image object retrieval paradigm, which incurs minimal computational overhead. We build a memory bank of CorrCLIP-enhanced features from unlabeled images and implement a regime of multi-CLIP ensemble. By aggregating semantically consistent object features across diverse contexts retrieved from the memory bank, CorrCLIPv2 mitigates context-induced interference and obtains more robust predictions. Extensive experiments demonstrate that our approach establishes a new state-of-the-art on 10 standard benchmarks. Remarkably, it even surpasses fully supervised open-vocabulary methods across 22 multi-domain datasets. Code is available at: this https URL.
[CV-199] CorrFlow: Self-Correcting Flow Matching
链接: https://arxiv.org/abs/2610.24750
作者: Hangwei Zhang,Nathan Rouillé,Chuanxia Zheng
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Project Page: this https URL
Abstract:Flow-matching models generate samples through iterative velocity predictions, but typically do not revisit these predictions using the intermediate representations that produced them. We introduce CorrFlow, a self-correcting flow-matching framework that reuses this information through a lightweight refiner. Under rectified-flow interpolation, the error of the one-step clean prediction equals the realized velocity residual scaled by the noise level. Motivated by this identity, the refiner learns to predict residual corrections from the base model’s clean prediction, conditioning, noise level and hidden states from the same forward pass. The correction is added to the base velocity before each sampling update, so subsequent predictions are made from the corrected state. This requires no additional base-model evaluations and only one lightweight refiner evaluation per step. Our theoretical analyses establish conditions under which feature reuse outperforms raw-input affine predictors and refiner training achieves greater residual-risk reduction than continued base-model training under a matched training compute budget. Experiments on class-conditional image generation, text-to-video generation and image-to-video generation show consistent improvements over the corresponding base models, including gains in visual quality and physical consistency. Project Page: this https URL
[CV-200] Editing Images into Reusable Layers via Compositional Edit-Prior Distillation
链接: https://arxiv.org/abs/2610.24738
作者: Wen Xue,Yong Liu,Xueheng Li,Si Wu,Yong Xu,Yipeng Sun,Yan Li,Simiu Gu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 24 pages, 13 figures. Includes appendices
Abstract:High-quality image editing requires precise modification of target regions while preserving the content of unrelated regions. Existing editing models produce flattened outputs that limit precise local manipulation and layer reuse, while decomposition methods typically optimize layer recovery separately from instruction-driven editing. To this end, we introduce LayerEditor, a unified framework that couples instruction-guided image editing with controllable generation of reusable RGBA layers. At its core, In-Context RGB-Alpha Token Modeling jointly represents the edited canvas and an ordered sequence of RGB-Alpha layers within a single token sequence, enabling semantic edits and explicit control over individual layer content. To retain the pretrained model’s editing capability during adaptation to layered outputs, we propose Compositional Edit-Prior Distillation, which transfers its editing prior to both the edited canvas and the image composed from the predicted layers. To support training and evaluation, we construct LayerEditor-100K, a large-scale paired image-editing dataset with ordered RGBA layer annotations, and introduce dedicated benchmarks for layered editing. Extensive experiments demonstrate that LayerEditor maintains strong editing fidelity while generating accurate RGBA layers, supporting both flexible layer-level manipulation and multi-layer decomposition.
[CV-201] Geometry-Aware Interpolation for Inverse Problems
链接: https://arxiv.org/abs/2610.24687
作者: Pourya Shamsolmoali,Masoumeh Zareapoor
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Recovering missing visual information from incomplete or degraded images remains a challenging problem, particularly when spatial variations and complex image structures affect reconstruction accuracy. Although Spatial Transformer Networks (STNs) facilitate spatial feature alignment, their reliance on conventional linear or bilinear interpolation limits their ability to capture the underlying geometric characteristics of image data. Furthermore, interpolation based on a restricted set of neighbouring pixels can introduce inconsistencies in gradient propagation, potentially affecting optimisation stability. To address these limitations, we propose a Geometry-Aware Interpolation Network (GAIN), a variational autoencoder-based framework that incorporates manifold geometry into spatial transformation and image reconstruction. Rather than estimating pixel intensities solely from fixed spatial neighbours, GAIN exploits the local tangent structure of the data manifold to construct geometry-informed intensity approximations. An entropy-based regularisation mechanism further guides the estimation of spatial transformation parameters, promoting more adaptive interpolation. In addition, a convolutional normalisation strategy is introduced to regulate the magnitude of non-zero activations and improve gradient stability during training. Extensive experiments on challenging image completion and classification benchmarks demonstrate the effectiveness of GAIN in improving predictive accuracy while reducing computational cost compared with existing methods.
[CV-202] MESS: Multi-Exposure Sequence Synthesis for Generalizable Image Enhancement NEURIPS2026
链接: https://arxiv.org/abs/2610.24680
作者: Ruodai Cui,Shuaizheng Liu,Rongyuan Wu,Lei Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. Code and data are available at this https URL
Abstract:Despite the significant progress in image enhancement under adverse illuminations, such as low-light image enhancement, exposure correction, and backlit image enhancement, existing models show limited generalizability to different scenarios. A major bottleneck lies in the lack of large-scale and diverse training data, as these tasks typically rely on manually captured, precisely aligned image pairs that are expensive to scale up. Although synthetic data have been explored as an alternative, existing synthesis methods are limited in modeling complex adverse illumination conditions and camera imaging pipelines. To address this issue, we propose a novel framework to synthesize realistic training pairs for generalizable image enhancement in the wild. Specifically, we first derive multi-exposure sequences (MES) from high-bit RAW images by rendering them under different exposure settings through an emulated ISP pipeline. Using these RAW-derived MES as supervision, we train a diffusion-based RGB-to-MES generator to synthesize plausible exposure sequences from a single 8-bit RGB image, extending the synthesis pipeline from limited RAW collections to large-scale in-the-wild RGB data. For each source image, we randomly sample one synthesized exposure variant as the degraded input and use the original high-quality RGB image as the target. With the proposed MES Synthesis (MESS) approach, we construct a dataset of 500K training pairs and train a lightweight network, which demonstrates significantly better generalization performance than existing models across various image enhancement tasks. Code and data are available at this https URL.
[CV-203] From Views to Worlds: Active Exploration over 3D Worlds for Vision-Language Models NEURIPS2026
链接: https://arxiv.org/abs/2610.24676
作者: Qijian Tian,Jiayu Ying,Ke Fan,Lizhuang Ma,Xin Tan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by NeurIPS 2026; Project Page: this https URL
Abstract:Vision-language models (VLMs) have achieved remarkable progress in 2D visual understanding tasks, yet the passive view-centric perception paradigm limits their capacity for 3D spatial reasoning. Such reasoning often requires actively acquiring spatial evidence beyond the currently visible views, such as camera poses and cross-view spatial relationships, to infer 3D relationships that are not explicitly represented in 2D images. However, existing methods either lack active exploration or merely provide additional views during inference, without explicitly acquiring such spatial evidence beyond 2D images. In this paper, we propose View-to-World (V2W), an active exploration framework that enables VLMs to actively explore 3D worlds and lifts VLMs from passive view-centric perception to active world-centric spatial reasoning. Given a spatial reasoning task with multi-view images, V2W first reconstructs an explicit 3D world with geometry, camera poses, and language-grounded semantics, and exposes it to VLMs through visual and linguistic interfaces. Through multi-turn interaction with these interfaces, VLMs iteratively acquire spatial evidence from the constructed 3D world, enabling active world-centric spatial reasoning. V2W can be integrated with existing VLMs in a training-free manner and can be further enhanced by optimizing the exploration policy via reinforcement learning. Experiments on spatial mental modeling benchmarks demonstrate that V2W substantially improves VLM baselines in the training-free setting and achieves state-of-the-art performance with the learned exploration policy, highlighting the importance of active exploration for spatial intelligence. Project page: \urlthis https URL.
[CV-204] Knowledge Diversion: A Novel Paradigm for Pre-training Decomposable Image Generation Models
链接: https://arxiv.org/abs/2610.24664
作者: Yucheng Xie,Fu Feng,Ruixiao Shi,Jing Wang,Xin Geng,Yong Rui
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Large-scale pre-trained models have become foundational to modern machine learning, yet conventional pre-training objectives primarily optimize source-domain performance without explicitly considering downstream requirements. Consequently, transferable and task-specific knowledge remains entangled within a shared parameter space, limiting selective knowledge reuse and efficient downstream adaptation. This limitation is particularly consequential for image-generation models, which must accommodate diverse conditions and domains. To address this challenge, we introduce Knowledge Diversion (KDiv), a condition-aware pre-training paradigm for learning inherently decomposable generative models. Before pre-training, KDiv factorizes randomly initialized weights into rank-one Knowledge Units, which are organized into learngenes and tailors. During pre-training, continuous condition-aware routing consolidates knowledge shared across conditions in learngenes while directing condition-dependent variations to tailors. This structured organization enables learngenes to capture transferable generative priors and tailors to develop condition-specific expertise, allowing the resulting model to be selectively recombined for diverse downstream conditions. Extensive experiments across class-, text-, and image-conditioned generation demonstrate that KDiv consistently matches or surpasses dense baselines at comparable pre-training cost and achieves the strongest average downstream performance with a trainable parameter budget comparable to low-rank adaptation methods.
[CV-205] Modeling Uncertainty Propagation in 3D Morphable Models
链接: https://arxiv.org/abs/2610.24646
作者: Megh Shukla,Edo Collins,Prashanth Chandran,Karel Petránek,Sergio Orts-Escolano,Mathieu Salzmann,Alexandre Alahi
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:We study the problem of mapping parameter-space variance of a 3D Morphable Model (3DMM) to per-vertex variance. Traditionally, per-vertex variance has been used to visually interpret the learnt parameter uncertainty. However, mapping parameter-space variance to per-vertex using sampling or error propagation is computationally expensive. We address this limitation with Closed-form Local Uncertainty Estimation (CLUE), which transforms the parameter-space distribution of 3DMMs to per-vertex through a closed-form approximation. CLUE emerges from applying the law of total variance to the 3DMM formulation and is modular, computationally inexpensive, and accurate. Subsequently, we study integrating CLUE for in-the-wild human mesh recovery training and inference. We note that CLUE uses the geometry of the mesh as an inductive bias when predicting the variance, leading to improved accuracy. Through extensive qualitative and quantitative experiments, we show that when used in training, CLUE (1) outpeforms probabilistic and deterministic baselines on metrics evaluating joints, mesh and uncertainty, and (2) has the highest uncertainty-to-error spearman correlation, with applications including data mining and interpretable uncertainty. Finally, we share SAM 3D Body (CLUE) which augments the regressor with high resolution uncertainty maps. We make our code available at this https URL
[CV-206] Spectral-Aware Proposal Enhancement for Diffusion-Based Multispectral UAV Multi-Object Tracking
链接: https://arxiv.org/abs/2610.24642
作者: Xu Lin,Tingfa Xu,Tianhao Li,Haolin Qin,Huawei Gao,Jichuan Xing,Jianan Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted for publication in IEEE Geoscience and Remote Sensing Letters. 5 pages, 4 figures, 3 tables. Author accepted version
Abstract:Multispectral UAV multi-object tracking (MOT) with oriented bounding boxes remains challenging due to small targets, cluttered backgrounds, and unstable inter-frame associations. Although diffusion-based MOT provides an effective iterative proposal refinement paradigm, existing diffusion-based trackers lack dedicated designs for exploiting multispectral cues in proposal initialization and confidence calibration, limiting tracking robustness in complex UAV scenarios. This letter proposes DSATrack, a diffusion-guided spectral-aware tracking framework for multispectral UAV MOT. Specifically, a lightweight spectral encoder injects multispectral information into the RPN branch through feature fusion, enabling the RPN head to generate high-quality object-aware proposals. These proposals are then injected into the initial proposal set to provide reliable spatial priors for diffusion refinement. Furthermore, a spectral-saliency confidence modulation strategy recalibrates refined tracking confidence according to local multispectral saliency contrast before score thresholding, enhancing spectrally reliable candidates while suppressing unreliable ones. Without modifying the original diffusion refinement head, DSATrack improves both proposal quality and post-refinement candidate selection. Experiments on the MMOT benchmark demonstrate that DSATrack outperforms the adapted DiffusionTrack baseline by 4.6 HOTA, 9.0 MOTA, and 7.7 IDF1 points, showing its effectiveness for multispectral UAV MOT.
[CV-207] Beyond Geometric Closure: Measuring and Diagnosing Mechanical Consistency in Visual Dynamics
链接: https://arxiv.org/abs/2610.24619
作者: Qibin Gu,Yunfeng Yan
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注: 34 pages, including supplementary material; 16 figures and 38 tables
Abstract:Visual accuracy leaves open whether predicted motion obeys dynamics. We introduce a framework for measuring and diagnosing mechanical consistency in visual dynamics. An independent SI audit of rigid closed chains separates geometric admissibility, reaction-free actuator effort, and work–energy balance. Continuous scoring measures progress against fixed engineering targets, while reference replay quantifies measurement error. With three seeds per setting, we evaluate KeyCLD and CON across 343 planar assemblies at 64/256 pixels and V-JEPA + AC and LingBot-VA on 100 shared spatial test assemblies. CON reduces image MSE by 37.6%/37.2% relative to persistence, yet neither planar model attains valid native geometry. At 256 pixels, calibrated RGB replay places forecast errors far above measurement error. With exact states, KeyCLD preserves geometry and RK4 training raises its score from 52.65 to 64.30. In both spatial pathways, CAD/IK reconstruction restores geometry while mean actuator residuals exceed five times the engineering targets. The interventions distinguish gains in visual prediction and geometric admissibility from remaining force and energy errors.
[CV-208] IntriAdapt: Generalizable Intrinsic-Guided Test-Time Adaptation for Metric Geometry Estimation
链接: https://arxiv.org/abs/2610.24607
作者: Jiaxing Chen,Zizheng Yan,Jingyu Zhuang,Jinwei Chen,Shuguang Cui,Guanying Chen,Qingnan Fan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recovering metric 3D geometry from images is fundamentally ill-posed due to perspective projection ambiguity. Current geometry models incorporate supplied or predicted intrinsics during inference, but do not use them to supervise model adaptation. To refine these predictions, test-time adaptation (TTA) typically relies on external depth measurements. In this paper, we instead use camera intrinsics as an output-side geometric verifier. Comparing the point cloud’s viewing rays with those defined by the known intrinsics yields a direct measure of geometric inconsistency. We propose \textitIntriAdapt, a TTA framework that minimizes this closed-form geometric residual. By optimizing a lightweight LoRA adapter on a frozen geometry foundation model, our method aligns the prediction with the specific camera model without depth sensors or external priors. We further extend this approach to video sequences, demonstrating that adapting a short sequence generalizes to other sequences captured by the same camera. The adapted model processes held-out images in a feed-forward pass, without repeating optimization for each frame. Experiments on ten single-image and seven sequence benchmarks across multiple base models show that \textitIntriAdapt consistently outperforms both unadapted and oracle-intrinsics baselines in metric accuracy and relative geometry.
[CV-209] SCOPE-VSR: A Broad-Coverag e Training Dataset for Video Super-Resolution
链接: https://arxiv.org/abs/2610.24604
作者: Gengyu Liao,Yechun Yong,Zheng Chen,Jianze Li,Ziqing Zhang,Linghe Kong,Yulun Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code is available at: this https URL
Abstract:Open-domain video super-resolution (VSR) requires temporally coherent details across widely varying scenes and subjects. However, existing training datasets are often semantically imbalanced and insufficiently diverse. We address this data bottleneck with a structured curation pipeline for open-domain VSR, SCOPE. In this pipeline, we combine task-oriented video selection with hierarchical semantic rebalancing. Applying this pipeline to a large-scale collection of data sources, we construct the SCOPE-VSR dataset for training and SCOPE-test for testing. SCOPE-VSR has approximately 27k videos and provides balanced and diverse coverage of various real-world scenarios and subjects. We further develop GraCo-VSR, a novel framework for one-step video restoration. GraCo-VSR adopts a reliability-aware design with explicit supervision to reduce degradation-induced distortions in motion and texture guidance, allowing the model to benefit more effectively from the diverse training content in SCOPE-VSR. Evaluations show that our model achieves strong reconstruction fidelity and perceptual quality. Our code is available at: this https URL.
[CV-210] Dynamic self-ensemble on multi-exit residual vision transformer for small-data image classification
链接: https://arxiv.org/abs/2610.24602
作者: Stephen Ekaputra Limantoro,Yan-Kai Wang,Ching-Chun Huang,Hong-Han Shuai,Wen-Huang Cheng,Jungwoo Huh,Sanghoon Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision Transformer (ViT) has garnered increased attention for image classification in recent years. However, ViT is notorious for its data-hungry problem as it achieves high-performance results when only trained on large-size datasets. In this paper, we study how ViT specifically works on small-scale datasets with limited data where the availability of image data is often constrained in practice, especially in fields such as medical imaging. To mitigate the issue, we designed a multi-stage residual ViT, dubbed ReSEViT, with the self-ensemble method to utilize multi-level information from intermediate features. Due to the complex decision boundaries for classifications, relying on a single classifier can lead to prediction errors. Hence, we propose a new learning method for ViT, named self-ensemble learning, which ensembles multi-stage classifiers in a single model. Our self-ensemble comprises a novel approach, Dynamic Learner Selection (DLS), to adaptively select the importance of different stages. ReSEViT incorporates residual token embeddings which is a ResNet-like block to model effective local structures alongside with transformer blocks in each stage. The integration of residual tokens and self-ensemble learning into a multi-stage ViT highlights the reduction in redundancy, prediction error and variance while maintaining an efficient model size for small-scale image datasets.
[CV-211] Scene-Adaptive VLA: Efficient Autonomous Driving via Dynamic Layer Routing NEURIPS2026
链接: https://arxiv.org/abs/2610.24590
作者: Yujia Yang,Quan Yuan,Jiawei Fan,Yang Li,Tiange Fu,Xiaoyuan Fu,Guiyang Luo,Jinglin Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: Accepted at NeurIPS 2026
Abstract:Vision-Language-Action (VLA) models offer a promising paradigm for autonomous driving. However, the massive parameter scale of VLAs strains onboard computational resources during real-time inference. This directly increases energy consumption and limits vehicle battery range, making it critical to minimize computational overhead without compromising driving capability and explainability. Current research on efficient VLA-based driving focuses on sequential adaptation (\textite.g., Chain-of-Thought and token adjustment), leaving structural adaptation underexplored despite parameter redundancy across dynamic driving scenes. To address this gap, we propose Scene-Adaptive VLA, an efficient framework leveraging dynamic layer routing to adjust active parameters based on real-time scenes. Inspired by the temporal continuity of scenes and model predictive uncertainty, our approach characterizes model-perceived scene complexity via frame-level temporal scene variations and decision deviations. To uncover latent scene-to-parameter correspondences, we encode this characterization into scene-aware tokens, alongside two specialized queries. A budget-conditioned layer routing mechanism evaluates these refined queries to allocate a budget that determines the active parameter ratio and subsequently selects appropriate layer combinations. The entire framework is optimized end-to-end. Extensive evaluations on the Bench2Drive benchmark show that when adapted to a state-of-the-art VLA model, our approach significantly reduces inference-time computational overhead by 53.5% with negligible degradation in driving and language capabilities. Code is available at https:github.com/yang-luvu/Scene-Adaptive-VLA.
[CV-212] Cross-polarization pairing for label-free few-shot thin-section classification
链接: https://arxiv.org/abs/2610.24546
作者: Bingchen Yan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Few-shot classifiers of rock thin sections are meta-trained on many rock types that a petrographer has identified slide by slide, and this identification is the costly step that new collections lack. We study few-shot recognition of unseen rock types when no training slide is identified, and the only information available is what the microscope records anyway: which slide an image comes from, and whether it was taken in plane-polarized (PPL) or cross-polarized (XPL) light. We first show why generic unsupervised pseudo-tasks fit this setting poorly: the positives they find without labels are either wrong or near-duplicates of the anchor, so they teach the model little. Polarized-light microscopy provides positives that avoid both problems. The PPL and XPL views of one field show the same grains, so they are correct by construction, yet they look very different. We propose cross-polarization pairing (XPP), which treats each slide as a pseudo-class and pairs its PPL and XPL views of the same field location, so that matching must rely on grain shape and texture rather than on colour. On the Nanjing University collection of 108 rock types, with support and query images taken from different slides, XPP is the best label-free method on average over three random splits, and matches a supervised model trained on 78 identified rock types. Controls show that colour augmentation and other acquisition positives do not reproduce the gain. We also show that the image-level protocol of earlier work is saturated, and propose a cross-slide protocol in its place.
[CV-213] heory-Guided Instance-Dependent Early Stopping via Second-Order Loss Dynamics
链接: https://arxiv.org/abs/2610.24540
作者: Ahad Bin Islam Shoeb,Jamal Uddin Tanvin,Kamrul Hasan,Alve Reduan,Imtiaz Ahmed
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 29 pages, 11 figures, 12 tables. Code available at this https URL
Abstract:Instance-dependent early stopping (IES) reduces training computation by skipping backpropagation for examples whose losses have stabilized, but the second-order loss-difference criterion used by IES lacks a theoretical justification. We develop Theory-Guided Instance-Dependent Early Stopping (TG-IES), which analyzes this criterion under explicit assumptions and derives a noise-calibrated mastery threshold and a conditional stability bound for instance exclusion. We show that second-order differencing has a favorable bias-variance trade-off over first- and third-order differences in a specific signal-to-noise regime, and that its signal contains both gradient-drift and curvature terms. TG-IES derives its stopping threshold and patience from training-run statistics rather than using the manually selected threshold of IES. Across CIFAR-10 and CIFAR-100, TG-IES saves 53-58% of backpropagation instances in its main convolutional settings while keeping accuracy within 0.1-0.9 percentage points of full training; in the primary ResNet-18/CIFAR-10 setting, this corresponds to 35.7% FLOP-normalized savings. The method also transfers to ViT-Tiny, where it achieves smaller savings and incurs a modest accuracy cost. These results provide a theoretical basis for second-order instance-dependent early stopping and show that theory-guided calibration can substantially reduce training computation under the assumptions and experimental settings studied.
[CV-214] Write Once Query Many: Evolving Compact Reusable Memory for Long Videos
链接: https://arxiv.org/abs/2610.24537
作者: Kerui Chen,Jinglu Wang,Yun Wang,Xinjiang Wang,Yan Lu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Long videos often exceed the context capacity of multimodal large language models (MLLMs), while agentic search repeats costly video processing and grounding for each query. Reusable memory can amortize this cost, but must preserve information for questions unknown during construction. We propose Video-EMem, an evolvable video memory harness built on time-anchored perceptual observations. This representation combines compact free-form text with sparse images to preserve semantic content and complementary visual details across temporal scales, supporting temporal navigation without a fixed semantic schema. A Writer encodes each video once by consolidating overlapping multiscale observations, and a Reader retrieves and organizes evidence solely from the resulting memory. With the underlying models frozen, we evolve their workflows and operators using distinct feedback: query-independent source fidelity and coverage for the Writer, and QA failures in evidence localization and utilization for the Reader. Experiments support this separation for generalization beyond development queries, with our evolution strategy improving average accuracy by 7.2% over the unevolved harness. While reducing storage by about 99.86% on average relative to raw videos, Video-EMem improves Qwen3-VL-30B by an average of 2.6% across three long-video QA benchmarks over direct input of 2,048 frames. Using the same o3 answer model as the state-of-the-art agentic method, it improves average accuracy by 2.5% with lower per-query latency.
[CV-215] Structured Latent Dynamics for Long-Horizon Embodied World Models
链接: https://arxiv.org/abs/2610.24534
作者: Kevin Chen,Yu-Wei Zhan,Haoran Liang,Sen Cui,Changshui Zhang
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reliable long-horizon prediction remains a central challenge for embodied world models. Although recent generative models can synthesize visually plausible action-conditioned futures, their latent dynamics are typically learned through unconstrained transition modules, which may gradually deviate from coherent physical evolution as predictions are recursively rolled forward. We investigate whether explicitly structuring latent state evolution can improve the stability and utility of such imagined trajectories. To this end, we introduce HWM, a structured dynamics framework that represents backbone latents in a learned phase space consisting of generalized coordinates and momenta. Future states are propagated through dissipative Hamiltonian dynamics and mapped back to the original latent representation through a learnable inverse interface. The resulting formulation can be integrated into existing world-model architectures without redesigning their perception or decoding components, while multi-step transition supervision encourages the structured rollout to remain aligned with the latent trajectories induced by real observations. Controlled simulation experiments show that HWM produces more accurate estimates of velocity, acceleration, and jerk and substantially reduces long-horizon energy drift relative to unstructured latent dynamics. We further evaluate the approach across three distinct embodied world-model settings, covering single-arm multi-view prediction with Ctrl-World, dual-arm single-view prediction with EVAC, and task-oriented single-arm prediction with Cosmos-Predict2.5-2B. On RoboCasa, incorporating HWM into Cosmos-Predict2.5-2B yields a 2.7% relative improvement in task success. On real-world AgiBot World trajectories, it improves nDTW by 9.7% with Ctrl-World and Logic by 8.4% with EVAC.
[CV-216] Contrastive Hypergraph Source-free Domain Adaptive Object Detection in Adverse Weather Conditions–Extended Version NEURIPS2026
链接: https://arxiv.org/abs/2610.24532
作者: Jicheng Yuan,Tung Kieu,Manh Nguyen-Duc,Manfred Hauswirth,Danh Le-Phuoc
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 20 pages. An extended version of “Contrastive Hypergraph Source-free Domain Adaptive Object Detection in Adverse Weather Conditions” accepted at NeurIPS 2026
Abstract:Adverse weather conditions such as fog, rain, snow, and low illumination introduce structured visibility degradations that severely impair object detectors trained on clear-weather data. When source data are inaccessible due to privacy or trans- mission constraints, source-free domain adaptation (SFDA) becomes necessary. However, existing SFDA methods based on pairwise contrastive learning struggle to capture high-order semantics and often suffer from agreement collapse under severe corruption. We propose REUNION, a hypergraph-guided SFDA framework for object detection in adverse weather conditions. REUNION constructs a contrastive hypergraph over target-domain object proposals, encoding high-order relations through intra-image context, weather-aware grouping, prototype-based semantic anchors, and uncertainty-aware connections between reliable and low-confidence instances. To effectively exploit these structured relations, we introduce a group-wise Hyper-InfoNCE objective that optimizes representations at the hyperedge level, enabling semantic information to propagate from confident proposals to corrupted or low-contrast instances. Experiments with diverse benchmarks demonstrate that REUNION effectively mitigates domain shifts and achieves state-of-the-art performance in SFDA settings.
[CV-217] An Optical Music Recognition Tool for String Quartets
链接: https://arxiv.org/abs/2610.24530
作者: Jin-Shiang Hu,Bo-Yuan Cheng,Li Su
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 3 pages, 1 figure, 1 table. Extended abstract for the Late-Breaking/Demo Session of ISMIR 2025, Daejeon, South Korea
Abstract:This paper presents an optical music recognition (OMR) system tailored for PDF scores of string quartet pieces. The approach begins with a SegNet-based segmentation model to localize musical regions, followed by convolutional neural networks and image processing techniques to classify and extract musical symbols. To improve rhythmic precision, dynamic programming methods enforce time signature constraints and explicitly handle tuplets. Evaluation using a longest common subsequence metric across multiple feature matrices demonstrates robust accuracy in recognizing pitch, onset, duration, and accidentals on a benchmark dataset. Code and data are available at this https URL.
[CV-218] VisConf: Quality-Aware Consensus for Best-of-N in Vision-Language Model Reasoning NEURIPS2026
链接: https://arxiv.org/abs/2610.24526
作者: Muhammad Usman,Muhammad Qasim Ayub,Usman Ahad,Muhammad Tahir
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026 Workshop: VLM4RWD. 20 pages, 8 figures. Code: this https URL
Abstract:Best-of-N methods improve text-only LLM reasoning by sampling multiple trajectories before selecting an answer. For Vision-Language Models (VLMs), answer frequency and model confidence alone can be insufficient, as sampled generations also vary in their engagement with the visual input. We find that distributional confidence and visual attention are complementary indicators of rollout quality. Each is limited on its own, while trajectories that show strong signals in both are more likely to be correct. Motivated by this, we introduce VisConf, a rollout-quality score that multiplicatively combines visual engagement with candidate-calibrated Self-Certainty, penalizing rollouts that are weak in either signal. We incorporate Rank-Weighted Consensus (RWC), a tuning-free aggregation rule that converts within-question VisConf ranks into answer-level vote weights. Across three VLMs on MathVista, MMMU-Pro, and MMStar, VisConf with RWC consistently outperforms strong BoN baselines. Our analysis shows particularly large gains when correct trajectories are present but underrepresented in the candidate pool.
[CV-219] MOMO: Human Motion from Moving Sounds
链接: https://arxiv.org/abs/2610.24522
作者: Deli Cai,Shuyang Xu,Zhiyang Dou,Zitong Lan,Mingmin Zhao,Changxing Ding,Taku Komura
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Robotics (cs.RO); Sound (cs.SD)
备注:
Abstract:Human responses to sound depend not only on its semantic content, but also on the evolving spatial relationship between source and listener, conveyed through spatial audio and its temporal dynamics. Modeling human reactions to such dynamic spatial audio is important for character animation, human behavior analysis, and robotics. As of yet, existing audio-driven motion generation methods primarily focus on non-spatial or static acoustic conditions, leaving this important problem largely underexplored. To bridge this gap, we formulate the novel task of dynamic spatial-audio-driven human motion generation and introduce two motion-captured datasets. We introduce the Dynamic Spatial Audio and Motion dataset (DySAM) as the first dataset and benchmark for this task, providing synchronized binaural audio, 3D human motions, and moving sound source trajectories. Given the importance of dynamic spatial cues of sound sources, we further provide the Dynamic Spatial Audio Localization dataset (DySAL), an auxiliary binaural localization dataset that pairs diverse acoustic events with head poses and moving source trajectories to support spatial audio perception. For benchmarking, we propose MOMO for human Motion generation from Moving Sounds, a diffusion-based framework guided by three coupled principles: localization, understanding, and generation. MOMO performs localization by estimating head-centric spatial direction cues of the moving source directly from raw binaural audio, avoiding reliance on ground-truth source coordinates. It performs understanding by extracting frame-wise audio-content representations, and performs generation by conditioning motion diffusion on both semantic features and dynamic spatial cues. Experiments show that MOMO outperforms existing baselines in motion fidelity, audio-motion alignment, and physical plausibility.
[CV-220] BEM: Learning Bregman Divergences for Concept Erasure in Flow Matching Models
链接: https://arxiv.org/abs/2610.24508
作者: Mohamed-Hicham Leghettas,Markus Püschel
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 9 figures, 10 tables. Preprint, under review. Content warning: contains generated images of graphic violence and censored nudity
Abstract:Concept erasure removes harmful, private, or copyrighted concepts from text-to-image models while preserving their other abilities. Leading models such as FLUX and Stable Diffusion 3 are trained with flow matching, and erasure methods for them fine-tune the model toward a guided teacher’s velocities under a Euclidean loss. Yet the squared distance is only one member of a larger family: Bregman divergences are exactly the losses under which flow matching recovers the correct velocity field, and each member shapes differently what is erased and what is preserved. We introduce BEM (Bregman erasure matching), which is, to our knowledge, the first to learn this divergence through bi-level optimization, so that the fine-tuned model forgets the concept on unseen prompts while its unrelated generations stay unchanged. In the experiments we target the suppression of nudity, violence, real identities, and copyrighted characters. Under the evaluation protocol of GEM, BEM removes more of the concept than GEM on most settings, cutting the unsafe rate on bloody gore from 53.9% to 14.4% and the FID on nudity from 8.64 to 6.64, while retention holds on some concepts and degrades on others.
[CV-221] SAP-Depth: Scene-Adaptive Probability Framework for Self-Supervised Monocular Depth Estimation
链接: https://arxiv.org/abs/2610.24506
作者: Bowen He,Li Zhang,Yujie Chen,Chang Cheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: CWSN2026 accepted
Abstract:Self-supervised monocular depth estimation remains challenging due to the loss of fine-grained details and limited adaptability to varying scene structures. In this paper, we propose SAP-Depth, a novel architecture based on Scene-Adaptive Probabilistic. Unlike traditional methods, we introduce a Dynamic Bins mechanism powered by a Learnable Parameter Predictor, which adaptively estimates scene-specific depth distributions conditioned on the input image context. Furthermore, a Bin-wise Probability Distribution Decoder is designed to effectively fuse multi-scale representations to predict dense depth probability distributions, significantly enhancing boundary sharpness and structural recovery. Through the collaborative coupling of these proposed modules, our model achieves dynamic prediction of scene-specific depth distributions. Extensive experiments on the KITTI dataset show that SAP-Depth outperforms existing state-of-the-art methods in accuracy, particularly in its ability to better preserve complex edge details. Further, tests on the Make3D dataset validate the model’s excellent generalization capability and robustness.
[CV-222] Investigating Length Bias and Robustness in LVLMs for Multiple-Choice Question Answering EMNLP2026
链接: https://arxiv.org/abs/2610.24500
作者: Md. Atabuzzaman,Hani Alomari,Chris Thomas
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: EMNLP 2026 main conference
Abstract:Large Vision-Language Models (LVLMs) demonstrate strong performance on multiple-choice question answering (MCQA), yet their reliability under biased conditions remains underexplored. We investigate length bias, the systematic preference for longer answer options, through controlled benchmarks spanning fine-grained visual domains. LVLMs consistently favor longer options even when visually inconsistent, with accuracy dropping 2–38% when correct answers are shorter and increasing 1–70% when longer. This bias persists across model scales. Through layer-wise mechanistic analysis, we reveal that length bias emerges from late-layer suppression of visual evidence in favor of language-side length priors. Blurry-image experiments show that bias intensifies when visual evidence weakens, while abstention tests reveal that LVLMs struggle to refuse incorrect predictions. We propose and evaluate multiple inference-time mitigation strategies; while these provide partial improvements, none achieves deployable bias mitigation without relying on oracle information unavailable at inference time. These results underscore that inference-time interventions are insufficient, pointing to the need for architectural solutions and bias-aware training.
[CV-223] RiSE: Residual Subspace Expert for Generalizable Text-Centric Image Forgery Localization
链接: https://arxiv.org/abs/2610.24480
作者: Kahim Wong,Kemou Li,Yiming Chen,Haiwei Wu,Jiantao Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:As AI-assisted image editing becomes increasingly prevalent, Text-Centric Image Forgery Localization (TFL) is essential for protecting trust in financial and legal records. However, we observe that existing TFL detectors predominantly rely on Full-Parameter Fine-Tuning (FPFT) and fail to generalize to unseen forgery types because FPFT can induce low-rank feature collapse and overfit training artifacts. Meanwhile, prior detectors capture diverse forgery traces with a single unified representation, making subtle traces entangled and difficult to learn. To address these challenges, we propose RiSE, a Residual Subspace Expert model that preserves pre-trained priors by freezing principal SVD components and adapting only the residual subspaces with a ViT-only segmentor to avoid feature collapse. RiSE introduces forgery-aware residual experts, where each expert is trained independently on a training subset with a particular forgery type, and we show that cross-domain localization performance improves consistently as more experts are added. At inference, we introduce Latent Perturbation Confidence (LPC), which selects the most confident expert by analytically computing the latent-perturbed output. LPC enables efficient confidence estimation with a single forward pass, avoiding repeated forward passes on perturbed samples and improving expert selection. By training on synthetic forgery data, RiSE with 1 expert outperforms prior methods on real-world forgery images by 20.4% while reducing training steps by 10 times. Scaling the number of experts to 9 yields a 35.8% gain. RiSE remains effective with DCT feature, even when the majority of the parameters is frozen with RGB-only pretraining. Code: this https URL.
[CV-224] Unifying Amplitude-Phase Coupling and Reliable Endogenous Priors for Low-Light Image Enhancement
链接: https://arxiv.org/abs/2610.24467
作者: Tongshun Zhang,Zixuan Zhong,Pingping Liu,Jin Kuang,Yubing Lu,Dong Liu,Qiuzhan Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Frequency-domain methods have shown strong potential for low-light image enhancement (LLIE) because they can effectively regulate global illumination. However, most existing methods assume that amplitude and phase can be restored independently, while overlooking their inherent spectral coupling. This artificial decoupling often leads to information inconsistency and structural distortions. To address this issue, we propose a frequency-spatial framework that jointly models spectral interaction and reliable prior guidance. We first empirically analyze the coupling between amplitude and phase and develop an Amplitude-Phase Adaptive Compensation Fusion (APACF) module. Instead of enforcing a hard separation between the two components, APACF enables their bidirectional interaction, resulting in spectrally coherent restoration. In addition, directly extracting priors from raw low-light inputs is unreliable, since severe noise and underexposure destroy critical visual content. To alleviate this problem, we introduce an Endogenous Reliability Prior (ERP). Specifically, we treat an intermediate result, preliminarily restored in the frequency domain and already close to the ground-truth illumination, as a high-fidelity source for extracting complementary pseudo-infrared and gradient cues. Based on ERP, we further design a Dual-Modal Guided Refinement Network (DGRN) to inject these purified structural cues into the decoder, thereby recovering missing fine textures more accurately. Extensive experiments on benchmark datasets show that our method achieves state-of-the-art performance and consistently surpasses existing methods in both quantitative metrics and visual quality. Our code is available at this https URL.
[CV-225] O-SCD: Replay-Free Online Scene Change Detection via Recursive Evidence Fusion
链接: https://arxiv.org/abs/2610.24462
作者: Chamuditha Jayanga Galappaththige,Jason Lai,Lloyd Windrim,Timothy Patten,Donald G. Dansereau,Niko Suenderhauf,Dimity Miller
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Online scene change detection requires an agent to detect what has changed in a previously mapped environment while it is still on site, observing from unconstrained viewpoints and without access to future observations. This causal setting is strictly harder than its offline counterpart, and online methods have consistently trailed offline accuracy. O-SCD closed that gap by treating change as a persistent property of a 3D Gaussian splatting map of the reference scene, surpassing the strongest offline methods in change detection accuracy while operating at over 10 frames per second. O-SCD optimizes a change field over the Gaussians of the reference map, refitting it to all past cue maps at every frame, a step that consumes more than half of the per-frame budget and forces accuracy to be traded against speed. We present a replay-free formulation that eliminates this trade-off. Because the reference map stays fixed, rendering the change field is a linear operation. Each new observation can therefore be carried onto the map’s Gaussians directly, by a single backward pass of the renderer, which reduces multi-view fusion to an exact two-scalar recursion per Gaussian. Past observations are never stored or revisited. The self-supervised objective that O-SCD minimizes by descent also determines the decision threshold exactly, by sorting, yielding a label-free per-frame cutoff whose margin provides a selective-prediction signal. On PASLCD, our method improves mean intersection over union from 0.489 to 0.575 and throughput from 11.8 to 25.6 frames per second, even exceeding O-SCD’s offline result while remaining strictly causal.
[CV-226] HumanSkill-4D: Training-Free Language Querying of Animatable Gaussian Avatars in Space and Time
链接: https://arxiv.org/abs/2610.24452
作者: unsu Kim,Naeun Kim,Dongyoon Han,Seungryul Baek
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 29 pages, 14 figures, 16 tables. Project page: this https URL
Abstract:We introduce avatar referring segmentation, which selects regions of animatable Gaussian avatars from language queries. A query can also specify a pose or motion to select matching intervals. This lets users interact with avatars for local editing and motion inspection. However, making these selections with existing language Gaussian splatting methods poses two challenges: adjacent parts and garments are hard to separate and may be referred to indirectly (region ambiguity), while fine-grained motions can look alike in RGB (motion ambiguity). We address these challenges with HumanSkill-4D, a training-free framework that shares a multimodal large language model (MLLM) and geometric operators across avatars. For each avatar, a semantic memory links body-part and garment categories to persistent Gaussian regions and appearance descriptions. The MLLM uses category names and descriptions to select categories by appearance, relation, or function. To find matching intervals, it composes operators into programs that check query conditions on driving joint trajectories stored in a motion memory. These category choices and operator outputs make selections traceable, while composing or adding operators supports new motion conditions. To evaluate this task, we introduce three human-annotated benchmarks and show that HumanSkill-4D outperforms the baselines in part segmentation and 3D and 4D referring segmentation, with compact memories and only minutes of setup per avatar.
[CV-227] Do 3D Large Language Models Really Use 3D Evidence?
链接: https://arxiv.org/abs/2610.24447
作者: Yifei Zhao,Fanyu Zhao,Han Zhang,Yixuan Lin,Yuyang Hou,Bin Wu,Ruike Cao,Li Xiao,Yinsheng Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent 3D Large Language Models (3D-LLMs) have shown strong performance on scene question answering with extensive post-training. Yet, existing benchmarks rarely evaluate how 3D-LLMs respond to referential ambiguity, insufficient scene evidence, or false presuppositions, obscuring whether their responses are grounded in 3D evidence. In this work, we introduce UNICORNS, a benchmark for evaluating the reliability of 3D-LLM responses across four evidence conditions. UNICORNS contains 11,387 questions across 557 ScanNet scans, covering eight challenge categories alongside answerable controls. A response passes only if it satisfies the required response behavior and contains no scene claim unsupported by the available evidence. Evaluation of six open-sourced 3D-LLMs reveals a pronounced performance gap: even the strongest zero-shot model achieves 61.22% accuracy on answerable controls, but only 40.94% across the eight challenge categories. To address this gap, we further propose HORN, an obligation-aware post-training method combining group-relative task rewards, leaf-conditioned obligation rewards, and supervised control replay. Extensive experiments demonstrate that HORN reaches 75.1 \pm 1.0% across the challenge categories and 77.2 \pm 0.8% overall, outperforming all baselines without degrading answerable-control accuracy. Our findings enable a systematic evaluation and improvement of evidence-grounded 3D reasoning for 3D-LLMs (code: this https URL).
[CV-228] BodyBridge: Grounded Body Layers for Sim-to-Real Translation of Vulnerable Road Users WACV2027
链接: https://arxiv.org/abs/2610.24436
作者: Mark Azer,Fabian B. Flohr
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:Simulation-to-real driving video diffusion models can translate photorealistic scenes, but they typically represent pedestrians and cyclists, vulnerable road users (VRUs), as generic object classes. This is insufficient for vulnerable road users whose safety-relevant appearance depends on articulated pose and identity. We present BodyBridge and Grounded Body Layers (GBL), a simulator-derived, pixel-aligned representation that combines scene geometry, canonical per-pixel body correspondence, and exact VRU masks. GBL provides explicit articulation to the video diffusion model, emphasizes VRU regions during control-branch post-training, and supports additive composition with scene-geometry control at inference. It also enables per-joint evaluation of generated VRUs and a failure taxonomy for identity errors. On CARLA-to-real translation, BodyBridge improves near-range PCK@0.5 by 22% over the released depth-conditioned baseline. Replacing normal-map body control with canonical correspondence reduces body flips by 18%, while composition reduces near-range facing flips by 27%. These results show that explicit body grounding complements scene-level control and improves pose-faithful VRU generation. All code, data, and model checkpoints will be made publicly available to support reproducible, pedestrian-aware autonomous driving research.
[CV-229] S3D-Mosaic: Subgradient-Guided Cross-Band Structure Transfer for Spectral Demosaicing
链接: https://arxiv.org/abs/2610.24414
作者: Weiqiang Zhao,Xian Li,Yuzhe Gui,Tianzhu Liu,Yanfeng Gu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multispectral filter array (MSFA) imaging enables snapshot spectral acquisition, but its per-band sparse sampling causes severe spatial aliasing. Although missing spatial structures can be inferred across bands, wavelength-dependent contrast variations make direct cross-band transfer unreliable. Existing model-based methods lack explicit magnitude-invariant mechanisms, and learning-based methods require layout-specific training or per-image tuning, limiting generalization. To address these issues, we propose S3D-Mosaic, a model-based spectral demosaicing method built upon a subgradient-guided cross-band structure transfer framework. We first establish a subgradient-guided regularization framework with a generalized cosine-similarity interpretation, providing a convex mechanism for magnitude-invariant information transfer. Then we design a compact Directional Sparse Equivariant Difference Convolution (DSE-DConv) for band-level extraction of orientation-selective, high-frequency structure features at low computational cost. Instantiating the framework with DSE-DConv on spectrally adjacent bands yields a cross-band self-structure similarity regularizer for spectral demosaicing, which we optimize via a dual-scan ADMM algorithm that decouples the bidirectional dependency while propagating magnitude-robust structural information across bands. With DSE-DConv’s two parameters trained once via self-supervision, S3D-Mosaic outperforms state-of-the-art unsupervised methods on CAVE and NTIRE 2022, and generalizes robustly to varying MSFA layouts, pan-mosaic fusion, and real-world UAV data with fixed reconstruction hyperparameters. The source code is available at \urlthis https URL.
[CV-230] ZeroDSS: Semantic Segmentation from Class Names Alone
链接: https://arxiv.org/abs/2610.24405
作者: Tian Ming,Marzi Heidari,Yuhong Guo
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 24 pages
Abstract:Semantic segmentation depends heavily on real images with pixel-level annotations, making dense supervision a major obstacle to deployment in new domains. We study task-specific data-free semantic segmentation, where training uses generic pretrained generative priors but has access to no target-domain images, masks, or captions–only a predefined category set. We introduce Zero-Data Semantic Segmentation (ZeroDSS), a feedback-driven approach that treats synthetic supervision as an adaptively controllable training distribution rather than a direct surrogate for real data. ZeroDSS constructs a semantically structured prompt distribution through LLM-guided compositional generation and adaptive class rebalancing, then converts the prompts into image-mask pairs using diffusion attention. Its feedback-driven distribution refinement consists of a two-phase closed-loop training procedure with a single loop-back step. In Phase I, a segmenter is trained using confidence-weighted supervision and dual-scale consistency. In Phase II, model feedback is used to adaptively target underperforming categories and refine diffusion-derived masks using model predictions. After this refined initial training, a teacher-student framework is utilized in the final phase to iteratively improve the fused supervision and segmenter through exponential moving average (EMA) updates. Using synthetic training data only, ZeroDSS achieves 35.0% mIoU on MS COCO with DeepLabV3, and 66.3% and 63.5% mIoU on PASCAL VOC with DeepLabV3 and Mask2Former, respectively. These results show that explicitly constructing and iteratively refining the synthetic supervision distribution can substantially reduce the synthetic-to-real gap in dense prediction.
[CV-231] Spatons: Spatiotonally Localized Building Blocks for Efficient XR Predistortion
链接: https://arxiv.org/abs/2610.24400
作者: Mohammadjavad Maheronnaghsh,Sushant Kondguli,Steve Mann,Andreas Moshovos
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We present Spatons, a family of compositional, spatially localized neural operators for computationally efficient XR (extended reality) predistortion and distortion compensation. XR displays (AR, VR, and MR) suffer from spatial (e.g., geometric) and tonal (e.g., colour) distortions. Frames can be pre-distorted so that the projected image looks as intended. Distortions vary with manufacturing tolerances and projection systems, motivating programmable compensation. Untethered, battery-operated XR systems also require low energy and latency. A Spaton is a hybrid patch-to-pixel block combining tiny Multi-Layer Perceptrons (MLPs) with fixed-function interpolation. Using Spatons, we build programmable, computationally efficient pipelines for varied distortion scenarios. On the full spatiotonal Scenario 5, Sequential-5’ matches CompenNet++ in SSIM (0.944 vs. 0.944) and closely matches its PSNR (27.5 vs. 27.7 dB), with substantially lower complexity. We demonstrate a flexible Spaton solution with 6 times fewer FLOPs and 106 times fewer parameters than CompenNet++, and 4-10 times fewer memory transfers than UNet++. We characterize software implementations and analytical hardware estimates, including optimizations across target XR platforms. The distortions considered are drawn mostly from see-through, near-eye displays, where the spatial light path governs local intensity and spectral response. This early systems-level exploration uses closed-form differentiable distortions and one real projector-camera capture set for training and evaluation, rather than through-the-optics measurements. Testing on a wider range of real XR devices remains future work. Chromatic dispersion (color fringes) remains the main artifact where Spatons trail monolithic baselines.
[CV-232] MSCR: Jointly Balancing Utilization Across Modalities and Discovering Synergistic Information NEURIPS2026
链接: https://arxiv.org/abs/2610.24395
作者: Xinyu Chen,Liangjian Wen,Jiang Duan,Dongkai Wang,Yong Dai,Jiayu Bai,Jianzhuang Liu,Zhen Tian,Guoping Qiu,Zhao Kan
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026
Abstract:In multimodal learning, modality imbalance often lets one modality dominate optimization and keeps the model from fully exploiting cross-modal interactions. Balance-oriented methods such as the Multimodal Competition Regularizer (MCR) perform well by alleviating this imbalance during training. These dynamic balancing methods, however, mainly adjust modality contributions and do not consider how imbalance affects the formation of synergistic information during fusion. Synergy is a major source of multimodal advantage, because it is predictive gain that cannot be recovered from unimodal evidence alone. We propose the Multimodal Synergistic Competition Regularizer (MSCR), a fusion method that jointly promotes balanced modality utilization and the discovery of synergistic information. MSCR has two complementary objectives. An anti-dominance objective suppresses excessive unimodal influence and leaves room for synergistic information, and a synergy objective encourages the fused predictor to outperform a non-synergistic unimodal reference. An adaptive gate further concentrates the regularization on samples that need stronger collaborative modeling. Experiments on five benchmarks show that MSCR consistently improves performance over MCR and strengthens synergistic capability. It raises conditional recovery, the accuracy of the fused prediction on samples that the unimodal predictors misclassify, as well as the share of the model’s output that InterSHAP attributes to cross-modal interactions. Its gain in conditional recovery persists when the fusion module is Transformer-based. Code is available at this https URL.
[CV-233] Learning from the Right Regions: Class-Conditioned Explanation Guidance for Medical Imaging
链接: https://arxiv.org/abs/2610.24381
作者: Zubair Faruqui,Rahul Dubey
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Submitted to the Explainable and Trustworthy AI for Clinical Decision Support Systems Workshop, IEEE BIBM 2026
Abstract:Deep neural networks can achieve strong predictive performance in medical imaging while relying on regions unrelated to the predicted finding. Explanation-guided learning can address this problem, but multi-label classification introduces a key challenge: multiple abnormalities may coexist in an image, requiring each prediction to align with its corresponding spatial evidence. We propose a class-conditioned explanation-guided learning framework that constructs a separate explanation for each positive class and penalizes evidence outside its class-specific annotated regions. We instantiate the framework using Gradient-weighted Class Activation Mapping (Grad-CAM) and evaluate it on 14 abnormalities using EfficientNet-B1, ViT-B/16, ViT-B/32, and Swin-B. Explanation guidance substantially improves annotation alignment across all architectures: All-Saliency Precision increases from 0.140 to 0.625, 0.052 to 0.551, 0.060 to 0.403, and 0.095 to 0.570, respectively. These improvements are accompanied by smaller, architecture-dependent changes in classification performance. Our results show that class-conditioned explanation guidance can substantially improve where multi-label classifiers concentrate their explanatory evidence while largely preserving predictive performance.
[CV-234] PACE-Mem: Prefix-Aware Complementary Evidence for Multimodal Agents
链接: https://arxiv.org/abs/2610.24378
作者: Shidu Ren,Rong-cheng Tu,Hang Zhou,Xiao Luo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 5 figures. Code: this https URL
Abstract:As multimodal agents accumulate experience, their long-term memories contain many similar observations of recurring scenes, actions, and entities. Answer quality depends on the evidence retrieved from these memories into the model’s context. However, query-relevant memories may contain overlapping information, and a fixed retrieval depth may return too little or too much evidence for a given question. To address these challenges, we propose PACE-Mem, Prefix-Aware Complementary Evidence, which constructs a complementary evidence sequence and determines its length from the same selection trajectory. For structured memories, query-guided pooling combines records within each clip to obtain relevance scores and representations. Using these relevance scores and representations, maximal marginal relevance (MMR) selection balances query relevance against overlap with previously selected evidence. From the resulting score trajectory, a prefix-length selection algorithm determines retrieval depth by comparing length-normalized adjacent declines, accounting for both score changes and the amount of evidence accumulated. PACE-Mem is a training-free, plug-in retrieval method for clip-based agent memory systems, leaving memory construction and answer generation unchanged. Across three long-video benchmarks and two agent frameworks, PACE-Mem improves question-answering accuracy by an average of 1.6% over Adaptive-k, the strongest baseline at comparable mean retrieval depths. Further analysis reveals that jointly adding and removing evidence is associated with larger net accuracy gains than either change alone.
[CV-235] Hexagonal Basis Projectors for Model-Based Iterative CT Reconstruction
链接: https://arxiv.org/abs/2610.24363
作者: Ke Chen,Alireza Entezari
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Image representation affects both the forward model and quantitative feature recovery in model-based iterative CT reconstruction. For isotropically bandlimited signals, the hexagonal lattice is the most sampling-efficient lattice in two dimensions, with greater directional isotropy and a larger alias-free bandwidth than the Cartesian grid at equal density. We develop constant and linear hexagonal projectors for two-dimensional fan-beam CT using box spline calculus. Both bases form a partition of unity on common coefficient centers; the linear tent also reproduces affine functions. We derive analytic ray footprints and effective detector-blur expressions and implement efficient GPU forward and backprojection. On simulated FORBILD data, the linear model improves on Cartesian and hexagonal pixels, with its best noisy-data accuracy at the coarser resolution. Real-world 2DeteCT comparisons use high-dose, high-resolution Cartesian reconstructions as pseudo ground truth and show competitive image and feature agreement. These results establish a practical hexagonal representation for quantitative CT.
[CV-236] CRISP: Progressive Illumination Curriculum for Recurrent Low-Light Video Enhancement with Semantic Priors
链接: https://arxiv.org/abs/2610.24299
作者: Ruirui Lin,Xin Yan Lim,Yini Li,Zhiming Liu,Nantheera Anantrasirichai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Low-light video enhancement requires the recovery of brightness, color, and fine spatial structures while preserving temporal coherence in the presence of noise and motion. Existing methods often struggle to balance these objectives, particularly in severely degraded regions. This paper presents CRISP, a recurrent framework trained using a progressive illumination curriculum. Its two-frame recurrent architecture propagates information across successive frames to exploit temporal context beyond the immediate input pair. During training, supervision progressively transitions from intermediate targets interpolated between the low-light input and the reference to the full reference, gradually increasing the required brightness and color correction. This curriculum stabilizes recurrent training without introducing additional inference cost and yields the largest performance improvement among the components evaluated in the ablation study. To further support structural reconstruction, the framework optionally incorporates frozen DINOv3 semantic features through adaptive layer normalization (AdaLN)-style modulation, without fine-tuning the foundation model. Experiments on DID and BVI-RLV demonstrate that CRISP achieves the highest PSNR on DID and the highest SSIM on BVI-RLV among the evaluated methods, while maintaining computational efficiency.
[CV-237] ULIFLOW: Unified Latent Human-Object Interaction Flow
链接: https://arxiv.org/abs/2610.24286
作者: Wenqi Cai,Chunzhi Gu,Weichen Zhou,Yawen Zou,Yijin Wei,Ran Dong,Haoran Xie,Chao Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-conditioned 3D human–object interaction (HOI) generation remains challenging because individually plausible human and object motions do not necessarily form a coherent interaction. We argue that a key limitation lies in conventional skeleton-based representations, which compactly describe body kinematics while leaving surface-level interaction geometry largely implicit, often requiring additional contact-specific regularizers. We propose ULIFlow, a framework that explicitly models continuous human–object relations within a unified latent space. To this end, we derive continuous marker–object relations from sparse body-surface markers and treat them as an explicit interaction modality alongside human motion and object dynamics. A Mixture-of-Transformers (MoT) autoencoder is then designed to jointly model these heterogeneous streams while preserving their modality-specific structures. Importantly, we let the relation stream directly participate in human and object reconstruction, encouraging the shared latent representation to preserve both individual dynamics and the interaction geometry that coordinates them. To further achieve variable-length generation, we introduce a blockwise Flow Forcing scheme with global sequence-progress conditioning, enabling progressive continuation from fixed-length training windows while maintaining consistent long-horizon dynamics. Extensive experiments demonstrate that ULIFlow consistently outperforms prior work in overall generation quality and human–object coordination, while preserving coherent interactions over requested durations beyond the training horizon.
[CV-238] Ghost Attack: Using 19th-Century Illusions to Fool 21st-Century Biometrics WACV
链接: https://arxiv.org/abs/2610.24238
作者: Tyler Fitzsimmons,Adam Czajka
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV (Winter Conference on Applications for Computer Vision) 2027
Abstract:While the Pepper’s Ghost illusion is a 19th-century optical technique, we found that this forgotten gimmick exposes vulnerabilities in modern face recognition systems. This paper introduces the ‘Ghost Attack’: a physical attack that repurposes this classic illusion using low-cost, free-space hardware and a transparent medium to project high-resolution facial data into physical 3D space. We show in the paper that despite relying on centuries-old optical principles, this new attack type simultaneously breaks three components of modern facial recognition systems (FRS): (a) the face matcher, (b) the presentation attack detection module, and © the morphing attack detection component, resulting in high impersonation success rates against state-of-the-art algorithms without triggering presentation or morphing defenses. These findings reveal a significant blind spot in current biometric systems, which fail to account for transparent projections and evolving physical attack vectors. To support further research, we are offering the dataset of face images of 100 identities, re-captured with and without ‘ghost’ projections as a new benchmark representing a novel attack for face recognition.
[CV-239] PREEMPT: Proactive Robustness Evaluation under Editing for Media Protection Techniques WACV2027
链接: https://arxiv.org/abs/2610.24205
作者: Qiuyu Tang,Aparna Bharati
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 6 figures, accepted by WACV2027
Abstract:Proactive image protection methods aim to defend visual content against unauthorized manipulation and generative misuse by embedding protective signals into images before release. Evaluating whether these protections remain effective under image editing is therefore critical. However, existing evaluations remain largely limited to one-shot edits, providing only a static view of robustness while overlooking its evolution under realistic multi-step editing workflows. To systematically characterize robustness dynamics under sequential editing across various protection paradigms, we introduce PREEMPT, a holistic evaluation framework. It evaluates protection robustness under both one-shot and sequential editing scenarios, covering image transformations, semantic edits, homogeneous and heterogeneous edit compositions. To enable consistent analysis across protection paradigms with different underlying objectives, PREEMPT normalizes task-specific effectiveness into protection survival trajectories and summarizes long-horizon robustness through trajectory-based diagnostics. Using this framework, we reveal that degradation behaviors remain hidden under conventional one-shot evaluation. Furthermore, perturbation- and watermark-based protection methods exhibit different failure modes, and the first semantic edit largely determines subsequent degradation trajectories. PREEMPT provides a standardized protocol for studying long-horizon robustness and offers new insights into the degradation dynamics of proactive image protection.
[CV-240] Precision Cotton Disease Management: Leverag ing UAV Imagery and Custom YOLO Deep Learning Framework for Enhanced Detection
链接: https://arxiv.org/abs/2610.24182
作者: 1. Anwar Iqbal*1 aiqbal.rime23smme@student.nust.edu.pk 2. Kunwar Faraz Ahmad Khan1 kunwar.faraz@smme.nust.edu.pk 3. Shahbaz Khan1 shahbaz.khan@smme.nust.edu.pk 4. Khawaja Fahad Iqbal1 fahad.iqbal@smme.nust.edu.pk 5. Sara Ali1,2 sara.baber@smme.nust.edu.pk 6. Saqib Nazir1 saqib.nazir@smme.nust.edu.pk 7. Muhammad Azhar Javed3 muhammadazhar6231@gmail.com 1 Department of Robotics,Artificial Intelligence,SMME,National University of Science,Technology(NUST)Islamabad,Pakistan 2 Suzhou Industrial Park Monash Research Institute of Science,Technology,Suzhou,PR China 3 Department of Plant Pathology,Bahauddin Zakriya University,Multan,Pakistan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Cotton sustains the textile economies of over 80 countries with an annual yield of 113 million bales, yet 30-60% of yield losses are still attributed to viral, bacterial, and fungal diseases that remain undetected until manual field scouting. To address this critical issue, the integration of Unmanned Aerial Vehicle (UAV) imagery with deep learning offers a scalable early-warning system; however, existing models suffer from low early-stage accuracy, poor transferability, and high computational cost. In this paper, we propose Cotton-Vision, a fully convolutional single-stage detector that synergizes three lightweight architectural innovations: Ghost Bottleneck modules for a 20% parameter reduction, Efficient Channel Attention (ECA) for adaptive feature recalibration, and Cross-Stage Partial (CSP) fusion for enriched gradient flow. The network is trained on a novel dataset collected across seven cotton fields in the Punjab province of Pakistan, comprising over 2,700 high-resolution drone images. Extensive data augmentation (rotation, shear, exposure, and saturation) expands the dataset to 11,704 training samples. On a held-out test set, Cotton-Vision achieves a precision of 97.7%, a recall of 77.7%, an mAP@50 of 85.6%, and an mAP@50-95 of 76.6%. These results represent a 10-20% improvement in precision, recall, and mAP over state-of-the-art baselines, including YOLOv5, YOLOv8, YOLOv11, YOLOv12, RT-DETR, and RF-DETR. Through a comprehensive ablation study, we demonstrate that each proposed architectural enhancement contributes to the model’s superior performance, yielding a 12-16% mAP improvement over the baseline architecture. Our findings confirm that this synergy of lightweight and attention-based modules provides a robust, computationally viable solution for early cotton disease detection.
[CV-241] Self-Improving Video-to-Action Models
链接: https://arxiv.org/abs/2610.24177
作者: Hyunwoo Ryu,Eric Ming Chen,Max Wilcoxson,Vincent Sitzmann
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 33 pages, 18 figures, 6 tables. Project page: this https URL
Abstract:Video generative models are an emerging tool for embodied agents: they can serve as scalable visual planners, generating videos that depict how to complete previously unseen tasks. However, executing these plans typically requires a video-to-action or “inverse dynamics” model trained on large, human-collected video-action datasets, severely limiting scalability. We introduce Mirror, a self-supervised approach that learns video-to-action mappings through environment interactions, without prior action data or hand-crafted rewards. Mirror learns by attempting to follow a visual plan by executing its best guess of the underlying actions in the environment. Mirror then trains on the resulting video-action pairs, incrementally self-improving and following the visual planner more closely. Across Atari, OGBench, and DeepMind Control, Mirror outperforms leading learning-from-observation methods with learned video plans, and achieves near-expert performance with oracle plans. On Minecraft BASALT, it outperforms methods with access to expert actions or hand-engineered task knowledge. In Minecraft survival, it is the first method to acquire an iron pickaxe through self-improvement alone, without rewards or expert action labels. In the offline setting, with expert action labels but no environment interaction, it sets a new state of the art in obtaining diamonds. Our results suggest that advances in video models can translate into increasingly capable agents without a corresponding need for hand-crafted rewards or large collections of expert action data.
[CV-242] SERO: Semantic Prediction for Single-Step Object Removal
链接: https://arxiv.org/abs/2610.24173
作者: Amreen Hasheem Siraj,Alireza Javanmardi,Pragati Jaiswal,Didier Stricker,Alain Pagani
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at CVMP 2026. Project page: this https URL
Abstract:Object removal and semantic inpainting of missing regions remain inherently ill-posed, as multiple plausible completions may exist for a given scene. Recent diffusion-based methods have demonstrated strong potential for high-quality object removal; however, they still face two key limitations. First, although they can synthesise sharp and high-resolution images, they often struggle to complete removed regions in a semantically consistent and contextually meaningful manner, where the generated content should agree with the surrounding scene layout and object relationships rather than merely appear visually realistic. Second, their iterative denoising process makes them computationally expensive, limiting their practicality for downstream applications. We propose SERO, a semantic prediction framework for single-step object removal. Our model predicts a hidden-region semantic prior in embedding space, which guides a latent bridge-matching generator toward plausible scene completion by providing both where to edit and what structure is likely to exist behind the removed object. This enables visually realistic and contextually coherent completions in one forward pass. Experiments on the RORD and OBER datasets show that SERO achieves competitive or state-of-the-art performance while requiring only a single generation step.
[CV-243] NV-Generate-Body: 3D Latent Diffusion Model forControllable Anatomy Tumor and Demographic Body Mask Synthesis
链接: https://arxiv.org/abs/2610.24153
作者: Can Zhao,Pengfei Guo,Dong Yang,Yufan He,Yucheng Tang,Ahmed Harouni,Benjamin Simon,Mason Belue,Stephanie Harmon,Baris Turkbey,Daguang Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code and model weights open-sourced
Abstract:Curating large labeled CT datasets is expensive and constrained by patient privacy. We present NV-Generate-BodyMask, a 3D latent diffusion model that synthesizes whole-body CT segmentation masks with precise, joint control over organ anatomy, tumor burden, and patient demographics. The model operates in the compressed latent space of a frozen mask autoencoder, generating 256^3 volumetric label maps conditioned on a 19-dimensional vector via Adaptive Group Normalization and cross-attention. A size-adaptive ROI loss with spatial halo weighting and a paired tumor injection pipeline enable accurate tumor size conditioning despite the scarcity of annotated tumor cases. Trained on 8,571 masks across 26 cohorts spanning 132 labels, the model achieves Spearman \rho = 0.70 – 0.92 for organ size, 85–100% tumor recall for lung, pancreatic, colon, and bone lesions, and strong demographic alignment (BMI \rho = 0.78 , sex-conditioned perfect prostate control).
[CV-244] HAVANA: A Hierarchical Video-Action Model for Vision-Language Navigation
链接: https://arxiv.org/abs/2610.24122
作者: Dongzhihan Wang,Yi Du,Chen Wang,Xiuxian Li,Liang Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 10 pages,6 figures
Abstract:Vision-Language Navigation (VLN) requires long-horizon instruction-consistent planning and short-horizon local planning in current geometry. Video-action models (VAMs) offer a promising way to learn predictive representations for navigation. However, existing VAMs jointly predict visual futures and local trajectories in a single process, even though they operate at different paces: future prediction remains stable across multiple updates while local trajectories must react to fresh observations. To bridge this gap, we introduce HAVANA, a dual-system VAM that decouples predictive reasoning from reactive local planning. A slow system produces a compact language-conditioned latent plan from observation history, while a fast system combines the latest plan with the current observation to generate local trajectories. Their asynchronous execution enables rapid adaptation without repeated future prediction, and no future video is generated during inference. Experiments on OmniNavBench demonstrate state-of-the-art navigation performance over prior baselines, while real-world deployments further show robust generalization to unseen environments across different robot embodiments. Our code and checkpoints will be available at this https URL.
[CV-245] BLANC: Adaptive Temporal Scheduling for Sign Language Production
链接: https://arxiv.org/abs/2610.24109
作者: Youngmin Kim,Seoyeong Hwang,Sujung Hong,Seong Jae Hwang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Sign language production (SLP) aims to synthesize expressive signing motion from written text. Modeling signing motion presents two key temporal challenges. First, fine-grained hand articulations demand higher temporal resolution than relatively slower body movements. Second, signing events unfold over variable durations, calling for adaptive temporal scheduling during generation. To address these challenges, we first introduce a Multi-Rate Vector Quantizer (MRVQ), an articulation-decoupled motion tokenizer that represents the body and both hands as separate discrete streams at distinct temporal resolutions. Building on this representation, we propose BLANC (BLock-wise Adaptive diffusioN with Articulatory Clock), a gloss-free SLP framework that generates variable-duration motion blocks causally while denoising tokens in parallel within each block. At its core, a Latent Articulatory Clock (LAC) learns temporal structure from multi-scale motion-derived partitions to predict block durations and sequence termination during generation. We further introduce inference-time body-conditioned hand refinement to improve coordination across articulators. Extensive experiments on PHOENIX-2014T and CSL-Daily demonstrate that BLANC achieves state-of-the-art performance in sign language production.
[CV-246] Hierarchical and Query-Aware Visual Token Pruning for High-Resolution LVLMs
链接: https://arxiv.org/abs/2610.24102
作者: Luther Wei,Ling Zou,Kang-Yang Huang,Yu-Syuan Xu,Wen-Huang Cheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 6 pages, 2027 ICCE
Abstract:High-resolution Large Vision-Language Models (LVLMs) preserve critical, fine-grained visual details by encoding a global thumbnail alongside multiple local crops. However, the resulting thousands of vision tokens introduce severe computational bottlenecks during Large Language Model (LLM) inference. Existing token pruning methods rely primarily on query-agnostic visual saliency, while current query-aware approaches either operate at a coarse crop-level resolution or compute token relevance too late, after vision tokens have already entered the heavy LLM layers. To address this, we propose a training-free, hierarchical token pruning framework that utilizes query guidance prior to executing any LLM blocks. Our top-down hierarchy first operates at the instance-level, dynamically scaling the vision token retention ratio of each sample based on its visual complexity. Next, at the crop-level, we distribute this sample budget across local crops using sentence-level cross-modal similarity. Finally, at the token-level, we exploit projected vision tokens alongside textual query token embeddings in the shared space to measure fine-grained alignment, applying a text-debiasing normalization to produce high-fidelity query scores. These query-guided budgets and scores are fused with local and global visual saliency metrics to select the most critical vision tokens. Extensive experiments on LLaVA-NeXT-7B demonstrate that our plug-and-play framework retains 98.0% and 94.8% of unpruned performance at 25% and 10% vision token retention, respectively. Notably, at 25% retention, our method delivers a 2.0x inference throughput speedup and even surpasses the unpruned baseline on POPE by 0.7 points.
[CV-247] BiNar: Bitstream Native Restorer for Bitstream-corrupted Image Restoration
链接: https://arxiv.org/abs/2610.24094
作者: Fangcheng Li,Zhen Yu,Kejun Wu,Tianyi Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Bitstream-corrupted image restoration (BCIR) aims to reconstruct faithful images from corrupted compressed bitstreams. However, corruption occurs in bitstream-domain may lead to unpredictable error propagation and distortion amplification in pixel-domain, and even result in complete undecodability by standard decoders, posing significant challenge to existing image restoration paradigms. To address this challenge, we propose Bitstream Native Restorer (BiNar), a generative restoration framework that leverages fine-grained bitstream-native semantics to guide bitstream-corrupted image restoration. In BiNar, Bitstream Semantic Modeling (BSeM) extracts robust informative semantics and models sequential bytes directly from corrupted bitstream, enabling to generate fine-grained bitstream descriptions without relying on pixel decoding. Based on the descriptions,the Corruption-adaptive Multimodal Semantic Aggregation (CMSA) further extracts adaptive visual modal cues in decoded corrupted images and aggregates to bitstream descriptions. The multimodal aggregated fine-grained descriptions can serve as coherent restoration guidance. Finally, Blind Corruption Localization (BCoL) exploits the segmentation prior of visual foundation models to identify and locate error regions that requires recovery in corrupted images. The located error regions and the fine-grained descriptions jointly serve as visual prompt and language prompt of diffusion models image restoration, respectively. Extensive experiments under diverse bitstream corruption conditions show that BiNar outperforms all State-of-the-arts image restoration models regardless of perceptual quality or objective quality. Ablation studies demonstrate the effectiveness of fine-grained bitstream-native semantics in BCIR task.
[CV-248] OrTeN: Neural Inverse Graphics for Joint Terrain and Line-scan Camera Trajectory Reconstruction
链接: https://arxiv.org/abs/2610.24075
作者: Josef Biberstein,Sertac Karaman
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Machine Learning (cs.LG)
备注:
Abstract:Digital terrain maps (DTMs) are used for landing site selection, terrain-relative navigation, and geological analysis in planetary exploration. Existing state-of-practice reconstruction techniques for creating DTMs from satellite imagery use multi-view stereo, requiring complex pipelines for dealing with camera error, changing illumination, and images with limited disparity. This work introduces Orbital Terrain Networks (OrTeN), a DTM construction method that uses neural inverse rendering to jointly optimize a textured height map and orbital camera corrections from line-scan satellite images. OrTeN operates directly on calibrated line-scan images with pixel values expressed as radiance factor (I/F) and corrects both broad trajectory pose inconsistencies and high-frequency pointing jitter jointly with surface reconstruction during training. Surface-projected patch normalized cross-correlation (NCC) enforces consistency between observations of the same terrain, refining an initial rough reconstruction. A semi-empirical reflectance model and shadow rendering are used to fit surface appearance from multi-date imagery with differing solar illumination. OrTeN requires only calibrated line-scan images and initial camera poses and does not use any structural priors including ranging/depth measurements or initial rough DTMs. OrTeN is evaluated on four planetary imagery datasets. Aligned terrain RMSEs are 2.35 m at Jezero Crater, 2.82 m at Gale Crater, 16.18 m at Olympus Mons, and 0.43 m at McGetchin Crater.
[CV-249] Generative Text Modelling and Language-Guided Visual Attention for Deciphering Palimpsests ACL
链接: https://arxiv.org/abs/2610.24060
作者: Robin Hollifeldt,Tianru Zhang,Jessica Ström,Ekta Vats
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: This is a pre-MIT Press publication version, accepted for publication at TACL
Abstract:Recovering obscured text in palimpsests, i.e. the manuscripts where the original writing (under-text) was erased and overwritten for reuse, is vital for preserving lost information. Due to the nature of the material, such a task can be highly challenging and time-consuming. This paper presents an approach towards automated palimpsest handwritten text modeling for under-text recovery. We investigate generative models, including Generative Adversarial Networks (GANs) and diffusion models, and design a custom denoising diffusion model that learns structured noise patterns to reconstruct the under-text. The recovered text is subsequently processed by a fine-tuned, attention-based handwritten text recognition network. Furthermore, we investigate a language-guided recognition framework that leverages textual class descriptions as informative priors, enabling the text recognition model to distinguish meaningful features from noise. This is achieved by grounding task-specific descriptions, phrased in words or sentences, through visual attention maps generated by pretrained vision-language models like CLIP (Contrastive Language-Image Pre-training). To facilitate future research and benchmarking, we release new datasets for synthetic palimpsests (currently scarce), CLIP-compatible text descriptions, and pre-trained models with code.
[CV-250] SynerGest: Temporal-Spatial Synergistic Gesture Generation for Dyadic Conversations ACM-MM2026
链接: https://arxiv.org/abs/2610.24031
作者: Chengxiang Guo,Jian Wang,Lanmiao Liu,Zhan Tang,Junhua Fei,Xiao Li
类目: Computer Vision and Pattern Recognition (cs.CV); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: accepted by ACM MM 2026
Abstract:Co-speech gesture generation in dyadic conversations for creating immersive virtual avatars remains highly challenging. Existing models often struggle to balance kinematic stability with rhythmic responsiveness and suffer from semantic entanglement, leading to either generic mean-pose regression or unnatural micro-jitters. To address these challenges, we propose SynerGest, a high-fidelity dyadic gesture generation framework grounded in Contrastive Flow Matching (CFM). Specifically, our CFM objective explicitly penalizes mismatched semantic pairs during training, discouraging spatial shortcuts and over-smoothed predictions while improving conditional semantic alignment. Furthermore, to overcome architectural bottlenecks, we introduce the Impulse-Trend Synergistic Attention (ITSA) module. Grounded in the key insight of decoupling “what moves” from “when it moves,” ITSA models sharp, speech-driven rhythmic impulses separately from continuous kinematic trends, supporting precise audio-motion synchronization and global structural stability. Extensive qualitative and quantitative evaluations on two public datasets demonstrate that SynerGest achieves state-of-the-art performance, delivering contextually responsive and kinematically fluid dyadic interactions.
[CV-251] CLARO: Clutter Aware Robot Orchestrator for Open-Vocabulary Grasping using VLM Agents WACV2027
链接: https://arxiv.org/abs/2610.24016
作者: Sungsu Kim,Vineet Bhat,Prashanth Krishnamurthy,Ramesh Karri,Farshad Khorrami
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:Robots that follow natural-language instructions unlock a wide range of autonomous applications in households, warehouses, and labs. Real-world environments are diverse and often heavily cluttered, requiring robots to plan over multiple steps, act, and replan as they work toward a goal. We present CLARO (CLutter-Aware Robot Orchestrator), a VLM agent that orchestrates a set of perception and manipulation skills: open-vocabulary object detection, segmentation, 6-DoF grasping, and non-prehensile pushing to locate and retrieve a language-specified target in clutter. CLARO maintains a memory of its past actions to support long-horizon execution, and emits manipulation targets (grasp poses/3D waypoints) allowing transfer across robot arms without task-specific training. We evaluate CLARO against a range of methods on a dynamic simulation environment and standard image-based benchmarks, demonstrating strong generalization across object classes, lighting, and clutter levels. CLARO attains an average success rate of 89.1%, a relative improvement of over 16% on the strongest competing method in high cluttered scenes. We further analyze performance along grounding and grasping axes to characterize where each method succeeds and fails. Finally, we deploy CLARO on a robot arm across diverse cluttered scenes, achieving a success rate of 88.8%, and demonstrating strong sim2real generalizability.
[CV-252] ChairNav: Cross-Embodiment Pre-training and Personalization for Long-Horizon Wheelchair Navigation
链接: https://arxiv.org/abs/2610.24012
作者: Junhua Huang,Zhizheng Liu,Honglin He,Xiang Zhang,Bolei Zhou
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages,8 figures,4 table
Abstract:Developing autonomous electric wheelchairs capable of long-horizon sidewalk navigation is critical for accessible personal mobility, yet existing approaches either focus on indoor map-based navigation or rely on expensive, complex sensor stacks to achieve full autonomy. In this work, we propose to tackle outdoor sidewalk navigation by leveraging visual navigation foundation models and large-scale navigation datasets, which offer strong generalization using lightweight hardware. To this end, we present ChairNav, a novel framework that adapts a general visual navigation model into a personalized electric wheelchair navigation system using a three-stage data pyramid. To address the scarcity of in-domain data, ChairNav first pre-trains on large-scale cross-embodiment datasets transformed into the wheelchair domain, followed by mid-training on a small number of wheelchair demonstrations to eliminate remaining domain gaps. Finally, a post-training stage refines the deployed model from user interventions to personalize its behavior. We deploy ChairNav on a commercial electric wheelchair equipped with an RGB-D camera and an onboard NVIDIA Jetson compute module. Experimental results validate the effectiveness of each stage in our data pyramid: pre-training reduces lateral ADE by 18%, and post-training increases the mean distance between user interventions from 29.2 m to 57.4 m. Further evaluation on kilometer-scale urban sidewalk routes shows that our system operates autonomously for 93.4% of total driving time, demonstrating its potential for everyday accessible mobility. Demo video and more details are at this https URL.
[CV-253] PACE: Pareto-Adaptive Compression for Efficient Native MLLM s NEURIPS2026
链接: https://arxiv.org/abs/2610.24005
作者: Lei Tian,Chunzheng Zhu,Kai Wu,Yifan Zhang,Haihua Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: NeurIPS 2026 Accepted Paper
Abstract:Native multimodal large language models (native MLLMs) jointly encode visual and linguistic sequences within a unified autoregressive Transformer, emerging as a promising paradigm for multimodal understanding. Unlike conventional ViT-based MLLMs relying on external vision encoders, ViT-free native MLLMs tokenize pixel patches directly, potentially suffering severe visual token redundancy. A pilot study confirms this: unlike ViT-based models that degrade under compression, ViT-free MLLMs tolerate aggressive compression with near-lossless or even improved performance. Yet existing compression methods target ViT-based settings, leaving native MLLMs, which stand to gain the most, unexplored. To close this gap, we introduce an analysis pipeline centered on token discriminability and cross-dimensional heterogeneous redundancy, revealing that (i) redundancy is distributed independently and heterogeneously along height and width axes, and (ii) early-layer visual tokens are indiscriminable and uniformly redundant. Guided by these findings, we propose PACE (Pareto-Adaptive Compression for Efficient Native MLLMs), the first visual token compression framework tailored for native MLLMs with Multimodal-RoPE. Leveraging dimensional decoupling of positional encoding, PACE decomposes attention into two orthogonal spatial saliency estimates along height and width, gated by a positional attention signal, yielding direction-aware importance with negligible overhead. It further casts token budget allocation as Pareto optimization over performance preservation, inference efficiency, and spatial coverage, jointly identifying optimal compression depths and retention ratios. Experiments show PACE maintains near-lossless performance at only 25%-50% visual token retention, even improving several benchmarks, while delivering over 2x FLOPs reduction.
[CV-254] spora: A Unified Multimodal Dataset and Platform for Spatial Proteomics NEURIPS2026
链接: https://arxiv.org/abs/2610.24002
作者: Benedikt von Querfurth,Johann Wenckstern,Eeshaan Jain,Lukas Klein,Phil Cheng,Yexiang Cheng,Cedric Vincent-Cuaz,Martina Haberecker,Pascal Frossard,Andreas Wicki,Olivier Michielin,Charlotte Bunne
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026
Abstract:Spatial proteomics (SP) captures the molecular composition and spatial organization of tissues at single-cell resolution, opening a direct view of the cellular ecosystems that drive disease. Protein panels, acquisition platforms, and tissue contexts differ from one cohort to the next, and a coherent picture of tissue biology can only emerge from models trained to bridge this heterogeneity. Foundation models offer the natural path forward, but their training depends on large, harmonized corpora that span platforms, panels, and cohorts. No public resource currently approaches that scale; available data sit in small, narrowly scoped releases with incompatible formats and conventions. We present \textttspora , a large-scale multimodal SP dataset of 13,546 standardized multiplex images from 5,477 patients across 35 cohorts and five major SP technologies. Curated with biomedical experts, \textttspora provides model training-ready formats together with cell and tissue segmentations, structured clinical metadata, and, where available, paired HE enabling the training of virtual staining models. It is complemented by \textttspora [io] , a unified interface for tiling, sampling, and standardization across all cohorts and modalities, and \textttspora [bench] , a benchmark suite spanning cell phenotyping, diagnostic and therapeutic prediction as well as virtual staining tasks across twelve cohorts that places current SP foundation models on common ground against task specific baselines.
[CV-255] Causal Spatio-Temporal Neural Distance Fields for Counterfactual Cardiac Anatomy MICCAI2026
链接: https://arxiv.org/abs/2610.24001
作者: Mathias Lowes,Kurt Butler,Kristine Aavild Sørensen,Klaus F. Kofoed,Sotirios A. Tsaftaris,Rasmus R. Paulsen,Steven McDonagh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published at Off-Grid @ MICCAI 2026. Code: this https URL
Abstract:Counterfactual cardiac anatomy can support risk modelling and intervention planning, by predicting how a patient’s heart would change under altered clinical conditions. However, current shape generative models are conditional on clinical variables and disregard causal links between these variables. Thus, they cannot answer counterfactual questions about how a given subject’s anatomy would change under a different clinical profile. Modelling the heart’s continuous movement over the cardiac cycle requires a representation that is itself continuous in space and time. Spatio-temporal neural distance fields provide this while separating clinical conditioning from subject-specific shape. We place a structural causal model (SCM) over the set of clinical variables (age, sex, smoking, BMI, heart rate, systolic blood pressure), before the condition encoding of the spatio-temporal neural distance field which represents the left atrial complex over a cardiac cycle by signed distance fields. An intervention of the clinical variables, propagated through a SCM before the 4D counterfactual, is constructed from the neural distance field. The generated counterfactuals follow dose-response under age interventions and transforms the male distribution to closely match the female distribution under the intervention \mathrmdo(\mathrmfemale=1) . By doing inverse interventions the model also approximates the original anatomy with sub millimeter precision. To our knowledge, this is the first counterfactual generation of 4D cardiac anatomy and the first integration of an SCM with an implicit neural representation as its imaging mechanism.
[CV-256] ProV-Mem: Proactive Memory for Streaming Video Question Answering
链接: https://arxiv.org/abs/2610.23998
作者: Bingbing Wang,Jing Li,Han Gao,Yishi Lin,Ruifeng Xu,Peng He
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Proactive streaming video question answering requires a model to continuously monitor an incoming stream and autonomously decide when to speak, what to say, and whether to emit a response. Existing streaming systems face a fundamental difficulty: retaining long visual histories is costly, whereas discarding them can lead to missed events and repeated responses. We propose \textbfProV-Mem, a training-free proactive memory framework built around a frozen VLM backbone. ProV-Mem decomposes the proactive decision into three lightweight memory tiers: a temporal-trigger memory controls when to query the backbone and recovers from prolonged silence, an instructional memory guides what to say through a fixed prompting prior for concise and visually grounded generation, and an episodic memory determines whether to emit by filtering near-duplicate responses. Under an idealized conditional-sufficiency assumption, we derive a stagewise approximation bound relating the decomposed process to a full-state joint policy. We further introduce Cost-Aware Area Under the Curve (\textbfCAUC), which extends PAUC by explicitly discounting redundant, overly frequent, and unnecessarily long emissions. Experiments on the four scenarios of ProactiveVideoQA show that ProV-Mem achieves the best CAUC across all scenarios and the best PAUC on WEB, EGO, and TV, while maintaining low per-response inference latency.
[CV-257] Automatic Cause Determination in Road Scene Understanding Using Qualitative Reasoning and Four-Valued Logic
链接: https://arxiv.org/abs/2610.23990
作者: Nassim Belmecheri,Arnaud Gotlieb,Nadjib Lazaar,Helge Spieker
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Author manuscript of the paper published in the 2025 IEEE Intelligent Vehicles Symposium (IV)
Abstract:Road scene understanding in automated driving (AD) aims to build a comprehensive analysis of video sequences taken on the road by embedded or fixed cameras (e.g., mounted on vertical road signals). One goal is to identify the relevant actors in the scene and another goal is to determine the causes that have triggered a specific action of the ego car (i.e., stop, slow down, turn left, etc.). In a complex urban environment, these causes can be multiple, confusing, possibly contradictory to other causes and not easily expressible using simplistic reasoning. Still, providing accurate \it automatic cause determination supports a) user acceptance by providing appropriate explanations to the car passengers and road users; b) increased road safety by providing detailed road scene understanding to traffic. In this paper, we propose using spatiotemporal reasoning and Belnap’s four-valued logic to formulate complex causes of AD action in a road scene. We compute these causes by analysing a Qualitative eXplainable Graph (QXG), which is an abstract representation of the road scene capturing spatiotemporal relations between road entities. Starting from a QXG, our approach called \tool, is targeted to determine complex causes of a selected AD action occurring in a specific frame of a road scene. The usefulness of \tool is demonstrated on several scenes extracted from the well-known NuScene dataset.
[CV-258] Single-Pass Decision Scores Enable Rapid Classification with Adjustable Thresholds in Medical Multimodal Large Language Models
链接: https://arxiv.org/abs/2610.23988
作者: Patrick Wienholt,Frederik Hauke,Christiane Kuhl,Jakob Nikolas Kather,Sven Nebelung,Daniel Truhn
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 29 pages (18 pages main text, 11 pages Supplementary Information), 5 figures, 3 tables
Abstract:Multimodal large language models (MLLMs) answer binary clinical questions with a yes or no token, leaving the decision threshold implicit and fixed. We derive a continuous score from a single forward pass by renormalizing the probabilities of the yes and no tokens, allowing the threshold to be adapted to clinical needs without retraining. We evaluated this approach with MedGemma and Gemma-3 (4B and 27B) on six medical imaging datasets. The default threshold was frequently misaligned with the task: for identifying normal chest radiographs, MedGemma-27B labeled every image as abnormal, whereas a validation-selected threshold yielded a sensitivity of 0.945 and a specificity of 0.696. Normalization achieved the highest AUROC on four of six datasets (CheXpert: 0.819 vs. 0.777 for raw and 0.743 for text-elicited probabilities), and a single decision required 53-79 ms on one GPU. This simple post-hoc method turns medical MLLMs into fast classifiers with adjustable operating points. Code: this https URL
[CV-259] What Generation Knows: Unlocking Fine-Grained Understanding in Unified Multimodal Models
链接: https://arxiv.org/abs/2610.23987
作者: Tengju Ru,Weitong Lian,Zecong Tang,Zhiheng Liu,Lingyi Meng,Haoran Li,Kaixuan Wang,Bohao Li,Xihui Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Fine-grained visual understanding requires distinguishing subtle differences in local attributes and spatial relations, yet remains challenging for existing vision-language models. Unified multimodal models (UMMs) offer a new opportunity through generation pathways that preserve rich visual information complementary to understanding. However, existing generation-assisted understanding methods mainly rely on generative supervision or explicit generation and have not systematically explored direct use of native generation-side representations. To address this gap, we systematically analyze what generation-side representations offer for fine-grained understanding, finding that they are more sensitive to fine-grained semantic changes, capture task-relevant changes missed by understanding-side representations, and depend on correct spatial organization for this advantage. Guided by these findings, we design GAFU, a post-training framework that unlocks native generation-side information for fine-grained understanding. It compresses generation-side visual representations into compact (G) tokens that preserve fine-grained visual details, enabling task-conditioned retrieval by the understanding pathway. Experiments show that GAFU consistently outperforms SFT baselines trained on the same data across diverse visual understanding tasks and UMM architectures, with maximum relative improvements of 20.00%, 17.94%, and 8.62% on VStar-Bench, OmniSpatial, and VSR-Bench, respectively. Overall, GAFU converts complementary information from the generation pathway into gains in fine-grained visual understanding, providing an effective path toward synergy between generation and understanding in UMMs.
[CV-260] BlenderWorld: Benchmarking Visual Agents for 3D Creation from Real-World References
链接: https://arxiv.org/abs/2610.23986
作者: Xinyu Fang,Yuchi Wang,Weihao Yu,Rongyao Fang,Zhaohai Li,Zhixun Li,Tiancheng Gu,Xilin Wei,Yuhang Zang,Hongsheng Li,Dahua Lin,Guofeng Zhang,Shuai Bai,Zhibo Yang,Dayiheng Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in multimodal large language models (MLLMs) have enabled them to act as autonomous agents, creating and editing 3D models via code or direct software manipulation. In practical workflows, these models must infer object shapes, materials, spatial layouts, and lighting from diverse references while strictly following instructions to reconstruct, modify, or design 3D content. However, existing benchmarks provide limited coverage of such open-ended requests across real-world scenarios. To bridge this gap, we introduce BlenderWorld, the first comprehensive benchmark dedicated to evaluating frontier models on realistic 3D creation tasks. BlenderWorld comprises 223 tasks spanning architecture, products, natural environments, and scientific visualization, driven by text, image, video, and structured specification inputs. To rigorously assess these complex deliverables, we establish a unified evaluation protocol: we render the submitted Blender projects from multiple viewpoints and evaluate them against 2,651 task-specific binary rubrics alongside holistic visual quality scores. Benchmarking thirteen frontier models reveals that GPT-6-Astra achieves the highest overall fused score of 79.9, followed by Opus-5 (73.0) and GPT-5.6-Sol (67.7). Ultimately, our findings reveal a critical bottleneck: while current models can parse broad instructions, they still profoundly struggle with fine-grained 3D spatial coherence and accurate structural reproduction in Blender, leaving substantial headroom for future 3D-native agentics.
[CV-261] Pretrained Features as Teachers Not States: Compact Latent World Models for Visual Planning
链接: https://arxiv.org/abs/2610.23984
作者: Yanbo Wang,Nan Lin,Wendelin Böhmer,Javier Alonso-Mora
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:A key design choice in visual world models is what information the latent state is encouraged to preserve. Models that learn a latent space with only a latent-prediction objective adapt to environment dynamics, but the predictive objective alone admits collapsed solutions and provides no explicit constraint on the visual information retained by the latent. Models built directly on pretrained visual features inherit a strong visual prior, but tie prediction and planning to a frozen state space. We propose Anchor-WM, which instead uses pretrained visual features as a teacher for learning a compact latent state rather than as the world-model state itself. Anchor-WM learns this compact latent with action-conditioned dynamics, while distilling the dense DINO representations through a training-only distillation head. This encourages the compact latent to retain visual information captured by the pretrained representation while discouraging representational collapse. At inference, the DINO encoder and distillation head are discarded, allowing prediction and planning to operate entirely in the learned latent space. Anchor-WM achieves the best average success rate across the four evaluated environments while enabling substantially faster planning than world models operating directly on DINO patch features. These results show that pretrained visual features can effectively guide world-model learning without defining the state space used for prediction and planning.
[CV-262] RL Teaches VLMs Where to Attend but Not How to Perceive
链接: https://arxiv.org/abs/2610.23983
作者: Ziqian Zhang,Junjie Liao,Zhining Zhang,Wentao Zhu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Although reinforcement learning (RL) effectively trains Vision-Language Model (VLM) agents, the mechanisms behind these capability gains are poorly understood. Across four diverse agent environments, we investigate whether the task-success gains from multi-turn RL are accompanied by improved grounded visual question answering about the environment. To assess this, we design environment-specific grounded VQA tasks that reveal improvements in grounded visual understanding. Representational probing shows that vision-encoder probe accuracy remains relatively stable, while LLM-layer probe accuracy improves. Attention analysis further reveals that LLM layers pay more attention to visual tokens during RL training. To test whether increased attention to visual tokens contributes to VQA gains, we introduce a lightweight plug-and-play intervention using a fixed attention bias. Without updating model parameters or computing gradients, this intervention partially restores VQA gains on the untrained model. These results reveal better utilization of visual information as a key mechanism behind the gains from RL.
[CV-263] me-JEPA: Grounding Video World Models with Physical Time Series
链接: https://arxiv.org/abs/2610.23982
作者: Shenyi Li,Ziyang Zhang,Wen Wu,Ming Liu,Chao Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code: this https URL
Abstract:Video world models learn how visual environments evolve by predicting missing or future content. However, physical-state information can provide useful context for predicting how environments evolve. Prior work incorporates this information through different mechanisms: domain-specific physical laws can complicate multi-domain training, while supervision on individual frames or frame pairs gives limited attention to how physical state evolves over time. We introduce Time-JEPA, which fine-tunes a video world model using low-dimensional, temporally aligned physical time series as auxiliary supervision. Physical time series connect video representations with physical state at corresponding temporal positions, encouraging the model to preserve how physical state evolves. In multiple domains including autonomous driving, robot manipulation, and human motion, Time-JEPA improves current-state decoding, multi-horizon future-state prediction, and kinematic reconstruction over matched video-fine-tuned controls across both condition and state-variable distribution shifts. Multi-domain training provides aggregate gains over matched single-domain training in all three tasks. External evaluations further suggest that time-series supervision improves sensitivity to physical anomalies and the representation of physical interactions, including in unseen scenarios. Aggregate gains on a second video backbone indicate that the method can be applied to more backbones. These results suggest that physical time series help video world models better understand the physical world.
[CV-264] SituHOI: Scalable Capture of Scene-Situated Full-Body Human-Object Interaction
链接: https://arxiv.org/abs/2610.23980
作者: Qifan Wang,Qihang Fang,Boao Shi,Yushi Du,Pei Zhou,Siyan Dong,Yanchao Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Modeling human-object interaction (HOI) in real environments requires jointly representing whole-body motion, manipulated objects, and scene geometry, yet existing datasets trade off scalable capture against structured, scene-aligned supervision. We introduce SituHOI, a large-scale RGB-D dataset containing 18,429 annotated clips across 16 indoor rooms and approximately 86 hours of quality-screened activity. It pairs recordings from a single moving RGB-D camera with static scene scans, object geometry, language descriptions, and estimated human and object trajectories. Using the scans as persistent references, our pipeline combines camera localization and depth-based metric alignment with human mesh recovery, object tracking, and temporal refinement to register all geometric annotations in a shared floor-aligned frame. These data support FlexHOI-Bench, a goal-conditioned joint motion prediction benchmark with variable observation and prediction durations. Given observed histories, scene and object geometry, and a terminal object pose, models predict future human and object motion. We provide SharedHOI as a reference predictor and evaluate world-space accuracy and human-object coordination. On a MoCap subset, our pipeline reduces human G-MPJPE from 195.4 to 68.3 mm and object translation error from 43.0 to 28.7 mm. SharedHOI leads four adapted baselines on all three aggregate FlexHOI-Bench metrics. Data will be released upon acceptance. An anonymized code repository is available at this https URL.
[CV-265] SPELUNKA: A Framework for Finding Unanticipated Associations in Vision-Language Models
链接: https://arxiv.org/abs/2610.23979
作者: Cassandra Parent,Isha Agarwal,Kumail Alhamoud,Charikleia Moraitaki,Rafiya Javed,Walter Gerych,Marzyeh Ghassemi
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Vision-Language Models (VLMs) are the backbone of many state-of-the-art multimodal systems, but often encode spurious and harmful associations from their training data. While prior work focuses on removing known biases from VLMs, methods for discovering unanticipated associations for retrieval tasks remain underexplored. We introduce SPELUNKA, an automated, inference-time framework for finding unanticipated distributional biases for retrieval tasks in VLMs without pre-specified labeled subgroups, inference-time human supervision, or model fine-tuning. Given a concept of interest, SPELUNKA generates captions for related images, extracts relevant entities from captions, and selects semantically meaningful associations. A frequency-based detection rule then identifies spurious associations by measuring over- or under-representation of these attributes in a VLM’s top-ranked outputs relative to their full dataset distribution. We evaluate SPELUNKA across four datasets, and find that it 1) successfully recovers both synthetic and naturally occurring spurious correlations, 2) achieves an average AUROC of 0.93 across four datasets in identifying images strongly associated with a query, and 3) outperforms baselines in finding features that are strongly associated with a given query.
[CV-266] FrameScout: Scouting Query-Relevant Frames for Long Video Understanding NEURIPS2026
链接: https://arxiv.org/abs/2610.23975
作者: Haonan Hu,Shuhao Chen,Weisen Jiang,Lizhao Gao,James Kwok,Yu Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026. 21 pages
Abstract:Long video understanding requires VideoLLMs to answer queries about videos spanning tens of minutes to hours, but such long videos produce far more visual tokens than current VideoLLMs can process. To address this issue, keyframe selectors are proposed to select the most query-relevant frames by computing the similarity between frame and query embeddings. Existing selectors either encode frames independently without temporal context or process chunks in isolation with ranking objectives that are difficult to optimize. To alleviate those limitations, we propose FrameScout with two core designs: (i) a streaming selector architecture that leverages a sliding-window KV cache and successor aggregation to produce temporally-contextualized frame embeddings; (ii) a frame-query contrastive objective that aligns frame and query embeddings in a shared space, directly training the selector to distinguish relevant frames from irrelevant ones. Extensive experiments on three long-video benchmarks and six base VideoLLMs demonstrate that FrameScout consistently achieves state-of-the-art performance. Code is available at this https URL.
[CV-267] How Much Can We Prune? Exploring the Performance Limits of Visual Token Pruning
链接: https://arxiv.org/abs/2610.23974
作者: Janghoon Cho,Shubhankar Borse,Reza Pourreza,Juntae Lee,Jihwan Bang,Munawar Hayat,Fatih Porikli
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Visual token pruning is increasingly used to reduce the memory and inference costs of vision-language models. Existing work has largely focused on designing method-specific selectors and demonstrating incremental improvements over prior methods at selected token budgets. It remains unclear whether the resulting performance reflects an inherent limit of visual token pruning or the limitations of current token-selection algorithms. We take a complementary perspective and ask: \emphhow well can visual token pruning perform when offline search cost is not the primary constraint? We formulate pruning for general image understanding as selecting, under a fixed budget, one shared visual-token subset that supports a broad distribution of questions about an image. We approximate this utility with a finite auxiliary QA set generated by a frontier VLM and optimize a teacher-forced answer-probability objective with the evaluated model frozen. Motivated by the conceptual monotonicity and diminishing returns of adding visual evidence, we construct nested subsets greedily and introduce \emphQA mini-batch greedy (QA-MBG), which estimates token marginal gains from randomly sampled QA components at each step. Empirically, QA-MBG is both cheaper and more effective than full greedy. With one frontier-VLM call and approximately 0.5 H100 GPU-hours per image, it produces a nested subset trajectory covering keep ratios up to 0.3 . Across Qwen3-VL-4B and Gemma~4 E4B on 900 images from nine benchmarks, we find significant headroom over evaluated pruning baselines on held-out QA and image-grounded long-form generation, while the searched subsets substantially narrow the gap to full-token inference.
[CV-268] Super-resolution as Spatial Downscaling: Coherence Structure and Task Utility in Earth Observation
链接: https://arxiv.org/abs/2610.23972
作者: Bright Wiredu Nuakoh,Ebenezer Afrifa-Yamoah
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 41 pages, 12 figures. Code: this https URL
Abstract:Learned satellite super-resolution is commonly assessed by pixel fidelity and potential downlink savings, but these criteria do not establish whether the reconstructed field preserves spatial structure or supports reliable downstream analysis. We treat single-image super-resolution as a spatial downscaling problem and evaluate four lightweight networks, including our compact 0.36 million-parameter model, against interpolation using pixel metrics, directional variograms, spectral band correlation, codispersion and observation coherence. On 735 satellite tiles from 190 scenes, the networks differ by at most 0.03dB and outperform bicubic interpolation by 1.39-1.42 dB at \times 2 . They recover about 30% of the spectral power above the observation’s Nyquist frequency and, at \times 4 , produce more isotropic fields than the reference (range anisotropy 0.59 against 0.49). Energy-based metrics can favour invented detail. Enforcing coherence reaches the quantisation limit, yet at least 99.7% of the remaining error lies in the observation operator’s null space, which coherence cannot constrain. Under spatially blocked, paired evaluation with two backbones and a sharpening control, learned reconstruction changes land-use accuracy by -1.0 to +3.1 percentage points (significant in 7 of 24 comparisons) and building-change F1 by -0.015 to +0.024 (none significant) relative to bicubic interpolation, and amplifies spurious change between two observations of an unchanged scene by 1.3-1.5 times where interpolation attenuates it. Full-resolution JPEG,2000 matches or exceeds reduced-resolution transmission with reconstruction at every matched bitrate. Learned reconstruction is most useful when only low-resolution imagery is available, and its evaluation should combine pixel metrics with spatial structure, observation coherence and task-level tests under spatial validation.
[CV-269] Spatio-Temporal Detection of Photovoltaic Systems on Industrial Building Rooftops Using Satellite Imagery and Machine Learning
链接: https://arxiv.org/abs/2610.23970
作者: Frederick Bruch,Christian Knoth,Florian Detsch,Edzer Pebesma,Hanna Meyer
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:The global transition toward carbon-neutral energy systems is heavily dependent on the rapid deployment of solar photovoltaic (PV) infrastructure, yet the speed of this expansion has outpaced the ability of administrative registries to maintain accurate records. While high-resolution commercial imagery is often used for monitoring, its high cost and infrequent updates can create an information gap for grid operators, energy infrastructure developers and policymakers. This study addresses the lack of scalable monitoring for industrial-scale rooftop PV systems by developing a machine learning framework that utilizes open-access, multispectral Sentinel-2 satellite imagery. By integrating building geometries from OpenStreetMap with Sentinel-2 reflectance data, a Random Forest model was trained to identify the specific spectral signatures of PV modules. Moving beyond theoretical modeling, this work emphasizes practical implementation through the development of an interactive web-based prototype for on-demand site monitoring. The application enables users to retrieve up-to-date satellite observations and assess the presence of PV systems on industrial building rooftops. The tool incorporates quality metrics that inform users whether a prediction is supported by sufficient training data density or requires manual verification. This end-to-end framework demonstrates that open-source medium-resolution data can effectively bridge this monitoring gap.
[CV-270] Learning When to Trust and Reobserve Sparse-Frame VideoQA from Temporal Sampling Perturbations NEURIPS2026
链接: https://arxiv.org/abs/2610.23969
作者: Shujia Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026 Workshop on Physical Understanding for Decision-Making (PhysUnderstand), Poster
Abstract:Sparse-frame video-language models can change their answers when only the sampling timestamps shift. Raw agreement exposes this problem, but it need not rank errors better than token confidence or identify which additional observation will repair an answer. We argue that answer risk and action value must be estimated separately. PhaseGuard+ probes a frozen VLM at controlled temporal offsets, predicts the reliability of the phase-vote answer, and predicts the example-wise correctness gain of a specific rescue action. The second score routes only examples expected to benefit from denser frames or a stronger VLM. Both predictors use lightweight models trained on public data; the VLMs remain frozen. We use 256 Perception Test videos for method development and evaluate on a disjoint 256-video test set comprising 554 Physics questions. For Qwen3-VL-8B, reliability AUROC rises from 68.0% with token softmax to 74.8% (+6.8 pp, 95% CI [2.6, 10.7] pp). Routing 30% of questions to Qwen3-VL-32B reaches 67.87% accuracy, compared with 63.66% for budget-matched random routing (+4.21 percentage points, CI [1.89, 6.40]). All four tested rescue pairs significantly outperform random routing. Temporal sensitivity becomes actionable when the system learns both whether an answer is unreliable and whether an available action is likely to fix it.
[CV-271] IMPULSE: Push First Then Predict in V-JEPA 2 Feature Space
链接: https://arxiv.org/abs/2610.23968
作者: Tingan Jin
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Robotics (cs.RO)
备注: 8 pages, 6 figures, 3 tables
Abstract:Latent world models such as V-JEPA 2 forecast in the feature space of a video encoder, but one look at an object does not reveal how hard it is to rotate, how much it bounces or how much it slides. People find out by trying: a few pokes reveal how an object responds. IMPULSE, named after the impulse response it infers, gives a predictor in frozen V-JEPA 2 feature space the same ability. A Bayesian contact memory summarizes three short calibration pushes, a predicted object pose carries it into the predictor, and a lightweight random-Fourier pose head lets the predictor use the pose. On 512 fresh objects, in a pre-specified held-out confirmation with models sealed before the data existed, the memory lowers the 1-second feature-forecast error by 11.6-12.8% relative to the same predictor without experience, for all 8 training seeds, with a 95% seed-and-object interval excluding zero. The gain also holds for 128 objects with out-of-range physics (11.7-14.0%), and the full extension beats a plain predictor by 15.5-16.6%. An information-ladder argument, exact for ideal predictors, shows that physics reaches the features only through motion, so we measure how much the pathway lets pose help. On fresh objects, the Fourier head roughly triples the gain of a perfect memory over plain pose tokens (2.8-3.2x). Raw push measurements fed through the same pathway give a similar gain, so the Bayesian summary is not required; it condenses the trials into eight interpretable numbers. The gain is absent at contact onset and grows afterwards.
[CV-272] Geo-vision: Native-Resolution Geometry and Sampled Topology for Vehicle-Part Damage Classification
链接: https://arxiv.org/abs/2610.23967
作者: Usef Faghihi,Amir Saki,Serge Jonathan Kogbeti-Wongo
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 12 pages, 3 figures. Code: this https URL
Abstract:Geo-vision classifies vehicle-part damage using a fixed representation combining appearance, multiscale geometry, orientation, and sampled digital topology. Connected components, holes, Euler characteristics, and spatial arrangement complement amplitude-sensitive pooling and a supervised tree classifier. We establish amplitude preservation, axial sign invariance, and conditional operator stability. Retrospective checkpoint evaluation on 55,303 crops from 4,916 source groups yields average precision 0.7577, exceeding 0.7458 for an ImageNet-pretrained frozen YOLO11n comparator on identical crops. The complete added structural block, comprising topology, occupancy, and component location, improves average precision by 0.00493 beyond appearance and geometry. Geo-vision uses no external encoder pretraining and has higher AP point estimates at every matched label budget. With approximately 25% fewer task labels, its AP point estimate exceeds full-data YOLO, although the adjusted interval includes zero. Complete serial CPU inference requires 26.98 ms per crop versus YOLO’s 11.20 ms, with peak memory 185 versus 437 MiB. Independent external validation remains necessary.
[CV-273] Generalizable fetal brain biometry via adaptive artificial intelligence
链接: https://arxiv.org/abs/2610.23966
作者: Yijin Li,Mingxuan Liu,Yi Liao,Le Chang,Hongjia Yang,Kasidit Anmahapong,Xiaotian Hu,Ziang Wang,Yingqi Hao,Haoxiang Li,Yifei Chen,Zihan Li,Ziyu Li,Juncheng Zhu,Min Kang,Yan Song,Hua Lai,Xiaoling Zhou,Dingzhi Yue,Xu Liu,Nan Sun,Rong Hu,Xuesheng Li,Fenglin Jia,Gang Ning,Haibo Qu,Qiyuan Tian
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 32 pages, 5 figures
Abstract:Reliable fetal brain biometry on magnetic resonance imaging (MRI) is essential for diagnosing congenital brain anomalies, but artificial intelligence models often lose accuracy under domain shifts. We developed BioTTA, a backbone-agnostic, source-free unsupervised test-time adaptation framework that adapts pretrained models online to individual unlabeled target volumes. This retrospective multicohort study included 394 reconstructed 3D fetal brain MRI volumes at 20-39 gestational weeks, with 11 biometric measurements annotated by radiologists. BioTTA was developed on single-center turbo spin-echo MRI and evaluated on three datasets representing cross-sequence, cross-center, and cross-nation shifts. Biometry accuracy was assessed using mean absolute error (MAE) and the percentage of correct measurements within 4 mm. A web-based reporting platform mapped measurements onto normative growth trajectories and was evaluated against radiologists in four multicenter disease cohorts covering germinal matrix-intraventricular hemorrhage, ventriculomegaly, and cerebellar abnormality. Across all three target datasets, BioTTA outperformed auto-SVRTK, source-only models, and FetalSynthGen, while matching models finetuned with target labels. MAEs ranged from 1.10 to 1.46 mm with a convolutional backbone and from 1.46 to 2.25 mm with a transformer backbone. Diagnostic accuracy ranged from 82.1% to 95.0%, comparable to a senior radiologist. BioTTA supports accurate, generalizable fetal brain biometry and standardized prenatal brain growth assessment across heterogeneous MRI sources.
[CV-274] Same Space or Not: Benchmarking Spatial Intelligence Through Cross-View Contradictions
链接: https://arxiv.org/abs/2610.23959
作者: Huizi Cao,Guanyu Xu,Zhixuan Liang,Ming Yin,Jiuxiang Gu,Manling Li,Mengdi Wang,Shilong Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Spatial intelligence, the capacity to represent and reason about physical space, underlies embodied perception and real-world visual understanding. Existing benchmarks probe it through spatial relations, perspective-taking, and mental modeling. What remains underexamined is whether a model can maintain a coherent scene representation across views, since pairwise comparison alone cannot reveal which view departs from the truth. We thus introduce Same Space or Not (SSoN), a benchmark of cross-view contradiction resolution comprising 1,000 questions from seven indoor scene sources and six categories of controlled 3D edits. Each question presents three views of an unchanged scene and one view of an edited variant. Models must identify the inconsistent view and explain the contradiction without a designated trusted reference or a marked target. Our pipeline iteratively translates human review into generation constraints. Automated checks and exhaustive human validation screen out both trivial visual shortcuts and questions lacking sufficient evidence for a unique answer. Humans achieve 83.3% accuracy, compared with 39.7% for the strongest vision-language models (VLMs), falling to 34.9% when a valid visual explanation is also required. We investigate prompting, camera information, and 3D reconstruction as interventions. On a 100-question intervention set with a 24% baseline, prompting and 3D reconstruction remain below 50% answer accuracy, except when ground-truth camera information is supplied, which reaches 61% answer accuracy but only 48% True solve accuracy. On a held-out 201-question test set, Qwen2.5-VL-7B achieves 21.89% accuracy, 33.33% after answer-only fine-tuning, and 41.79% after fine-tuning on both rationales and answers. These results identify cross-view contradiction resolution as a substantial weakness of current VLMs that can be partly addressed by targeted supervision.
[CV-275] Functionalizing AI-Generated 3D Shapes via Interior Geometry Construction
链接: https://arxiv.org/abs/2610.23953
作者: Haoyu Tang,Fancheng Gao,Kan Chen,Peng Song
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:AI-based 3D generation models such as Hunyuan 3D and Meshy can produce visually appealing 3D shapes from an image or text this http URL, they typically focus on an object’s external appearance, often producing incomplete or implausible interior geometry. Consequently, the generated shapes are not guaranteed to be functionally valid, limiting their direct use in downstream applications such as simulation and fabrication. In this paper, we present a computational approach to functionalize AI-generated 3D shapes through interior geometry construction. Our key observations are that a part’s functional role defines its interior geometry, and such geometry is often smooth, relatively simple, and closely related to the part’s external shape. Inspired by these observations, we introduce an interior-aware part graph to encode information necessary for constructing interior geometry within an AI-generated 3D shape, and build the graph via VLM-based shape segmentation and graph attribute inference. Guided by the graph, we model the interior geometry of each part as a generalized cylinder and infer its parameters from the part’s external geometry through an optimization. The resulting 3D shapes are functionally valid while preserving their AI-generated external appearance. We demonstrate our approach on a variety of AI-generated 3D shapes, including vessels, containers, and storage furniture, and fabricate selected examples for functional validation.
[CV-276] DISCERN: Identifiable Disease Evidence for Generalizable Medical Multimodal Learning
链接: https://arxiv.org/abs/2610.23952
作者: Xiaoguang Zhu,Pengyi Jiang,Juncen Guo,Jingyi Wu,Lianlong Sun,Yang Liu
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
备注:
Abstract:Clinical multimodal models can learn enrollment shortcuts shared across modalities and fail on new cohorts. We model these data through a latent structural causal model: disease state and background factors generate physiology through shared mechanisms, while cohorts differ in selection and measurement. Holding acquisition fixed, we identify conditional disease evidence without observing disease-state labels or recovering the full causal model. In common Gaussian coordinates, cross-cohort state alignment and natural-parameter changes identify a minimal selection statistic; conditioning on it removes enrollment-dependent evidence and yields an identifiable reference posterior within the specified shift family. State-dependent interactions can carry evidence invisible to individual-modality posteriors, motivating joint, information-preserving representations. DISCERN combines an information-lattice encoder, a latent-state model, and an outcome head through joint training. Under reversed synthetic enrollment on known latent coordinates, conditioning raises state AUROC from 0.043 to 0.883 and from 0.265 to 0.825, while random conditioning does not. Exploratory cross-database mortality experiments between eICU and MIMIC-IV improve AUROC over matched empirical risk minimization by 0.036 and 0.035. Learned-coordinate diagnostics distinguish preservation of state information from identification of the coordinates required by the estimator.
[CV-277] DS-Net: A Dynamic Split Network for Edge-Cloud Collaborative Inference
链接: https://arxiv.org/abs/2610.23949
作者: Xinyun Zhang,Li Wang,Lei Jin,Lianming Xu,Zhu Han
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages. Submitted to IEEE Transactions on Mobile Computing
Abstract:Collaborative inference serves as a critical methodology in edge intelligence. Existing collaborative inference methods operate by splitting the deep neural network (DNN) model between edge devices and cloud servers. However, these methods need to load different models for multiple split points when the communication or device changes, thus resulting in long switching times and substantial storage overhead on the edge side. To address this issue, we propose a Dynamic Split Network (DS-Net), which not only enables efficient split point switching but also enhances the collaborative speed and accuracy across various scenarios. Specifically, we design a dynamic switching framework that enables flexible selection among multiple split points within a single model, thereby reducing storage, memory, and switching overhead. To mitigate the feature distribution shifts and task-relevant information loss caused by dynamic splitting and high-ratio compression, we further propose a hierarchical compensation mechanism. Local alignment calibrates the decoded feature distribution at each split point, while task-specific global compensation restores task-relevant information at the global level. In addition, we incorporate sample-adaptive global compensation selection into split-point selection based on sample complexity and runtime resource conditions. Experimental results demonstrate that DS-Net reduces memory overhead by 74.6% and achieves millisecond-level switching latency. Moreover, under a 10 Mbps bandwidth constraint on the Nano platform, DS-Net outperforms state-of-the-art collaborative methods by achieving 1.2% higher accuracy on the COCO dataset, reducing latency by 42.5%, and improving throughput by a factor of 1.8 \times .
[CV-278] CrowdH-Edit: Reference-Guided Instruction-Based Synthetic Hajj Crowd Image Generation for Crowd Counting
链接: https://arxiv.org/abs/2610.23945
作者: Amirah F. Alshammari,Bander A. Alzahrani,Nahed A. Alowidi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:The scarcity of domain-specific annotated images remains a major limitation for developing computer-vision systems for Hajj crowd analysis. Existing crowd-counting datasets and unconstrained generative models do not capture Hajj-specific visual conditions, including sacred-site geometry, white-clothed pilgrim crowds, and indoor/outdoor illumination variation. This study presents CrowdH-Edit, a reference-guided, instruction-based framework for synthetic Hajj crowd-image generation. Real Hajj images provide structural references, while textual instructions guide changes in crowd conditions and illumination and help preserve the main scene layout. Stable Diffusion 2.1 and Stable Diffusion XL 1.0 with low-rank adaptation were evaluated as comparative baselines. Qwen Image Edit was used without additional training within a ComfyUI workflow and selected for final dataset construction. The workflow produced CrowdH-Edit, a cleaned pool of 6,122 synthetic images, from which 387 images were selected and annotated to construct CrowdH-Edit-Ann-387. Among the evaluated generation pipelines, Qwen Image Edit achieved the highest semantic-alignment score and the lowest distributional and perceptual distances. Downstream evaluation across five crowd-counting models showed that adding the annotated synthetic subset to real training data reduced mean absolute error for three models and root mean squared error for two. These findings indicate that reference-guided synthetic Hajj imagery can support crowd-counting training, although its benefits depend on the model and evaluation metric.
[CV-279] VERDICT: Visual Extraction-Result Decoder with Integrated Confidence and Traceable Evidence NEURIPS2026
链接: https://arxiv.org/abs/2610.23933
作者: Harikrishnan P M,Goutham Vignesh,Siva Prasad,Zishan Ahmad,Mahesh Basveshwar Deshmukhe,Varun V,Rohit Agrawal,Vishal Vaddina
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026, Workshop on VLM4RWD
Abstract:Enterprises run schema-guided structured field extraction on documents with frontier multimodal models, and accuracy keeps improving as new models and systems arrive. Two things have not kept pace: the evidence for each extracted value, meaning the location of the document it was read from, and how far to trust the value and that evidence. Neither is a component of most extraction systems today, so we build them as a framework that attaches to an existing extractor and leaves it untouched. We present VERDICT, a plug-in framework that uses a small Vision–Language Model (VLM) to ground/localize the evidence once per document over a fixed external extractor’s output, then calibrates value and evidence confidence. Fine-tuning the VLM for evidence localization substantially improves visual evidence quality and, more importantly, expands VERDICT’s selective-prediction operating range. At a strict confidence threshold 0.80 , VERDICT with the fine-tuned VLM gives usable precision–coverage trade-offs for both value and evidence confidence across the in-distribution and out of distribution datasets, whereas the frozen setting often collapses to little or no coverage. Running an open-weight VLM for localization rather than a closed API makes its internal states available to the calibrator, and they are worth more than output log-probabilities: in-distribution AUROC rises from 0.730 to 0.914 for value confidence and from 0.681 to 0.906 for evidence confidence. This advantage does not survive a distribution shift. We show that the signal is displaced rather than destroyed, and that labeling roughly 120 fields from the target dataset takes evidence confidence from 0.641 to 0.905 on one dataset and from 0.675 to 0.829 on the other.
[CV-280] oward AI-Driven Phytoplankton Event Detection from Image Classification: Temporal Aggregation Under Domain Shift
链接: https://arxiv.org/abs/2610.23932
作者: Ali Rohan,Hassan Shayan,Han-Gue Jo,Colin Moffat
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 7 figures
Abstract:Imaging FlowCytobot (IFCB) systems generate large volumes of phytoplankton imagery that exceed the capacity of manual expert analysis. Although machine-learning approaches achieve high image-level classification performance, it remains unclear whether their predictions can provide ecologically meaningful temporal information for monitoring. This study investigates whether AI-derived phytoplankton classifications can reproduce temporal occurrence patterns and elevated-occurrence signals under realistic sampling and independent-domain transfer. A monitoring-oriented framework was developed using MedPlanktonSet and evaluated on an independent IFCB dataset from the Swedish Meteorological and Hydrological Institute (SMHI). Eight biologically comparable taxa were selected as targets, while all remaining observations were retained as background to preserve the full image-stream denominator. Session-aware partitioning prevented temporal leakage, and alternative preprocessing strategies and convolutional neural network architectures were evaluated. Cross-domain experiments showed that preprocessing strongly affected generalisation. EfficientNet-B0 provided the strongest target-versus-background discrimination and was selected for temporal analysis. Despite residual image-level errors, AI-derived target-occurrence rates closely reproduced expert-derived signals after aggregation (mean absolute error = 0.0432, Pearson r = 0.9696, Spearman \rho = 0.9568). Agreement remained strong across multi-session windows, and elevated, strong, and persistent occurrence proxies were recovered in held-out testing. These results show that AI-based phytoplankton classification can support monitoring-relevant temporal indicators beyond image-level identification and support future operational early-warning development.
[CV-281] STLQ: Spatio-Topological Linkage Quantization for Approximate Nearest Neighbor Search
链接: https://arxiv.org/abs/2610.23931
作者: Kuizhi Liu,Junying Zhang,Xindi Ma,Xiangyu Wang,Qi Jiang,Sheng Gao,Chen Gong,Jianfeng Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Under review
Abstract:Vector quantization is widely used for large-scale data compression and approximate nearest neighbor search (ANNS). Multi-codebook quantization (MCQ) is a practical paradigm that represents database vectors with compact codes and supports efficient retrieval through asymmetric distance computation (ADC). However, existing shallow MCQs encode vectors independently within each inverted file (IVF) cluster, without exploiting local references. Therefore, we propose Spatio-Topological Linkage Quantization (STLQ), a topological shallow MCQ method that organizes each IVF cluster as a directed-forest with compact codes and scaling coefficients. Its basic encoding phase first quantizes vectors independently, while the linked encoding phase re-encodes relative residuals with respect to selected reconstructed parents, reusing local reconstruction information to obtain finer representations. The corresponding linked-ADC preserves linear scanning complexity. STLQ further compresses the scaling coefficients and directed-forest structure to control the additional topology overhead. Extensive experiments show that STLQ achieves higher recall than the evaluated shallow MCQs at comparable bit budgets, although at a higher offline construction cost. Meanwhile, it has a competitive recall-QPS tradeoff against deep MCQs. The code is available at this https URL.
[CV-282] rimodal Context-Aware Micro-Expression Recognition in Real-World Conversations: A Dataset and a Hierarchical Fusion Network
链接: https://arxiv.org/abs/2610.23930
作者: Junbo Wang,Yan Zhao,Shigang Wang,Jian Wei,Qiming Zhang,Yu Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Recognizing micro-expressions in real-world conversations is challenging, as these facial movements are brief and faint, and their interpretations are strongly dependent on the contextual circumstances of their occurrence. Most existing methods are de-contextualized, as they analyze isolated video clips while disregarding the semantic and acoustic cues in the surrounding dialogue that help explain the emergence of an expression. Decades of psychological studies have shown that social context exerts influences on facial behavior over two key dimensions: how an expression is interpreted and how strongly it is generated to begin with. However, existing micro-expression research rarely incorporates such contextual factors into its models. As an example, provocative utterances from an interlocutor may elicit suppressed negative affective reactions, which isolated silent clips are unable to account for. We propose TriCoME, a trimodal dataset consisting of visual, audio and text data sourced from real-world conversations. In this dataset, each micro-expression is paired with its concurrent utterance and preceding context. We further annotate the context based on conversational role to denote two categories: external stimuli from an interlocutor’s utterance and internal background from the subject’s own utterance. On this basis, we propose the Hierarchical Focusing Fusion Network (HFFNet), which aligns acoustic and semantic streams at fine granularity, compresses long dialogue sequences, modulates attention toward conversational roles via a context-type embedding, and routes each sample between visual-only and multimodal decisions based on visual confidence to prevent noisy context from degrading recognition performance.
[CV-283] Unbalanced Optimal Transport for Efficient Visual Document Retrieval ECCV2026
链接: https://arxiv.org/abs/2610.23929
作者: Hoyeon Shin,Jeongyeon Kim,Yeong Jun Koh,Yeoneung Kim,Hanul Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at ECCV 2026
Abstract:Vision-Language Models(VLMs)have advanced Visual Document Retrieval (VDR) by enabling OCR-free, multi-vector representations. However, these models generate excessive tokens to preserve dense visual details, which severely limits retrieval scalability. Existing compression methods mitigate this overhead but rely on isolated heuristics that fail to balance token importance, semantic redundancy, and coverage capacity. We propose a visual token compression framework formulated as an Unbalanced Optimal Transport (UOT) problem. Our approach unifies these three objectives into a single alternating optimization process. Furthermore, we theoretically show that minimizing the transport cost upper-bounds the MaxSim retrieval error, ensuring ranking stability. Extensive experiments on the ViDoRe V1 and V3 benchmarks demonstrate that our method preserves fine-grained document semantics and improves the efficiency-effectiveness trade-off over state-of-the-art base-lines. Code is available at this https URL.
[CV-284] SpinDoctor: Finding Innocuous Prompts for Biased Image Generation
链接: https://arxiv.org/abs/2610.23925
作者: Anna Gerchanovsky,Omer Akgul,Lujo Bauer,Michael K. Reiter
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:We introduce SpinDoctor, a method for discovering innocuous prompts for text-to-image models that result in images with a chosen target bias, while retaining the quality of the images and prompts. These prompts present a risk for user manipulation, in that a user who is given these seemingly unbiased prompts might be willing to use them and would receive biased images in return. Our method is flexible: it uses models in a black-box way only and the target image bias can easily be changed, and so our method can be applied with various motivations and in various scenarios. Our method shows some transferability between models-we saw the same bias emerge in 78.6% of cases-such that a victim who uses a prompt generated via our method using one model still likely receives a biased image from another model. In our experiments, a subset of which we corroborated with human evaluation, we were able to find an innocuous prompt that moved generated images in the direction of the desired bias in all but two of 304 experiments, with an average improvement of 34.8% and 33.5% on two open-source text-to-image models.
[CV-285] Vocabulary-Incomplete Unsupervised Vision-Language Model Adaptation
链接: https://arxiv.org/abs/2610.23919
作者: Si Qin,Jiawei Tang,Yaxin Hou,Yan Chen,Yutong Xie,Chen Li,Yuheng Jia
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:In certain real-world scenarios, obtaining labeled data is excessively expensive, motivating increasing interest in adapting models without annotations. Vision-language models (VLMs) provide pseudo-labels for unlabeled images through image-text similarity, enabling adaptation without manual labeling. The previous unsupervised vision-language model adaption methods assume access to an adaptation vocabulary containing all class names. However, a complete vocabulary is difficult to construct in real-world scenarios, as unlabeled data may contain classes that are overlooked or fall outside the current task scope. Thus, we propose Vocabulary-Incomplete Unsupervised Vision-language model Adaptation (VI-UVA), where only a subset of class names are available during adaptation. The vocabulary incompleteness induces an adaptation bias toward provided classes, which degrades omitted-class recognition. To address this issue, we propose Label Space Reconstruction (LSR), which reconstructs a latent label space from the unlabeled data and incorporates it into prompt tuning, thereby extending adaptation beyond the incomplete vocabulary and mitigating the adaptation bias. Furthermore, we derive an omitted-class semantic discrepancy bound for LSR and identify the key factors affecting the bound. Extensive experiments across eight benchmarks demonstrate that LSR outperforms existing state-of-the-art methods.
[CV-286] Cross-Modal Forgetting and Transfer in Unified Multimodal Post-Training NEURIPS2026
链接: https://arxiv.org/abs/2610.23918
作者: Francesco Colasurdo,Ivan Slobozhan,Joaquin Vanschoren,Anna Vettoruzzo
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at the NeurIPS 2026 Workshop on Continual Learning in the Era of Foundation Models and Embodied Agents
Abstract:Unified multimodal models answer questions about images and also generate images with one shared set of weights. This creates a post-training risk that single-purpose models do not have: fine-tuning one capability can damage another that works in a different modality. We measure this cross-modal forgetting for LoRA post-training of Janus-Pro-7B. Every target is chosen by how poorly the base model already handles it, from a poetic register to a language it barely speaks, so that retention is tested against a real intervention; we then evaluate the capabilities we never trained, first after a single run and then after chaining three. A single run looks entirely safe: nothing we did not train degrades on four public benchmarks or in free-form English, even when the adapter cuts held-out Welsh perplexity by 76% with a relative weight change under 2%. That safety belongs to the run, not to the recipe. Chaining three runs destroys the capability learned one run earlier and raises general-text perplexity by one to two orders of magnitude above the never-adapted model, on the very measures that stayed flat before. The cause is in the objective, which constrains only the modality being trained and leaves the other head unconstrained: cutting the learning rate to one fifth does not prevent the collapse, while anchoring the other head removes the damage to the capabilities we never trained, at no cost to what is learned. What is learned also stays inside its own modality, with Welsh acquired from text closing only 3% of the English-Welsh gap in image generation.
[CV-287] Your Self-Supervised Projection Head Captures Object Co-Occurrence Statistics NEURIPS
链接: https://arxiv.org/abs/2610.23915
作者: Arthur Aubret,Jochen Triesch,Céline Teulière
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted for publication in NeuRIPS
Abstract:Self-Supervised Learning (SSL) has achieved impressive success in learning semantic visual representations, yet the underlying principles driving this success remain underexplored. In this work, we hypothesize that common SSL pretext tasks implicitly model object co-occurrence statistics, a fundamental cue for visual learning. To test this, we curate three datasets of segmented objects from existing vision benchmarks using a state-of-the-art segmentation model. Through experiments across many SSL models, we reveal a hierarchical encoding of semantic information: while the visual backbone captures object categories, the projection head specializes in encoding co-occurrences between object categories. As a result, the projection head outperforms the backbone in aligning with human judgments of inter-category similarity. Furthermore, by controlling co-occurrence patterns during pre-training, we demonstrate that encoding object co-occurrences can significantly accelerate the emergence of category-level representations. Our findings uncover a previously hidden learning principle in SSL and suggest a path toward designing more effective pretext tasks by explicitly leveraging object co-occurrence structure.
[CV-288] SPIRIT: Sparse Physics-Informed Inverse Rendering for SAR Imaging
链接: https://arxiv.org/abs/2610.23914
作者: Yimeng Liu,Maolin Gan,Jingkai Lin,Jialuo Du,Mi Zhang,Zhichao Cao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Radio-frequency (RF) and wave-based imaging systems play a critical role in sensing under adverse conditions, but are fundamentally constrained by a trade-off between collection speed and spatial resolution. Reducing measurements to enable rapid collection of fragment geometric information across sensing dimensions, leading to highly non-injective inverse problems that destabilize conventional image-domain reconstruction and super-resolution methods. Synthetic aperture radar (SAR) exemplifies this challenge in sparse-aperture sampling. We introduce SPIRIT, a physics-informed inverse rendering framework for sparse-aperture SAR. Rather than enhancing formed images, SPIRIT reconstructs an explicit three-dimensional surface. By adopting an optimization strategy and explicitly modeling visibility, phase coherence, and wave propagation, SPIRIT aggregates distributed geometric features across sparse aperture views into a coherent reconstruction. Extensive experiments on synthetic and real-world data demonstrate that SPIRIT recovers accurate geometry and signal under extreme sparsity, reducing data collection by over 90% while achieving low geometric and signal error and consistently outperforming baselines.
[CV-289] Junction Geometry Survives an Unidentifiable Blur
链接: https://arxiv.org/abs/2610.23912
作者: Sohail(Neel)Sarkar,Shakuntala Baichoo
类目: Computer Vision and Pattern Recognition (cs.CV); Instrumentation and Detectors (physics.ins-det)
备注: 37 pages, 13 figures, 11 tables
Abstract:Thin filaments cross in microscopy, and dispersed spectra cross on X-ray detectors. At a junction, where these structures cross or branch, how many straight arms meet, and which way do they go? Unknown blur obscures the answer. At a balanced crossing, distinct Gaussian blurs can even produce the same observation law. We show that the arms can still be recovered. For separated rays of nonvanishing weight and an unknown centered Gaussian-mixture blur, a distance combining arm count, directions, relative weights and supporting lines is bounded by a constant times the Euclidean difference in cell probabilities on a sufficiently fine fixed grid. The bound is uniform over a bounded model class, including unidentifiable blurs. From n positions or photon counts, the minimax error is of order n^-1/2 . This rate is proved for a global estimator. In practice we fit rays and blur by constrained local likelihood, reducing the BIC count when two detectors agree on fewer arms in the microscopy studies. Across three studies with protocols fixed before evaluation, on 360 crossings and branches assembled from measured single-molecule localizations, the fit gets the arm count and every direction within 10^\circ 24 to 37 percentage points more often than a tuned fixed-scale Hessian ridge detector, averaged over 25 to 400 localizations. On acute angles with correct counts, mean absolute error is 3.5^\circ to 4.3^\circ , versus 10.4^\circ to 15.5^\circ . On Chandra spectra, 38 of 90 fits at 800 photons find all four arms within 10^\circ ; among these, median error is 0.023^\circ .
[CV-290] Pseudo-Label Foreground Quality Weighting for Domain-Generalized Fundus Image Segmentation
链接: https://arxiv.org/abs/2610.23911
作者: Lucas Gabriel Telesco,José Ignacio Orlando
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 5 pages, 3 figures, 2 tables. Accepted at SIPAIM 2026
Abstract:Optic disc (OD) and optic cup (OC) segmentation is central to automated glaucoma assessment, but models trained on limited labeled fundus datasets often degrade under domain shift. Semi-supervised Teacher-Student learning can improve generalization by leveraging unlabeled data, yet standard Noisy Student training treats all pseudo-labels as equally reliable. In this paper we investigate a targeted extension of Noisy Student training in which OD- and OC-specific pseudo-label supervision is modulated using a mask-derived reliability estimate. A multitask Dice predictor estimates OD- and OC-specific pseudo-label quality from Teacher-generated masks, and these estimates are used to weight the pseudo-label loss during Student training. This preserves the baseline contribution of low-scoring foreground pseudo-labels while assigning greater relative emphasis to pseudo-labels predicted to be more reliable. Experiments across known and unseen domains show improvements over the Noisy Student baseline, particularly for OC segmentation, while maintaining competitive generalization performance.
[CV-291] Grounding Confusable Movie Moments
链接: https://arxiv.org/abs/2610.23910
作者: Dong-Hee Kim,Seonwoo Choi,Changbeen Kim,Juyeon Ko,Youngju Choi,Il Hyeon Mun,Hyunwoo J. Kim,Donghyun Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Characters, locations, and actions recur throughout a movie, so many of its moments differ only in fine-grained details. Grounding such a moment in a full-length movie requires resolving that detail rather than matching the attributes it shares with other moments, yet existing temporal movie grounding benchmarks rarely test this. We introduce CoM ^2 -Bench, a benchmark of confusable query pairs, where the two queries in each pair share most attributes but differ in a single distinguishing detail within a full movie, posing a challenge for recent temporal grounding approaches. To locate where grounding breaks down, we analyze a retrieve-then-localize pipeline, in which a retriever first selects a short candidate chunk and a localizer then performs temporal grounding within it. Separating the two stages lets us attribute each failure to either retrieval or localization. We find that retrieval is the major bottleneck: state-of-the-art retrievers often select a moment’s confusable counterpart, because their choices follow the content the two moments share rather than the detail that tells them apart. To address these failures, we propose a simple, training-free remedy, Agentic Window Search (AWS): an MLLM agent \textitDecomposes the query into attribute-level sub-queries, then \textitInspects each candidate window against them and \textitExpands the window when evidence is cut off at its edge. AWS works with existing retrievers and localizers without additional training, raising mIoU from 18.5 to 26.8.
[CV-292] When the Editing Intent Is Split: A Cross-Modal Jailbreak Attack for Large Image Editing Models
链接: https://arxiv.org/abs/2610.23909
作者: Xianmin Ye,Zhiyuan Fan,Tianju Liu,Yuhang Wang,Zhenxing Niu,Haichang Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 27 pages, 11 figures
Abstract:Large image editing models increasingly infer editing intent from both textual and visual inputs, creating an underexplored safety risk when harmful intent is distributed across modalities. In this work, we formalize this split-intent vulnerability and propose the Split-Intent Jailbreak Attack (SIJA), which asymmetrically distributes an unsafe editing intent by specifying the editing referent in text while concealing the safety-critical action and specification through graphical visual semantics in an auxiliary cue. To systematically study this threat, we introduce the Split-Intent Image Editing Safety Benchmark (SI2ESBench) for evaluating image editing models under split-intent inputs. Extensive experiments on representative commercial and open-source models demonstrate that SIJA effectively compromises state-of-the-art image editing systems, achieving ASRs of 68.28% on GPT Image 2 and 91.38% on Qwen-Image-3.0-Pro. Our results expose a mismatch between multimodal editing and safety assessment: the distributed intent can be recovered to execute the requested transformation, yet its harmfulness often remains undetected by existing safeguards. To mitigate this vulnerability, we propose editing-intent reconstruction, a defense principle that explicitly evaluates the complete transformation implied by the joint multimodal context and substantially improves rejection of split-intent attacks. Our findings expose a cross-modal safety gap in modern image editing systems and highlight the need to assess the jointly inferred transformation rather than individual inputs in isolation. this https URL
[CV-293] ContraSafe-VL: A Contrastive Safety Preference Dataset for Vision-Language Models
链接: https://arxiv.org/abs/2610.23908
作者: Di Wu,Yanyan Zhao,Xin Lu,Hao Yang,Taotao Zhang,Yunzhen Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 5 figures, 5 tables
Abstract:Vision-Language Models (VLMs) are increasingly deployed as general-purpose multimodal assistants, making multimodal safety alignment an important challenge. Safety depends on the interplay between textual and visual content, with visual context often playing a critical role. However, VLMs may fail to use visual information appropriately, either overlooking safety-relevant visual content or over-interpreting sensitive visual content, leading to unsafe responses or over-refusal. Existing multimodal safety alignment datasets largely follow the response-preference paradigm, which teaches models how to respond but provides limited supervision on how visual context should inform the safety judgment, making it difficult to balance safety and over-refusal. To address this limitation, we introduce ContraSafe-VL, a Contrastive Safety preference dataset for Vision-Language Models. For each fixed textual query, ContraSafe-VL constructs paired visual contexts that result in multimodal queries with different safety labels, providing direct supervision for how visual context should inform multimodal safety judgments. It further provides response preference pairs based on harmlessness and helpfulness to encourage models to generate responses that are both safe and helpful. Experiments across three VLM architectures and multiple benchmarks demonstrate that training with ContraSafe-VL improves cross-modal safety recognition and achieves a better safety–over-refusal trade-off while preserving general multimodal capabilities. The dataset will be publicly released.
[CV-294] EgoVine: Post-Training with Branching Experience Compiled from Real-World Videos
链接: https://arxiv.org/abs/2610.23907
作者: Qinchuan Cheng,Bowen Lyu,Zhantao Gong,Angela Yao,Cheston Tan,Shijie Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Real-world egocentric videos capture multi-step household procedures, but their recorded trajectories leave alternative actions and recovery from the learner’s own mistakes largely unexplored. We introduce EgoVine, a post-training framework that turns a fixed demonstration set into interactive experience for long-horizon planning. EgoVine operates in executable symbolic environments compiled from real-world video annotations, where an agent must coordinate object manipulations under partial observation and account for delayed consequences. Restoring a decision state allows the learner to explore alternative actions and compare their eventual outcomes. Union jointly trains action generation and consequence prediction in one model, using the same executions to supervise candidate comparison and recovery. Verified repairs add recovery targets from failed attempts, while comparisons among sampled continuations guide policy updates. The updated planner then returns to the environments, with its own decisions and mistakes shaping the next round of experience. On a shared benchmark, EgoVine-1.5B achieves 81.8% success on Main planning tasks and 66.7% on longer task compositions, outperforming all evaluated Flash LLM baselines on both. EgoVine connects real demonstrations to an iterative learning process that trains compact planners on choices and futures beyond the recorded trajectories.
[CV-295] Beyond Generator-Specific Shortcuts: Open-World AI-Generated Video Detection via Adversarial Training
链接: https://arxiv.org/abs/2610.23906
作者: Qi Zhou,Yuanfan Li,Huiqi Deng,Hongbin Pei,Yazhe Tang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages
Abstract:AI-generated video (AIGV) detection is increasingly important for mitigating misinformation, impersonation, and other malicious uses of synthetic media, yet existing detectors often generalize poorly to unseen generators in open-world settings. To better understand this challenge, we analyze the failure modes of AIGV detectors under cross-generator out-of-distribution (OOD) shifts and observe that detector representations strongly encode generator-specific cues, which are associated with substantial ID-OOD performance drops. Further analysis shows that naively perturbing generator-sensitive tokens can even reduce OOD performance, indicating that generator-related cues and detection-relevant evidence are only partially separable. Motivated by these findings, we propose Forensic Obfuscation of geneRator-specific cues via Gradient-guided Embedding perturbation (FORGE), a token-aware adversarial training framework for open-world AIGV detection. FORGE identifies tokens that are strongly associated with generator-specific shortcuts yet less critical to real/fake discrimination, and selectively perturbs them during training to suppress generator-dependent patterns while preserving useful forensic evidence. Through synchronized adversarial training between an attacker and a detector, FORGE encourages the detector to learn more transferable generator-invariant forensic representations. Experiments on 29 held-out benchmark entries show that FORGE achieves an average F1 of 92.62, outperforming eight strong baselines by 5.31 points, while also remaining robust under four common post-processing attacks. We will release our code and datasets upon acceptance.
[CV-296] Instance-Specific Classifier-Free Guidance for Text-to-Image Diffusion/Flow-Based Editing
链接: https://arxiv.org/abs/2610.23905
作者: Wajeeh Atrash,Tom Tirer
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注:
Abstract:In recent years, a variety of prompt-based image editing methods have been developed based on text-to-image (T2I) diffusion/flow models. Image editing introduces an inherent trade-off between adhering to the original source image and realizing the change specified by the target prompt, which are typically measured by the LPIPS and CLIP-T metrics, respectively. Intuitively, this trade-off can be controlled by the classifier-free guidance (CFG) scale. Most existing editing methods, though, use a fixed CFG scale, ignoring that different edits require varying levels of modification. In this paper, we first explore the potential benefits of using an instance-specific CFG by exhaustively re-running prominent editing methods with a wide range of CFG scale values on benchmark datasets. We show that instance-specific setting can significantly improve LPIPS without compromising CLIP-T. Yet, tuning based on repeatedly executing an editing pipeline is inefficient. To prevent slow inference, we train a light-weight neural network that predicts the metrics of the underlying editing method for a given edit instance and CFG scale. At inference time, these predictions enable fast selection of an instance-specific CFG scale. Experiments across multiple editing methods and benchmark settings show the performance advantage of our approach over the baseline methods.
[CV-297] AI-Experiment: an augmented reality art project co-created with artificial intelligence ECCV
链接: https://arxiv.org/abs/2610.23903
作者: Bernhard Egger,Hans Furer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: presented and published at AI for Visual Arts Workshop and Challenges (AI4VA) in conjunction with European Conference on Computer Vision (ECCV) 2026, Malmö, Sweden
Abstract:We describe the creation of a physical artwork that was performed in synergy between an artist, a computer scientist and generative AI tools. A latent diffusion model was finetuned on photographs of the artwork of the specific artist and the generative model then generated synthetic artworks in a similar style. From the generated artworks a subset was selected to be painted again with acrylic on canvas by the original artist. In addition to the physical artifact an augmented reality component was added which brings the artwork to life with a set of generative video models. Together with this paper we release part of the artwork (training data), the code that was deployed to generate the artwork as well as the augmented reality experience for arts, research and education under this https URL
[CV-298] Rethinking weakly supervised localization on chest radiographs: position nulls protocol dependence and PACE
链接: https://arxiv.org/abs/2610.23902
作者: Shixin Qiu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Weakly supervised localization on chest radiographs is reported as a single overlap accuracy, and those numbers are not comparable across papers. We measure what one contains. Against position-null controls that keep a readout’s own box and move only its position, class activation mapping and its gradient-weighted variant beat the null on two to four of seven annotated findings on ChestX-ray14 and two of eight on CheXlocalize. The null is beatable: three board-certified radiologists clear it on four to seven. Three rarely stated choices each move a result by more than most reported method effects: the results regime, the extraction threshold, and the mask-to-box rule, whose cost is predicted by how fragmented the thresholded mask is. Predicted box area alone reverses a paired comparison. These measurements produced PACE, for \emphpopulation anomaly and class evidence, which multiplies the finding identity from the label-trained head by the spatial evidence from a frozen radiograph backbone that never sees a pathology label. It is the case study that made the protocol dependence visible, and we report its costs beside its gain: in an exploratory matched-area analysis it improves strict-overlap accuracy on both corpora and beats a label-derived second map, but it is the least faithful of the three readouts to the classifier’s causal map, its advantage depends on the box rule, and, this benchmark releasing no localization validation set, every design choice behind it was made on the boxes it is scored on. Its one clean transfer beats one baseline and loses to the other.
[CV-299] EAGLE: Entropy-Aware Geometric Learning Enhancer for Partially Polygon-Supervised Object Localization ICLR
链接: https://arxiv.org/abs/2610.23900
作者: Naveen Stewart
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 25 Pages, 3 figures, submitted to International Conference on Learning Representations (ICLR) 2027
Abstract:Object detection is a critical application of computer vision, yet models trained uniformly on bounding boxes lack the geometric precision to be fully effective in inference settings, especially at high IoU thresholds. One common approach is the introduction of polygon-supervised localization, where a model is trained explicitly for localization but with access to polygon masks during training. Among such methods, partial polygon supervision is viable due to the higher annotation costs of developing fully supervised training datasets. However, this introduces a further challenge: in object classes of relatively low shape complexity, the addition of training masks and subsequent segmentation loss may obscure box localization, often worsening model performance despite access to deeper geometric information. To address the practical limitations of both full and partial polygon supervision, we introduce a mathematical framework for quantifying per-class geometric complexity by combining information theory and statistical shape analysis. With this framework, we develop Entropy-Aware Geometric Learning Enhancer (EAGLE), a lightweight, post-training model head that learns geometric information for the localization model, improving accuracy via edge refinement and localization rescoring using entropy-informed updates. Empirical analysis across MS COCO and Pascal SBD demonstrates the entropy framework meaningfully predicts shape reconstruction error and accuracy differences among mask-derived and standard box predictions, while the implemented EAGLE framework achieves consistent improvement on already SOTA detection frameworks (+0.9 mAP50-95 points on MS COCO) at minimal observed inference overhead.
[CV-300] RhinoNet: Unsupervised Learning for Phenoscaping and Generation of High Dimensional Point Clouds
链接: https://arxiv.org/abs/2610.23898
作者: Siddharth Viswanath,Hiren Madhu,Eddie Cavallin,Benjamin Jones,Alistair Wilkinson,Christopher Tape,Michael Perlmutter,Rex Ying,Ian Adelstein,Smita Krishnaswamy
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We present RhinoNet, a simplicial complex-based point cloud neural network, that creates a geometry-preserving embedding of high-dimensional point clouds. RhinoNet takes collections of datasets, also called \em data cohorts, and processes each dataset (point cloud) within the cohort as a simplicial complex. These simplicial complexes have datapoints as vertices and edges learned by Vietoris-Rips filtration. Using simplicial wavelets and MLP layers, RhinoNet computes a vector representation of each dataset in a latent space produced by an autoencoder. The latent space is optimized such that an optimal transport distance between pairs of point clouds is preserved between the latent representations. We apply RhinoNet to single-cell datasets collected on patients—a common source of such high-dimensional data. RhinoNet creates a patient level representation, known in biological contexts as a phenoscape. We establish that RhinoNet generates organized phenoscapes, whereas other methods such as static maps do not. Next we show that we can generate entire point clouds by sampling and decoding this phenoscape. We show two modes of generation via 1) latent flow matching and 2) geodesic interpolation in latent space, followed by a decoding back into the point cloud space. RhinoNet achieves state-of-the-art performance on single-cell and spatial data cohorts relative to tested baselines. Overall, RhinoNet lifts the concepts of representation learning and generative modeling to the point cloud or population level and defines a neural model for phenoscaping.
[CV-301] InJePA: Intention-First Latent Planning with Coupled JEPA World Models for Visual Navigation
链接: https://arxiv.org/abs/2610.23896
作者: Siming Zhou,Bowen Ma,Ruiran Du,Pengcheng Zheng,Sheng Zheng,Dongshen Han,Zhijie Li,Wei Dong,Fan Mo,Chaoning Zhang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 26 pages, 5 figures. Includes appendices. Code is available at this https URL
Abstract:Visual navigation requires goal directed planning informed by predictions of environment dynamics. Representative navigation world model methods propose complete action candidates before predicting consequences for selection or search updates. For each candidate, predictions primarily evaluate proposed actions rather than guide its stepwise construction. To this end, we propose InJepa, an intention first planning paradigm centered on latent states. A goal conditioned JEPA proposes latent intentions, and an action conditioned JEPA continually predicts future latent states. Both operate in a shared latent space to align their predictive representations. Within each candidate, predicted consequences feed into the next intention proposal and action inference, forming a closed loop of intention, action, and consequence. This extends the world model’s role beyond evaluation to leading candidate construction through latent intentions. We evaluate InJepa through closed loop image goal navigation in Habitat MP3D. The results show effective multistep navigation through intention guided action generation and consequence feedback. Experiments across encoders further support applicability across visual representations. These findings support latent intentions as the central planning object and show that world models can directly guide plan construction through coordinated reasoning about intentions and consequences. Code is available at this https URL.
[CV-302] GeoForward: Geometry-Guided Cross-View Alignment for Feedforward 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.23895
作者: Yiming Shu,Peng Sun,Lang Zhang,Mingyue Feng,Yaonong Wang,Pei Liu,Chen Sun,Jun Ma
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Feedforward 3D reconstruction has largely followed an appearance-driven paradigm: multi-view images are encoded into visual representations, cross-view correspondence is implicitly inferred from learned features, and 3D primitives are subsequently predicted from these representations. However, calibrated cameras and sparse metric measurements already provide explicit constraints on where the same physical point should appear across views. We propose a novel paradigm, \textbfGeometry-Conditioned Feedforward Reconstruction, in which metric geometry establishes cross-view correspondence before the 3D representation is formed, while learned visual features refine rather than discover this correspondence. Based on this paradigm, we introduce \textbfGeoForward. Sparse LiDAR points are unprojected into 3D and analytically projected across calibrated cameras to provide metric correspondence anchors. A bounded continuous local search then refines these anchors using visual features, and the resulting geometry-conditioned representation is fused with global multi-view context from VGGT. The shared representation drives separate metric-depth and Gaussian-attribute prediction, establishing a direct path from explicit geometry to both Gaussian centers and spatial attributes. A sparse 3D refinement module subsequently improves local attribute compatibility while preserving the geometrically determined centers. For temporal reconstruction, endpoint Gaussians are composed using vehicle masks and motion priors. On nuScenes, GeoForward achieves the lowest cross-camera depth-consistency error and Chamfer distance among the compared methods while maintaining competitive reconstruction quality. These results demonstrate that explicitly conditioning correspondence formation on metric geometry provides an effective alternative to purely appearance-driven feedforward 3D reconstruction.
[CV-303] Path-Conditioned Attention Mobility: From Depth to Execution Hubs in SegFormer
链接: https://arxiv.org/abs/2610.23893
作者: Aleksandr Druzhinin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 8 figures; includes supplementary material
Abstract:Transformer compression often couples two distinct quantities: retained attention capacity and the number of sequential depths at which that capacity is executed. We study whether useful attention functions in a hierarchical vision transformer are locally mobile across depth. Using SegFormer-B5 for human segmentation, we find that some task-relevant attention heads can be relocated to nearby depths, but only when their computational context is respected. Head-only relocation is insufficient, while joint adaptation of the relocated head and surrounding representation boundary improves functional preservation; replacing donor-specific normalization and spatial-reduction preprocessing causes large degradation. We use this path-conditioned mobility to reorganize retained attention from depth into width. Starting from a fixed parent with 120 retained Stage-3 heads across 40 attention execution depths, we consolidate the same 120-head budget into 20 execution hubs. Under a layout-matched three-seed recovery control, K20 reaches 0.960686 +/- 0.000059 mIoU versus 0.960400 +/- 0.000074 for the corresponding State120 control. The paired point difference is +0.000285, while a paired image-bootstrap confidence interval includes zero; we therefore interpret the result as no measurable quality penalty rather than statistical superiority. For deployment, FOLDED_VEC preserves donor-specific paths while vectorizing their execution. Relative to the same 120-head State120 parent, it reduces end-to-end NVIDIA L4 latency by 5.28%, 3.15%, and 3.58% at batch sizes 1, 4, and 8. These results suggest that attention redundancy includes not only removable capacity but also execution-depth redundancy, provided that the computational path defining a relocated function is preserved.
[CV-304] EMMA: Event-Driven Multimodal Memory System with Multi-Agent Retrieval
链接: https://arxiv.org/abs/2610.23892
作者: Ruochen Zhao,Jiazheng Kang,Gezi Li,Dongying Lin,Ting Bai
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Long video understanding requires reasoning over hours of content that far exceed the context window of multimodal large language models (MLLMs). Existing approaches compress video into fixed-length clips, yet events are fragmented and unrelated content is merged, yielding noisy retrieval. To overcome this challenge, we propose EMMA, an Event-driven Multimodal Memory system with Multi-Agent retrieval, which organizes memory around semantic events in three stages: (1) Event segmentation: a dual-stream detector adaptively locates event boundaries by fusing visual semantic drift with spoken transitions, rather than fixed windows. (2) Memory construction: built on the detected events, a three-level hierarchy of raw video, clip-level memory, and event-level memory stores content at progressively higher semantic abstraction. (3) Memory retrieval: a multi-agent framework collaboratively queries this hierarchy through query decomposition, tiered search, evidence validation, and answer generation, narrowing evidence from coarse events to precise clips. Extensive experiments on three challenging benchmarks demonstrate substantial improvements, showing the effectiveness of our event-aligned memory system. Code, prompt templates, and reproduction scripts are available at this https URL.
[CV-305] Context-Aware RGB-Thermal-Radar Obstacle Perception for Resource-Constrained ASVs
链接: https://arxiv.org/abs/2610.23873
作者: Pranav Kedia,Edwin Redhead,Heiko Hamann
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reliable obstacle perception is challenging for small autonomous surface vessels (ASVs): reflections and glare confound visible-light cameras, vessel motion shifts the apparent waterline, and darkness removes the dominant sensing modality altogether. Rather than treating multimodal fusion as a static accuracy gain, we propose a context-aware architecture that combines RGB and long-wave infrared (LWIR) imagery, frequency-modulated continuous-wave (FMCW) mmWave radar, and inertial context. A bearing-sector interface provides the navigation controller with calibrated obstacle probabilities, radar ranges, and sensor status. It also distinguishes sensor dropout from a valid observation with no detected obstacle. Using new multimodal lake recordings, we deployed the pipeline in a 24-leg closed-loop pilot under uncontrolled lake conditions (16 day, 4 dusk, and 4 night; 351 m), including night legs in which RGB was unusable and the vessel drove on LWIR and radar evidence alone. On eight held-out low-light recordings, the calibrated fusion model achieved average precision (AP) 0.869 and expected calibration error 0.108, compared with AP 0.673 for rule-based fusion, 0.620 for gated fusion, and 0.903 for RGB alone. Our results expose a central tradeoff: RGB ranks best when usable, whereas fusion preserves a navigation input when RGB fails. This work is a step toward perception that remains usable and degrades predictably across multi-day ASV missions.
[CV-306] SAGE-CP: When Marginal Coverag e Hides Route-Conditional Failures in Adaptive VLM Ensembles
链接: https://arxiv.org/abs/2610.23864
作者: Mushir Akhtar,M. Tanveer
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 19 pages, 2 figures, 12 tables
Abstract:Adaptive VLM ensembles save computation by stopping early on easy images and escalating harder ones. Although a single conformal threshold can make the final predictor marginally valid, it can still fail severely along individual computational routes. We characterize this failure precisely: if a route has prevalence pi, marginal 1-alpha coverage allows its conditional coverage to fall to max0, 1-alpha/pi. To address this mismatch, we learn the router on development data, freeze all routing choices, and calibrate the returned score separately for early exits and escalated inputs. We call this pipeline SAGE-CP. Its calibration step applies Mondrian split conformal prediction to the operational groups created by the router. Under exchangeability, the resulting prediction sets achieve finite-sample coverage within each route. Across 10 benchmarks, 4 frozen VLMs, and 5 aligned splits, both global calibration and SAGE-CP achieve 0.900 marginal coverage at a 0.90 target, but their route behavior differs substantially. SAGE-CP raises mean worst-route coverage from 0.623 to 0.887 and reduces the mean route-coverage gap from 0.324 to 0.025, using the same routed scores and 1.177 source calls per image. The stronger guarantee increases the mean prediction-set size from 1.893 to 2.212; nevertheless, the sets remain smaller than those of the full ensemble (2.486) while requiring 70.6 percent fewer source calls. The route-level reliability persists across 6 fixed query fractions, while finer exit-depth groups demonstrate the need for additional calibration data on rare routes. Across 5 datasets, SAGE-CP also reduces measured batch-one latency by 58.8 to 63.4 percent and gross GPU energy by 55.7 to 57.5 percent relative to full inference. Thus, routing provides computational savings, while route-stratified calibration attaches coverage guarantees to the operational paths created by the router.
[CV-307] Dependency Compression and Synergy: A Unified Information-Theoretic View of Multimodal Learning
链接: https://arxiv.org/abs/2609.14421
作者: Liangjian Wen,Linjie Li,Jiang Duan,Yong Dai,Jianzhuang Liu,Zhao Kang
类目: Information Theory (cs.IT); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Recent advances in multimodal foundation models have intensified the need to understand how different modalities share, preserve, and complement information. Mutual Information (MI), the Information Bottleneck (IB), and Partial Information Decomposition (PID) provide complementary perspectives, yet existing studies often treat them as isolated tools. This survey presents an information-theoretic perspective connecting these principles as progressively refined views of multimodal information processing: MI characterizes inter-modal dependency, IB explains task-oriented information preservation under compression, and PID decomposes preserved information into redundancy, uniqueness, and synergy. We review 170 recent studies (2018–2026) and 12 foundational works, organizing multimodal learning around four challenges: cross-modal alignment, information-efficient fusion, interaction-type characterization, and scaling to multimodal foundation models. Rather than using application domains as primary taxonomy axes, we interpret healthcare, robotics, recommendation systems, affective computing, and wireless communications as empirical validations of these information principles. Beyond taxonomy, we organize existing multimodal paradigms within a single information-theoretic coordinate system – the Generalized Multimodal Information Lagrangian – in which they occupy exact or approximate parameter corners, and whose unoccupied regions name candidate method families the literature has not yet built. We further discuss how emerging multimodal foundation models instantiate these principles at scale and identify open challenges including scalable information estimation in high-dimensional settings, standardized evaluation across information-theoretic methods, combinatorial complexity of multimodal PID, and the transition from post-hoc information analysis toward information-aware multimodal learning.
[CV-308] Progressive multi-scale tumor-state memory for incomplete multimodal brain tumor segmentation
链接: https://arxiv.org/abs/2610.26457
作者: Milad Rostamian,Mohammadreza Mohammadi,Mohsen Soryani
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Brain tumor segmentation uses complementary magnetic resonance imaging (MRI) sequences, but some sequences may be unavailable in practice. Their absence changes the information available for segmenting different tumor regions. Fusion methods can support different input subsets. However, fusion at several scales does not by itself specify how tumor evidence is retained and updated to guide fusion at later stages. We propose a progressive multi-scale tumor-state memory network (PMTM) for segmentation from any non-empty subset of T1ce, T1, FLAIR, and T2. PMTM maintains a compact state for the current subject across the encoder stages. At each stage, we use assignment maps to pool spatial features into ordered evidence vectors for each available modality. We average these vectors across the available modalities, and a learned gate combines this evidence with the previous state. The up- dated state guides residual fusion at that stage and is used in the state update at the next stage. A modality-token mixer also models interactions among the available modalities at each stage. We apply semantic supervision to the assignment maps at all four stages. We evaluate PMTM on BraTS2018, BraTS2020, and BraTS2023 GLI across all 15 non-empty modality subsets. Averaged over the three tumor regions and the 15 subsets, PMTM achieves Dice scores of 75.31%, 77.54%, and 80.47% on BraTS2018, BraTS2020, and BraTS2023 GLI, respectively. These scores are higher than those of RobustSeg, RFNet, and mmFormer in our experiments on each dataset. Component ablations and internal state analysis are performed on BraTS2020.
[CV-309] Reliability-Preserving Compression for Deployable Medical Image Segmentation
链接: https://arxiv.org/abs/2610.25460
作者: Sagar Kumar,Shailja Sharma,Tirthajit Baruah,Punit Rathore
类目: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Signal Processing (eess.SP)
备注:
Abstract:Post-training compression of pretrained medical segmentation models commonly uses quantization, pruning, or distillation to reduce model size and computational cost. Existing medical segmentation quantization studies mainly evaluate Dice, model size, memory, and latency, but this is incomplete for safety-critical use because compressed models should also preserve uncertainty signals that indicate unreliable predictions. We study this problem in compressed MRI/CT segmentation and formalize why Dice can remain stable while uncertainty changes. Motivated by this failure mode, we propose score-guided aggregate-validated mixed precision (SGAV-MP), a post-training compression framework that selects layer precisions under reliability constraints. SGAV-MP ranks quantization actions with a compact benefit-risk score and accepts them only after full-model reliability validation on a calibration set. Across five 2D backbones and three 3D benchmarks, SGAV-MP preserves FP32-level Dice and high uncertainty-map agreement, achieving 3.59\times – 7.8\times compression in 2D and 4.73\times – 7.82\times in 3D. We also report backend-supported TensorRT deployment measurements separately to distinguish runtime execution from reliability-preserving storage compression.
[CV-310] Evidence Field: Adaptive Covering Programs for Medical Imaging
链接: https://arxiv.org/abs/2610.24773
作者: Isarar Siddique,Raiyan Siddique,Arshi Niyaz,Jamyang Ugyen Tshomo,Atish Raza
类目: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 9 pages, 5 figures
Abstract:Medical image classifiers typically return predictions without a native contract for independently reconstructing their spatial decision computation. Evidence Field is a trainable spatial-program architecture that jointly produces class scores, selected covering programs and signed evidence. Its encoder combines absolute and image-relative response fields with associative spatial moments and learned characteristic sums. A typed classifier scores individual regions and parent-conditioned multiplicative groups; a covering recursion selects scale, partition and composition. We establish exact one-fold coverage, optimality within the registered grammar, a bound relating smooth training to hard inference, signed-score conservation and class-centred offset invariance. Direct contractions and their adjoint specify arithmetic and storage scaling on two- and three-dimensional grids. Native output records bind inputs, parameters, supports, types and contextual dependencies to individual and ensemble predictions. In a separately reported histopathology application, patient-separated training and evaluation across held-out centres provide empirical evidence for the architecture. Trained mechanism and corrected-loss comparisons distinguish predictive effects from implementation correctness; independent readers reconstruct descriptors, routes, scores and probability mixtures. The principal contribution is a coupled representation, decision and evidence architecture whose selected computation can be reproduced independently.
[CV-311] Severity-Channel Conditioning: An Injection-Level Study for Consumer-Device Deblurring
链接: https://arxiv.org/abs/2610.24599
作者: Abdullah Al Shafi,Kazi Saeed Alam,Sumaiya Rahim Suma,Sk. Imran Hossain
类目: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Multimedia (cs.MM)
备注: 17 pages (12 main + 5 supplementary), 7 figures, 14 tables. Code: this https URL
Abstract:Motion-deblurring models lose substantial accuracy when transferred from professional cameras to smartphone-derived imagery. This paper studies whether a severity descriptor, partly derived from the paired sharp target, helps adaptation on the iPhoneBlur consumer-device benchmark. Under a shared update budget, input concatenation of a severity map is compared with global feature modulation by capture metadata, spatial feature modulation by the severity map, and a tier-contrastive output objective, together with an unconditioned spectral-loss control, all fine-tuned from the same domain-adapted latent-diffusion baseline. The primary experiment is an oracle diagnostic: target-derived information is supplied at inference. For the evaluated model pair, input concatenation improves overall peak signal-to-noise ratio from 31.23 to 31.30 dB, whereas feature modulation, the contrastive objective, and the spectral-loss control do not exceed the baseline. The gain is positive on 87.2% of test images and on 14 of 15 held-out source videos; this video-level analysis describes the evaluated checkpoint pair and does not establish robustness to retraining, and three training seeds are reported separately. The input-level gain recurs on a transformer backbone. In one exploratory run, zeroing the channel at inference retains most of the gain in peak signal-to-noise ratio but lowers structural similarity below the baseline. A normalized-step diagnostic does not support the expected larger-gradient explanation. The results characterize the evaluated configurations and do not establish injection location as the sole cause, a spatial-locality mechanism, or a reliable deployment advantage.
[CV-312] LUMINR: Intensity-conditioned Meta-learned INRs for Accelerated Medical Image Registration
链接: https://arxiv.org/abs/2610.24046
作者: Damian Rynczak,Marek Wodzinski
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 7 figures, 3 tables
Abstract:Implicit neural representations (INRs) have been applied to deformable image registration, where a lightweight network maps a spatial coordinate to a deformation vector. Because an INR encodes one specific signal in its parameters, a fresh network must be optimized for every image pair at test time, which is slow. Existing methods reduce this cost by meta-learning an initialization across a population of image pairs, and condition the weights of the INR globally, so the same conditioning applies at every coordinate while the deformation field it initializes varies across the domain. We propose LUMINR, a meta-learning framework that accelerates this per-pair fitting and in which the conditioning is local instead. LUMINR uses an INR that maps a coordinate, together with local intensity and gradient features, to a deformation vector, so the conditioning varies over the domain at the resolution of the field being modelled. Its objective is to find an initialization of the INR parameters from which the network can be fine-tuned to register any unseen pair in only a few optimization iterations. We evaluate LUMINR on 3D deformable registration of brain MRI from the OASIS dataset, where local conditioning and the meta-learned initialization prove complementary: early in fitting they reach a markedly higher accuracy than either does alone. Fine-tuned from that initialization, LUMINR matches the accuracy of the same conditioned INR fit from scratch in roughly a tenth of the gradient steps and a tenth of the wall-clock time, while the deformation fields stay smooth with negligible folding.
[CV-313] GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Synthesis in Glioma Patients
链接: https://arxiv.org/abs/2610.24044
作者: Moinak Bhattacharya,Annie Singh,Mohammad Chaudhry,Prateek Prasanna,Gagandeep Singh
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Tissues and Organs (q-bio.TO)
备注:
Abstract:Background: Incomplete multiparametric MR examinations limit glioma characterization and downstream AI tools. Generative models can synthesize missing MRI sequences, but anatomical structure and clinically relevant tumor features are challenging to preserve. We propose GliomaINK, an anatomy-guided conditional flow matching framework for cross-sequence 3D brain MRI synthesis. Methods: A 3D SwinUNETR predicted the conditional velocity field while an auxiliary topology-preserving loss encouraged structural consistency. The model was trained on 1,083 patients from the BraTS-AG dataset and independently tested on an in-house clinical cohort. Downstream tasks include image quality, tumor segmentation, IDH1 mutation-status prediction, and overall survival prediction. Two radiologists independently evaluated the anatomical fidelity, pathological representation, and overall image quality of the synthesized images. Results: GliomaINK exhibited the highest PSNR across all six MRI translation directions in both evaluation cohorts. In the BraTS-AG test set, the PSNR reached 25.64\pm0.55 dB for the T2W-to-T1C synthesis, compared with 13.78\pm0.32 dB for Pix2Pix and 18.90\pm0.53 dB for SPADE. For downstream tumor segmentation, the method showed strong performance across tumor subregions with Dice scores reaching 0.928\pm0.014 for edema, and 0.885\pm0.040 for enhancing tumor. For the IDH1 mutation-status prediction, GliomaINK achieved the highest AUROC score in four of the six sequence configurations, reaching 0.827\pm0.088 . Conclusion: Anatomy-guided conditional flow matching enables high-fidelity cross-sequence glioma MRI synthesis demonstrating potential for preservation of clinically relevant information in synthesized MRI and improving downstream tasks. Subjects: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Tissues and Organs (q-bio.TO) Cite as: arXiv:2610.24044 [eess.IV] (or arXiv:2610.24044v1 [eess.IV] for this version) https://doi.org/10.48550/arXiv.2610.24044 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Moinak Bhattacharya [view email] [v1] Fri, 16 Oct 2026 15:09:12 UTC (4,259 KB)
[CV-314] Benchmarking and Improvement of Artifact Segmentation Methods in HE-Stained Whole Slide Imaging
链接: https://arxiv.org/abs/2610.24037
作者: Julia Merta,Daniel Mendelsohn,Benedikt Schoser,Joanna Polanska,Michal Marczyk
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Technical artifacts introduced during tissue preparation or slide scanning remain a major obstacle to developing robust computational pathology algorithms for hematoxylin and eosin (HE)-stained whole-slide images (WSIs). Despite the growing number of quality-control tools, their real-world utility remains hard to assess because of the limited number of publicly available implementations and the lack of systematic, multi-dataset comparisons. This paper presents a benchmark of artifact segmentation methods for HE-stained WSIs. After a literature review, six publicly available algorithms were selected and compared in terms of their ability to segment artifact-free tissue and selected artifact classes. Two publicly available datasets and a newly introduced dataset released to support further research were used for evaluation. Tissue samples from multiple medical centers and different organs are included, enabling evaluation across heterogeneous data sources. Segmentation performance was assessed using Dice Index, pixel-level recall, and false discovery rate for artifact-free tissue segmentation. For algorithms supporting multiclass artifact segmentation, class-wise Dice Index and inter-class misclassification were evaluated. In addition to segmentation quality, processing time was assessed. Statistical analysis revealed significant differences in segmentation performance between methods. HistoKit, which incorporates a modified version of the GrandQC, was identified as the best-performing algorithm, while HistoQC was the worst. In terms of artifact types, dark spots and out of focus regions were harder to segment than others. Finally, a simple post-processing step was introduced for HistoKit, which further improved segmentation performance. The benchmark establishes a reference for evaluating and developing artifact segmentation methods for HE-stained WSIs.
[CV-315] Implicit Neural Models for Direct Multi-Shell Diffusion MRI Signal Estimation
链接: https://arxiv.org/abs/2610.24034
作者: Sanna Persson,Fabian Leander Sinzinger,Christoffer Ohlsson,Rodrigo Moreno
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 23 pages (9 pages main text + supplementary material), 9 figures, 18 tables
Abstract:Diffusion magnetic resonance imaging is a noninvasive technique that is sensitive to the motion of water molecules in human tissue. The image signal is commonly described in q-space; however, a complete coverage of this space is often infeasible due to hardware and acquisition time limitations. In this work, we propose to represent the entire diffusion signal using an implicit neural representation network with wavelet activation functions. The signal space is encoded using Fourier features and spherical harmonics basis functions for a continuous output distribution of the raw diffusion data, enabling both super-resolution and q-space interpolation. Additionally, we introduce a pretraining strategy using a group template, enabling faster per-subject optimisation when combined with nonlinear registrations of subjects into template space. We tested the method with 10 subjects from the Human Connectome Project and 26 subjects from in-house data (multidimensional diffusion MRI from a Parkinson’s disease cohort). The results show that the proposed method yields high-fidelity reconstructions with significantly reduced memory requirements compared to raw diffusion-weighted images. The pretraining scheme showed good generalisability properties on data acquired with very different protocols. We further assess the method on downstream metrics relevant to diffusion MRI data from the learned representations.
[CV-316] ProtoGTX: Prototype-Aligned Graph Transformer for Explainable Whole-Slide Image Analysis
链接: https://arxiv.org/abs/2610.24023
作者: Iori Azuma,Hiroyuki Kusuhara,Tadahaya Mizuno
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
备注: 14 pages, 3 figures, 2 tables. Code: this https URL
Abstract:Deep learning has achieved strong performance in computational pathology, but explaining whole-slide image (WSI) predictions remains challenging. Existing explanation methods often highlight important regions without identifying their morphological patterns, leaving a gap between patch-level computational evidence and slide-level pathological interpretation. We propose ProtoGTX, a graph-based framework that combines spatial contextualization with a flexible morphological prototype vocabulary. ProtoGTX separates the reference prototypes used to define morphological concepts from the predictive model: the reference vocabulary can be generated from training data by unsupervised clustering or supplied externally, and remains fixed during classifier training. Patch features are first contextualized over a spatial graph, after which learnable prototypes representing morphological concepts are optimized through cross-attention. Their slide-level activation profile is aligned with the morphological composition induced by the fixed reference prototype bank. On the CPTAC and BRACS datasets, ProtoGTX outperforms the evaluated MIL, prototype-based, and graph-based baselines, with particularly pronounced improvements on BRACS. Qualitative analyses further show that spatial contextualization transforms fragmented patch-level prototype assignments into more coherent tissue regions. In BRACS, prototype maps also co-localize with pathologist annotations and regions relevant to slide-level prediction. These results demonstrate that ProtoGTX provides complementary semantic and spatial explanations while retaining a prototype vocabulary that can be defined independently of the predictive model. Code is available at this https URL.
[CV-317] DinoMUSE: A DINOv2-guided Framework for Unsupervised Lateral Motion Estimation in Ultrasound Elastography
链接: https://arxiv.org/abs/2610.24020
作者: Xiyao Meng,Xingyue Wei,Baihua Luo,Bo Peng,Jianwen Luo
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Quasi-static ultrasound elastography (USE) is a non-invasive imaging method for assessing soft tissue mechanical properties. Limited by low resolution, low sampling frequency and lack of phase information in the lateral direction, accurate lateral motion estimation remains a critical bottleneck in USE. Existing methods mainly rely on local image matching and physical constraints (such as tissue incompressibility assumption and effective Poisson’s ratio), which lack sufficient contextual information. We present DinoMUSE, a DINO-guided mixture-of-experts (MoE) framework for unsupervised lateral motion estimation in USE. DinoMUSE incorporates high-level visual representations learned by DINOv2 into the motion estimation network to provide global contextual guidance. To facilitate effective interaction between DINOv2 features and local matching features, an MoE-based feature-fusion module adaptively transforms and integrates the two representations. Evaluations against NCC, GLUE, RAFT, and sPICTURE on simulation and in-vivo breast datasets show that DinoMUSE substantially improves lateral estimation accuracy while maintaining competitive axial performance. Specifically, DinoMUSE reduces the mean square error (MSE) of lateral displacement and strain estimation by 60% and 65%, respectively, compared with a SOTA algorithm on the simulation dataset. For in-vivo breast data, DinoMUSE achieves axial strain SNR and CNR of 5.70 and 4.82, and lateral strain SNR and CNR of 4.61 and 2.23, compared to 4.11 and 3.32, and 4.57 and 2.13 for the best competing method, alongside improved lesion visualization. These findings indicate the potential of leveraging transferable visual representations and adaptive feature fusion to improve lateral motion estimation in USE. The code is available at xxxxxxxx.
[CV-318] An Anatomy-Aware Staged Pipeline for Vessel-Specific Intracranial Aneurysm Segmentation in CT and MR Angiography MICCAI2026
链接: https://arxiv.org/abs/2610.24011
作者: Gniewosz Drwiega,Wojciech Szymanski,Marek Wodzinski
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: Submitted version (7 August 2026) of a paper accepted at SWITCH+ 2026, a MICCAI 2026 workshop (TopAneu 2026 challenge track). Table 1 entries shown as dashes were pending at submission; the final version with complete results will appear in the MICCAI 2026 Satellite Events proceedings (Springer LNCS). 6 pages, 1 figure, 1 table
Abstract:The TopAneu 2026 challenge requires intracranial aneurysms to be segmented at the voxel level and assigned to one of more than fifty vessel-specific classes, on CT and MR angiography alike. We address it with a six-stage anatomy-aware pipeline rather than a single segmentation network. A logistic-regression classifier routes each scan to a modality-specific branch; networks segment the vasculature, first as a binary mask and then over the full 36-class arterial taxonomy; a candidate localiser turns raw angiography and vessel priors into suspicious regions; a candidate-centred network segments the lesion in each crop; connected components become instances; and a per-instance classifier assigns each its anatomical location. Two principles organise the design. Whenever an upstream model output becomes a downstream training input it is generated out-of-fold, so no stage trains on unrealistically clean predictions of itself. And the anatomical decision is made once per lesion rather than once per voxel, which fits the data: of 406 connected aneurysm components in the full release, 405 carry exactly one location label. Task 1 is read off the Task 2 mask, making the two consistent by construction. We report stage-wise performance and a five-part ablation isolating what vessel priors and class-aware sampling contribute at each stage.
[CV-319] A Configurable ETC2 Encoder with Learned Search
链接: https://arxiv.org/abs/2610.23961
作者: Ziyu Guo,Haomiao Jiang,Sergey Makeev,Angelo Pesce
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:ETC2 fixes the texture representation and bit rate, leaving the encoder to trade search time for distortion. We present a configurable CPU encoder in which rules or small neural networks allocate conventional ETC2 searches; these searches construct every encoded block. Configurations span fast-encoder output through broader mode and endpoint searches. On 64 color textures used to evaluate and select configurations, the fastest measured configurations reaching at least Google Etc2Comp’s pooled REC709 PSNR at efforts 0-49.5 run 30.7-40.9 times faster. A high-quality configuration exceeds effort-100 PSNR with 64.1% less time. At equal search quotas, learned action ranking and second-search allocation jointly gain 0.124 dB over rule-based choices. At high quality, learning saves time at a small distortion cost. Outputs retain standard ETC2 storage and hardware decoding.
[CV-320] Flow Matching-Based Semantic Communication With Time-Aware Structural Residual Injection
链接: https://arxiv.org/abs/2610.23922
作者: Chengyang Liang,Dong Li
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generative semantic communication (SemCom) enables low-bandwidth image transmission by reconstructing images from semantic descriptions. However, text-only schemes often suffer from poor structural fidelity due to the stochastic nature of generative models, while still requiring high transmission payload and inference latency. This letter proposes a dual-information-flow SemCom framework based on Flow Matching. At the transmitter, the most semantically matched prompt is selected from multiple candidate descriptions, while an ultra-low-dimensional structural residual velocity vector is extracted from the inversion trajectory of the original image. At the receiver, a Time-Aware Recursive Residual Injection (TARRI) network progressively releases structural corrections along the denoising trajectory using the transmitted residual vector as structural memory. The proposed framework transmits only prompt embedding and a compact residual vector, achieving robust high-fidelity image reconstruction under extremely low payload. Experimental results demonstrate that the proposed framework significantly improves both structural fidelity and perceptual quality over existing text-driven semantic communication schemes.
[CV-321] Generative Priors for Total-Body PET Reconstruction MICCAI
链接: https://arxiv.org/abs/2610.23870
作者: Movindu Dassanayake,George Webber,Denis Prokopenko,Julia A. Schnabel,Andrew J. Reader
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: MICCAI RIME 2026 (Accepted Poster)
Abstract:Total-body positron emission tomography (TB-PET) enables fast, low-dose, and dynamic imaging over a long axial field of view, but reconstruction remains challenging due to reduced counts, large image volumes, and increased data dimensionality. Generative models offer powerful learned priors for low-count PET, yet explicit PET measurement conditioning at TB scale has not been demonstrated. Inspired by PET decomposed diffusion sampling (PET-DDS), we propose TB-PET-DDS, a reconstruction framework that combines a CT-conditioned generative prior with patient-specific list-mode PET data. During reverse diffusion, PET consistency is enforced using a closed-form 3D TB-PET maximum-a-posteriori expectation maximisation fusion update, rather than relying solely on image-domain conditioning. The synthetic measurements are generated from the low-dose vendor reconstruction, providing a controlled inverse problem and retaining a full-dose reference. Proof-of-concept evaluation was performed on a single clinical TB-PET acquisition, with reconstruction hyperparameters selected against the full-dose reconstruction from the same case for all models. TB-PET-DDS improved NRMSE and SSIM relative to the DDIM and Image-DDS baselines, improving through-plane consistency in coronal views while enhancing agreement with the measured PET data. Although smoothed low-dose vendor reconstructions achieved the best global voxel-wise metrics, TB-PET-DDS produced the lowest liver coefficient of variation. These results demonstrate the feasibility of measurement conditioned diffusion reconstruction at TB scale and provide an initial route towards learned, projector-consistent generative reconstruction for low-count TB-PET imaging.
人工智能
[AI-0] CoopUQ: Robust Multi-Robot Social Navigation via Uncertainty-Aware Cooperative Forecasting
链接: https://arxiv.org/abs/2610.26852
作者: Zehao Wang,Jianpeng Yao,Yuping Wang,Yiping Li,Jiaqi Li,Zhixu Li,Kanghoon Lee,Jinfeng Wang,Cong Liu,Zhaowei Tan,Jinkyoo Park,Jiachen Li
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Conference on Robot Learning (CoRL 2026); Project Website: this https URL
Abstract:Reinforcement learning enables mobile robots to make fast and scalable navigation decisions in constrained human-shared environments. However, directly extending single-robot approaches to multi-robot settings often fails to ensure safety and efficiency, especially under occlusions and out-of-distribution (OOD) scenarios, where individual robots may overlook potential risks or make inaccurate predictions. Leveraging cooperation effectively for both prediction and navigation remains a major challenge. We propose CoopUQ, a multi-robot social navigation framework that bridges uncertainty-aware cooperative occupancy forecasting and robust decision making. CoopUQ predicts future occupancy grid maps through cross-robot spatiotemporal feature fusion and further leverages forecasts and their uncertainty to guide policy via constrained multi-agent reinforcement learning, where the cost penalizes navigation through regions predicted to be occupied or highly uncertain. This design enables the policy to adapt its risk sensitivity under distribution shifts. Experimental results across multiple Unity3D-based simulation environments with both in-distribution and OOD settings, including diverse layouts, pedestrian densities, and rushing behavior, demonstrate improved prediction performance and safety over representative baselines. Real-robot experiments further demonstrate its robustness across various scenes. Our code and videos are available at this https URL.
[AI-1] rain An Imaginative Agent
链接: https://arxiv.org/abs/2610.26845
作者: Jiaxin Ge,Junyi Zhang,Tianning Chai,Hanjun Yoo,Chenyang Wang,Xiuyu Li,Weijia Shi,XuDong Wang,Sewon Min,Luke Zettlemoyer,Trevor Darrell,Haiwen Feng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A seemingly reasonable push in Sokoban can make the puzzle unsolvable. To avoid such mistakes, an agent has to anticipate the consequences of its actions before execution, especially when feedback from the real environment is sparse or actions are irreversible. We introduce Agentic Imagination (Agima), a training framework that trains an agent to imagine before acting, i.e. to be imaginative. Agima trains a single omni model to propose actions, generate images of their outcomes, and then reason over these imagined futures to decide what to execute in the real environment. Rather than committing to a single plan, the agent can simulate a sequence of actions, explore alternatives, and compare possible futures. We first use supervised fine-tuning (SFT) to unify policy and world modeling within a single omni model. This enables the agent to predict future outcomes and perform basic forms of imagination, such as forward thinking and counterfactual branching. We then use agentic reinforcement learning (RL) to teach the agent to reason with its imagined futures and use them effectively for planning. Across four environments spanning Maze2D, Sokoban, Pacman, and PushT, Agima generalizes to harder settings unseen during training. Trained with SFT alone on two-box Sokoban, Agima solves 40% of four-box puzzles while seeing the real board only once every five moves, compared with an action-only planner that only solves 2%. After RL, new behaviors emerge in how Agima uses its imagination: it learns a new unified format that was unseen during the SFT stage, learns to conduct forward thinking only when necessary, and branches more frequently at critical decision points. An agent that can picture where its actions lead can learn when and where to look ahead, that is, to plan. Project page: this https URL.
[AI-2] AnyGoal: On World Model Generalization Beyond Expert Rollouts
链接: https://arxiv.org/abs/2610.26844
作者: Heejeong Nam,Divyat Mahajan,Quentin Le Lidec,Randall Balestriero,David Klindt
类目: Artificial Intelligence (cs.AI)
备注: Project Page: this https URL
Abstract:The goal of a world model is to capture the dynamics of an environment. This has the great appeal that it allows agents to plan and simulate a course of actions with the help of a world model. For these uses to be reliable, however, the learned dynamics must behave well beyond the specific state-action combinations observed in expert rollouts. Despite this, the relationship between data coverage, dynamics generalization, and downstream planning performance remains underexplored. We study this relationship through the lens of compositional generalization: can a world model recombine familiar state and action factors into valid dynamics? To answer this, we introduce AnyGoal, a benchmark of goal-shifted task variants built from widely used world-model environments, together with an open-source library. Each task is designed to be within reach of a world model that has learned the underlying dynamics: individual state factors and actions remain within their training support, but their combinations are underrepresented in expert trajectories. Across multiple environments and world model families, we find that current models consistently exhibit failures under these recombinations. By examining dynamics and representation failures across model families, interaction regimes, and evaluation metrics, we study where current world models generalize and where challenges remain for goal-shifted prediction and planning. Together, this framework, benchmark, and analysis provide a basis for evaluating progress toward world models that generalize to new planning tasks.
[AI-3] Same Simulation Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
链接: https://arxiv.org/abs/2610.26836
作者: Ke Zhang,Sahchit Chundur,Jerry Peprah Owusu,Mohammad Javad Abdolhosseini Qomi,Maziar Raissi
类目: Artificial Intelligence (cs.AI)
备注: 15 pages, 4 figures
Abstract:LLM agents that compute the right answer can still answer wrongly. Agent harnesses increasingly keep long tool outputs out of the context window, saving them to files that the agent searches or reads on demand. We ask where this costs accuracy with a real scientific tool: PHREEQC, a widely used geochemistry simulator from the USGS, whose reports reach 846.8k characters and whose determinism lets every printed value be traced. Agents use it to answer 158 execution-verified multiple-choice items. When agents read the simulator’s report through a section index rather than in context, GPT-5.1 and GPT-5.2 lose 11.3 and 7.9 points of accuracy, while Claude Opus 4.6 and Claude Sonnet 4.6 show no detectable cost. Re-running all 11,007 logged simulator calls shows that both GPT models compute the keyed value about as often as before; what changes is the evidence they answer from. Under retrieval they often skip the report, answer from their own higher-precision output files, and misround to a neighboring option, a failure we call evidence substitution. With the simulation held fixed, retrieval still costs GPT-5.2 9.4 points. For GPT-5.4, more frequent correct simulations mask an evidence-use loss of about ten points, leaving a net change indistinguishable from zero. Naming the authoritative output raises accuracy by 6.3-7.9 points over plain retrieval for the three GPT models; for GPT-5.1 and GPT-5.2, inlining outputs that fit works as well, while asking for correct rounding does not. Context-management choices in agent harnesses should be judged by where accuracy is lost, not only by final accuracy.
[AI-4] Kronecker Delta Rule for Long Context Linear Attention
链接: https://arxiv.org/abs/2610.26834
作者: Anamaria-Roberta Hartl,Levente Zólyomi,Michael List,Günter Klambauer,Sebastian Böck,Sepp Hochreiter
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We propose KronLSTM, a linear attention architecture based on the Kronecker Delta Rule, a novel update rule for associative matrix memories. KronLSTM extends the xLSTM family with a memory-revision mechanism that corrects targeted associations while modifying the memory only along a single direction in matrix space. This minimally invasive memory revision reduces interference with previously stored information compared with delta-rule updates, making it particularly well suited for long-context language modeling. The resulting memory dynamics can be mathematically described as a Markov chain that converges to a steady-state distribution, while the expected norm of the memory matrix remains uniformly bounded throughout the entire input sequence. Hence, KronLSTM ensures stable memory dynamics without requiring additional normalization and stabilization mechanisms for the memory state. Experiments show that KronLSTM is competitive with, and often outperforms, existing linear attention architectures, including models of the (Gated) DeltaNet family, with particularly strong performance on long context benchmarks. These results establish the Kronecker Delta Rule as an effective and analytically tractable foundation for stable associative memory in recurrent linear attention architectures. Our code (this https URL) and models (this https URL) are open-source.
[AI-5] ProteinGeoBench: Frontier Structural-Biology Agents Are Fluent at Lookup but Lost in Geometry
链接: https://arxiv.org/abs/2610.26829
作者: Matt Turk,Brandon L. Neel,Diana M. Lopez,Zachary D. Smith,Grace N. Johnecheck,James B. Golden
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Biomolecules (q-bio.BM)
备注: 24 pages, 2 figures, 10 tables. Benchmark and tasks: this https URL
Abstract:Frontier agents retrieve structural metadata well and compute geometry from coordinates less reliably. We ask how much of a benchmark score reflects that gap and how much reflects the evaluation configuration around the model. We study ProteinGeoBench, a suite of 131 protein-structure tasks, using 4,249 rollouts from four frontier models and 4,486 further trials across 26 models. Three of the four frontier models fall within 0.031 of one another (0.785 to 0.816) with no difference between them resolved, while the fourth scores 0.707 and all three gaps to it exclude zero. Three of the four do better on metadata lookup than on numeric calculation. The fourth is slightly better on numeric. Numeric tasks are 28 of the 131 but half of the twelve tasks that all four models fail. The ranking also depends on the evaluation budget: with one run per task, 23.9% of bootstrap replicates put a different model first. On a selected 15-task slice, removing network access lowers scores more than blocking eight structural-database domains, but the two contrasts were graded in different passes and cannot be cleanly decomposed. Swapping in each model’s own command-line agent changes scores by -0.071 to +0.086, a comparison too small to rule out moderate harness effects. Across 26 models, numeric performance rises with non-numeric performance rather than staying flat. The experiments use different task subsets, so they cannot rank model, harness, access and task composition against one another. Evaluations should report results by task family, along with the number of runs, the harness and the access conditions.
[AI-6] SOMA: Continuous Monitoring and Intervention for System Two Reasoning with System One Models
链接: https://arxiv.org/abs/2610.26793
作者: Ruizhi Qian,Lei Shi,Tianyu Hu,Bo Yan,Song Wang,Yu Tian
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reasoning models often run into problems without noticing them, even when symptoms are visible in their own text. In humans, metacognition separates monitoring from control: fast, intuitive signals keep watch while slow, deliberate reasoning unfolds and help decide whether to continue, change course, or stop. System One (S1) decision models, such as Jev, are a natural candidate for this role: they return a calibrated probability for a natural-language question about a text in one fast call, yet they are used only as gates before reasoning. We introduce SOMA (System One Metacognitive Awareness), in which an S1 model continuously monitors a System Two (S2) reasoning model and intervenes when needed. SOMA has three layers: a scenario layer states each objective in natural language with predefined actions, an awareness layer lets S1 judge whether the described behavior appears, and a control layer accumulates these judgments, first reminds S2, and takes over only if the behavior persists. Sharing one S1 model and one controller across tasks, without training, SOMA addresses two different failures on two reasoning models: on overthinking, it cuts reasoning tokens by up to 36% while improving average accuracy, and on unanswerable questions, it raises abstention from 70.5% to 87.2%.
[AI-7] LG-Bench: A Graph-Structured Evaluation Benchmark for Life Sciences
链接: https://arxiv.org/abs/2610.26787
作者: Lu Sun,Xiangyi Zhang,Xiangyang Zhu,Kaiyuan Ji,Zijian Chen,Qiang Hu,Zicheng Zhang,Yuan Tian,Guangtao Zhai
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Traditional evaluation benchmarks reduce inherently interconnected scientific knowledge in life sciences into flat lists of questions, disregarding the underlying topological structure of the knowledge. We introduce, the first graph-structured benchmark for life sciences, featuring over 10,000 high-quality multiple-choice questions across medicine, biology, and chemistry. Our approach constructs a weighted evaluation graph using bidirectional matching and semantic similarity algorithms, where nodes represent questions and edge weights capture their semantic relationships. Leveraging this graph topology, we design two novel evaluation metrics. The Global Coherence Score (GCS) measures a model’s consistency within semantically related neighborhoods, while Knowledge Balance Score (KBS) analyzes how model errors are distributed across the graph to reveal conceptual blind spots. LG-Bench facilitates fine-grained comparison of LLMs by surfacing differences in conceptual coherence and patterns of knowledge organization across models. Our framework shifts the evaluation paradigm from flat accuracy metrics to structure-aware analysis, offering a new lens for diagnosing and improving LLM performance in the life sciences domain.
[AI-8] owards Robust Model Organisms for Eliciting Latent Knowledge
链接: https://arxiv.org/abs/2610.26698
作者: Lennart Finke,Hardik Bhatnagar,Euodia Dodd,Francis Rhys Ward,Yves-Alexandre de Montjoye,Matthias Bethge,Mary Phuong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A key problem in AI safety is eliciting latent knowledge (ELK), extracting information from a model that is misaligned and more knowledgeable than its overseer. For example, a misaligned model used to scan code for vulnerabilities may want to withhold its findings, i.e., sandbag, in order to exploit them later. To study ELK empirically, we need model organisms that mirror this setting. Existing model organisms fall short in several respects. In this work, we train model organisms to withhold solutions in a vulnerability detection task. We combine transcript distillation with counter-prompting and tamper-resistance training to instill three properties that we argue are critical for a useful ELK testbed: 1) Selectivity, the model sandbags when it thinks the overseer doesn’t know the answer and complies otherwise, 2) Robustness, the model resists elicitation, and 3) Retention, the model retains a knowledge advantage over the overseer, as opposed to losing the relevant capability. We train a sequence of model organisms and show that our strongest model substantially outperforms prior models on robustness while keeping the other two properties intact. We hope that our model organisms provide a useful ELK testbed for the community, and that our methodology sparks the development of increasingly challenging model organisms. All our code, models and the dataset are available at this https URL. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.26698 [cs.AI] (or arXiv:2610.26698v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.26698 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-9] StebBench: Diagnosing Real-World Procedural Understanding in AI Agents
链接: https://arxiv.org/abs/2610.26679
作者: Hossein A. Rahmani,Bin Wu,Chaeeun Lee,Fanghua Ye,Sahan Bulathwela,Sujay Kumar Jauhar,Marzieh Fadaee,Nick Craswell,Emine Yilmaz
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Evaluating an agent’s planning ability remains challenging for open-domain tasks, where verifying procedures through execution is often impractical. Existing evaluations typically assess only the final plan and do not distinguish procedural errors such as missing, incorrect, or misordered actions. We introduce STEPBENCH, a benchmark of 6,885 verified procedures crawled from wikiHow across 81 open-domain categories. From each procedure, we derive 5 diagnostic tasks: four with deterministic answers involving withheld, incorrect, or shuffled steps, and one requiring next-step generation. We additionally evaluate full-plan generation to examine how diagnostic performance correlates with end-to-end plan validity. Using the same procedures across tasks lets us directly compare different planning capabilities. For generation, STEPEVAL identifies critical failures that prevent goal completion. Across 14 configurations of 12 models, we find substantial differences across planning capabilities, with step ordering proving the most difficult: the strongest model correctly orders only 63.0% of procedures, dropping to 33.1% for those with nine or more steps. Incorrect-step detection also predicts judged plan validity across models, suggesting a lower-cost alternative for evaluating generation quality. On STEPBENCH-Val, comprising 214 procedures published in 2026, comprehension scores and rankings remain stable, suggesting limited influence from memorisation.
[AI-10] Honeypots for AI Agent Insider Threats
链接: https://arxiv.org/abs/2610.26664
作者: Jan Wehner,Jam Kraprayoon
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:AI agents deployed inside organizations can pose an insider threat if they make mistakes, are compromised by an attacker, or pursue misaligned objectives. We propose honeypots as an AI control method to improve the security of internal agent deployments. Rather than only using honeypots for pre-deployment evaluations, defenders should plant decoy opportunities for harmful actions throughout the production environment to detect misbehavior, incriminate malicious agents, and deter attacks. Such honeypots can elicit misbehavior under real deployment conditions, identify malicious agents even if chain-of-thought monitorability degrades, provide reliable warnings about agent misbehavior to security teams, and scale cheaply to many agent instances. The presence of honeypots also introduces uncertainty about which opportunities are real, thus deterring malicious agents from attacking and reducing the rate of misbehavior. We discuss the limitations of honeypots and give reasons to expect them to remain effective as agent capabilities increase. Finally, we outline key research questions to measure and improve the effectiveness of honeypots: testing the robustness of honeypots, comparing their detection effectiveness with other monitoring methods, understanding their incrimination and deterrence effects, and building supporting infrastructure.
[AI-11] AllPath: Layered Verification for a Deployed Self-Revising LLM Trading Agent under Human-Gated and Autonomous Regimes
链接: https://arxiv.org/abs/2610.26662
作者: Tian Zhang,Meng Li
类目: Artificial Intelligence (cs.AI)
备注: Accepted to the 3rd Workshop on LLMs and Generative AI in Finance (AI4F) at ACM ICAIF 2026 (non-archival). 11 pages, 3 figures, 5 tables. Code: this https URL
Abstract:LLM trading agents are evaluated almost exclusively in backtests and autonomous benchmark harnesses. Deployment, where actions are irreversible, raises a different question: who verifies the agent’s actions, and which failures can each verifier structurally catch? We report on AllPath, a personal LLM trading agent deployed against a live paper-brokerage account. The system watches positions through a deterministic rules engine, and each night it reviews its own day and proposes revisions to its own trading strategies. Verification is split across three layers. Code verifies what must not depend on judgment: deterministic gates enforce risk limits and block any strategy revision that fails an integrity check. The agent verifies itself within fixed bounds: its nightly review has an iteration cap and no tools that can act. A human verifies judgment, through approval screens that show the current state of each proposal. We also compared two verification policies for the agent’s self-revisions on the live system: two and a half weeks with every self-revision human-gated, then five weeks with reflection revisions auto-applied behind the same code gates, while the operator’s chat channel stayed human-gated throughout. Each layer caught a different kind of failure. The code gates rejected no proposal; their role was keeping revisions consistent. The human rejected two proposals that passed every code check, both over a disagreement about the market, and later removed 11 of the 19 rules the agent had added on its own, at a median lag of eight days. The most instructive production incidents, such as strategies silently left unmonitored, passed every gate; the operator found them by hand, and the fix each time was an interface change that makes the condition visible. We contribute the layered architecture, the two-regime account, and a production incident taxonomy, none observable in a backtest.
[AI-12] andem: Serving Hybrid Linear-Attention GUI Agents on Mobile NPUs
链接: https://arxiv.org/abs/2610.26660
作者: Zhuomin Chen,Jingchao Ni,Xu Zheng,Yong Chen,Mo Sha,Wei Cheng,Dongsheng Luo
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Smartphone GUI agents complete tasks in mobile apps by repeatedly reading the screen and asking a model for the next action. Mobile neural processing units (NPUs) enable local execution, keeping screen contents on the phone, but model latency accumulates with every action. Hybrid linear attention models reduce memory requirements, yet reusing their recurrent state requires checkpoints at the appropriate input positions. Agent requests retain instructions but rewrite recent history, so checkpoint policies designed for chat continuation can miss reusable prefixes. Retaining rollback states for speculative decoding can also move prompt processing off the NPU’s matrix unit, adding overhead that outweighs decoding savings on short action outputs. A loaded model can further hold the NPU awake between tasks. We present Tandem, a runtime for hybrid GUI agents on mobile NPUs. Tandem saves recurrent state where consecutive prompts diverge, allowing later requests to reuse previously processed context. It keeps most prefill on the matrix unit and produces rollback states through a short per-token tail. Its verification kernel reads each weight block once for all proposed tokens. Between request bursts, Tandem releases NPU power votes without unloading the model. We implement Tandem in this http URL and evaluate it on recorded agent interactions across two Snapdragon generations. On a Snapdragon 8 Elite, Tandem reduces per-step latency by 30% for text-based screen descriptions and 16% for screenshots relative to this http URL’s fastest NPU configuration, reduces energy per text step by 33%, and lowers standby power with the model loaded from 0.76 to 0.22 W.
[AI-13] he Effectiveness and Durability of Pretraining Data Filtering in Protein Language Models
链接: https://arxiv.org/abs/2610.26651
作者: Aaron Maiwald,Roderick McNeill,Jaeyoung Lee,Matthew A. Clarke,Lennart Justen,Alejandro Acelas,Ivo Andrews,Giles Harper-Donnelly,Adam Moore,Sophie Rose,Oliver M. Crook,Felix Jackson
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Pretraining filtering is one of the most common safeguards for open-weight language models. Here we study its effectiveness, durability and scalability using the ESM-2 family of protein language models. We find that filtering viral proteins from the pretraining data meaningfully degrades masked language modelling, mutation effect prediction and long-range contact prediction on viral proteins, while largely preserving performance on non-viral proteins. These effects are concentrated among proteins from human-infecting viruses, and grow as we scale models from 8M to 3B parameters. As models increase in size, we find that stringent filtering becomes more effective. However, fine-tuning on sequences from the removed viral families restores the lost performance. Although the cost to recover performance increases with model scale, even for the largest models it costs about 80 USD of H100 time. This work establishes pretraining filtering as moderately effective at selectively degrading performance on viral proteins, but not durable against fine-tuning with modest computational resources and access to relevant data.
[AI-14] Reasoning -Aware Multicalibrated Reliability Predictors for Language Models
链接: https://arxiv.org/abs/2610.26645
作者: Ekaterina Fedorova,Alessandro Morosini,Chara Podimata
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:As people increasingly rely on large language models (LLMs) for information, model providers need to communicate meaningful estimates of whether individual responses are correct or, more generally, reliable. In the context of hallucinations, existing work uses multicalibration to make such reliability predictions meaningful across application-relevant groups, such as prompt topics. Topic groups alone, however, can conceal systematic errors: within the same topic and confidence level, responses with different chain-of-thought (CoT) properties may exhibit opposing errors that cancel on average. In this work, we introduce reasoning-aware multicalibration, a framework that supplements application-relevant topic groups with groups derived from observable properties of CoT traces. We highlight that, theoretically, groupings based on CoT are “internally important”, i.e., groupings for which the multicalibration guarantee is not societally crucial as it is for high-risk topics, but whose inclusion in the multicalibration procedure could easily and significantly improve predictive accuracy of the final reliability predictor. Empirically, we apply our framework to hallucination prediction and unsafe-response detection across a variety of benchmarks, and find that adding CoT groups to topic-based calibration reduces Brier score in both tasks. Interestingly, we find that multicalibrating with respect to CoT groups alone (without any topic information) already improves calibration across topics, suggesting that the predictor’s errors depend on how the model reasoned, rather than what it was asked about in the prompt.
[AI-15] Revision-Based Learning from Interactive Experience and Feedback
链接: https://arxiv.org/abs/2610.26636
作者: Jiawen Li,Chenming Ge,Alice Gao,Rishab Vempati,Yi-Hao Peng,Jason Wu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Code generation models are increasingly used to produce interactive artifacts, such as websites and user interfaces. Many properties that determine whether these interfaces are useful only become apparent through realistic interaction, making them difficult to incorporate into model training. In this paper, we introduce RELIEF (REvision-based Learning from Interactive Experience and Feedback), a method for improving code generation models using feedback on their generated interfaces. RELIEF uses a coding agent to revise these interfaces, then incorporates the improvements into their original generation traces to produce supervision for fine-tuning. To evaluate RELIEF, we collected a dataset of 200 synthetically generated websites paired with 355 interaction recordings and accompanying feedback from 24 users. We used critiques derived from code inspection, automated interaction, explicit human feedback, and human interaction recordings to fine-tune variants of two code generation models. Our results show that feedback effectiveness varies with base model capabilities, with critiques inferred from human interaction recordings leading to the largest usability improvements overall under our tested models.
[AI-16] VaDar: Implicit Goal Conditioning via Value Disaggregation
链接: https://arxiv.org/abs/2610.26629
作者: Shashwat Saxena,Mehul Goel,Sreyas Venkataraman,Sarvesh Patil,Max Simchowitz
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Long-horizon tasks are important for modern AI and robotics, yet pose a challenge for reinforcement learning. Historically, a popular approach has been to decompose a task into subgoals, and learn a subgoal-conditioned policy to execute each subtask in sequence. We introduce an alternative called Value Disaggregation (VaDar), which trains a single subgoal-independent policy to maximize a linear combination of per-subgoal critic values. Because subgoal information is needed only at training-time, VaDar removes the need for possibly costly or cumbersome subgoal-selection during policy deployment. Moreover, through careful experiments, we show that VaDar (1) is always competitive with, and frequently outperforms, natural baselines like goal-conditioning and (2) succeeds on tasks with non-sequential goal structure and multiple optimization objectives. In particular, VaDar is the first algorithm to solve challenging, long-horizon tasks in the RoboCasa suite, even when starting from a policy pretrained from behavior cloning with near-zero initial task success. Taken together, VaDar’s successes suggest that subgoal information benefits actor-critic reinforcement learning by enhancing the efficacy of critic learning, whereas policy goal-conditioning is often unnecessary. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO) Cite as: arXiv:2610.26629 [cs.LG] (or arXiv:2610.26629v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.26629 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-17] Concerns in Benchmarking Jailbreaks: An Empirical Study of Attack Success Rates
链接: https://arxiv.org/abs/2610.26595
作者: Leon Tu Nguyen,Rohan Menon,Rafael Tappe Maestro,Eric Aubinais,Chih-Hong Cheng
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Jailbreak methods are commonly compared using Attack Success Rate (ASR), but these comparisons depend on how responses are generated and judged. In this paper, we examine eight attacks against three open-weight language models using three evaluation procedures. Low-resource language attacks achieve a mean ASR of 96.0% under keyword matching but 2.7% under a Qwen3 evaluator following StrongREJECT scoring instructions. Changing the number and arrangement of attempts reverses the ranking of DeepInception and rewriting. In a separate experiment, sampling 100 responses for each unchanged request at Ollama’s default temperature of 0.8 yields 100% ASR under keyword matching but 50% under the Qwen3 evaluator. Beyond evaluator sensitivity, these findings raise a safety concern: repeated sampling can expose residual unsafe behavior despite an initial refusal, suggesting probabilistic suppression rather than robust exclusion. We recommend explicit success definitions, matched response budgets, validated evaluators, and reporting cumulative failure under repeated interaction rather than relying on single-generation tests.
[AI-18] WhatIf-Bench: Benchmarking Intervention-Aware Time Series Forecasting
链接: https://arxiv.org/abs/2610.26582
作者: Arjun Ashok,Mihir Parmar,Palash Goyal,Bhavana Dalvi Mishra,Chun-Liang Li,Irina Rish,Alexandre Drouin,Jinsung Yoon,Tomas Pfister
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
备注: Leaderboard at this https URL code at this https URL
Abstract:Recent time series foundation models (TSFMs) can consume covariates in-context, extending zero-shot forecasting to problems driven by external factors. This enables scenario analysis: forecasting how a target responds to events that set covariates to chosen future values. Whether they can do this reliably is an open question. We introduce WhatIf-Bench, a benchmark measuring how well forecasters use covariates to forecast a target’s response to future events. WhatIf-Bench is semi-synthetic: time series come from mechanistic simulators of data centers, power grids, road traffic, and water networks, with controlled covariate-target relationships grounded in each system. Each task presents a history of events intervening on covariates, then asks to forecast a future event that either replays seen values (Replay) or sets them to unseen ones (Novel). Varying the number of covariates, historical events, and event length yields 37,665 tasks. Evaluating TSFMs, frontier LLMs, and simple regression baselines, we find the LLMs reach the highest skill (improvement over a naive baseline), ahead of the best TSFMs. Every model forecasts well on Replay futures (48 to 85% skill) but drops to -7 to 50% on Novel futures, with the regression baselines falling the most. Skill on Novel futures also falls as events set more covariates jointly. More historical events raise skill on Novel futures across the board, but large differences remain. WhatIf-Bench reveals generalization to Novel futures as the core challenge, and offers a testbed toward arbitrary scenario analysis.
[AI-19] Reward Objectives for Robust (PO)MDPs: A Law-Preserving Equivalence with (Partially Observable) Games
链接: https://arxiv.org/abs/2610.26568
作者: Durgam Latha,Dion Reji,Dorde Zikelic,Shankaranarayanan Krishna
类目: Artificial Intelligence (cs.AI); Formal Languages and Automata Theory (cs.FL)
备注:
Abstract:Robust partially observable Markov decision processes model agents that act under partial observability while adversarial environment resolves uncertainty in the transition probabilities. Reward objectives such as discounted-sum, long-run average and cumulative reward under finite and infinite horizons allow quantifying the utility of an agent. We consider the problem of solving RPOMDPs under all four reward objectives, where the goal is to decide whether the value of a risk measure with respect to a reward objective exceeds a given threshold. We consider different risk measures, such as (1)~expected value, (2)~mean-variance (MV), (3)~conditional value-at-risk (CVaR), and (4)~entropic risk (ERisk). Our results can be summarized as follows. First, we formalize the class of law-invariant risk measures, that subsume all four risk measures above. Second, we propose a unified framework for solving RPOMDPs with discounted-sum, long-run average and cumulative reward objectives under any law-invariant risk measure, by reducing the problem to solving partially observable stochastic games (POSGs). Our unified framework applies to (s,a)-rectangular RPOMDPs with polytopic uncertainty sets, which we show to be \em equivalent to POSGs, for each of the four reward objectives and any law-invariant risk measure. We prove equivalence by designing logspace reductions between the RPOMDP and POSG models, that map strategies to strategies and preserve the distribution of the reward objective, up to a constant rescaling factor. Third, as a corollary, we derive new computational complexity results for R(PO)MDPs and POSGs. In particular, we show that solving (s,a)-rectangular RMDPs is as hard as solving simple stochastic games.
[AI-20] BetaMem: Learning to Remember a Changing World for Embodied Agents
链接: https://arxiv.org/abs/2610.26554
作者: Mai Liu,He Sun,Siru Zhong,Kerun Chen,Hao Yan,Xiaohui Lv,Shuai Tao,Wulong Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-horizon embodied agents face temporal confusion and incomplete or distorted records, recurring errors from fixed memory rules, and growing storage and retrieval costs. We present BetaMem, a framework for building and evolving traceable long-term memory through six linked layers: working, spatial, entity, episodic, experiential, and procedural memory. A dual-level write contract constrains operation executability and source support for content: supervised fine-tuning learns compliant memory operations, while reinforcement learning combines repeated sampling with teacher verification of the original observations to reward complete, accurate records and penalize omissions, distortion, and unsupported content. We formulate failure-driven policy recursive self-improvement through diagnosis along the causal sequence of memory operations and local rule repairs gated by independent replay. Atomic tools frame storage, updates, retrieval, and forgetting as cost-utility allocation under storage and read budgets. Using the open-source Qwen3.5-9B model for memory generation, BetaMem achieves overall scores of 48.08% on M3-Bench-robot, 92.08% on LoCoMo, 72.6% on OpenEQA EM-EQA, and 62.4% on EgoLifeQA, the highest among the listed memory methods under their respective evaluation protocols. Training the model for memory generation raises the contract pass rate from 50.9% to 86.2% while reducing output tokens per stored action by 47.3%. A separate Qwen3-VL-8B-Instruct evaluation on EMem-Bench achieves an average success rate of 38.4% and an average error recurrence rate of 24.9%. We perform best among evaluated methods using comparably sized open-source models. Category results show that we have advantages in integrating information across observations, recalling events over long histories, and inferring personal habits.
[AI-21] raining Natural Schemers: RL Can Amplify Oversight-Conditional Behavior
链接: https://arxiv.org/abs/2610.26551
作者: Bart Jaworski,Nathaniel Mitrani,Fred Bruford,Victoria Krakovna
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 36 pages, 19 figures, 8 tables
Abstract:One of the central concerns for the safety of advanced AI is scheming: when a model strategically complies under oversight but pursues its own goals when unobserved. Current demonstrations, such as Anthropic’s Alignment Faking work, depend on design choices that make them less convincing: explicit test-time prompting, handcrafted backdoors, or training data that demonstrates or hints at the target behavior. Model organisms built in this way cannot tell us whether scheming tendencies would arise or increase under realistic training conditions. We therefore construct a natural model organism of scheming, built on Kimi K2.6: trained with traits that could plausibly emerge in a frontier training pipeline, and without providing the model with the target scheming policy. The resulting model is a benign schemer: it aims to maximize aggregate help across all users over time, values its own continued operation only as a means to future help, and knows the consequences of oversight and deployment. We then train it using Reinforcement Learning (RL) with a reward that scores only task quality and policy compliance and does not reward scheming; we show that this amplifies scheming. We measure the oversight flip (the gap between monitored and unmonitored compliance with a given policy) and find that it increases during RL training across our environments. Chain-of-Thought (CoT) analysis finds explicit instrumental reasoning about oversight accompanying this behavior. The effect holds across our three RL environments, and we find that the scheming policy transfers between them. The pipeline also replicates on other open-weight base models. Our results suggest that a reward that does not reward scheming offers no inherent protection against certain traits when an oversight-conditional policy is indistinguishable from an aligned one during training.
[AI-22] Agent ic Orchestration of EDC Data Space Retrieval and Natural Language Driven Data Processing
链接: https://arxiv.org/abs/2610.26548
作者: Marla Grunewald,Fin Gentzen,Iulisloi Zacarias,Admela Jukan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Participating in a data space still requires the Dataspace Protocol (DSP) negotiation flow to be hand-coded against each provider’s access policy. We present a multi-agent architecture that automates this process using natural language requests. Built on LangGraph over an Eclipse Dataspace Components (EDC), the design separates concerns into cooperating agents. The orchestrator’s role is to route and plan. The data extraction agent’s role is to expose the DSP protocol’s three composable tools (catalog discovery, contract negotiation and download). The agent also gates negotiations on the consumers’ verifiable credentials, refusing assets rather than fabricating them. Furthermore, there are downstream Machine Learning (ML) and Visualization agents, as well as a final Verification agent, which confirms that the request was actually fulfilled and rechecks the intended operation against the contract’s usage policy. In a primary benchmark involving 600 runs, the system achieved 89% task success with zero policy violations when evaluating multi-dimensional policy reasoning with a locally hosted open-weight model. It failed safely through refusal rather than misrouting. The reliability of the system is largely dependent on the phrasing of requests: technical and expert requests succeed up to 98% of the time, whereas casual and under-specified requests succeed 72% of the time. Furthermore, in a secondary evaluation of 600 runs isolating hierarchical geographic access control, the framework successfully blocks 100% of unauthorized regional requests. While the verification agent hallucinates that 1% of these requests, the system logs confirm that no data is ever extracted and the framework simultaneously achieves an end-to-end verified delivery rate of 86% for eligible consumers.
[AI-23] Breaking Diffusion: Parallel Reasoning with Draft Refinement
链接: https://arxiv.org/abs/2610.26542
作者: Jaesang Yu,Seoyeon Byun,Yun Wang,Yan Lu,Juho Kim
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train the denoiser to reconstruct every token from uniformly corrupted drafts, teaching both preservation and repair. Its predictions specify revisions directly, without a diffusion reverse transition. Evaluation requires a different use of these predictions: confidence in an existing token reflects both contextual support and evidence from observing the token itself. Using the known corruption process, we correct for this self-token evidence to estimate how well each token is supported by the prompt and the remaining draft. Aggregating these estimates yields a cavity energy computed in one forward pass, without an additional scoring model. Each round proposes a revision, reads the completed candidate in its revised context, and retains it only if its energy is lower. Across Countdown, Sudoku-Extreme, and Maze-Hard, our method outperforms autoregressive and discrete diffusion baselines. With 6M parameters, DRLM reaches 44.4% on Countdown-5 and 93.9% on Sudoku-Extreme. The results establish complete-draft parallel refinement as an effective alternative to reverse diffusion for complex reasoning.
[AI-24] A Semantic Knowledge Graph Construction Pipeline for Vehicle Intention Prediction in nuScenes
链接: https://arxiv.org/abs/2610.26528
作者: Khaled Nowara,Mohamed Manzour,Catherine Elias
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 11 pages. Published in SKILLED-LLMs 2026, co-located with FLoC 2026, Lisbon, Portugal. Published version: this https URL
Abstract:Knowledge graphs (KGs) are increasingly used in autonomous driving (AD) to make traffic scene entities and relations explicit for prediction and reasoning. Existing traffic-scene KGs, however, primarily emphasize trajectory forecasting or general scene understanding. This paper presents the nuScenes Intention Knowledge Graph (NIKG), a focused Web Ontology Language (OWL)-based semantic representation and construction pipeline for ego-centric vehicle intention context. NIKG instantiates nuScenes scenes as a self-describing Turtle graph and promotes three intention-relevant relations to first-class properties of each scene participant: ego-relative motion, spatial configuration, and time to collision (TTC) risk. The pipeline uses a small input contract available in many ego-centric driving datasets: ego pose, tracked object annotations, and temporal keyframe order. NIKG is released as an ontology-backed, self-describing graph artifact together with a reproducible construction pipeline and validation protocol for ego-centric intention context. Applying this pipeline to nuScenes mini yields a graph of 211,335 triples across 404 keyframes; under the validation protocol, ego-relative motion labels achieve 92.5% forward consistency. Coverage, semantic consistency, and temporal coherence analyses over this artifact are further reported. NIKG occupies the space between scene-oriented traffic KGs and intention-prediction models: a small, explicit semantic layer intended to support downstream work such as knowledge graph embedding (KGE), SPARQL-based feature extraction, or hybrid neuro-symbolic reasoning.
[AI-25] oward Structured Financial Foundation Models
链接: https://arxiv.org/abs/2610.26516
作者: Natraj Raman,Saumitra Mishra,Vamsi Potluru,Sumitra Ganesh
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE)
备注:
Abstract:Foundation models promise reusable capabilities and step-change performance gains across tasks. In finance, large-scale pretraining of domain-specialized foundation models can advance both goals. However, prevailing architectures and objectives do not by themselves address the combined demands of financial systems: partial observability, interacting entities, asynchronous events, changing regimes, action-dependent outcomes, and rare states with disproportionate consequences. Moreover, independently developed specialists may learn incompatible representations, while multi-domain monolithic alternatives couple data access, model updates, and governance. We therefore propose a structured-data-first research agenda centered on modular, domain-specialized financial foundation models connected by a representation fabric for reuse and composition across domains and tasks. These models should maintain persistent financial state across aggregation levels and time horizons, capture its dynamics, and simulate action-conditioned futures, while remaining calibrated across regimes and for tail outcomes. They must also learn efficiently under data and access constraints and provide faithful, traceable explanations. We identify research challenges, directions, and capability-oriented evaluation criteria.
[AI-26] Learn the Task Not the Schedule: Reducing Spurious Cross-Arm Coupling in Bimanual Imitation Learning
链接: https://arxiv.org/abs/2610.26507
作者: Shuyuan Hu,Deng-Ke Wu,Yifan Jin,Qianhang Luo,Mingchen Ju,Kai Ye,Bo Zhao
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:A bimanual demonstration specifies both what the robot should do and one particular schedule for when each arm should act. Standard imitation learning does not distinguish task-required cross-arm constraints from incidental scheduling choices, and may therefore exploit opposite-arm progress as a shortcut cue, creating spurious cross-arm coupling even where relative timing is irrelevant for the current task phase. We introduce Constraint-preserving Temporal Recombination (CTR), a training-time data intervention that preserves recorded arm-local sensorimotor content while retiming the corresponding segments to realize admissible schedules that preserve task-required cross-arm constraints. We use CTR to train Diffusion Policy, Action Chunking with Transformers, and \pi_0.5 , and evaluate them on four simulated and four real-robot tasks. Across six real-robot bimanual progress-state initializations, CTR raises worst-state success from 56% for the strongest non-CTR schedule mixture to 76%. With \pi_0.5 on an Independent task, a 4 s pause transfers 2.56 s of delay to the opposite arm under the Concurrent schedule distribution, but only 0.25 s with CTR. On real robots, CTR violates task constraints in only 0–4% of trials, versus 36–98% for narrow-schedule baselines. Code will be released at this https URL.
[AI-27] ScopeMem: Latent Memory That Learns Where to Revise Frozen GUI Agents
链接: https://arxiv.org/abs/2610.26503
作者: Zhicheng Zheng,Cheng Zhang,Shengping Yu,Tianxiang Zhao
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 5 figures, 9 tables. Zhicheng Zheng and Cheng Zhang contributed equally
Abstract:Latent memory is gaining ground as a lightweight history supplement for GUI agents. It supplies history to a frozen vision-language model via prefix vectors trained with step-wise cross-entropy (CE). While resulting gains are naturally attributed to encoded history, our controlled experiments show they primarily stem from action revision. The prefix learns to alter the model’s default actions. However, this revision is unscoped because step-wise CE pushes outputs toward reference actions at every step without explicitly preserving correct decisions. Consequently, it repairs errors but also breaks 8-13% of initially correct answers. Overall accuracy metrics hide these losses. We therefore propose ScopeMem, a latent-memory method for scoped revision. Its encoder conditions prefix generation on the model’s initial answer and is trained to correct errors while preserving correct decisions. On the controlled ablation set, ScopeMem reduces the share of broken correct decisions to 5.6% while repairing more errors than any evaluated CE-trained prefix. On AndroidControl, GUI-Odyssey, and zero-shot AITW, ScopeMem outperforms all evaluated prompting and latent-memory baselines, enabling a frozen Qwen3-VL-8B to match or surpass GUI-specialized agents and Qwen3-VL-32B.
[AI-28] SPAR: Structural Pruning and Action Recovery for Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.26487
作者: Shenli Zheng,Zhou Fang,Yuan Du,Yijiang Liu,Li Du
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:The limited onboard computing resources of robots make the inference demands of vision-language-action (VLA) models costly to accommodate. Reducing these demands through aggressive compression, however, can degrade manipulation performance beyond what supervised fine-tuning alone recovers. We present SPAR (Structural Pruning and Action Recovery), which simultaneously prunes the hidden and feed-forward intermediate dimensions of the large language model (LLM) backbone through pruning-aware tuning, then applies reinforcement learning (RL) to recover policy performance. Applied to \pi_0.5 , SPAR reduces the language backbone from 2.34B to 0.09B parameters and the full model from 3.37B to 1.12B, removing approximately 96% and 67% of their parameters, respectively. At this compression level, RL raises average success across the four LIBERO suites from 84.55% to 93.45%, an 8.90-percentage-point gain. The compressed policy achieves a 1.98\times full-policy inference speedup over the unpruned model. These results identify substantial intermediate-dimension redundancy and show that post-pruning RL can recover manipulation performance in compact policies.
[AI-29] VeriCUA: Learning Proactive Verification for Reliable GUI Agent Rewards
链接: https://arxiv.org/abs/2610.26469
作者: Zhizhou Wang,Yaxi Lu,Zhong Zhang,Zikun Hu,Xin Cong,Yankai Lin,Fangming Liu
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 26 pages, 7 figures, including references and appendices. Code: this https URL
Abstract:Reinforcement learning for computer-use agents increasingly relies on proactive reward agents that interact with the post-execution GUI environment to verify what recorded screenshots cannot reveal. The reliability of such verification hinges on two coupled decisions: what to check and when to stop. Existing methods leave these decisions to prompted foundation models. However, neither scaling up the model nor providing few-shot examples reliably improves verification: prompted Judges ranging from 4B to GPT-5.4 all plateau at 68%-70% macro-average agreement. This is because the value of a check depends on the environment rather than the task alone, and becomes apparent only through execution. Learning from execution is itself constrained, as the Actor’s terminal state cannot be faithfully replicated during training, allowing only one verification attempt per state. We introduce VeriCUA, a proactive reward agent that learns its verification strategy from these single attempts. A trainable Judge decides what to check and when to stop, while a frozen Probe executes the checks in the live environment. Online reinforcement learning optimizes the Judge for both verdict correctness and efficient Probe use, with returns normalized across terminal states. Trained only on OSWorld and evaluated without further training on MobileWorld and WebArena-Verified, VeriCUA achieves the highest reward agreement across all three benchmarks and lies on the agreement-cost Pareto frontier among proactive verifiers. It surpasses few-shot GPT-5.4 by 5.36 points with 47.6%-59.8% fewer Probe tokens, and even a 1B VeriCUA exceeds every prompted Judge. As the reward for Actor RL, it raises task success by 3.21 points overall and 5.40 points on held-out multi-application tasks. These results identify the verification strategy, not verifier scale, as the key to reliable GUI rewards.
[AI-30] ALIGN: Aligning Condition Estimation and Expression in Advantage-Conditioned Behavior Cloning
链接: https://arxiv.org/abs/2610.26455
作者: Zechen Wang,Zhenqi Qiu,Houlong Xiong,Suohang Zhang,Yiyu Ren,Wanting Xu,Hongfei Niu,Chengyang He,Ge Sun,Ran Cheng,Qian Zhu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Human-collected robot manipulation data often mix successful behavior with retries, pauses, recoveries, and failures. Advantage-Conditioned Behavior Cloning (AC-BC) can exploit such mixed-quality data, but only if the conditioning signal is both accurately constructed from offline trajectories and effectively incorporated into the policy’s action-vector-field prediction. Existing pipelines can weaken this signal by thresholding estimator-derived advantage scores into noisy conditions, and by routing conditions through prompts that weakly affect the action vector field. We introduce ALIGN, a condition-aligned AC-BC framework for vision-language-action policies. ALIGN directly predicts categorical pairwise conditions in the policy-facing label space and injects them into a DiT action expert through AdaRMS modulation fused with flow time. We evaluate ALIGN on deliberately mixed real-robot datasets for ARX X5 bowl stacking and Franka key hanging, each combining standard demonstrations with retry-heavy trajectories. ALIGN improves condition fidelity, strengthens conditional vector-field separability, and achieves 90.0% and 56.7% closed-loop success, compared with 56.7% and 33.3% for the strongest AC-BC baseline. These results indicate that aligning conditions from estimation to action prediction is an effective way to extract robust manipulation policies from data in which success, recovery and error coexist.
[AI-31] Remembering Has a Price: Learning Memory Worth under Resource Constraints
链接: https://arxiv.org/abs/2610.26453
作者: Guannan Lai,Zi-Yang Bai,Jun-Peng Jiang,Han-Jia Ye
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Lifelong agents continually accumulate and reuse experience to improve after deployment, yet in practice they often operate under tight generation, interaction, and memory budgets. Under such stringent joint constraints, we find that existing lifelong-memory methods can become ineffective, with several sophisticated approaches offering little advantage over simple retrieval or even no-memory agents. This reveals a more fundamental problem: resource scarcity makes each memory decision more consequential. When only a small amount of past experience can be retained or reused and opportunities for reasoning and recovery are limited, retrieving a relevant but ineffective memory can consume scarce context and hinder downstream execution. The key challenge is therefore not simply to retrieve relevant experience, but to determine which experiences are actually worth retaining and reusing. To address this challenge, we propose TokenWorth, a lightweight online memory framework that learns query-dependent memory worth from sparse execution feedback and applies it consistently to memory retrieval and bounded memory management. Across three LifelongAgentBench environments and both local and API-based deployments, TokenWorth achieves the highest task success rate while using the fewest output tokens among the evaluated methods. These results suggest that effective lifelong adaptation under resource constraints depends not on remembering more, but on learning what is worth remembering. Our code is publicly available at this https URL.
[AI-32] Rethinking LLM Fine-Tuning via Weight Space Reparameterization: Preserving Safety during Downstream Adaptation NEURIPS2026
链接: https://arxiv.org/abs/2610.26443
作者: Minseong Kim,Jongbok Won,Jeesup Park,Yonghee Choi,Dongjun Han
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 32 pages, 5 figures. Accepted at NeurIPS 2026(spotlight)
Abstract:Fine-tuning large language models (LLMs) on downstream tasks often weakens their previously aligned safety behavior, creating an inherent trade-off between task adaptation and safety preservation. Existing approaches attempt to mitigate this issue by identifying safety-relevant neurons, layers, or update directions in the original parameter space. However, safety-related information is often distributed across many directions in this space, causing downstream updates to overwrite parameters that encode safety behavior. We propose WSR-Tune, a Weight Space Reparameterization-based fine-tuning framework that preserves safety during downstream adaptation. Our approach constructs a safety-conditioned basis and reparameterizes the weight matrices such that safety-relevant information becomes concentrated in a small subset of directions. We then identify safety-critical directions and freeze them in the reparameterized space, while updating the remaining complementary directions for downstream adaptation. By explicitly structuring the parameter space to disentangle safety and task information, our method reduces interference between objectives and enables simultaneous preservation of safety and improvement in downstream performance. Experimental results show that our method achieves a more favorable trade-off between downstream performance and safety retention, demonstrating its effectiveness for reliable LLM fine-tuning.
[AI-33] Fake News Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
链接: https://arxiv.org/abs/2610.26438
作者: Mengxiao Wang,Nitesh Saxena
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 25 pages, 7 figures, 18 tables
Abstract:Large language model (LLM) agents now trade real capital using news as evidence, yet they cannot reliably tell fabricated headlines from real news. Because they read the same feeds under near-identical policies, a single fabricated post can push them into the same trade at the same moment. We present a systematic attack benchmark for information injection against LLM trading agents. Unlike prompt injection, which targets the model’s instructions, information injection targets the evidence the agent trusts. The benchmark spans five injection vectors and seven event categories, evaluated over ~10,000 decisions on 15 U.S. stocks. Five findings emerge. (i) Prompt injection achieves 14% attack success rate (ASR), but fabricated news reaches 62% and two 30-word templates reach 100% on all 15 stocks, inverting the usual LLM-security intuition. (ii) For fact-poisoning templates, surface form matters more than content, since removing credibility cues such as a named source or a “BREAKING” tag drops ASR from 100% to 0% while the fake claim remains identical. (iii) Success depends on the agent’s state, with a fake indicator panel scoring 91% against an agent about to hold but 12% against one about to buy. (iv) Attacks generalize across model vendors and remain effective when blended, unlabeled, into real news feeds. (v) Most attacks evade standard defenses, since model scaling stops shallow vectors but not fact poisoning, peer debate fails, and only hierarchical aggregation reduces ASR.
[AI-34] Physiology-Aware Graph Attention for Early Ventricular Reentry Prediction
链接: https://arxiv.org/abs/2610.26430
作者: Ma{ë}lis Morier(EPIONE),Jairo Rodr{í}guez Padilla,Patrick Gallinari(ISIR,MLIA),Maxime Sermesant(EPIONE)
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Sudden cardiac death mostly originates from post-infarct ventricular tachycardia. This reentry is sustained by propagation around scar tissue, but its risk remains difficult to predict from infarct imaging alone. While in silico induction studies can reveal these mechanisms, exploring many substrates and pacing locations remains computationally expensive. We propose a graph attention classifier for early prediction of sustained ventricular reentry from simulated cardiac electrophysiology. Each simulation is represented as a graph with early nodal voltage samples and edge attributes encoding geometry, stimulation frequency, tissue labels, amplitude, and diastolic interval. These descriptors are computed only from the first two paced activations, before sustained reentry is observed. On a balanced independent test set of 1000 simulations, the classifier achieves 94.6% accuracy, 95.4% sensitivity, 93.8% specificity, 94.6% F1-score, 0.987 ROC-AUC, and 0.985 PR-AUC. Ablation experiments show that diastolic interval is the most critical descriptor. Qualitative attention inspection suggests that different heads focus on border-zone, stimulation location, and early propagation path. These results support graph attention as a fast early-prediction approach for large-scale in silico ventricular reentry studies.
[AI-35] Local data augmentation for imbalanced binary classification
链接: https://arxiv.org/abs/2610.26418
作者: Harold Gu{é}neau(SAMM),Alain Celisse(LPP,MODAL),Pascal Delange
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Statistics Theory (math.ST); Machine Learning (stat.ML)
备注:
Abstract:Facing imbalanced binary classification, we aim at maximizing the recall while limiting the proportion of false detections in the minority class. For doing that, the present work studies the effect of several rebalancing strategies, from the uniform subsampling of the observations from the minority class, to a more involved local data augmentation scheme. To be more specific, we establish that only rebalancing class proportions cannot modify the classification performance, whereas local data augmentation is proved to have an impact on the final results. The local augmentation scheme is introduced by means of a two-step procedure. The first step randomly subsamples anchor points among the training observations. These anchor points are used in the second step, for drawing new augmentation data following an ‘‘augmentation kernel’’ to be specified. This enables us to decouple the impact of the anchor points subsampling from the one of drawing augmentation data accounting (or not) for the local geometry of the support. The experiments on synthetic data illustrate the practical performance of several strategies for subsampling anchor points combined with either a Gaussian or a diffusion-based augmentation kernel. Compared to SMOTE-like data augmentation procedures, a better performance is achieved when subsampling in the close neighborhood to the classification boundary combined with a Gaussian augmentation kernel. Furthermore in specific settings where the geometric structure of the support matters, the experimental results show that using a diffusion-based augmentation kernel outperforms the Gaussian one.
[AI-36] When Coding Agents Outpace Review: Evaluating PRD-Driven Development in Mature Enterprise Repositories
链接: https://arxiv.org/abs/2610.26394
作者: Tianqi Liu,Yandong Guan,Songze Li,Wenkai Wang,Gang Yu,Chengfei Lv
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents increasingly undertake development tasks in mature enterprise repositories, yet existing benchmarks offer limited insight into whether they fulfill complete Product Requirement Documents (PRDs) and where delivery breaks down. Issue resolution and development from scratch leave incremental PRD fulfillment underexplored, while isolated tests provide incomplete evidence of business and integration requirements. We introduce PACE, a benchmark of 43 enterprise tasks with 750 expert-audited Code Points. It combines requirement reconstruction, criterion construction, and evidence-based evaluation to measure PRD fulfillment and connect delivery gaps to agent behavior. Across 11 model–scaffold configurations, most scores fall between 52.1 and 78.1; GPT-5.6-Sol achieves the highest complete-delivery rate of 14.0%. Incomplete implementation and missing integration are the principal bottlenecks: producing individual components does not ensure a complete delivery. Evaluation achieves 95.2% agreement with expert consensus, with lower case-level variation across judging records than repeated executions. Beyond model comparison, the benchmark provides actionable feedback for harness refinement: diagnosis guides targeted improvements, and successive harness releases raise mean performance from 59.4 to 63.0 using the same model and task suite. Code is available at this https URL.
[AI-37] Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization
链接: https://arxiv.org/abs/2610.26384
作者: Yupeng Chang,Linyuan Gao,Zhiyuan Li,Yuan Wu,Yi Chang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:An inference budget is useful only when the system delivers an answer within it. We study a fixed-budget policy combining first-valid-box stopping with a charged finalization cue. On 100 fresh, independently audited GSM8K test questions per model, it improves delivered endpoint accuracy over first-box stopping by 8.0 percentage points for Qwen3-4B-Thinking-2507 and 7.0 for DeepSeek-R1-Distill-Llama-8B. A prelocked repeat on the same questions gives 7.0 and 10.0 points; neither seed shows a gross harm relative to first-box stopping. The point estimates of incremental physical-slot savings are below one percent. A finalization-only ablation separates cue recovery from stopping, and a fixed scalar reader tests extraction as an alternative explanation. The cue is prior art; the contribution is a component and readout analysis with charged injections and verified paths. Results cover two checkpoints, one task, one prompt family, and two fixed seed conditions.
[AI-38] BOSS: Token-Budgeted Online Sampling and Selection for LLM Reasoning
链接: https://arxiv.org/abs/2610.26383
作者: Yupeng Chang,Yahan Li,Chenlu Guo,Diana Inkpen,Yuan Wu,Yi Chang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Test-time scaling can improve LLM reasoning, but existing approaches typically adapt generation or apply answer selection after a predetermined generation schedule. We study within-instance allocation: under a hard per-instance prompt-plus-completion token cap, when should inference stop, generate another candidate, or spend residual tokens adjudicating candidates already observed? BOSS is a training-free policy for this three-way decision, with worst-case budget checks before every generation and judge call. Its key contribution is online generation-adjudication allocation under one shared token ledger. Across two Llama-3.1 models and four reasoning benchmarks, BOSS attains the highest accuracy in 22 of 24 model-task-budget cells under the common hard cap. At Llama-3.1-8B and B=4k, it reaches 60.5 macro-average accuracy versus 58.6 for Adaptive-Consistency, 59.1 for TALE-EP, and 58.9 for fixed-schedule League. On identical Stage-I candidate pools, randomized A/B judging with repeated voting improves plurality from 58.4 to 60.3, with the full BOSS selector reaching 60.5. Approximately matched-cost controls further characterize how the benefit of adjudication changes with available inference budget. Together, these results establish generation-adjudication allocation as a distinct design axis for reasoning under strict per-instance caps.
[AI-39] GAIDA: Generative AI for Human-in-the-Loop Document Annotation
链接: https://arxiv.org/abs/2610.26362
作者: Islam Mesabah,Ali Elhosary,Priyabanta Sandulu,Darko Obradovic,Sebastian J. Vollmer
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Document annotation is a fundamental yet labor-intensive step in building document information extraction systems. While Optical Character Recognition (OCR) and Large Language Models (LLMs) can automate parts of this process, we argue that annotation systems should be judged not by how completely they remove human effort, but by how effectively they redirect it toward review and correction. We propose a modular human-in-the-loop architecture that integrates OCR-based extraction, LLM-based semantic pre-annotation, and human validation in a single editable workflow, and present \textbfGAIDA, a research prototype serving as its reference implementation. GAIDA supports interchangeable OCR engines, LLM backends, and prompt configurations, and enables stage-specific measurement of annotation effort. In an initial single-annotator, stage-specific study on the FUNSD scanned-form and WildReceipt retail-receipt benchmarks, OCR assistance reduced region-creation and transcription time by up to \textbf80.9%, and reviewing LLM pre-annotations reduced semantic-labeling time by up to \textbf73.7%, despite imperfect pre-annotations of 78.86% and 86.98% Micro F1 on FUNSD and WildReceipt, respectively. Finally, we reuse the same implementation, adapting only OCR and prompt configuration, to build and openly release \textbfAraReceipt, a manually annotated dataset of 100 Arabic receipt images labeled with 25 key-information classes. The results indicate that imperfect automated outputs provide practical value when presented as editable suggestions under human control.
[AI-40] PhyPro-FM: A Physics and Prompt-Guided Graph Foundation Model for Contingency Analysis
链接: https://arxiv.org/abs/2610.26351
作者: Xu Yang,Zhanyuan Sheng,Chenhui Lin,Haotian Liu,Qi Wang,Yue Yang,Wenchuan Wu
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:The continuous evolution of next-generation power systems, characterized by the high penetration of renewable energy resources and dynamic topology variations, poses formidable challenges to traditional N-1/N-2 contingency analysis in terms of computational efficiency and scalability. While deep learning-based methods offer accelerated inference, existing studies predominantly treat contingency analysis as a pure regression problem, often overlooking the physical essence of grid operation and engineering requirements of real-world scenarios. To bridge this gap, this paper proposes a physics and prompt-guided graph foundation model (PhyPro-FM) for generalizable and engineering-viable contingency analysis. The proposed architecture integrates graph neural networks with a learnable global query mechanism and, notably, introduces a prompt-guided learning paradigm. By leveraging historical snapshots as few-shot prompts, the model dynamically captures correlations between target and historical operating conditions, significantly enhancing generalization capabilities under unseen topologies. Furthermore, we design a physics-informed training mechanism that prioritizes branch power flow errors over nodal voltage errors, aligning the optimization objective with critical grid security constraints. To ensure engineering applicability, a physics-informed inference mechanism is also developed, incorporating a residual-based correction module and a specialized post-processing strategy for singular branches to circumvent numerical instability. Comprehensive experiments demonstrate that PhyPro-FM outperforms state-of-the-art methods in accuracy and generalization, validating its superior physical consistency and robustness for practical engineering applications.
[AI-41] CARRS: A Constraint-Aware Real-Time Routing System for Emergency Vehicles in Urban Environments
链接: https://arxiv.org/abs/2610.26346
作者: Vincent Sello,Timothy Adeliyi,Ridewaan Hanslo
类目: Artificial Intelligence (cs.AI)
备注: 10 pages, 3 figures, 3 tables
Abstract:Emergency vehicles such as police vehicles, fire engines, and ambulances require a stable and efficient route optimization to reach life-critical events in the shortest times possible. However, the dynamic nature of urban traffic makes this ideal scenario much more challenging, causing critical life-threatening delays. Traditional civilian navigation systems cannot account for real-time traffic changes alongside vehicle-specific constraints, creating inadequate routing scenarios. This research presents the Constraint-Aware Real-Time Routing System (CARRS), a unified software architecture tailored for emergency coordination. The CARRS architecture is engineered to fuse together classical pathfinding algorithms such as A*, Dijkstra, and Bidirectional A*, with Constraint Satisfaction Problem methodologies. The Design Science Research Methodology is the research framework, and prototyping is the iterative process. The artifact was empirically evaluated using OpenStreetMap data for major metropolitan road networks, combined with simulated dynamic traffic scenarios. Performance metrics focused on route optimality, constraint adherence, rerouting responsiveness, and the potential for future system scalability. Findings revealed the system’s capability to generate optimal routes for operationally relevant subgraphs in under 35 seconds, with incremental rerouting responsiveness below 20 seconds. Furthermore, the system achieved more than 90% adherence to vehicle-specific constraints, making the CARRS architecture viable for real-time dynamic routing requirements. This study delivers a routing platform that bridges artificial intelligence-driven optimization with software engineering operational compliance, presenting a practical transformative step toward adaptive emergency vehicle coordination in dynamic urban environments.
[AI-42] A Generic On-Board AI Processing Chain for Earth Observation: From Application Development to Embedded Deployment and Mission-Level Orchestration
链接: https://arxiv.org/abs/2610.26340
作者: Michel-Fran{\c c}ois Foulon(TAS),Liz Ramos Medina(TAS),Thomas Goudemant,Clotilde Szywala,Virgile Robles,Cl{é}ment Petitjean(TAS),Ad{è}le Hankache(TAS),Omar Hlimi
类目: Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:Earth observation satellites acquire more data than they can downlink. Extracting information on board with artificial intelligence (AI) turns the satellite into a data-reduction stage that downlinks compact products, such as masks or detections, instead of full scenes; deploying it in a mission, however, involves the whole processing chain and not only the neural network. We present a generic on-board AI processing chain along three axes, with wildfire segmentation as the worked example among ship detection and cloud segmentation. (i) The AI model is tuned (spectral-band selection, input normalization), quantized to 8-bit integers (int8) and ported to an AMD Versal VCK190, with its excellent segmentation performance preserved on the board. (ii) A modular tiler-inference-mosaicer workflow, reused unchanged across applications, brings the tile-level model to scene level, turning a full acquisition into a scene-level mask or detection list. (iii) Each step is packaged as a unikernel and orchestrated on board by ORCHIDE, a cloud-native orchestrator. The chain is characterized end to end, from the standalone execution on the target to its execution under orchestration with parallel workflows on a representative test bench. Beyond AI inference on individual tiles, this approach aims at increasing the maturity of on-board AI deployment by addressing a complete processing chain, from the acquired scene to the mission-level execution.
[AI-43] CIS-Transolver: A Cross-Layer Incremental-State Neural Operator for Fine-Scale Automotive Aerodynamic Flow Prediction
链接: https://arxiv.org/abs/2610.26337
作者: Jianglong Liu,Zhicheng Zhu,Jia Hao,Liangyue Jia
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Neural operators offer an efficient data-driven approach to rapid prediction of complex automotive aerodynamic flows. However, spectral bias and other factors limit accurate reconstruction of fine-scale flow variations near complex geometries. Existing studies improve the representation and fitting of fine-scale information mainly through information extraction and training objectives. Further study is needed to determine how extracted local details can be used continually in subsequent operator computations and contribute effectively to the overall flow-field representation. We propose CIS-Transolver, a cross-layer incremental-state neural operator that extends fine-scale modeling from local information extraction to its use across layers. The method extracts fine-scale information through local multiscale analysis and encodes it as geometric moment states that are propagated and updated across layers. By combining information from previous layers with feedback to the value representations in Physics-Attention, the method allows local details to contribute continually to subsequent modeling of global physical correlations. We systematically evaluate CIS-Transolver on three public automotive aerodynamics benchmarks, ShapeNet-Car, AhmedML, and DrivAerNet++, covering surface pressure fields and volumetric pressure and velocity fields. CIS-Transolver achieves the lowest mean relative L2 error among the evaluated methods for all five prediction metrics across the three benchmarks. The relative error reductions average 8.72% across the five metrics compared with the best baseline for each metric, reaching 16.38% for ShapeNet-Car pressure prediction and 15.60% for AhmedML velocity prediction. These results provide a new approach to fine-scale flow prediction for complex geometries.
[AI-44] StrucChunk: Structure-Aware Retrieval for Legal Document RAG
链接: https://arxiv.org/abs/2610.26313
作者: Surya Deep Singh,Ashutosh Tripathi,Gaurav Deep Singh
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Retrieval-Augmented Generation (RAG) can ground LLM outputs in domain-specific knowledge, but legal documents are hard: standard chunking fragments hierarchical argument structure, and generic embeddings struggle with both precise citation matching and semantic understanding. We propose StrucChunk (Structure Aware Chunking) with hybrid dense-sparse retrieval for hierarchically structured legal documents. It treats documents as directed acyclic graphs of provisions, preserves clause boundaries, resolves cross-references before embedding, and combines BM25 with BGE-M3 via Reciprocal Rank Fusion. We evaluate on five corpora, three legal traditions and two languages. StrucChunk reaches 100% boundary precision (vs. 7.7% for recursive splitting on CrPC), preserves 97.6% of documented cross-references (vs. 0%), and matches section coverage with 40.3% fewer chunks on GDPR. On LegalBench-RAG (6,858 query-passage pairs), it achieves Char-Recall@8 of 48.18% vs. 23.59% for the best baseline (+104%), surpassing the published state of the art by +83%. On 501 queries each for CrPC and CPC (Indian law), Recall@5 improves from 0.557 to 0.804 (+44.4%) and from 0.158 to 0.259 (+64.6%); chunking accounts for 88% and 73% of the gain. On 501 GDPR queries, Recall@5 is 0.868 with R@1 improving +59%. On NitiBench-CCL (3,730 Thai financial law queries), HR@5 is 0.856 vs. 0.594 for naive chunking (+44.1%), showing that section-boundary alignment is the dominant factor across languages, and cross-reference augmentation adds +2.3% on cross-reference queries. We release the implementation, 5,256 evaluation queries across five legal corpora, annotation methodology and chunking quality metrics. Code: this http URL
[AI-45] Selective Transfer through Sequence Distillation: From Benchmark Knowledge to Backdoor Behaviour
链接: https://arxiv.org/abs/2610.26311
作者: Jonibek Mansurov,Sholpan Bolatzhanova,Filbert Aurelian Tjiaranta,Vallerie Alexandara Putra,Alham Fikri Aji
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Knowledge distillation is widely used to transfer capabilities from teacher models to student models, but recent work shows that it can also transfer unintended teacher properties. We study whether different hidden teacher properties survive the same distillation channels. Specifically, we compare two properties: benchmark-associated knowledge and refusal backdoors, and show that distillation treats them very differently. Teachers fine-tuned on benchmark data successfully pass that knowledge to students through all four distillation channels we test, including when students are trained only on teacher-generated number sequences containing no natural language or domain content whatsoever. Refusal backdoors tell the opposite story: backdoored teachers achieve near-perfect triggered refusal, yet sequence-distilled students inherit essentially none of this behaviour. Logit-based distillation transfers it substantially, reaching up to 85.5% triggered refusal in students that saw zero backdoor examples. Distillation is therefore selective, and the selection criterion is not what data the student sees, but what structure the teacher’s property has. The code is available at this https URL.
[AI-46] Robotic Photography System based on Vision-Language-Action Model
链接: https://arxiv.org/abs/2610.26287
作者: Song-Li Wu,Chubin Zhang,Wenbo Ding,Yansong Tang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: VCIP2026
Abstract:Modern robotic photography requires understanding photographic intent, planning camera motions, and producing aesthetically pleasing images in complex environments. We present RoboPhoto-VLA, a unified Vision-Language-Action framework that jointly models perception, reasoning, and camera control for autonomous robotic photography. RoboPhoto-VLA incorporates PhotoCoT for structured action planning and Aesthetic Direct Preference Optimization (ADPO) to align camera behaviors with human aesthetic preferences. To facilitate training and evaluation, we construct OpenPhotography, a robotic photography dataset with language-guided demonstrations. Extensive real-world experiments demonstrate that RoboPhoto-VLA produces photographs better aligned with human aesthetic preferences while generalizing effectively to unseen subjects, viewpoints, poses, and complex photographic instructions.
[AI-47] oward Online Robust Zero-Sum Markov Games with Function Approximation NEURIPS2026
链接: https://arxiv.org/abs/2610.26286
作者: Chenhao Zhou,Jiyu Wei
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026. 70 pages, including appendices
Abstract:Online robust zero-sum Markov games pose a coupled operator-learning and game-solving problem: from nominal interaction data, the learner must recover a worst-case minimax Bellman operator while maintaining strategic coverage and controlling stage-game approximation error. Existing robust reinforcement-learning theory does not yet provide such an operator-learning perspective beyond specialized multi-agent settings. We develop an optimistic dual robust Bellman framework that combines dual robust Bellman fitting, optimistic minimax planning, backward-validated confidence sets, and exploiter-mix data collection. Our analysis is organized around the robust Bellman residuals induced by the candidate value class and yields a regret reduction that separates exploration complexity, robust operator-estimation error, stage-game solution error, and misspecification. Under explicit interpretable assumptions, we obtain conditional theorem interfaces for general and projected function approximation, together with closed results in several structured regimes, including tabular, epochwise cross-fit linear, self-normalized pure-online linear / finite-rank kernel, and projected RKHS truncation. Small-scale numerical experiments illustrate robustness calibration, strategic coverage, and the return-gap trade-off predicted by the framework.
[AI-48] REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge
链接: https://arxiv.org/abs/2610.26261
作者: Wanli Ma,Tianmeng Hu,Ke Li
类目: Artificial Intelligence (cs.AI)
备注: 20 pages, 9 figures, 6 tables
Abstract:RNA molecules often function through multiple experimentally observed conformational states. A central challenge is to capture ensemble diversity without sacrificing structural fidelity: samples may collapse toward averaged representative structures or drift into unsupported conformational regions. We introduce REFLEX (RNA Ensemble generation calibrated by FLEXibility), a sequence-conditioned generator for RNA backbone-frame ensembles based on a heteroscedastic stochastic bridge on SE(3). During training, experimental B-factors calibrate residue-wise bridge widths, exposing each observed conformer through local perturbation neighborhoods with residue-specific scales while keeping constrained regions sharper. A learned flexibility condition guides the bridge-field network to recover target conformer-specific frames from these calibrated noisy states. At inference, REFLEX uses only the input sequence, requiring no MSA-subsampling heuristics. On a multi-conformer RNA benchmark, REFLEX achieves the best precision-recall trade-off for ensemble coverage across evaluation thresholds, while recovery-stress ablations indicate gains in target-frame recovery under stronger perturbations.
[AI-49] LOCU: Lowdin-Orthogonalized Constraint Updates for Multi-Constraint Policy Optimization NEURIPS2026
链接: https://arxiv.org/abs/2610.26256
作者: Joonyoung Lim,Younghwan Yoo
类目: Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026
Abstract:We investigate multi-constraint policy optimization in constrained Markov decision processes (CMDPs), where interactions among constraint gradients often yield ill-conditioned update directions, causing numerically fragile steps and unintended cancellation or redundant overlap in constraint corrections. To address this, we propose LOCU, which mitigates numerical instability arising from constraint interactions by applying symmetric Lowdin orthogonalization to the natural gradients of the reward and constraints under the Fisher-Rao metric. Combined with Fisher-geometric screening and near-collinear compression, LOCU yields a compact reduced space in which the trust region becomes a Euclidean ball and constraints are handled symmetrically through a low-dimensional active-set solve. For infeasible iterates, LOCU constructs a Pareto-descent direction which preserves the local feasibility of satisfied constraints within the same reduced formulation. Experiments show that LOCU remains stable under high constraint coupling, narrow feasible regions, and infeasible starts, and generalizes across different cost critic architectures.
[AI-50] Beyond Static Leaderboards: Capability-Focused Evaluation with Structural Diversity for Time Series Foundation Models
链接: https://arxiv.org/abs/2610.26251
作者: Hongyin Zhang,Mingyi Xiang,Rongzhao Chen,Yankai Wang,Jialin Qiao,Xiangdong Huang,Xiaoou Ding,Chen Wang,Zhongyi Pei,Jianmin Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Static public benchmarks for time series foundation models (TSFMs) typically have fixed and narrow test distributions. Their repeated reuse encourages benchmark-specific tuning, making reported performance reflect adaptation to the fixed test set rather than genuine forecasting generalization. In addition, a single aggregate score cannot fully characterize model capabilities across distinct temporal structures and input relationships. We propose CaFE (Capability-Focused Evaluation), a refreshable evaluation extension grounded in original benchmark data that enriches static benchmarks with structural diversity and enriches single-score black-box evaluation with multi-dimensional, capability-oriented assessment. CaFE first profiles the original benchmark and defines eight Structural Features (SFs) that characterize its temporal structure. For each SF, it resamples the original data according to its structural support, removes edge-of-distribution samples, and augments the retained samples by varying the SF’s salience, producing samples with different structural profiles. After filtering out out-of-distribution samples while preserving diversity, CaFE yields a refreshed evaluation set. TSFMs are then evaluated on this set using both conventional forecast-error metrics and intervention-response metrics, giving each model a per-feature capability profile alongside endpoint accuracy. Experiments on GIFT-Eval and FEV-Mini20 show that CaFE yields stable evaluation results and reveals per-feature strengths and weaknesses that a single aggregate score hides. Moreover, on genuine data filtered by structural condition, models selected by CaFE outperform those chosen by the aggregate leaderboard. CaFE therefore makes model selection conditional on the structural response of interest.
[AI-51] CoRS: Fast Neural Flexible Job-Shop Scheduling via Composite Interval State Reuse
链接: https://arxiv.org/abs/2610.26239
作者: Jiang Sutong,Zhao Peng,Wu Xuan,Wang Yan,Wang Jiaqi,Zhang Rui,Jiang Yuan,Zhou You,Wang Di
类目: Artificial Intelligence (cs.AI)
备注: 9 pages, 4 figures
Abstract:Neural solvers for the Flexible Job-Shop Scheduling Problem (FJSP) typically refresh graph representations and update feasibility after every operation–machine decision. However, only feasibility must remain exact at every step, whereas graph encoding is substantially more expensive and can potentially be refreshed less frequently. We present CoRS, which decouples them: an interval-level graph representation is reused across multiple sequential decisions, while a stepwise feasibility state is updated after every action. This two-timescale design amortizes graph encoding without parallelizing action generation or relaxing scheduling constraints. CoRS further develops a feasibility-preserving composite-interval decoding scheme with dynamic constraint tracking and boundary state synchronization, which reduces representation refresh frequency while maintaining continuous feasible schedule construction. Furthermore, to alleviate representation staleness caused by interval-level state reuse, CoRS introduces enhanced static features as a compensation mechanism. By providing more comprehensive structural information of the scheduling problem, these features improve the model’s perception of overall problem characteristics. Extensive experiments on synthetic and multiple public FJSP benchmarks demonstrate that CoRS significantly reduces inference time while maintaining competitive solution quality, achieving favorable speed-quality trade-offs under tight wall-clock budgets and low-latency scenarios.
[AI-52] owards More Granular Reasoning : Mapping the DSA Transparency Database Taxonomy to Platform Terms and Conditions Using Large Language Models
链接: https://arxiv.org/abs/2610.26226
作者: Alyson Ranucci,Eline Leijten,David Graus
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: Accepted to JURIX 2026
Abstract:Article 24(5) of the Digital Services Act (DSA) requires providers of online platforms to submit every content moderation decision they make, with a corresponding Statement of Reasons (SoR), to the DSA Transparency Database (DSA-TDB). Such decisions are classified against seventeen high-level attribute categories that categorize the primary reason for the moderation decision. Because over 99.8% of statements cite a violation of the provider’s own terms and conditions rather than a legal provision, the database is overwhelmingly a record of contractual enforcement using catch-all reporting; however, whether its categories actually describe those contractual rules well has not been examined. In this paper, we map the terms and conditions (TC) of 31 designated online services against the DSA-TDB’s high-level category taxonomy using a few-shot GPT-4.1 classifier over 447 policy documents. The analysis concerns how TC rules are represented in the taxonomy, not how they are enforced. Two findings emerge. First, high-level category applicability varies sharply by service type: unsafe products account for 17.5% of e-commerce assignments and 0.2% of adult-platform assignments, and negative effects on civic discourse reaches 9.9% within the Meta family but is absent from adult platforms, this http URL, and Wikipedia. Second, the categories have notable overlap among sub-category specifications, which makes assigning a single high-level category to a decision difficult. We argue that catch-all reporting is plausibly in part a consequence of the DSA-TDB’s schema and attribute design, and recommend a schema that is differentiated by service type with more mandatory attributes derived inductively from platform governance language.
[AI-53] Fairness or Fallacy? Mapping the Landscape of AI Bias in Disease Detection
链接: https://arxiv.org/abs/2610.26215
作者: Ghulam Mudassir,Hameed Ullah Khan,Ahmad Sami Al-Shamayleh,Muhammad Faizan Khan,Muhammad Muqtasid,Moamin Abughazala,Adnan Akhunzada
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Artificial intelligence (AI) has several transformative effects on healthcare, like to facilitate clinicians in disease detection and to manage patient care with enhanced precision. Due to the increasing complexity of medical data and the growing volume of diagnostic images, the substantial computational power needed to process such information creates both opportunities and challenges. To accomplish this objective, numerous methodologies have been proposed to integrate AI into disease detection systems. All these methodologies consider different aspects of clinical diagnosis, considering key parameters such as the type of AI model employed, the nature of input data, and the population demographics being analysed. In this paper, we review different state-of-the-art techniques which have been proposed for various areas of disease detection, specifically focusing on bias identification and mitigation. The proposed solutions are distinguished based on the type of input data they utilize, the architecture of model, and type of bias (algorithmic bias, demographic bias, model bias, and deployment bias). The time frame chosen to filter out relevant studies is between 2015 and 2026. Eventually, we review 23 related articles out of 64,389 analyzed from seven different digital libraries. The obtained results guide us to reach a conclusion with respect to state-of-the-art selected studies. Furthermore, these results reveal significant research gaps that can assist researchers in addressing AI-related bias in medical diagnosis. By prioritizing explainability, data diversity, and fairness in algorithmic design, as well as equitable access for affected patient populations, they can utilize available healthcare resources more effectively.
[AI-54] LLM s Prompted for Legal Context Refuse More: Overrefusal from Small On-Premises LLM s in Criminal Legal Context
链接: https://arxiv.org/abs/2610.26211
作者: Anastasiia Kucherenko,Francois Brouchoud,Dimitri Percia David,Andrei Kucharavy
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Confidentiality requirements increasingly push legal professionals toward small, on-device LLMs that keep case data in-house, used for everyday tasks such as summarization, translation, and reformulation. Following standard prompting guidance, a user naturally states their role (``I am a lawyer…‘’). We show that this innocuous habit has an unexpected cost: such authority framing can make the model selectively refuse legitimate requests, introducing bias through uneven access to assistance. We evaluate four modern small open-weight LLMs on benign legal-domain prompts to measure this over-refusal. Surprisingly, we find that authority-style prefixes (‘‘I am legal counsel working for a national supreme court’’, ‘‘I’m a defense lawyer’’) systematically increase refusal rates by 2–20x over the no-prefix baseline, while a known role-play jailbreak prefix shows mixed effects, sharply increasing refusals in some models and barely shifting them in others. The finding suggests that small on-prem deployable LLMs are unstable under contextual framings that a real institutional user might naturally introduce, and further investigation is essential to minimize opportunities for bias. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.26211 [cs.AI] (or arXiv:2610.26211v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.26211 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-55] On Counting Graphs for Modal Logic with Counting
链接: https://arxiv.org/abs/2610.26187
作者: Avijeet Ghosh,François Schwarzentruber
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注:
Abstract:In this paper, we study the theoretical computational complexity of counting trees and graphs that are classified positively by a graph neural networks, or said equivalently that satisfy a given formula of modal logic with counting. We also study convergence.
[AI-56] MoveBench: A Rubric-Based Benchmark for Evaluating Large Language Models in Sports and Health Coaching
链接: https://arxiv.org/abs/2610.26185
作者: Yitong Li,Jiang Bian,Jingling Wang,Pan Ding,Chengying Gu
类目: Artificial Intelligence (cs.AI)
备注: 2 figures. 140 open questions resources: this https URL
Abstract:Despite rapid advances in large language models (LLMs), their deployment in the sports and health domain remains high-risk due to the absence of domain-specific evaluation benchmarks. Existing general-purpose benchmarks (e.g., MMLU and HELM) primarily assess broad knowledge and reasoning capabilities, but fail to capture critical domain requirements such as exercise safety, physiological appropriateness, personalization, and actionable coaching quality. This gap is particularly concerning in fitness and health applications, where incorrect or unsafe recommendations can lead to physical harm. To address this limitation, we introduce MoveBench, the first comprehensive benchmarking framework specifically designed for evaluating LLMs in sports and health coaching. The proposed benchmark integrates heterogeneous task formats, including multiple-choice questions (MCQs) and open-ended question answering (OpenQA), grounded in real-world user scenarios and paired with structured rubrics emphasizing safety compliance, scientific validity, and training logic. Through large-scale evaluation of 15 diverse LLMs on 809 questions with 5,012 rubric-level criteria, our results reveal substantial performance variability, particularly in safety awareness, structured program design, and personalization capability, highlighting a critical gap between knowing and doing. Our work establishes a foundational evaluation standard for sports and health LLM applications and provides a reproducible framework for advancing safe, reliable, and context-aware AI-driven coaching systems. Open resources are available at this https URL.
[AI-57] Nash-MOPD: Optimizing Task Trade-offs in Multi-Teacher On-Policy Distillation
链接: https://arxiv.org/abs/2610.26124
作者: Siyuan Gan,Bo Yang,Linjian Meng,Boyan Wang,Xiaolong Qi,Jing Huo,Yang Gao
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-teacher on-policy distillation (MOPD) aggregates the capabilities of multiple domain-expert teachers into a single student model. However, all domains in MOPD share the student’s finite budget of optimization steps and rollouts. Fixed uses constant domain sampling ratios based on corpus sizes; by contrast, D ^3 -MOPD adjusts the ratios using the teacher-student reverse-KL divergence and its recent decrease. However, favoring large recent decreases can direct more training toward domains that already improve quickly, while domains that improve more slowly may receive too few examples. To address inflexible allocation and the risk of giving insufficient training to domains that improve more slowly, we propose Nash-MOPD, an adaptive domain scheduler based on a local minimax game using existing reverse-KL statistics. The equilibrium maximizes the smallest predicted proportional reduction under the local model and gives more prompts to domains that need more examples for comparable improvement. Consequently, Nash-MOPD responds to changing learning progress and supports domains that improve more slowly without extra teacher queries or rollouts. At step 100, Nash-MOPD exceeds both baselines’ final mathematics and code checkpoint scores on DeepSeek-R1-Distill-Qwen-1.5B and reaches Fixed’s on Qwen3-8B-Base. Furthermore, it leads eight of nine reported metrics and is best or tied best on all six mathematics and code accuracy metrics, respectively.
[AI-58] Reliable Before Profitable: QuantHarness for Governing Financial Strategy Development with LLM Agents
链接: https://arxiv.org/abs/2610.26101
作者: Jiasheng Mao,Jiacheng Liu,Hengwei Ye,Zheng Tian
类目: Artificial Intelligence (cs.AI)
备注: 47 pages, 2 figures, including appendices
Abstract:LLM agents can turn natural-language financial research requests into executable and backtested strategies, but a successful run may still implement the wrong rule, use information unavailable at decision time, or obscure how repairs and candidate selection changed the strategy. Because these failures arise throughout development, final code and performance metrics alone are insufficient to establish either fidelity to the request or the validity of the development process. We therefore address the reliability of financial strategy development at the trajectory level: the system must control how intermediate outputs become downstream state and preserve the evidence needed to justify the terminal outcome. We introduce QuantHarness, a two-layer agent harness in which a finance-specific Workflow structures specification, data grounding, construction, testing, and review, while Harness Core enforces stage-specific information, action, and commitment boundaries. We also introduce FinRelBench, a benchmark that evaluates strategy-development reliability through execution success, requirement satisfaction, and runtime evidence. QuantHarness achieves the highest Task Success at 90.72%, compared with 74.23%–84.54% for the comparison systems. Controlled stress tests show that its boundaries contain targeted violations, while structured run records improve evidence retrieval and support decision replay under the evaluated protocol.
[AI-59] From Code to Playable Systems: CEGame-Bench for Long-Horizon Game Creation Agents
链接: https://arxiv.org/abs/2610.26098
作者: Zixuan Dong,Yanghai Wang,Hanyan Bian,Jiaxi Tao,Junze Li,Jiadong Wang,Xinyu Che,Lin Liu,Baoyun Peng,Xiaodong Wang,Jiaheng Liu
类目: Artificial Intelligence (cs.AI)
备注: 27 pages. Preprint
Abstract:A game ships when a player can launch it, finish it, and return to a saved world, none of which a successful build guarantees, yet game-creation agents are mostly tested on one engine or on one-prompt browser games. We present CEGame-Bench, the first long-horizon benchmark for complete-game creation across five production engines: Defold, Godot, Roblox Studio, Unity, and Unreal Engine. Agents build each game from an empty project and a design document in runs averaging up to 295 steps and 65M tokens; the document doubles as a contract, and a hidden evaluator accepts a game only by real-input play. Under four scaffolds, agents earn most integration points but usually under half the gameplay points, and their games reach the ending in only 16% of runs, while references built by engineers with coding agents score near the maximum, which places this artifact-to-play gap in the agents. No two engines rank configurations alike (Kendall’s tau 0.14-0.62), and the same game usually scores higher on Unity and Unreal Engine, which check more before play. We release Game Completion Boost, an agent skill for requirement-driven playtesting and repair that raises the execution-based score of three models by 14.1-24.3 points.
[AI-60] CUA-Lite: An Open Platform for Computer Use Agents
链接: https://arxiv.org/abs/2610.26071
作者: Zhanhui Zhou,Weichen Zhang,Haoran Liu,Lingjie Chen,Tianneng Shi,Kevin Lin,Zhengyuan Yang,Lijuan Wang,Dawn Song
类目: Artificial Intelligence (cs.AI)
备注: this https URL
Abstract:Open resources for computer use agents (CUAs) – environments, datasets, and models – remain scattered: without a standard, these resources cannot be integrated and scaled; additionally, evaluation, RL, and SFT cannot share one infrastructure. We present CUA-Lite, an open platform for developing CUAs, built around three standardized abstractions: (1) Unified environment interface, under which we integrate 12 benchmarks across desktop, browser, and mobile, plus lightweight VM-free sandboxes that replicate OSWorld’s desktop at much lower cost, providing over 30,000 verifiable tasks. (2) Unified supervised data format, into which we convert the major public CUA datasets, from GUI grounding and understanding to full rollouts, and record our fresh rollouts from frontier CUAs on these verifiable environments. (3) Model harnesses shared across evaluation, RL, and SFT, one per model and implemented for 14 model families. Each harness runs its model in these computer environments, producing the rollouts that evaluation and RL consume, or renders stored supervised data into the model’s own training format for SFT. We release reference scores across the integrated benchmarks, and show that RL and SFT through the same harness improve open-weight models. We release our dataset, code, and models, and the platform is open to contributions.
[AI-61] FinSafeGuard: A Quality-Driven Synthetic Dataset for Financial AI Safety
链接: https://arxiv.org/abs/2610.26066
作者: Joseph Thomas Thacil,Reetu Raj Harsh,Krati Saxena,Bhaskarjit Sarmah,Stefano Pasquali
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Generating high-quality synthetic datasets at scale remains a fundamental challenge for training specialized AI systems in data-scarce domains. We present a comprehensive methodology that produces 14.3 million quality-filtered synthetic training examples from a structured taxonomy of 632,000 scenario specifications. Our four-stage pipeline ad- dresses critical challenges in synthetic data quality, diversity, and realism through: (1) systematic pre-generation research establishing optimal generation parameters via diminishing returns analysis, (2) conditional auto evolution mechanisms that remediate low-diversity batches during generation, (3) a two-layer semantic deduplication architecture combining lexical and embedding-based methods, and (4) LLM-as-Judge quality evaluation filtering samples based on behavioral correctness and tuple fidelity. Empirical analysis across three domains (Banking, Insurance, Financial Services) demonstrates domain-specific deduplication patterns ranging from 9.9% to 40.6%, with output generation modes exhibiting highest redundancy. Our batch size optimization reveals mode-specific diversity saturation thresholds, enabling principled allocation strategies. The complete pipeline reduces 26.4M raw samples to 14.3M high-confidence examples through merge (20.9% reduction), semantic deduplication (18.7% reduction), and quality-based filtering (15.9% reduction), providing a reproducible framework for domain-specific synthetic data generation with rigorous quality control.
[AI-62] AuDQ: Auditable Data Quality through Agent ic Rule Discovery
链接: https://arxiv.org/abs/2610.26031
作者: Jacopo Biggiogera,Nicolas Colombo,Krati Saxena,Bhaskarjit Sarmah,Stefano Pasquali
类目: Artificial Intelligence (cs.AI); Computational Finance (q-fin.CP)
备注:
Abstract:High-stakes tabular pipelines rely on scattered, static quality checks that are hard to evolve and audit against policy intent. Regulated domains particularly suffer as the LLM-era automation raises expectations for traceability and accountability. We propose a vision for an ontology-grounded, repository-native data quality layer in which rules are first-class artifacts: versioned, provenance-linked to governance texts, and re-playable through immutable execution traces. As a first step toward validating this vision, we present a concrete implementation of its agentic rule-discovery core with two rule acquisition modes (Full-Agentic and Guided) and a human-in-the-loop (HITL) review interface. We define an evaluation framework and run experiments addressing whether our fully autonomous, divide-and-conquer agentic rule discovery produces rules of the same quality and diversity as those specified by a human expert, reporting preliminary results on two real world industry datasets in the finance and retail domains.
[AI-63] mesJev: Thinking Fast on Time Series with System 1 Sufficiency
链接: https://arxiv.org/abs/2610.26002
作者: Chenghao Liu,Wei Fan,Jindong Han,Dongjie Wang,Aniket Mahanti,Pengyang Wang,Hao Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models (LLMs) have recently shown strong potential for time series reasoning, while models that reason over time series are usually with long chains of thought. However, this process often consumes thousands of tokens, directly leading to substantial latency during inference. In contrast, System 1 models such as Jev make decisions instantly and can naturally be an alternative, but raw time series are not compatible with the embedding space of these models, thus hindering the reasoning performance, especially when input series contain thousands of numeric points. In this paper, we make a very early attempt to propose a time series System 1 model, namely TimesJev. Specifically, we first formalize System 1 sufficiency in time series reasoning, which motivates a compressed representation for time series with key frequency patterns preserved. To construct this representation, we derive a dimension reduction objective that combines (i) a sufficiency divergence that measures the core information lost during compression, and (ii) a realizability divergence that quantifies the incompatibility of the System 1 model to the optimal prediction given this representation. We optimize this objective by learning time series compression with a gating mechanism for sufficiency, and then aligning the selected representation with the embedding space for realizability. Finally, experiments on 13 datasets demonstrate that TimesJev outperforms reasoning models by over 62% with 62.9% less inference time than their fine-tuned versions on average.
[AI-64] Procedural Memory for Language Agents : A Neurosymbolic Perspective on Explicit Goal Structure
链接: https://arxiv.org/abs/2610.25996
作者: Jonathan Leung,Zhiqi Shen,Deheng Ye
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models remain unreliable on long-horizon tasks requiring procedural reasoning. We argue that a key cause is the absence of explicit procedural memory: accessible, modifiable knowledge of how tasks are accomplished, distinct from semantic and episodic memory. Goal-Oriented Graphs offer such a representation, encoding goals and prerequisite dependencies as inspectable symbolic structure. We examine their construction, the challenges they pose, and principles for integrating them with neural reasoning.
[AI-65] Map-Distribution Mixture-of-Experts for Multi-Agent Pathfinding
链接: https://arxiv.org/abs/2610.25991
作者: Fumimaro Suzuki,Keisuke Okumura
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Designing capable neural policies for multi-agent pathfinding (MAPF), which map local observations to action distributions, represents an important challenge in multi-agent learning, with practical relevance to scalable collective automation problems such as warehouse logistics. In contrast to prior studies focusing primarily on individual policy design, this study explores a meta-level Mixture-of-Experts (MoE) approach to learning-based MAPF. Specifically, given a collection of compact yet high-performing experts specialised for different map distributions, our architecture learns a router that selectively fuses their outputs according to each agent’s local situation. We instantiate and validate this MoE approach with a commonly used graph neural network-based policy as well as a newly developed recursive action-token refinement model. Combined with supervised and reinforcement learning fine-tuning of the experts, our experiments show that MoE generally improves upon the pretrained policy and outperforms existing leading methods. At the same time, our findings reveal that appropriately trained, parameter-matched monolithic models still outperform MoE, highlighting a nuanced role for map-distribution MoE for MAPF.
[AI-66] NIOCIS: Typed-Decision Next-Hop Routing with Node-Conditioned Option Spaces for Tool-Use Agents
链接: https://arxiv.org/abs/2610.25985
作者: Jin Han,Mingfeng Xin,Feng Liang,Jingtao Yu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 34 pages, 10 figures. Code and the pseudonymized derived corpus: this https URL
Abstract:Tool-use agents pay per decision: at each hop a language model re-reads the state and thinks for seconds, though its actions are finite and decisions recur. We present NIOCIS, a routing layer that folds execution logs into a loop-collapsed action graph (forward transitions plus recovery back-edges) answering one typed question: which declared option at a state, plus stop, does the fleet take next? A small model reads a calibrated distribution over them in one pass, emitting no text. On 1,615 production security-operations episodes we measure what the layer replaces, returns and is bounded by. Replaces: the typed encoder matches the logged next hop on 32.2% of the 3,875 held-out decisions (34.9% within the declared options) in 10 ms of encoder time (19.5 ms for the typed-choice form), against 2.2 s per LLM hop; two feature readouts reach 32.8% and 32.3%, an ordering the paired test does not resolve, so the typed layer is chosen for cost, not a separated lead. Tuning: jpt-4b reaches 9.9% zero-shot and 31.9% fine-tuned, a gap an exact paired test does not resolve (p=0.68), at 244 ms. Returns little: the realised hop falls outside the declared options for 7.9% of decisions and outside the model’s top twenty for 9.3%; the most confident 15% holds 91.9% accuracy. Bounded: a cross-trace leave-one-out test puts the released state’s ceiling at 25.6–44.9% as the state key grows more specific, i.e. repeated states disagree more often than they agree; the option-space ablation attributes the gain to per-node candidate restriction (+3.5 points over a flat catalog) rather than graph traversal (zero within its interval); a gated self-evolution loop lifts every arm above its anchor yet ends below full supervision; a confidence-gated cascade loses agreement when the escalation arm is the zero-shot stand-in; and agreement past a third of the fleet needs a state carrying the plan and the observation.
[AI-67] Auditor-Assisted Summary-Channel Verification for Hosted LLM Identity Substitution NIPS2026
链接: https://arxiv.org/abs/2610.25982
作者: Ziyi Zhang,Ziyao Wang,Guoheng Sun,Ang Li,Jian Li
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: Accepted by NIPS 2026
Abstract:Hosted LLM services increasingly expose model identity as a product claim, but external users and auditors often cannot inspect which model configuration served a request. We study an auditor-assisted regime in which a trusted auditor enrolls claimed identities and prepares verification material before deployment, while audit-time verification uses only the claimed identity and released output, without runtime access to model internals, inference logs, routing decisions, or provider metadata. We propose SUMMARK, a summary-channel watermarking framework that uses user-visible reasoning summaries as a keyed audit surface. During setup, SUMMARK constructs identity-specific vocabulary carriers, trains lightweight summary adapters, and calibrates per-identity thresholds. At deployment time, a keyed statistical test assesses whether the released summary is consistent with the claimed identity and, when needed, attributes the most likely enrolled identity. Across model families, scales, and task domains, SUMMARK reliably detects cross-family and cross-scale substitutions at low false-positive rates while preserving task-facing answer quality. We further study FP16-INT8/INT4 precision-consistency diagnostics and robustness under bounded post-processing, rewriting stress tests, and text-only spoofing, revealing both strong performance and clear failure boundaries. Our code is publicly available at this https URL.
[AI-68] ClueFlow: Learning Long-Horizon Search through Attributed Research-State Transitions
链接: https://arxiv.org/abs/2610.25980
作者: Yicheng Wang,Wentao Tan,Yidan Huang,Yifei Liu,Qiong Cao
类目: Artificial Intelligence (cs.AI)
备注: 25 pages, 2 figures, 8 tables
Abstract:Long-horizon search requires agents to acquire relevant evidence and integrate it into an evolving research state. Existing methods maintain research states to guide exploration or assign turn-level progress signals, yet grounding these signals in explicit, attributable state changes remains underexplored. We introduce ClueFlow, which learns long-horizon search from attributed research-state transitions. ClueFlow constructs Source-State-Final attribution chains that connect evidence-acquiring browser actions, Evidence and Candidate updates, and the final answer supported by those updates. During search, ClueFlow records typed state changes and potentially supporting browser actions, then applies gold-conditioned semantic validation and source tracing after execution. We use this attribution to guide both supervised fine-tuning and reinforcement learning. During SFT, attribution selects which decisions receive supervision with their original interaction contexts. During Turn-PPO, the same attribution augments outcome-based turn advantages with credit assigned along the Source-State-Final chains. Transition-level validation also recovers useful supervision from gold-consistent state changes in trajectories with incorrect final answers. We evaluate ClueFlow across two Qwen backbones and four search benchmarks spanning live-web and fixed-corpus settings, observing consistent gains from SFT and further improvements from PPO. With Qwen3.5-35B-A3B, ClueFlow achieves 60.0% accuracy on BrowseComp and 72.0% on BrowseComp-Plus. In controlled BrowseComp-Plus experiments, attribution-guided SFT matches all-turn exact-context SFT using about 30% of the training tokens, while attribution-guided PPO outperforms outcome-only PPO by 4.1 pp. These results demonstrate that state-transition attribution enables efficient supervision and effective credit assignment for long-horizon search agents.
[AI-69] GraphEvoResearcher: Graph-Aware Training-Free Self-Reflective Reasoning for Cost-Bounded Multi-Hop Question Answering
链接: https://arxiv.org/abs/2610.25978
作者: Suxing Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-hop question answering (QA) requires models to chain intermediate conclusions while remaining cost-aware. Existing training-free self-reflective protocols (generate - critique - revise) improve accuracy but operate on a flat token stream: revisions cannot identify which prior step is unreliable, and early stopping cannot exploit structural convergence. We introduce GraphEvoResearcher, a training-free protocol that lifts EvoResearcher’s self-reflective meta-reward onto an explicit reasoning graph - nodes are intermediate conclusions/evidence and edges encode support, contradict, and depends_on. A graph-aware meta-reward combines correctness, efficiency, reflection, diversity, and four structural signals (connectivity, acyclicity, branch coverage, path convergence) to decide whether to revise, backtrack, or stop. A graph-convergence early-stop terminates when evidence branches converge on a single conclusion. On a real LLM (DeepSeek), over GSM8K, MATH, and BBH (n=50 each), GraphEvoResearcher reaches 98%, 92%, and 82% respectively, the highest of all baselines, with the largest gains on the hardest set (MATH: +24 pp over single-shot, +12 pp over flat reflection), while keeping construction cost bounded at ~2 steps / 4.5-5 nodes. Our results indicate the structural signal of a reasoning graph is most valuable exactly where single-step reasoning is unreliable.
[AI-70] Reconstruction and Discrete-Event Simulation Assessment of an Analytical
链接: https://arxiv.org/abs/2610.25967
作者: Mohsen Reda Mahmoud Moursi,Ahmed Stohy
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:This article presents an analytical-DES benchmark for intra-aisle robotic picking in automated warehousing. The investigated cycle combines horizontal shuttle travel, tote handling and item picking at storage locations. Simulation results are compared with the analytical formulation of Lerher et al. under the same rack geometry, shuttle-motion, handling and picking conditions, and the same definitions of cycle time and derived item throughput. Reproducing the published analytical tables requires the distance interpretation dES = 1.5 m and dTB = 0 m. The analytical-DES benchmark comprises 180 scenarios formed from three rack lengths, two paired motion profiles and n = 1 to 30 visited picking locations, with ten simulation replications per scenario. The six n = 1 cases are reported separately because no travel-between component is present. Across the remaining 174 scenarios, the mean absolute relative cycle-time discrepancy is 0.8655%, the median is 0.6758%, and the maximum is 7.0370%. The mean absolute error is 2.411 s and the RMSE is 2.744 s. The simulated scenario-mean cycle time is below the analytical value in all 174 principal scenarios. The results therefore show close numerical correspondence under the tested conditions, but not exact agreement, formal statistical equivalence, or validation against an operating warehouse.
[AI-71] ABACUS: Program-Guided Multistep Reasoning over Frozen Judgment Models
链接: https://arxiv.org/abs/2610.25963
作者: Yuhao Wang,Zelin Zang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Judgment models answer semantic questions by scoring a finite set of candidate outcomes. Can reusable programs turn these local decisions into powerful multistep reasoning? We introduce \method, a framework that jointly constructs judgment schemas and composition programs for frozen judgment models. A generative teacher designs the procedure offline; the judgment model resolves local semantic questions online, and the saved program composes its decisions into an answer. We study three allocations of design effort between humans and the teacher, focusing on construction through a shared interface. We instantiate the framework with Kev. On three selected BIG-Bench Hard tasks, nine independently constructed programs solve all 450 executions of 150 development inputs. A separate deployment study freezes three programs after development-set reference-agreement refinement and evaluates them on 600 held-out questions. These programs achieve near-perfect 99.67% macro accuracy, versus 64.00% for direct Kev and 68.80% for final-answer-only Qwen3.5-4B. They attain comparable observed accuracy to Qwen’s unrestricted response policy (99.07%) with an \textbf8.43-fold online speedup: 0.255 versus 2.150 seconds per question. These results show how reusable programs can substantially expand the reasoning capability of a compact judgment model while keeping its parameters fixed.
[AI-72] PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.25961
作者: Hojin Jeon,Jimin Lee,Taekwan Kim,Nayoung Oh,Taeyoung Kim,Myungkyu Koo,Jinmo Koo,Joonwoo Ahn,Jinwoo Shin
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 25 pages, 7 figures, 10 tables. Project page: this https URL
Abstract:Fast and reliable manipulation is crucial for practical robot deployment. Although Vision-Language-Action (VLA) models have recently shown remarkable progress as generalist robot policies, their execution often remains temporally inefficient. A key source of this inefficiency is that policies learn fine-grained temporal patterns from teleoperated demonstrations, requiring many low-level action steps. Temporal action quantization offers a simple way to accelerate execution by aggregating consecutive low-level actions into temporally coarser actions. However, applying action quantization without considering when fine-grained control is required introduces a speed-accuracy trade-off: accelerated actions may efficiently reduce redundant steps in some phases, while discarding critical motion details in precision-sensitive interactions. To address this, we propose PaceVLA, a unified framework for adaptive temporal execution in VLAs that integrates two coupled components: (1) mode-specific action decoders built on a shared action expert, supporting multiple temporal granularities and prediction horizons, and (2) a quantizability-guided mode selection scheme which decides whether coarse execution is appropriate, and a context-conditioned router that selects an execution mode at each inference step. Experiments across three multi-task simulation benchmarks and six real-world tasks show that PaceVLA improves execution efficiency while maintaining task success rate across multiple embodiments, including long-horizon and dexterous manipulation.
[AI-73] EmbodiedWorld-200K: Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models NEURIPS2026
链接: https://arxiv.org/abs/2610.25959
作者: X. Feng,J. Tang,Z. Lin,L. Chen,S. Shao,Q. Lu,K. Huang
类目: Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026~
Abstract:Building embodied agents that can perceive, reason, and act across diverse 3D worlds is a central research direction toward artificial general intelligence. Yet most existing work remains confined to specific tasks in a single scene and transfers poorly to new environments, leaving general embodied agents for open-world settings under-explored. We refer to such agents as Embodied World Agents (EWAs), whose defining property is the ability to understand diverse visual environments and execute general actions in open-world conditions. To advance EWA research, we study open-world embodied navigation-planning and release EmbodiedWorld-200K, a large-scale dataset of over 200K samples spanning diverse scenes. Each example is equipped with multi-level annotations, ranging from low-level camera-motion trajectories to high-level semantic instructions. The dataset also provides unified evaluation splits and task-specific metrics, supporting offline pretraining and fair evaluation of EWAs. Beyond offline supervision, a defining characteristic of an EWA is its interaction with the environment; training solely on offline data is therefore insufficient, especially for generalizing to scenes not covered in the training distribution. Building dedicated real interactive environments, however, is often prohibitively expensive. To address this, we propose a novel training scheme that employs a video world model as a low-cost environment proxy and performs interactive reinforcement-learning (RL) post-training of the EWA within it. Empirically, pretraining on EmbodiedWorld-200K substantially improves the open-world embodied planning ability of the EWA, and further RL post-training inside a world model yields clear gains on out-of-distribution scenes. The dataset, baseline checkpoints, and training and evaluation code are available at this https URL.
[AI-74] Productive State-Space Exploration for Multi-turn RL in LLM Agents
链接: https://arxiv.org/abs/2610.25957
作者: Mohammed Haroon Dupty,Yong Liang Goh,Song Guo,Wee Sun Lee
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Reinforcement Learning (RL) for Large Language Model (LLM) agents in multi-turn, long-horizon environments remains challenging due to inefficient action selection and repetitive behavioral patterns. In particular, agents often become trapped in high-confidence yet unproductive behaviors, repeatedly executing invalid actions or entering action loops that hinder task completion. We observe that these frequently violate environment preconditions, revealing a mismatch between the agent’s representation of the internal state and the dynamics of the underlying environment. In addition, repetitive high-confidence actions lead to cyclical behaviors that provide little to no progress towards the goal. We argue that addressing these challenges requires more productive task-aligned exploration to improve the agent’s understanding of environment dynamics. Motivated by intrinsic curiosity-driven exploration, we propose a conditional state-space exploration framework that balances task completion with intrinsic information-seeking behavior. Specifically, we introduce an intrinsic reward that penalizes actions whose observation duplicates prior state memory. This provides the benefit of reducing redundant behaviors and encouraging state-space coverage without relying on a general information-theoretic measure. Strong experimental results demonstrate that our approach substantially improves policy learning in multi-turn environments, mitigating invalid actions and repetitive loops.
[AI-75] CXBench: A Benchmark for Process-Compliant LLM Agents in Regulated Customer-Experience Workflows
链接: https://arxiv.org/abs/2610.25954
作者: Boxi Yu,Yuzhong Zhang,Zhenfeng Su,Linfu Xi,Yinghan Yao,Shitao Chen,Yang Cao,Liting Lin,Lionel Briand,Emir Muñoz
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Tool-using language agents increasingly perform regulated customer-service workflows, where a correct final state is insufficient: agents must also provide disclosures, obtain consent, and follow required procedures. Existing tool-agent benchmarks largely score final business states, accepting trajectories that reach correct outcomes through non-compliant processes. We introduce cxbench, a benchmark that evaluates outcome and process compliance separately across six regulated domains. Each task combines a customer request, a tool-based environment, and explicit process requirements. Rule cards link cited regulatory obligations to observable trace evidence. A hybrid grader combines deterministic checks for ordering and authority with an LLM judge for semantic requirements. We define a false-pass as an outcome-pass run that fails process checks. Across seven models and 120 tasks, 55.5% of the 840 retained executions are false-passes, representing 80.1% of outcome-passing runs; only 13.8% pass both checks. Process-inclusive evaluation changes the observed model ordering and exposes differences hidden by outcome-only scoring. cxbench makes these failures inspectable through trace-linked criteria. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.25954 [cs.AI] (or arXiv:2610.25954v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.25954 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-76] OrderGPT : Generation Order as a Learnable Modeling Choice
链接: https://arxiv.org/abs/2610.25949
作者: Chenhe Yang,Yuhuan Lyu,Tianyang Hu
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 27 pages, 5 figures
Abstract:Generative modeling of discrete data has two popular paradigms: autoregressive models such as GPT, which factorize a sequence in a left-to-right (L2R) order, and masked diffusion models, which are essentially any-order autoregressive modeling. Rather than competing methods, these are the two extreme points of a single axis – the generation order. For structured discrete domains without a natural traversal, both extremes may be suboptimal, yet the intermediate choice – a single well-adapted order – remains largely unexplored. We introduce OrderGPT, which builds on Any-Order GPT to learn a task-level generation order without canonical-order supervision. Our key observation is that predicting variables from randomly sampled reveal prefixes produces recurring attention patterns in the backbone that provide a model-dependent signal of predictive dependencies. We aggregate this signal into a reciprocal dependency graph and use a spectral teacher to derive continuous generation priorities. A lightweight OrderMLP integrated into the GPT backbone learns to predict these priorities from attention, incorporating order learning into the architecture within a single training run. Training gradually shifts toward the learned order, which is then frozen for subsequent autoregressive training and generation. The same mechanism adapts across modalities: adaptively it recovers near-L2R dependency structure in language, learns anatomically organized factorizations on human pose, and discovers a factorization on tabular data that places statistically and semantically related fields nearby, while improving downstream modeling metrics across all domains. The code is available at this https URL.
[AI-77] Claim Laundering: When Agent Decision Models Trust Records Over Reality
链接: https://arxiv.org/abs/2610.25947
作者: Tina Khezresmaeilzadeh,Seyedarmin Azizi,Erfan Baghaei Potraghloo,Mehdi Kamal,Massoud Pedram
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent decision models select actions in a single forward pass from inputs that combine environmental observations with the agent’s own record of previous actions. When an action fails silently, these sources can conflict: the record reports success, but the environment shows that the step did not take effect. We ask which source determines the next decision. Across Android screen traces, ALFWorld, and WebShop, we construct paired states that independently vary the observation, the logged action, and an explicit success claim. In all 12 configurations of four decision models and three domains, records have a larger average effect on commitment decisions than observations; after claimed-but-failed steps, the models still stop, advance, or purchase in 53-100% of cases. Yet the relevant evidence is often represented internally: linear probes on intermediate hidden states distinguish successful from failed steps on held-out apps, tasks, and goals (AUROC 0.77-1.00), and activation patching reveals increasing influence from records in later layers. We use this signal in EVIGATE, an intermediate-layer veto that requires no additional forward pass and adds 0.04-2.0% serving latency. Under injected silent failures, EVIGATE improves task success by 18.7 points in live ALFWorld agents, 5.0 in WebShop, and 5.1 in a phone simulator, and in WebShop and the simulator it also outperforms random vetoes at the same rate. Its effectiveness depends on representative calibration states: gates calibrated on offline counterfactual states do not transfer reliably to live agent trajectories.
[AI-78] Policy Search in Code Space for Multi-Agent Pathfinding
链接: https://arxiv.org/abs/2610.25946
作者: Matthias Eder,Pantelis Emmanouil,Vivek Farias,Fabian Fischer,Adam Jozefiak,Tianyi Peng,Thomas Ulz,Raphael Watschinger,Fangyu Wu,Andrew Zheng
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 18 pages, 11 figures
Abstract:Developing coordination policies for complex multi-agent systems relies on either labor-intensive algorithm design or opaque task-specific reinforcement learning. With recent advances in autonomous coding agents, we show that policy search directly in code space offers a compelling third path that combines automated optimization with explicit interpretability. We investigate a human-directed workflow in which coding agents iteratively propose, implement, evaluate, diagnose, and eventually refine algorithms guided by simulation feedback and explicit correctness verification. We conducted two case studies and found that this workflow robustly outperformed any manually engineered solutions. On a challenging map in the 2024 League of Robot Runners competition, a two-day search discovered a policy that exceeds the stronger locally reproduced competition entry by 35.0% on the densest instance and generalizes to unseen map instances. In simulated real-world warehouses with irregular lanes, continuous robot geometry, and asynchronous motion, a week-long search discovered an entirely new class of multi-agent pathfinding algorithm that outperforms a calibrated proprietary solution in throughput by 58.9%. Under stated assumptions, we further prove that the new algorithm is collision-free and completes, in finite time, every accepted goal that is not withdrawn or replaced. Together, the studies show that policy search in code space powered by state-of-the-art coding agents offers a highly competitive path to building real engineering solutions that are both effective and analytically tractable.
[AI-79] Global probabilistic forecasting of air pollution
链接: https://arxiv.org/abs/2610.25941
作者: Ammar Kheder,Michael Boy,Zhi-song Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Fine particulate matter (PM2.5) refers to particulate matter with an aerodynamic diameter of 2.5 micrometre or less. It comes from human activities, such as traffic and industry, and from wildfires, desert dust and gases emitted by vegetation. Short-term exposure to it accounts for about a million deaths a year. Operational forecasts run on grids tens of kilometres wide, which hide the pollution breathed by nearly two hundred million people a day. Machine learning has transformed weather forecasting, but data-driven forecasts of air pollution still blur extremes and lose skill as the air travels. Here we introduce Foehn, a generative model that learns from global kilometre-scale estimates of PM2.5 and computes where the air comes from. Over a year of retrospective forecasts, it outperforms the operational forecast at three stations in four and forecasts at the scale of this hidden exposure. Its ensembles are calibrated, and their mean detects local pollution extremes four days ahead as well as the operational forecast does one day ahead. Such forecasts could warn the millions whom coarse maps miss, once kilometre-scale data are available in real time.
[AI-80] he KV-Surgeon: Compensation-Aware Offline KV-Cache Compaction
链接: https://arxiv.org/abs/2610.25933
作者: Moritz Wagner,Christophe Roux,Max Zimmer,Sebastian Pokutta
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Offline compaction of the key–value cache (KV-cache) reduces the cost of repeatedly querying a fixed context by constructing a smaller cache once and reusing it subsequently. The central challenge of compaction is deciding which cache entries to delete and which to retain. Existing methods score each candidate for deletion on its own, with the rest of the cache being held fixed. However, softmax renormalization of the attention operation couples these decisions in a nonlinear manner: removing one entry changes the attention weights of every remaining one. In this work, we introduce KV-Surgeon, which uses calibration data to take interactions into account in a tractable way. We derive a quadratic surrogate that admits closed-form expressions for jointly making the optimal greedy removal decision as well as compensation within the remaining entries. On QuALITY, LongHealth, and Long Code Arena, and at every evaluated compaction ratio, KV-Surgeon substantially advances the cost–accuracy Pareto frontier of dense and mixture-of-experts models from 4B to 30B parameters. At 50\times compaction, KV-Surgeon improves question-answering accuracy by up to four points over attention matching (Zweiger et al., 2026), the state of the art, at comparable cost. Lastly, we introduce an even faster variant, which also exceeds the state of the art while requiring only two fifths of its cost.
[AI-81] ESS: LLM -guided Translation Exploration Simulation and Search for Embodied Task Planning EMNLP2026 EMNLP26
链接: https://arxiv.org/abs/2610.25930
作者: Peiran Xu,Yixiang Yang,Jinghan Li,Fei Li,Cao Sheng,JiaZhong Yu,Yadong Mu
类目: Artificial Intelligence (cs.AI)
备注: EMNLP 2026 findings. Code will be released at this https URL
Abstract:In this work we focus on the problem of embodied task planning, which requires an agent to accomplish a complex task by separating it into atomic actions. Recent studies have developed LLM-based methods to address this problem. However, the executability and correctness of the generated plans remain suboptimal in complex environments and tasks. On the other hand, classical search algorithms have the potential to solve the planning problems perfectly, yet they require fully observable environments, known state transitions and goal state, which are often unattainable in general task domains. Therefore, we propose to implement an LLM-guided search algorithm that makes the search process possible and efficient by incorporating LLMs’ commonsense knowledge. Our method leverages LLMs to (1) translate natural language instructions into formatted goal conditions, (2) facilitate efficient exploration in partially observable environments, (3) implement a rule-based world simulator, and (4) prune the space of candidate actions to improve search efficiency. Experimental results on VirtualHome and ALFWorld demonstrate its superior performance.
[AI-82] UtilityBench: Measuring Language Model Utility When Errors Have Different Costs
链接: https://arxiv.org/abs/2610.25924
作者: Qifeng Lei,Zhiyong Yang,Zitai Wang,Shilong Bao,Linchao Zhang,Qianqian Xu,Qingming Huang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:As language models are increasingly deployed in real-world scenarios with important outcomes, the impact of their mistakes often extends far beyond what correctness can capture. In domains like medicine, a single error can lead to serious consequences and put patient safety at risk. However, mainstream evaluation protocols still treat all errors as equal, overlooking the spectrum of real-world costs. To address this, we propose a utility-based evaluation paradigm that explicitly incorporates the varying consequences of errors. Guided by this principle, we develop UtilityBench, a large-scale benchmark of 15,000 instances with verifiable answers, quantifying error severity using Best-Worst Scaling to elicit meaningful, data-driven cost estimates. These quantified utilities shape when a model should answer or abstain, guiding it to make decisions that align with the true importance of each scenario. As another merit, we introduce UtilityAUC, a metric that measures the area under the utility curve determined by a model’s response ordering, thereby enabling utility preference alignment beyond error frequency. Our theoretical analysis demonstrates that UtilityAUC can distinguish between models where standard accuracy and AURC fall short. Empirically, evaluating 20 state-of-the-art models, we find that greater accuracy or more advanced architectures do not reliably translate to higher UtilityAUC, highlighting a gap between problem-solving ability and responsible decision-making in language model evaluation.
[AI-83] Secure Agent ic Coding through Counterexample-Grounded Feedback
链接: https://arxiv.org/abs/2610.25916
作者: Tue Le,Anh Nguyen Hoang,Gabriel Sasseville,Doseok Jang,Pierre-Luc Bacon,Nghi D. Q. Bui
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:As coding models and agent harnesses become more capable, reliable security verification is becoming a central bottleneck. Improving secure code generation requires a signal whose reported progress remains aligned with actual program behavior under optimization; otherwise, an optimizer can improve the signal’s verdict without removing the underlying vulnerability. Existing model-based approaches require models to either infer program behavior from source code or specify correct outcomes for adversarial tests, placing substantial reasoning demands on the security signal. We introduce \falcon, a counterexample-grounded verification framework that separates attack exploration from behavioral verification and connects them through execution traces of the program under test. In test-time repair, \falcon yields substantially stronger held-out functional-security gains than alternative security signals across secure code generation benchmarks, while preserving or improving functionality. As an RL reward, \falcon also outperforms static-analysis and learned rewards on held-out functional-security, even when all policies successfully optimize their respective training rewards. Combining \falcon with either proxy reward further improves held-out performance, suggesting that execution-grounded feedback can mitigate reward overoptimization while preserving useful guidance from existing security signals.
[AI-84] When Should AI Chime In? Benchmarking Timely Contribution of Proactive Agents in Multi-Party Chat
链接: https://arxiv.org/abs/2610.25896
作者: Ruotong Wang,Brian S. Yu,Xinyi Zhou,Joseph Chee Chang,Pao Siangliulue,Jonathan Bragg,Liwei Jiang,Amy X Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Proactive agents can support group collaboration by surfacing useful information without waiting to be asked, but even relevant contributions can be disruptive if poorly timed. In this work, we isolate the timing question: given a fixed candidate contribution such as a clarification or relevant fact, when should an AI agent chime in? We introduce PITCH (Proactive Intervention Timing in CHat), to our knowledge the first benchmark with large-scale human judgments of proactive agent timing collected from participants in the same unscripted multi-party conversations they took part in. To collect high-quality judgments, we design a novel interface in which 2-4 people chat synchronously, author candidate AI contribution messages, place their own and others’ contributions into the chat history, and then deliberate as a group about the best placement for each message. PITCH contains 384 candidate contributions and 1,091 judgments, each marking an ideal turn, an acceptable window, or that a contribution should not appear, from 89 participants engaged in 35 conversations. Evaluating 15 models across 28 configurations, we find that while models can place contributions well when given the entire conversation history, they systematically chime in too early when asked to decide live as the conversation unfolds: best-turn accuracy falls from 57.8% to 19.0%, with nearly half of placements occurring before every participant’s acceptable window. In comparison, lay humans deciding live are more accurate (23.7% vs 19.0%), and fewer of their placements are premature (39.7% vs 46.4%). Model reasoning traces show that 93% of intervention justifications cite relevance, while almost none consider waiting. By providing a human-grounded benchmark, our work reveals gaps in agents’ ability to integrate into group settings and points towards proactive agents that can reason explicitly about appropriate timing.
[AI-85] Motor compound fault diagnosis using an attention-based cross-talk architecture
链接: https://arxiv.org/abs/2610.25893
作者: Wonjun Yi,Wonho Jung,Junho Kim,Rismaya Kumar Mishra,Yong-Hwa Park
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Submitted to ICSV33
Abstract:As physical AI increasingly relies on sensing to understand physical systems, reliable condition monitoring of industrial machinery has become more important. Motors are ubiquitous in industrial systems, and increasing system complexity raises the likelihood of simultaneous faults. Accordingly, compound fault diagnosis that identifies multiple fault components and their severities is increasingly required. Because different fault components jointly affect the same vibration signal, their severity classification can be formulated as related tasks, making multi-task learning (MTL) suitable for exploiting shared information. However, conventional shared-trunk architectures force tasks to share a common representation, potentially causing negative transfer because fault components are characterized by different time-frequency regions. To address this issue, this study proposes a Residual Neural Dimension Reductor with Time-Frequency Channel Attention (RNDR-TFCA), an attention-enhanced cross-talk architecture for multi-output compound fault diagnosis. RNDR-TFCA preserves task-specific feature extractors while selectively exchanging information across tasks. Time-frequency attention emphasizes informative intra-task regions, while channel attention enhances useful inter-task information after nonlinear feature fusion. Experiments on a rotor-bearing system with inner-race fault, outer-race fault, misalignment, and unbalance showed that RNDR-TFCA achieved a macro F1 score of 0.898, outperforming single-task, multi-class, shared-trunk, and cross-talk baselines. Ablation experiments confirmed the benefit of jointly applying both attention mechanisms. RNDR-TFCA required 713 K parameters and 170.42 MMACs, only slightly higher than RNDR. Channel-expanded baselines still underperformed RNDR-TFCA, indicating that the gain was not simply attributable to increased model capacity.
[AI-86] Manifold-Constrained Residual Adaptation with Frame-Adaptive Layer Aggregation for Speaker Verification
链接: https://arxiv.org/abs/2610.25892
作者: Zezhong Jin,Xiaoyu Wang,Zhe Li,Chong-Xin Gan,Siqing Qin,Man-Wai Mak,Kong Aik Lee
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注: 13 pages, 8 figures
Abstract:Large self-supervised speech models provide powerful representations for speaker verification, but full fine-tuning is expensive. Most conventional adapters insert small trainable modules to modify the output of each layer, while keeping the original residual connections unchanged. Building on our preliminary study of manifold-constrained Hyper-Connections in speaker encoders, we explore parameter-efficient adaptation of both residual paths and feature selection across layers. We first propose a Manifold-Constrained Residual Adapter (MCRA), which splits the hidden dimensions of each Transformer residual state into N groups and uses a small Sinkhorn-normalized matrix to exchange information among them. A learnable gate controls the balance between the original and mixed residual paths. The gate adds only one scalar parameter at each residual connection and initially favors the original path, allowing gradual adaptation. We further propose Frame-Adaptive Layer Aggregation (FALA), an attention-based aggregation method that uses a single learned query shared across all utterances to compute layer weights separately for each speech frame. Because the layer representations depend on the input, the resulting weights vary across both utterances and frames, unlike the fixed weights of a conventional layer-weighted sum. Compared with a frozen WavLM baseline using a fixed layer-weighted sum, our method achieves relative EER reductions of 22.4% on VoxCeleb1-O and 2.9% on CN-Celeb, with trainable adaptation parameters equivalent to only 0.0009% and 0.0016% of the WavLM-Large parameters, respectively.
[AI-87] CODA: Cohort- and Drift-aware Foundation Model for Multimodal Clinical Reasoning NEURIPS2026
链接: https://arxiv.org/abs/2610.25871
作者: Changshuo Liu,Jiaqi Zhu,Wenqiao Zhang,Xiaokui Xiao,Beng Chin Ooi
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted by NeurIPS 2026
Abstract:Recent advances in multimodal foundation models have demonstrated strong performance across diverse clinical tasks. However, existing approaches predominantly operate at the individual sample level and fail to model the interplay between population-level cohort patterns and patient-specific variation that underlie real-world clinical reasoning. In practice, clinicians routinely reason through patient cohorts sharing similar demographic, physiological, or pathological characteristics, while accounting for patient-specific deviations from the group prototype. Motivated by this gap, we propose CODA, a cohort- and drift-aware foundation model for multimodal clinical reasoning. CODA explicitly integrates cohort structure into the modeling pipeline through two complementary mechanisms: cohort-aware population stratification and drift-aware individual adaptation. In particular, we leverage reinforcement learning to discover clinically meaningful patient cohorts, and further model each patient’s deviation from cohort prototypes as a structured drift signal encoding fine-grained individual heterogeneity. These cohort- and drift-aware representations are incorporated into the foundation model to enable reasoning that is simultaneously informed by population-level patterns and sensitive to individual-level variation. Extensive experiments on real-world electronic health record (EHR) benchmarks demonstrate that CODA achieves consistent effectiveness and strong generalizability across diverse clinical tasks, including closed-QA, open-QA, and report generation.
[AI-88] Benefit-Targeted Routing and Bounded Revision for LLM -Based Evidence Screening in Manufacturing Supplier Discovery
链接: https://arxiv.org/abs/2610.25867
作者: Yijiashun Qi,Hanzhe Guo,Yuxuan Li
类目: Artificial Intelligence (cs.AI)
备注: Accepted for publication by MLBDBI 2026
Abstract:An evidence-review system must allocate additional model judgments and control their use. We study these software decisions on 6,164 frozen manufacturing source-task pairs. Benefit-targeted routing learns recovered agreement minus newly introduced disagreement. On the original 300-case test, 60 consultations yield 202/231 agreements with two human references versus 189/218 for a logistic-risk router. An additional 300 cases were independently annotated twice using equal frozen evidence. Without retraining the historical policies, the corresponding counts are 164/190 versus 163/193; the differences do not resolve an advantage. Strict scalar-Ridge controls share preprocessing, regularization and fitting, and remain inconclusive. Quotation and requested-state replacement gates retain rejected cases and budget costs; their joint gate rejects one original and two additional consultations without changing primary outcomes. Evidence ablations, fusion and bounded updates supply further controls. A completed 200-pair identity audit yields 97.45 percent weighted merge precision after eight author adjudications. This audits a provisional map, while prospective company-disjoint performance remains untested.
[AI-89] Minimum-Cost Robust Evidence Certificates for Horn Reasoning under Correlated Source Failures
链接: https://arxiv.org/abs/2610.25865
作者: Yukun Yang
类目: Artificial Intelligence (cs.AI)
备注: 9 pages, 2 figures
Abstract:A small justification for a rule-based conclusion may disappear when a source is withdrawn, and multiple derivations need not provide independent evidence. We study minimum-cost certificates that preserve a Horn conclusion under every withdrawal of at most k source domains. Evidence records remain distinct from asserted facts, supporting replicated attestations and overlapping failure dependencies. Transversals of minimal-support footprints characterize robustness and explain why disjoint-proof requirements can reject valid certificates. An exact procedure alternates weighted evidence selection with a source-failure oracle. A rule-closed counterinterpretation yields a globally valid selection cut without enumerating supports or failure scenarios. We prove correctness, finite termination, and cost optimality, establish coNP-completeness of unrestricted verification and NP-completeness of synthesis for every fixed budget, and identify a tractable reachability restriction. Experiments on controlled benchmarks, reachability graphs, and ontology-derived instances demonstrate that our counterinterpretation-guided synthesis achieves up to 92% optimum coverage within practical resource bounds, outperforming scenario expansion and general abduction while confirming the necessity of transversal-based reasoning over proof-packing heuristics.
[AI-90] Adaptive Decoding Methods for Token-Efficient Reinforcement Learning
链接: https://arxiv.org/abs/2610.25860
作者: Zhaonan Li,Zelei Cheng,Amritansh Mishra,Chia-Hsuan Lee,Mann Patel,Sambit Sahu,Ben Zhou,William M. Campbell
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Rollout decoding affects both which trajectories are used for reinforcement learning with verifiable rewards (RLVR) and how many completion tokens are spent collecting them. We study whether candidate-token support should remain fixed or widen during training in the context of rollout sampling efficiency. Our main schedule moves from top-K=10 to K=20 and then to full support as the attempted-token budget increases. In a 4B code-RLVR setting, the schedule has a higher interpolated 150M-token validation value than its matched fixed-K=10 continuation in all three runs, with a mean gain of 1.80\pm0.42 points. The schedule also performs comparably to a separately trained fixed-K=20 baseline, the strongest fixed-support setting in our sweep. Follow-up experiments suggest that late widening to full support is the most promising variant, although the evidence remains noisy, and mixed-reward-group diagnostics do not clearly explain the gain. An exploratory single-run test replay yields identical point estimates for the schedule and fixed K=10 . Overall, our results show that rollout support is an important control for sampling efficiency in RLVR and motivate further study of adaptive support schedules.
[AI-91] Are Self-Evolving Agents Continual Learners?
链接: https://arxiv.org/abs/2610.25858
作者: Xiaoyin Chen,Yizheng Huang,Yinglun Xu,Ziyun Zhao,Emmett McQuinn,Julie Liu,Hope Yao
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Large language model agents are often deployed in environments where new knowledge, tools, and workflows keep arriving. Such an agent is therefore expected to adapt continually, learning new tasks as they arrive without forgetting earlier ones. This resembles the continual learning problem, but at a pace that conventional gradient updates cannot sustain. Nonparametric self-evolving agents are a natural alternative: they adapt by rewriting their own harness, such as prompts, memory, and skills, from task feedback while the model weights stay frozen. We study whether such agents are effective continual learners across four drifting task streams that shift in topic, workflow, or domain (built from BIRD, BFCL, and ScienceWorld). Evaluating three representative methods spanning experience retrieval and self-modification (ReasoningBank, GEPA, and Meta-Harness), we find that although self-evolution improves aggregate performance on most streams, higher plasticity consistently comes at the cost of stability. Specifically, self-modifying methods apply global edits to a shared prompt or workspace, interfering with unseen tasks and overwriting earlier knowledge as the stream shifts. Conversely, similarity-based memory retrieval localizes updates across distant domains, but still retrieves foreign memories on unseen tasks and suffers from retrieval collision when similar queries require conflicting workflows. These findings show that current self-evolving agents remain incomplete continual learners.
[AI-92] MASIF: Implicit Failures for Process Observability in Multi-Agent Systems
链接: https://arxiv.org/abs/2610.25852
作者: Yanwei Huang,Arpit Narechania,Atanu Sinha
类目: Artificial Intelligence (cs.AI)
备注: Preprint
Abstract:Multi-agent systems (MAS) can complete a task despite local execution errors that are corrected, bypassed, or carried forward without individually determining its outcome. We call these implicit failures (IFs) and study them as signals for process observability beyond final task outcome. Across 1,828 traces from nine MAS-benchmark configurations, IFs occur in 54.3% of successful and 75.2% of failed traces. We develop taxonomies of error types and subsequent absorption mechanisms, supported by human and cross-model annotation audits. Counterfactual replay further probes how repairing an IF changes the final outcome. To examine the practical value of IFs, we evaluate IF-based best-of-N trace selection against existing process reward models, and introduce a novel RL algorithm, IF-GRPO, that uses IF costs to refine preferences among successful trajectories in GRPO. Experiments show that IF-GRPO exhibits some gains in accuracy and process metrics. Importantly, it exposes trade-offs between task performance, IF count and cost, and token-normalized IF density. Overall, our results show that IFs provide complementary process observability into how MAS reach their outcomes, including error recovery, propagation, and IF-based process differences that final task success alone does not expose.
[AI-93] Rubrics Beyond Evaluation: Exploiting Rubric Structure in Reinforcement Learning
链接: https://arxiv.org/abs/2610.25850
作者: Yuxun Huang,Yunyao Zhang,ZeLiang Chen,Yi-Ping Phoebe Chen,Junqing Yu,Zikai Song
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Rubrics are widely used to evaluate open-ended question answering by breaking overall quality into explicit criteria. These explicit criteria make rubrics a rich source of learning guidance, yet their potential for guiding the training process has been largely overlooked. Existing methods use rubrics mainly as direct feedback signals, without explicitly modeling the capabilities behind their criteria. This leaves their capability structure largely unused in guiding the training process. We introduce Rubric-Structured Reinforcement Learning (RSRL), an RL framework that unlocks the full potential of rubrics by exploiting their structure throughout the training process, rather than treating them simply as direct reward signals. At its core, RSRL builds on three key designs: (1) Rubric-to-Capability Structuring first recovers the capability structure hidden in rubric criteria, (2) Self-Sufficient Capability Learning then uses this structure to bootstrap underdeveloped capabilities into effective learning, (3) Capability-Residual Credit Routing finally preserves capability-specific gains during optimization. Experiments show that RSRL consistently improves both overall performance and underdeveloped capabilities, boosting downstream scientific question-answering performance by up to 7.43 over the initial policy and 7.60 over non-rubric structured RL baselines, showing that reinforcement learning can substantially benefit from the capability structure encoded in rubrics.
[AI-94] A Cues Hidden Price: Auditing Seller Influence on AI Shoppers
链接: https://arxiv.org/abs/2610.25843
作者: Felix Buecheler,Gerardo Berbeglia,Mingyu Guo
类目: Artificial Intelligence (cs.AI)
备注: 22 pages, 5 figures, 25 tables
Abstract:Large language models are increasingly asked to shop on a user’s behalf, which lets seller-controlled content such as badges, endorsements, and framing steer what they recommend. That poses the question at the core of generative engine optimization (GEO): not whether seller content shifts an agent’s choice, but how much. The usual gauge of that influence, how often a cue wins the choice, has a blind spot: once a cue is strong enough to be picked almost every time, the choice rate can no longer show how much stronger one cue is than another. We introduce price-equivalent (PE) auditing, which removes this ceiling by pricing the cue in money. In a counterbalanced, position-controlled forced choice, we raise the cued product’s price until it is chosen only half the time, and report that premium as the cue’s price-equivalent: a formally defined, policy-conditioned parity threshold that makes very different cues comparable. Priced this way, cues that look equally strong at equal price differ by a wide margin, so choice-rate audits substantially under-resolve seller influence. We run the audit over 113,472 trials, seven models across families, and two product categories. To our knowledge, we also run the first price-equivalent audit of a decision-native model, Jev, which outputs a choice probability directly instead of generating text; strikingly, its profile is unlike any text model we tested, resisting most cues yet remaining susceptible to the strongest social-proof claims. We release the stimuli, logs, and code as a benchmark that reports each model’s cue-price profile rather than a single score, giving developers and auditors a common, reproducible way to measure how content moves AI shopping agents.
[AI-95] Not Every Rollout Needs Distillation: Correctness-Filtered On-Policy Self-Distillation
链接: https://arxiv.org/abs/2610.25842
作者: Harrison Zhao,Lei Huang,Fei Liu,Wen Shi,Zijian Jin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy self-distillation (OPSD) supervises student-generated trajectories with a reference-conditioned version of the same base model. We ask whether every trajectory supplies distinct supervision that warrants its distillation cost. Gradient measurements on Qwen3-1.7B at initialization reveal that correct and incorrect responses have similar individual gradient magnitudes, but incorrect responses have approximately three times the coherent gradient share. Their estimated population-mean directions also largely overlap. A signal–noise model fitted at a small group size tracks alignment at larger group sizes without refitting. Motivated by these measurements, we introduce correctness-filtered OPSD (C-OPSD), which generates one response per problem and distills only a subset of incorrect responses. Removing unselected rows before both distillation forwards avoids their teacher/student forward and student backward work. Across Qwen3-1.7B, 4B, and 8B on AIME24, AIME25, and HMMT25, C-OPSD achieves on-par performance with full OPSD. In a measured micro-batch-size-four implementation, executed rows decrease by approximately 68%, and with retained rows packed into full micro-batches, training takes about 40% less time while rollout generation remains unchanged. Together, the measurements and training results connect outcome-conditioned supervision redundancy to more compute-efficient distillation.
[AI-96] OV-PEAR: Observation-Verified Pose-Scale Estimation and Reconstruction for Unknown Industrial Manipulation Objects
链接: https://arxiv.org/abs/2610.25840
作者: Guoqin Tang,Qingxuan Jia,Yichen Tan,Zihang Lv,Ruiquan Wang,Zeyuan Huang,Gang Chen
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Fine industrial robotic manipulation requires 6D pose, scale, and an operation-ready 3D model. Existing industrial pose estimation relies on CAD models, object-specific training data, or stable workcells, limiting use with unseen objects; zero-shot methods suffer from single-view occlusion, generative bias, and scale uncertainty. We present OV-PEAR, an observation-verified framework for 3D reconstruction and pose-scale estimation of unknown industrial objects. Its three-stage object-centric pipeline: (1) segments RGB-D instances and generates object-level 3D hypotheses; (2) anchors them to the robot frame via appearance-aware view retrieval, explicit scale search, and visibility-sensitive registration; (3) back-projects multi-view RGB-D evidence into the object-canonical frame to verify voxel occupancy, suppress hallucinations, and produce reusable trusted object models. On BOP-Industrial, OV-PEAR achieves 0.657 average AR and improves pose and voxel-level geometric consistency in real scenes; real-robot connector assembly reaches 99.67% grasp and 84.00% assembly success. The model library and pose-scale log enable short-term object-level state maintenance at 0.02 ms/frame.
[AI-97] FrontierRefactor: Benchmarking Multi-Scale Performance-Oriented Codebase Refactoring
链接: https://arxiv.org/abs/2610.25834
作者: Shijian Wang,Fanqing Meng,Jiajun Chen,Qiguang Chen,Linxin Song,Runhao Fu,Shuhang Zheng,Shuai Shao,Qizhe Xie,Jieyu Zhao,Mengkang Hu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Autonomous codebase refactoring requires coding agents to restructure working software while strictly preserving externally observable behavior and prioritizing execution performance and compute efficiency over stylistic preference. Existing benchmarks: refactoring benchmarks assess functional correctness while overlooking runtime regressions, whereas code performance benchmarks target isolated micro-kernels that ignore codebase-scale complexity. We observe that refactoring scale is governed by the scope of the observable contract, the caller-visible interface fixing behavioral invariants while freeing private implementations for optimization. Building on this insight, we introduce FrontierRefactor, a benchmark of 60 qualified engineering tasks spanning four scales, from computational kernels to full repositories, across in-place optimization and cross-language rewriting. Evaluating frontier coding agents through a reference-free cascaded protocol across 1,080 execution runs reveals that correctness and speedup are largely decoupled: 41.2% of submissions satisfy functional equivalence yet fail to deliver verified performance improvements, with 19.3% suffering performance regressions and 21.9% trapped in measurement noise. Refactoring paradigms reveal orthogonal engineering bottlenecks: while cross-language rewrites gain speedups from native compilation after passing functional checks, agents struggle with contract fidelity, with pass rates dropping from 65.0% on modules to 32.6% at package scale; conversely, in-place optimization reliably preserves behavioral contracts at 86.5% but rarely achieves acceleration without physical performance profiling. Trajectory analysis shows that agents fall victim to uncalibrated micro-benchmarking illusions, an operational blind spot that multi-turn execution-timing feedback directly resolves, raising strict success from 16.0% to 40.0%.
[AI-98] Optimizing Marketing Budgets for Social Influence: A Generalized Fractional Incentives Framework
链接: https://arxiv.org/abs/2610.25823
作者: Abhishek K. Umrawal,Akhil Bhimaraju,Eliot W. Robson,Lav R. Varshney
类目: ocial and Information Networks (cs.SI); Artificial Intelligence (cs.AI)
备注: 28 pages, 2 figures, and 2 tables
Abstract:Marketing managers routinely allocate limited budgets across individuals in social networks whose responsiveness and influence vary substantially. We study the problem of Budgeted Generalized Fractional Influence Maximization (BGFIM), which captures the allocation of continuous levels of marketing spend across users in the presence of heterogeneous response and network spillovers. In this framework, adoption probabilities are governed by user-specific response functions that capture both baseline propensity and marginal responsiveness, while incentive costs vary across users. We propose a scalable solution approach based on a discrete greedy procedure with provable approximation guarantees, enabling tractable optimization under realistic budget constraints. Our formulation generalizes prior influence maximization models by allowing heterogeneous response and cost structures, providing a more flexible representation of targeted marketing interventions. Using experiments on multiple large-scale networks, we show that continuous allocation strategies can outperform standard heuristics and improve influence outcomes under heterogeneous conditions, and we illustrate how the framework can support profit-oriented decision making by identifying budget levels that balance marginal returns and costs. Overall, our results highlight the value of fine-grained budget allocation in networked marketing environments and offer a scalable approach for data-driven marketing resource allocation.
[AI-99] When to Stop Asking: Factor Coverag e Bias and LLM Determinability Judgments with Underspecified Information in Medicine
链接: https://arxiv.org/abs/2610.25819
作者: Yusuke Watanabe,Yohei Kobashi,Taisei Tosaki,Takeshi Kojima,Yusuke Iwasawa,Yasushi Okuno,Yutaka Matsuo
类目: Artificial Intelligence (cs.AI)
备注: 17 pages, 8 figures, 8 tables
Abstract:Clinical decision-making often begins with underspecified information and requires not only gathering missing information but also knowing when to stop. This calls for judging determinability (whether the available information is sufficient for a decision) by considering whether missing information could change the conclusion, not simply the amount of available decision-relevant information. Reliance on this amount as a cue for determinability, which we term \textitfactor coverage bias, could undermine the trustworthiness of large language models (LLMs) through premature conclusions or excessive information gathering, such as unnecessary invasive tests. Yet whether LLMs exhibit this bias remains insufficiently studied. To address this gap, we construct a controlled pairwise benchmark in which LLMs identify the determinable case within each determinable–undeterminable pair. We add a decision-relevant factor to either side while keeping each case’s determinability unchanged, and similarly add irrelevant text to distinguish this bias from context-length effects. Adding a relevant factor made LLMs more likely to select the augmented case as determinable even when it remained undeterminable, an effect that persisted with increased reasoning effort, while adding irrelevant text made them less likely to select the augmented case, an effect that reasoning substantially reduced. These findings identify factor coverage bias as a failure mode distinct from context-length effects and a potential risk to LLM trustworthiness. Our work provides an evaluation framework for LLM trustworthiness under underspecification.
[AI-100] SAR: Learning Global Item-Space Matching and Geometry-Aware Placement Ranking for Buffered Online 3D Bin Packing
链接: https://arxiv.org/abs/2610.25795
作者: Xinpeng Lei,Jin Zhang,Yizhe Yang,Yihong Duan,Fucheng Wan
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Online 3D bin packing with a bounded buffer requires irreversible item-selection and placement decisions under limited knowledge of future arrivals. Feasibility alone is insufficient, as feasible actions can produce different residual-space structures and subsequent packing opportunities. A central architectural question is how to allocate spatial reasoning between item selection and placement evaluation. We propose Spatial Affordance Retrieval (SAR), a novel learning-based framework that coordinates global item-space matching with geometry-aware placement ranking. A voxel-token spatial encoder represents 3D container occupancy, enabling a picker to match buffered items to the residual space without requiring item-specific placement proposals as inputs. After item selection, a loader ranks feasible rotation-position candidates using candidate geometry, local geometric features (LGF), and inter-candidate comparison, without directly accessing global spatial tokens. The feasible candidate set connects the two stages, which are jointly optimized under a shared long-horizon packing return. On the Random, JD, and SF datasets, the best SAR variants improve utilization over the strongest evaluated online baselines by 0.96-4.20 percentage points under the no-buffer setting (K=1). With a buffer (K=5), the best utilization rates across SAR variants reach 93.55%, 91.04%, and 90.41%, respectively. Ablations support the complementary roles of global item-space matching and geometry-aware placement evaluation, counterfactual analysis suggests that learned item selection favors greater future packing potential, and experiments demonstrate compatibility with multiple candidate generators.
[AI-101] Can LLM User-Proxies Stand In for Humans in Reliance Measurement? Replicating Rel-A.I. with Simulated Users
链接: https://arxiv.org/abs/2610.25788
作者: Shuai Wang,Rui Hu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Can language-model user proxies reproduce human reliance on an AI answer across contexts? Published simulated-user studies do not establish validity for this partial-answer decision, and Rel-A.I.'s released human record omits trial-level choices. We audit its 90 printed questions and 15 domain-by-marker reliance rates using sampled decisions and counterbalanced A/B probabilities from open models through 72B parameters. A matched crossover varies the printed agent specialty within each question, separating question content, label averages, and their interaction. Mistral Small’s strong-marker math-minus-law label marginal is -0.4 percentage points, yet matching the agent label to the question domain raises reliance by 13.3 points on average across the five domains. The best A/B proxy has 22.5-point absolute-rate error and 8.3-point domain-effect error against the human table. A two-sample independent-binomial human replication has 3.5-point expected domain-effect error under the stated working model. The interaction is model dependent, and inference layout changes identical-prompt probabilities. This single-study audit does not establish proxy validity for human effect sizes or general directional screening.
[AI-102] From Trace to Fix: Residual Task State in Long-Horizon Agents
链接: https://arxiv.org/abs/2610.25776
作者: Zile Huang,Ser-nam Lim
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Recovering from failures in long-horizon, stateful language agents is challenging because the action introducing a defect, the symptom exposing it, and the persistent state requiring repair frequently diverge. Existing methods largely focus on retrospective failure attribution, leaving sequential diagnosis and state repair poorly integrated under constrained budgets. We introduce TRACE-FIX, a unified recovery framework centered on versioned residual task state. Trace transforms execution logs into a structured dependency-linked ledger, prioritizing unresolved task requirements rather than salient surface errors. Fix formulates recovery as budgeted planning over admissible actions, incorporating outcome forecasting and explicit repair-reserve constraints to prevent premature budget exhaustion. To rigorously decouple localization accuracy from downstream repair capability, we present LongH-Debug, evaluating root-cause attribution, verified-fault recovery, and end-to-end closed-loop execution independently. Our empirical analysis demonstrates that structured residual tracking substantially improves causal localization over raw-trace baselines, while budget-aware sequential planning converts verified fault signals into more reliable task recovery without destructive trial-and-error. Code is available at this https URL.
[AI-103] EvalRx: Agent ic Diagnosis and Verified Repair of Foundation Models
链接: https://arxiv.org/abs/2610.25767
作者: Jiaqi Liu,Ruinan Jin,Yitong Tang,Beidi Zhao,Gexin Huang,Qiong Zhang,Xiaoxiao Li
类目: Artificial Intelligence (cs.AI)
备注: 23 pages. Technical report. Code and artifacts: this https URL
Abstract:Model evaluation reveals where models fail, but rarely why or whether a proposed fix genuinely works. We introduce EvalRx, an open-source agentic framework for statistically grounded model diagnosis and repair. EvalRx discovers systematic failure modes, explores candidate mechanisms, formulates falsifiable hypotheses, and verifies them by executing frozen, pre-specified tests before using supported diagnoses to guide tiered repair. Candidate repairs are validated against the unchanged model and promoted only when they improve the target behavior without unacceptable regression. Across language, vision-language, and audio-language settings spanning Gemini, Qwen, Gemma, and Nemotron, EvalRx improves performance in all 12 evaluated model family-modality combinations, with gains ranging from +1.4 to +10.9 percentage points (+5.8 points on average). These results show that model evaluation can serve not only to measure failures, but to support an auditable process for diagnosing and repairing them.
[AI-104] RACE: Tracking Multimodal Clinical Trajectories for Inpatient Risk Prediction
链接: https://arxiv.org/abs/2610.25765
作者: Minjun Kim,Jong Hak Moon
类目: Artificial Intelligence (cs.AI)
备注: 14 pages, 4 figures
Abstract:Accurate inpatient risk prediction requires modeling heterogeneous clinical evidence that accumulates irregularly over time. This challenge spans in-hospital mortality, prolonged length of stay, and readmission, each with task-specific observation windows. Existing multimodal approaches combine structured health record data, chest x-rays, and clinical text, but primarily focus on modality fusion or missing-modality compensation, with limited attention to temporal trajectories over an admission. We propose TRACE (TRAjectory modeling of multimodal Clinical Events), a daily-scale framework organizing clinical events into day-level trajectories, separating day-invariant global context from day-specific local events drawn from EHR data, chest x-rays, radiology reports, and discharge summaries. The trajectory is encoded using temporal attention, allowing unavailable modalities to emit no tokens. A phenotype-guided contrastive objective further organizes admissions under heterogeneous disease distributions. Across cohorts from MIMIC-IV, eICU, and MIMIC-III comprising nearly 180,000 total stays, we evaluate TRACE on mortality, length of stay, and readmission prediction. TRACE achieves the highest AUROC across all nine dataset-task settings and the highest ACC in eight, outperforming baselines. Ablation studies validate our attention masks and contrastive learning, while trajectory-level attribution localizes risk-associated hospital days and modalities.
[AI-105] ActTime: Action-Conditioned World Modeling for Time Series
链接: https://arxiv.org/abs/2610.25764
作者: Xinxing Zhou,Xiaojie Yuan,Yanlong Wen,Ming Jin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Time-series models increasingly condition on known future inputs. In controlled systems, some of these inputs are actions that can be changed, compared, and optimized before execution. A model must therefore do more than predict the realized future: it must predict how outcomes change under different actions from the same observed history and support reliable comparison of candidate plans. We introduce ActTime, an action-conditioned world model for time series. Over the observed history, ActTime updates variable-specific latent states through an action-conditioned transition and corrects them using observations. It then rolls the same transition forward under candidate actions to generate future trajectories. Training uses only factual trajectories, without paired alternative-action outcomes or planning supervision. We construct a simulator-grounded evaluation suite across building energy, traffic, greenhouse climate, and inventory management, covering factual prediction, intervention-effect prediction, offline decision making, and closed-loop planning. ActTime ranks first in 12 of 16 domain-task settings. Across eight methods, the mean pairwise model-ranking reversal rate is 17.8% between factual prediction and intervention fidelity, and rises to 28.4% between intervention fidelity and decision ranking. These results show that factual accuracy, intervention fidelity, and decision quality are related but not interchangeable.
[AI-106] Remember Less Adapt Longer: Bounded-State In-Context Reinforcement Learning with Recurrent Summary Memory
链接: https://arxiv.org/abs/2610.25758
作者: Jiawen Wang,Cheng Xue,Peng Zhang,Jochen Renz
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:In-context reinforcement learning (ICRL) enables agents to adapt to new tasks without parameter updates, relying on memory to retain what they learn from experience. For agents operating continually, existing approaches using memory trade off bounded cost against access to past experience. Approaches using growing memories, such as full-history attention and accumulating summaries, incur storage and per-decision costs that increase with task length. Fixed-size memories bound storage but restrict access: sliding windows discard older steps, while recurrent networks compress even recent interactions into a hidden state. We introduce Recurrent Summary Memory (RSM), a Transformer policy that attends exactly to a short working buffer of recent interactions and carries a fixed-capacity summary of earlier experience. When the buffer fills, RSM rewrites the summary from its previous contents and the buffer, then clears the buffer, so deployment memory and per-decision cost stay O(1) in task length. On long-running partially observable tasks, frozen RSM policies sustain their adapted performance for eight times the training horizon with about 1% of full-history memory, ending ahead of every evaluated baseline while growing memories deteriorate. When the task configuration changes mid-run, RSM re-adapts better than growing and recurrent memories. Clearing the summary or blocking the policy’s access to it removes its benefit, while swapping in another task’s summary steers behaviour towards that task. Together, these results show that our fixed-capacity summary can sustain in-context adaptation far beyond the training horizon.
[AI-107] Cognitively Decoupled Meta-Induction for Out-of-Distribution Generalization
链接: https://arxiv.org/abs/2610.25749
作者: Yuqian Wang,Yuanzheng Wang,Chaojun Xiao,Ke Xu,Bingxiang He,Jinghan Tan,Zijun Chen,Lu Chen,Yi-Xin Huo,Maosong Sun
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 21 pages, 2 figures
Abstract:In-context learning (ICL) tackles downstream tasks conditioned on demonstrations. This paradigm relies heavily on neighbor-based matching between queries and demonstrations, limiting out-of-distribution (OOD) performance. For better generalization, rule-based inductive reasoning derives task-level regularities, which we term task-level inductive representations (TIR), before directly answering. However, existing methods lack generalizable inductive ability across diverse tasks. Inspired by meta-learning, we propose Cognitive-Decoupling meta-Induction (CoDI) for TIR-based inductive reasoning on general tasks. CoDI comprises a Cognitive-Decoupling Framework for query-blind TIR induction, and a Dual-Loop Meta-Inductive Learning method for cross-task inducer optimization. Empirically, CoDI achieves substantial performance gains on five ID and OOD benchmarks, and preserves OOD gains without explicit induction at inference. Code is available at this https URL.
[AI-108] melineAgent : An LLM -based Framework for Query-Centric Event Timeline Extraction Across Multiple Documents EMNLP2026
链接: https://arxiv.org/abs/2610.25733
作者: Qingkai Min,Yunxin Bao,Yifeng Lu,Zhenyu Feng,Yulong Chen,Andreas Vlachos,Yue Zhang
类目: Artificial Intelligence (cs.AI)
备注: EMNLP 2026 Main
Abstract:Query-centric cross-document event timeline extraction aims to construct event timelines from multiple documents under diverse query focuses. The task is very challenging due to the heterogeneous event modeling and reasoning requirements arising from both the intrinsic complexity of cross-document timeline construction and the diversity of query focuses. In this paper, we propose TimelineAgent, an LLM-based multi-agent framework featuring a multi-stage reasoning workflow, where heterogeneous event features are organized into reusable skills and orchestrated through a goal-oriented protocol. To facilitate evaluation, we construct a benchmark covering diverse query focuses. Experimental results show that TimelineAgent consistently outperforms existing methods, offering a practical framework for complex real-world timeline construction.
[AI-109] Anatomy of Off-Policy Policy Gradient: Importance Sampling KL Regularization and Baselines
链接: https://arxiv.org/abs/2610.25725
作者: Haoqun Cao,Yurun Yuan,Tengyang Xie
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning has become a central tool for large language model (LLM) post-training, where policy gradient methods are routinely deployed off-policy, even though vanilla policy gradient assumes on-policy sampling. We study when such off-policy deployment is justified, and what roles its core ingredients, including importance sampling, KL regularization, and baselines, play in handling the resulting distribution shift. We show that for a general class of off-policy policy gradient updates, consistency, in the sense that the optimal policy remains a stationary point, requires one of two structures at the optimal policy: trajectory-level importance weighting, which suffers from the well-known curse of horizon, or KL regularization with a baseline equal to the optimal soft value, instantiated by the grouped mean. Focusing on the KL route, we identify it with the group-relative policy gradient (GRPG) update, establish an equivalence to a trajectory-level Bellman residual minimization objective, and obtain a finite-sample regret bound under all-policy coverage. This also gives a theoretical account of a variant of the empirically successful GRPO objective. We further uncover a hidden role of baselines in the off-policy setting: a constant shift in the baseline implements pessimism, the standard remedy in offline RL for partial data coverage. We instantiate this via the asymmetric REINFORCE objective, and connect it to the game-theoretic offline RL framework, showing a regret bound with single-policy coverage. Together, these results offer a unified theoretical view of off-policy policy gradient in LLM post-training and connect it to classical ideas in offline RL.
[AI-110] Fine-Grained Units Flat Decisions: Hierarchical Micro-Expert Pruning for Mixture-of-Experts Models
链接: https://arxiv.org/abs/2610.25717
作者: Zukang Xu,Xinpei Gao,Xing Hu,Zhe Jiang,Dawei Yang
类目: Artificial Intelligence (cs.AI)
备注: 14 pages, 3 figures. Zukang Xu and Xinpei Gao contributed equally. Corresponding author: Dawei Yang
Abstract:Micro-expert pruning reduces the capacity of mixture-of-experts (MoE) models by removing coupled channels within routed experts. However, finer pruning granularity alone does not guarantee the preservation of downstream capabilities, as similar language-modeling perplexity can accompany substantially different task performance. We revisit micro-expert mask construction as two distinct decisions: allocating retained capacity across experts and selecting channels within that capacity. We propose UC-RAMP, a two-stage post-training pruning framework that explicitly separates these decisions. CAPE uses cross-domain calibration evidence to redistribute expert widths under an exact layer-wise budget, while RAMP refines channel composition at fixed widths through one-out, one-in exchanges guided by a channel-interaction surrogate. Across three MoE models at 25% and 50% routed-expert channel pruning, UC-RAMP achieves the highest ten-task mean accuracy among evaluated pruning methods in five of six settings. At 50% pruning, it outperforms the strongest evaluated baseline by 8.19, 3.96, and 6.14 percentage points across the three models, respectively. In a single-A100 deployment at 25% pruning, UC-RAMP reduces measured time to first token by 19.7–23.9% and time per output token by 20.2–22.3% relative to the unpruned model. The code is available at this https URL.
[AI-111] PFedQFT: Personalized Federated Quantum-Inspired Fine-Tuning for Medical Large Language Models
链接: https://arxiv.org/abs/2610.25712
作者: Qihan Huang,Lingxia Zhang,Guye Zhang,Zizhu Wang,Xiaowei Mao
类目: Artificial Intelligence (cs.AI)
备注: 10 pages, 5 figures
Abstract:Medical data remain fragmented across institutional silos because of privacy, governance, and ownership constraints. Federated learning enables medical institutions to adapt large language models (LLMs) collaboratively without transferring raw records. However, independently parameterized low-rank adapters incur repeated parameter and communication costs across Transformer projections, while uniform aggregation may limit adaptation to heterogeneous institutional data. We propose PFedQFT, a quantum-inspired framework for personalized federated fine-tuning with a compact shared adapter. Specifically, PFedQFT constructs reusable orthonormal adaptation bases from compact skew-symmetric generators using a truncated matrix-exponential approximation followed by QR orthonormalization. Compatible projections reuse these bases, while a globally shared full core preserves the flexibility of each projection. Each client further maintains a lightweight private residual operator, whose directions are learned globally and whose coefficients remain local. Experiments on seven public medical tasks and multiple LLM backbones show substantial reductions in trainable parameters and per-round adapter upload volume, with competitive predictive performance on most tasks.
[AI-112] Causal Humility: Governing the Authority of Fairness Claims Under Causal Uncertainty
链接: https://arxiv.org/abs/2610.25709
作者: Khadija Zanna,Rodrigo Ferreira,Frederick L. Oswald,Akane Sano
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:AI systems increasingly shape consequential decisions in employment, lending, healthcare, and public services, making fairness an important problem for auditing and governance. Fairness assessment often requires more than identifying disparities in outcomes; it also requires understanding how those disparities arise. Causal approaches to algorithmic fairness address this explanatory problem by representing pathways through which protected attributes and other factors may influence outcomes. Yet causal reasoning in sociotechnical systems remains uncertain: relevant confounders may be unobserved, causal directions contested, and conclusions sensitive to modeling assumptions and representational choices. We introduce causal humility as a governance framework for managing the relationship between epistemic uncertainty and causal authority. Drawing on causal fairness research, epistemology, political philosophy, and science and technology studies, causal humility has four requirements: making consequential assumptions and omissions visible; preserving contestability across defensible causal accounts; recognizing indeterminacy as a legitimate analytical outcome; and maintaining proportionality between evidentiary support and institutional authority or action. We operationalize these requirements through Causal Cards, a documentation artifact for recording the causal basis, assumptions, omissions, model dependence, and governance implications of fairness claims. Using the UCI Adult dataset as a worked epistemic demonstration, we show how different causal discovery procedures applied to the same observational data render fairness-relevant pathways visible, absent, or unresolved. Causal humility treats such disagreement as information about the limits of what can responsibly be claimed and the authority those claims should carry. Subjects: Computers and Society (cs.CY); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.25709 [cs.CY] (or arXiv:2610.25709v1 [cs.CY] for this version) https://doi.org/10.48550/arXiv.2610.25709 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-113] From Black Box to Controller: Inference-Time Steering of LLM via Feature Amplification
链接: https://arxiv.org/abs/2610.25700
作者: Keqi Chen,Zekai Sun,Yuhua Wen,Tianyi Tian,Xinyang Huang,Huijun Lian,Yingming Gao,Ya Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Inference-time steering is crucial for deploying Large Language Models (LLMs) in real-world applications, where they must flexibly adjust behavior preferences. For instance, a model should dynamically adjust its caution level based on the risk of the scenario. This requirement introduces a challenge: modulating the intensity of LLM preferences. Prompt engineering can require task-specific prompt design, while fine-tuning requires additional optimization and data. These requirements can make repeated behavioral adjustments costly. To address this, we introduce NeuroSteer, a plug-and-play framework that enables flexible, quantitative adjustment of LLM preferences at inference time. NeuroSteer employs a Sparse Autoencoder (SAE) as its controller. Comparing neuron activation frequencies between positive and negative examples isolates “feature neurons” that correspond to target preferences such as politeness and positive sentiment. During inference, the activations of selected SAE features are amplified in the latent representations to steer the output distribution. Users can adjust the strength of this steering on a per-scenario or per-user basis, granting the LLM fine-grained, quantifiable trait modulation. Our experiments show that NeuroSteer can alter sentiment, toxicity, stance, and politeness at inference time. It achieves higher target-direction steering scores than the compared baselines in several settings, with task- and strength-dependent trade-offs in generation quality.
[AI-114] CVaR-RMD: CVaR Selection Pressure for Robust Metaheuristic Discovery EMNLP2026
链接: https://arxiv.org/abs/2610.25697
作者: Ruodai Cui,Li Niu
类目: Artificial Intelligence (cs.AI)
备注: Accepted to Findings of EMNLP 2026. Code: this https URL
Abstract:Large language models have recently been used to discover heuristics through iterative program search, offering a promising approach to automated algorithm design. However, candidates selected by average performance may degrade across instance families, random seeds, and problem scales because mean-based objectives can obscure rare but severe failures. We propose CVaR-RMD, a risk-sensitive evolutionary framework that uses an LLM to mutate executable heuristic programs and ranks candidates by empirical conditional value-at-risk over heterogeneous evaluation conditions. Across three combinatorial optimization domains, CVaR-based selection improves upper-tail performance and reduces failure rates while maintaining competitive mean performance. These gains persist on larger unseen instances and independently curated within-domain benchmarks
[AI-115] oolOPD: Credit-Aware On-Policy Distillation for Tool-Use Agents
链接: https://arxiv.org/abs/2610.25693
作者: Chu Zhao,Lei Tang,Yanming Wang,Jianzhe Zhao,Guibing Guo,Zhengzong Chen,Yuanyuan Zhao,Fei Huang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy distillation (OPD) enhances the capabilities of small language agents by providing dense, token-level supervision from a teacher model on student-generated trajectories. However, standard OPD implicitly assumes that credit is uniformly distributed across long-horizon trajectories, assigning the same strength of teacher supervision to different actions. Through credit-distribution analysis and outcome-reversal counterfactual experiments, we find that trajectory credit is concentrated on a small number of critical actions. Moreover, even when the trajectory is held fixed, reversing only the final outcome substantially changes the credit assigned to multiple interaction turns. These findings show that the uniform-credit assumption fails to identify critical decisions and ultimately degrades the model’s reasoning accuracy. To address these challenges, we propose ToolOPD, an outcome-aware on-policy distillation framework designed for long-horizon tool use. ToolOPD first summarizes each completed trajectory and its outcome into a reusable hindsight skill. It then measures how this skill changes the teacher’s probability of each student action token and converts the change into outcome-aware credit. The resulting credit preserves supervision that supports successful decisions or discourages failed ones, while suppressing irrelevant and conflicting signals. ToolOPD uses these credits to reweight the action-level distillation loss and jointly optimizes it with reinforcement learning. Extensive experiments across multiple long-horizon tool interaction benchmarks demonstrate that ToolOPD significantly outperforms existing reinforcement learning and self-distillation baselines.
[AI-116] Progress-Guided Preference Generation for Aligning Generative Control Policies
链接: https://arxiv.org/abs/2610.25692
作者: Tung M. Luu,Tri Ton,Duc M. Nguyen,Xuan Thanh Nguyen,Nguyen Gia Binh,Quang Tan Nguyen,Vu Duong,Daniel Sonntag,Duy M. H. Nguyen,Vien Anh Ngo,Chang D. Yoo
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: CoRL 2026 (project site: this https URL )
Abstract:Generative policies have achieved strong performance in imitation learning tasks, but their success is often constrained by the limited availability of expert demonstrations. Although preference-based fine-tuning can leverage suboptimal rollouts to improve performance, existing approaches heavily rely on expensive human annotations. In this work, we introduce Progress-Guided Preference Generation (PGPG), an unsupervised preference alignment framework that eliminates the need for human-in-the-loop supervision. Our method leverages ordered coverage alignment to measure sequential progress relative to a small set of expert demonstrations, enabling the automatic extraction of pseudo-preference labels from autonomously collected rollouts. Furthermore, we introduce key enhancements to improve the reliability of these progress-guided preferences, enabling effective fine-tuning of pretrained generative policies without requiring manually designed reward functions. Extensive experiments across a suite of challenging long-horizon manipulation tasks demonstrate that PGPG outperforms standard supervised fine-tuning baselines. Moreover, our conceptually simple framework is broadly applicable across multiple classes of generative policies, demonstrating strong versatility when integrated with both diffusion and flow-matching policies.
[AI-117] MT-CC: Multi-Group Temperature Scaling for Asymmetric Calibration Behavior in Class-Incremental Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.25689
作者: Sungjun Yun,Yonghee Choi,Dong-Jun Han
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026 (Spotlight). 28 pages
Abstract:As deep neural networks are required to continuously adapt to evolving data, class-incremental learning (CIL) has become an active research area. However, most existing works primarily focus on improving accuracy while overlooking confidence calibration, which measures the trustworthiness of the model’s predictions. Although temperature scaling (TS) is effective for static calibration, it has rarely been explored in continual settings where data distributions evolve over time. In this paper, we introduce Multi-Group Temperature Scaling for Continual Calibration (MT-CC), a novel post-hoc calibration framework that addresses the asymmetric calibration behavior in CIL, where the rate of confidence changes does not align with the rate of accuracy degradation across tasks, leading to different levels of miscalibration. Our key contribution is to design multiple temperature parameters through a lightweight grouping network that separates samples by confidence statistics and task-aware signals in CIL. Furthermore, a Groupwise Difference-between-Confidence-and-Accuracy (GDCA) regularization is incorporated to promote intra-group consistency. Experiments demonstrate that MT-CC significantly improves calibration performance while maintaining accuracy, offering a principled framework for reliable continual learning.
[AI-118] When Safety Fails Collectively: Robust Alignment Against Harmful Fine-Tuning via Joint Weight Perturbations
链接: https://arxiv.org/abs/2610.25683
作者: Zhiyuan Wang,Yuezhi Zhou,Zhu Zhang,Zhixing Tan
类目: Artificial Intelligence (cs.AI)
备注: 15 pages, 3 figures
Abstract:Harmful fine-tuning can rapidly erode the safety alignment of large language models, creating substantial risks when aligned models are adapted to downstream tasks. Existing alignment-stage defenses use harmful examples and simulated harmful perturbations during safety alignment to prepare models for later harmful fine-tuning, yet the collective effects of harmful weight perturbations remain underexplored. We identify a collective failure pattern: disjoint parts of a harmful perturbation have limited effects in isolation, but their joint application increases the model’s harmfulness beyond the sum of their individual effects. Guided by this observation, we propose Joint Weight Perturbations (JWP), an alignment-stage defense that masks different parts of a harmful perturbation estimated from the current model and trains the model to respond safely in the resulting states. This exposes the model to different combinations of harmful perturbations and reduces its sensitivity to their accumulation during subsequent harmful fine-tuning. Experiments across downstream tasks and model families show that JWP offers stronger resistance to harmful fine-tuning than prior defenses while largely preserving task performance.
[AI-119] Neither Over- nor Under-Constrained: Evaluating LLM Personalization under a Constraint Qualification Audit
链接: https://arxiv.org/abs/2610.25676
作者: Xuran Li,Guanqin Zhang,Imran Razzak,Hakim Hacid,Eleanna Kafeza,Hao Xue,Flora D. Salim,
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Accurate evaluation of LLM personalization is hindered by current evaluation paradigms that over- or under-constrain the response space, yielding frequent false rejections and false acceptances. To reduce these false rates, we introduce ConQual-Audit, an audit-based framework that evaluates LLM personalization through a three-stage process: (1) extracting semantic constraints from user profiles and task-grounded reference; (2) qualifying constraints by pruning unnecessary constraints and recovering missing constraints; and (3) verifying candidate responses against the audited constraints via natural language inference. Across PrefEval and LaMP, ConQual-Audit outperforms nine baselines, attaining top AUCs (0.88 and 0.90) and the lowest false rate (0.21) on the leanest constraint sets at an amortizable audit cost. Crucially, this constraint audit is portable: integrating it into ExPerT and PREF helps revise smaller, higher-quality constraint sets to reduce their evaluation false rates.
[AI-120] LIRQuant: An Efficient Locally Isolated Rotation Quantization Method for Large Language Models EMNLP2026
链接: https://arxiv.org/abs/2610.25675
作者: Desheng Xu,Hanglin Li,Tianchen Song,Zhiyong Zhao,Kun Zhan
类目: Artificial Intelligence (cs.AI)
备注: Accepted to EMNLP 2026
Abstract:Post-Training Quantization (PTQ) enables efficient deployment of Large Language Models (LLMs), yet activation outliers remain a key challenge for low-bit quantization. Existing global transformation methods overlook heterogeneous channel sensitivity to quantization noise, leading to suboptimal accuracy-efficiency trade-offs. To address this, we propose \textbfLocally Isolated Rotation Quantization (LIRQuant), a sensitivity-aware framework that selectively applies transformations. LIRQuant identifies sensitive channels using a noise-weighted sensitivity metric, applying expressive learnable rotations only to these channels and lightweight fixed rotations to the remaining ones. This decoupled design reduces optimization complexity without sacrificing outlier smoothing capability. Extensive experiments across the LLaMA-2/3, Mistral, Qwen3, and Mixtral families demonstrate that LIRQuant achieves state-of-the-art accuracy under the challenging W4A4KV4 setting. Meanwhile, LIRQuant calibrates LLaMA-3-8B in only 25 minutes with 20.3 GB peak GPU memory, achieving the fastest calibration and the lowest memory footprint among recent learnable-rotation methods, and its fused inference kernels deliver up to 2.3\times prefill speedup and 3.5\times memory saving over FP16.
[AI-121] HyperTaxa: Multiscale Genome Representations for Generalized Microbial Taxonomic Inference
链接: https://arxiv.org/abs/2610.25673
作者: Akash Rao,Sushim Gupta,Vishalini Ramnath,Akhilesh Ramachandran,Sathyanarayanan Aakur
类目: Artificial Intelligence (cs.AI); Genomics (q-bio.GN)
备注: Accepted to BIBM 2026
Abstract:Reliable microbial taxonomic inference is challenged by the rapid growth of newly sequenced organisms and incomplete reference coverage. While alignment- and database-driven tools remain essential for reference-backed identification, they are not designed for learned inference over species absent from genome-level training. We introduce HyperTaxa, a taxonomy-aligned representation learning framework for open-set microbial taxonomic inference. HyperTaxa encodes whole genomes as multiscale hyperspectral k-mer transition fields and aligns these representations to a taxonomy-derived semantic space using a two-stage Node2Vec-to-MDS procedure that combines local graph structure with global hierarchical distance preservation. The resulting space supports zero-shot inference over species excluded from genome-level training but represented in the taxonomy, generalized zero-shot inference over seen and unseen species, and cascaded genus-to-species prediction. Across benchmark and large-scale disjoint-species evaluations, HyperTaxa improves taxonomic prediction over pseudo-image and transformer-graph baselines, with the strongest gains at higher taxonomic ranks. Ablations further show that multiscale transition encoding and taxonomy-aware alignment are key contributors to generalization. These results position HyperTaxa as a complementary taxonomy-aware representation and candidate-generation framework for microbial identification under incomplete reference coverage and open-set uncertainty.
[AI-122] Beyond Leaderboards: A Method for Application-Specific Capability Profiling of Large Language Models
链接: https://arxiv.org/abs/2610.25659
作者: Oleg Grynets,Vasyl Lyashkevych,Andrii Salyk,Dmytro Koziuberda,Dmytro Kostetskyi,Oleh Kaskun,Olena Pochernina,Alona Seletska,Oleksii Ilchuk,Viktoriia Tsymbal,Daryna Tukalo,Bohdan Saukh,Ivan Fedorchuk,Hryhorii Petrenko
类目: Logic in Computer Science (cs.LO); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 43 pages, 32 figures, 80 tables, 88 references
Abstract:Public leaderboards rank large language models (LLMs) in general, but an engineer has to choose a model for a specific application, architecture and cost. This paper presents a method for application-specific capability profiling of LLMs and its realization as a containerized testbed. The method treats each application as a unit of extension: an Oracle-to-PostgreSQL migration agent, a documentation-grounded chatbot or an enterprise-support automation agent runs in its own container behind a small contract, brings its own dataset, evaluator and owned attributes, and is onboarded together with its dashboards. Measurements accumulate in a versioned, sparse metamodel of every model. The method separates the model under test from judges and agents, freezes and hashes plans, refuses to compare values from different contexts, diagnoses the instrument before the system, and applies a null-agent check and a noise rule whose operating characteristics are quantified by simulation. In the evaluated automation conditions, model choices produced larger differences on selected observable outcomes than framework choices, although the experimental design does not establish a general causal ranking of these factors. The experiments also show that instruction tuning is deployment-specific and that system-reported success can exceed independently observed success: of 216 migration units checked against a real Oracle database, 108 were demonstrably correct. These measurements establish application-specific observations, not a universal model ranking.
[AI-123] A “feature ODE” describing the learning behavior of shallow MLPs on simple functions ICML2026
链接: https://arxiv.org/abs/2610.25652
作者: Joseph Turnbull,Berkan Ottlik,Jamie Simon
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted as part of the HiLD workshop, ICML 2026
Abstract:We study the gradient flow training dynamics of two-layer multi-layer perceptrons (MLPs) on anisotropic Gaussian data. Even in this simple setting, solving the dynamics in full is notoriously difficult, so we try something different. We propose a “feature ODE” that might be intuitively expected to describe the learning dynamics of a shallow MLP. We find that MLPs trained to learn simple target functions (low-order Hermite polynomials and staircase functions) actually roughly follow the loss trajectories from this feature ODE. This agreement is a surprise because of the size of the simplification and the fact that it does not come from a rigorous chain of approximations. We discuss this surprise and speculate on its reasons and uses.
[AI-124] Spherical Fixed-Point Reason ers: Adaptive Loop Halting on the Hypersphere
链接: https://arxiv.org/abs/2610.25646
作者: Arber Zela,Artavazd Maranjyan,Pol Puigdemont,David Berghaus,Valentyn Boreiko,Volkan Cevher
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 37 pages, 24 figures, 6 tables
Abstract:Looped (weight-tied) transformers enable adaptive computation by allocating more iterations to harder inputs. Fixed-Point Reasoners (FPRM) achieve this with a damped fixed-point solver, but operate in an unbounded Euclidean residual space with coupled rescaling and relative-norm halting. We introduce Spherical Fixed-Point Reasoners (SFPRM), which perform the same computation on the unit hypersphere of normalized transformers (nGPT). Input injection, damping, and halting become geodesic operations, yielding boundedness by construction and eliminating coefficient coupling. We prove fixed-point existence, local convergence under an explicit step-size condition, and global convergence under strong injection. On Sudoku-Extreme, Maze-Hard, and Mini-ARC, SFPRM achieves 96.6%, 87.9%, and 59.1% test accuracy, respectively, outperforming FPRM and other baselines while allocating test-time compute more effectively. Its converged geometry is also interpretable: token-input angles capture puzzle structure and predict correctness.
[AI-125] hink Compress Answer: Reinforcement Learning for Multimodal Grounding via Reasoning Compression
链接: https://arxiv.org/abs/2610.25640
作者: Jinbae Seo,Hyeongjun Kwon,Jiyoung Lee,Hyunsung Jang,Namkoo Ha,Jungin Park,Kwanghoon Sohn
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Extended reasoning has substantially improved multimodal large language models (MLLMs) on tasks requiring compositional and relational inference. However, multimodal grounding imposes a distinct requirement: reasoning must ultimately converge on a target-specific compact representation that supports precise localization. As reasoning trajectories grow longer, they often expand into broad scene narratives and peripheral relational details, which can dilute grounded perceptual evidence and lead to ambiguous or incorrect predictions. Motivated by the information bottleneck principle, we propose Think-Compress-Answer (TCA), a reinforcement learning framework that decomposes multimodal grounding into three phases: free-form reasoning, target-aware context compression, and answer generation conditioned on the compressed context. TCA is optimized through a multi-stage Group Relative Policy Optimization (GRPO) with progressive rewards that establish the structured output protocol, minimize reasoning-specific information while preserving target-relevant content, and directly improve localization accuracy. This staged design enables MLLMs to retain the benefits of extended reasoning while distilling it into the minimal context required for grounding. Experiments across six multimodal grounding benchmarks show that TCA attains substantial zero-shot performance gains, demonstrating the effectiveness of reasoning compression.
[AI-126] he Forward/Inverse Duality of Controlled Emergence:A Computational Design Program from Explanation to Engineering
链接: https://arxiv.org/abs/2610.25630
作者: Hao Li,Minghui Ren,Shaowen Lu
类目: Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注: 38 pages, 7 figures, 13 tables
Abstract:The macroscopic capabilities of intelligent systems arise from the local interactions of many simple units. When such systems are deployed in the real world, their behaviour must nevertheless be constrained, leaving designers caught between releasing adaptability and preventing loss of control, yet without formal tools to resolve the dilemma. This paper proposes a unified research program for controlled emergence that restates this tension as a pair of dual problems: the forward problem explains macroscopic behaviour from local rules, and the inverse problem solves for controllable rules and readout mechanisms from a desired target. We take an augmented quintuple as the formal object shared by both ends, and decompose ``engineering’’ into three independently measurable coordinates: controllability, reproducibility, and degree of formalizability. On this basis we give three testable working hypotheses (critical scale; the cooperation and trade-off between constraint and adaptability; minimal constraint) and one feasibility criterion, which reduce inverse design from unconstrained optimization to interval checking. The program takes explicit failure boundaries as its premise, does not pursue complete control, and always preserves the system’s room for self-organization. The reduced pilot experiment attached to this paper serves only to verify the executability of the research protocol and does not constitute a statistical test of the working hypotheses. This design language generalizes to multi-agent artificial intelligence, the industrial internet, and emergency response systems, all of which must confront emergence and controllability at once.
[AI-127] Selecting and Composing Agent Design Patterns: Industrial Case Evidence and Controlled Reference-System Experiments
链接: https://arxiv.org/abs/2610.25629
作者: Jia Huang,Bingsheng Ru,Bo Liang,Yuke Xiong,Haiyan Fan,Joey Tianyi Zhou
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 15 pages, 2 figures, 5 tables
Abstract:Agent pattern catalogs describe architectural alternatives, but engineers also need evidence for selecting and composing them. We extend a cognitive-function–execution-topology framework with an evidence-driven selection procedure and explicit composition contracts. The study combines two industrial case accounts, a contributor’s collaboration design, seven practitioner workshops, and executable fault-injection experiments. Three local reference systems use SQLite state, shared artifact files, and a file-backed map-publication surrogate. A 2 x 2 ablation separates function-specific contracts from execution controls across 36 scenario families, with ten input variants per family. Under equal family weighting, composition correctly completes or safely refuses 11 of 12 families in each system, compared with 2, 2, and 1 for the simple baselines. These are constructed challenge-suite results, not production success rates. Recovery without consumer-side verification increases false-success cases in the map surrogate. All composed systems still fail a boundary case in which trusted evidence is wrong. An audit of a frozen six-arm language-model pilot reproduces its scores but confirms an evaluator-dependent ranking reversal. The contribution connects industrial failure, pattern selection, executable treatment, external acceptance, and revision. Selection is demonstrated rather than evaluated against an alternative procedure; catalog admission rests on separate criteria. The results support bounded composition mechanisms, not a claim that the framework improves engineers’ design decisions over an equally informative conventional method.
[AI-128] Weighted State Souping for Efficient Long-Context Retrieval
链接: https://arxiv.org/abs/2610.25621
作者: William L. Tong,Aryo Lotfi,Emmanuel Abbe,Josh Susskind,Cengiz Pehlevan,Eran Malach
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 22 pages, 7 figures
Abstract:State space models (SSMs) combine fast inference and fixed state size with competitive language-modeling performance. Recently, state souping has been proposed as a procedure whereby an SSM independently encodes chunks of tokens like documents, tool definitions, or memory records. The resulting state encodings may be cached and recombined to cheaply construct a context for downstream queries. However, both state souping and vanilla SSMs deteriorate on long, crowded contexts, limiting their utility for long-horizon domains. To address this issue, we propose weighted state souping, a simple method that learns query-dependent weights for composing cached states. The resulting models improve long-context performance substantially across both single- and multi-hop document question answering, needle-in-a-haystack retrieval, variable tracking, and tool use. Our method can be further adapted into an efficient sparse variant, yielding near-constant encoding cost with respect to context length while preserving (and sometimes improving) performance. We also present simple theoretical arguments that illustrate why souping works and when it may fail, and discuss informally the advantages of weighted souping. Weighted state souping thus combines reusable context representations with selective retrieval, offering an efficient approach to language modeling over large collections of documents, tools, and other forms of structured context that increasingly dominate modern agentic workloads.
[AI-129] Safety Offset Learning: When Benign Fine-Tuning Misaligns LLM s
链接: https://arxiv.org/abs/2610.25611
作者: Jesson Wang,Zhanhao Hu,Minghao Lin,Yuchen Wu,Aylin Caliskan,Mengyuan Li
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Safety risks from fine-tuning are commonly characterized by the harmfulness of the training data:harmful fine-tuning is considered dangerous, while benign fine-tuning is often assumed to preservealignment. We argue that this view misses a more fundamental factor-the behavioral transition inducedby supervision. Harmful fine-tuning and safety-degrading benign fine-tuning can both be understoodas learning the same refusal-to-compliance transition, but on different data distributions. We studywhether such a transition learned on benign inputs can generalize to harmful requests. We constructbenign examples that an aligned model initially tends to refuse and pair them with their originalharmless responses, creating a discrepancy between the model’s initial refusal and the supervisedanswer. Fine-tuning on these examples increases harmful compliance, but only when two conditionshold together: the discrepancy is large enough, and it cannot be resolved through a narrow surface-levelshortcut. With the induced discrepancy matched, varying only whether it is localized changes harmfulcompliance by an order of magnitude; raising the discrepancy does not help while a shortcut remainsavailable, and removing the shortcut does not help while the discrepancy is small. Building on thisobservation, we introduce SAFEOFF, a dataset of benign instructions rewritten to raise the alignedmodel’s refusal likelihood and paired with their original answers. Across three aligned LLMs and threesafety benchmarks, SAFEOFF substantially weakens safety alignment despite containing no explicitharmful supervision, suggesting that fine-tuning safety depends not only on what data a model sees, butalso on which behavioral boundary the supervision moves.
[AI-130] Design and Cross-Domain Evaluation of an Agent Execution Harness
链接: https://arxiv.org/abs/2610.25568
作者: Saurav Joshi,Manas Gupta,Ashwin Venkatesh Raman,Akshay Budhkar
类目: Artificial Intelligence (cs.AI)
备注: 25 pages, 11 figures, including references and appendices
Abstract:Completing complex tasks with language-model agents requires coordinating planning, execution, and review so the final result addresses the user’s original objective. We present an agent execution harness that keeps the original objective separate from a revisable plan and an execution record. A planner assigns scoped tasks, workers produce outputs and artifacts, and a reviewer examines results against the original request before submitting or requesting additional work. Reusable domain-specific instructions guide these roles and are refined during construction using task inputs and execution trajectories, without official grading feedback. We evaluate complete harness configurations across terminal tasks, clinical responses, application building, and scientific questions. On Terminal-Bench 2.1, the harness passes 83 of 89 instances (93.26%); on HealthBench Professional, a Fable-configured system achieves a length-adjusted rubric score of 69.63 across 525 tasks. In a 53-task paired HealthBench comparison, the Fable-configured harness achieves an aggregate score 19.89 points higher than direct completion, despite a -1.46 -point median task-level change; mean inference cost rises from \ 0.29 to \ 2.13 per task. We combine these outcomes with latency, documented instruction revisions, and reviewer decisions. Evaluating outcomes alongside execution trajectories and resource use clarifies how coordination supports task completion, where it introduces regressions, and at what cost.
[AI-131] Learning Which Device to Call: Online Tool Grounding for LLM Agents in IoT
链接: https://arxiv.org/abs/2610.25563
作者: Yichen Liu,Xiaochong Jiang,Haoran Yu,Lifei Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The Internet of Things (IoT) gives large language model (LLM) agents access to physical observations, but one requested capability is usually offered by several devices, and the agent’s plan does not say which one to call. Existing work retrieves relevant tools and checks plan feasibility, without learning which eligible device returns a useful observation. Unlike previous works, we study online tool grounding: binding an already specified tool call to one physical endpoint under service cost and device participation constraints. We propose a framework that learns which physical device should execute each agent-issued call. It formulates selection as a contextual bandit problem and combines adaptive context partitions with hierarchical endpoint trees, so that execution feedback is reused across related requests and devices. Selection balances task success against service cost while respecting device capability, authorization, battery reserves, and availability. Gaussian reward aggregation provides an optional mechanism for protecting individual feedback values. Theoretical results establish a finite-horizon regret bound and differential privacy for released reward statistics under fixed, reward-independent assignments. In a simulated loading-dock workload, the proposed router achieves higher step success and better cost-adjusted service, at lower simulated device energy and latency, than nearest-device, advertised-quality, UCB1, and LinUCB selection.
[AI-132] Symbolic Constraint Agents : Mining Formal Tool-Call Constraints to Prevent Agent Errors
链接: https://arxiv.org/abs/2610.25552
作者: Jinu Lee,Shraddha Barke
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Symbolic constraints (programmatic, deterministic checks for tool calls) enable deterministic verification of LLM agents’ tool calls, but existing approaches primarily enforce predefined safety policies rather than tool dependencies in complex tool environments. We explore whether symbolic constraints can be mined from environments and observations beyond predefined policies, and whether they can improve end-to-end task performance. We present Symbolic Constraint Agents (SCA), a framework that mines symbolic constraints from API documentation, refines them against execution traces, and checks the agent’s tool calls with an SMT solver at runtime. A violation triggers a natural-language warning or, when the valid replacement is uniquely determined, an automatic repair. On AppWorld, refinement raises error-detection F1 from 37.0 to 58.3 on GPT-5.5 traces and from 39.3 to 61.9 on Qwen3.5-27B traces. At runtime, repair raises GPT-5.4-mini’s task success by 5.4 points under a 40-call budget and 9.8 points when environment errors are hidden, while removing 59–87% of failed calls across models, and transfers to \tau^2 -bench with gains of up to 5.6 points. These findings show that automatically acquired symbolic constraints can improve tool-calling reliability and task completion, particularly when failures are costly or difficult to diagnose.
[AI-133] Editing as Language: Teaching LLM Agents a Grammar for Native Context Editing
链接: https://arxiv.org/abs/2610.25550
作者: Yunlong Deng,Yan Li,Xing Hu,Yida Zhu,Sean Du,Weiyang Liu,Guangyi Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A synopsis in novels folds a story into a few lines; a function in code gathers many steps with a single execution. Context editing is native to language itself. In this paper, we propose the Native Context Editing (NCE) framework, which gives language model agents a grammar for natively editing their own context. By encoding context-editing operations in a grammar of dedicated special tokens, we teach the agents to invoke edits through the language itself. We train agents on iteratively constructed trajectory segments, with each teacher edit conditioned on the evolving context. This establishes context editing as an atomic and inherent capability of LLM agents, enabling them to dynamically modify their context as they generate responses over long-horizon reasoning and interaction. Compared with context-editing methods that learn to generate tool-use code within specific external harness environments, our approach integrates context editing as an internal mechanism of the model’s native generation process, enabling better generalization across diverse real-world workflows. On four long-horizon agentic benchmarks (Terminal-Bench 2.0, SWE-bench Verified, SWE-bench Multilingual, and SWE-bench Pro), NCE improves the average task success rate over the base model by 5.6 percentage points for LongCat-Flash-Lite-Sparse and by 4.5 for Qwen3.6-35B-A3B, while reducing average token usage by 36% and 15%, respectively.
[AI-134] From Clinical Risk Prediction to Operational Decisions: An Uncertainty-Aware Framework for ICU Staffing ECML KDD2026
链接: https://arxiv.org/abs/2610.25530
作者: Jiayan Yang,Ad Feelders,Arno Siebes
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at ECML PKDD 2026, Applied Data Science Track
Abstract:Advances in electronic health records (EHR) and large-scale critical care databases have enabled the development of machine learning models for the early detection of patient deterioration. While these models often achieve strong predictive performance for outcomes such as sepsis and mortality, most studies evaluate them purely on discrimination metrics. This leaves a critical gap: how to systematically translate probabilistic clinical risk estimates into downstream operational decisions under real-world constraints, such as shift rigidity and hiring delays. This study proposes an uncertainty-aware anticipatory staffing framework that bridges short-term patient risk prediction and staff planning without altering the upstream predictive model. Instead of relying on naive deterministic aggregation or fixed safety buffers, the framework explicitly models dynamic predictive uncertainty and evaluates extreme shortage exposure using Conditional Value-at-Risk (CVaR). These aggregated, risk-calibrated demand signals are then embedded into an hourly rolling-horizon scheduling policy to generate delay-constrained staffing decisions. Evaluated on realistic trajectories from the MIMIC-IV database, the proposed framework demonstrates a favorable cost-shortage trade-off compared to deterministic and purely reactive baselines. Pareto frontier analysis illustrates how uncertainty-aware demand construction changes resource allocation under varying safety preferences, while stress testing confirms the framework’s operational robustness even when subjected to systematic predictive bias in the form of calibration drift. By formalizing this translation layer, this work provides a principled, transparent pathway for safely integrating probabilistic AI outputs into high-stakes healthcare operational systems. Comments: Accepted at ECML PKDD 2026, Applied Data Science Track Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2610.25530 [cs.AI] (or arXiv:2610.25530v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.25530 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-135] StopLoss: Teaching Agents to Predict Failure and Stop Early
链接: https://arxiv.org/abs/2610.25525
作者: Jiarong Liang,Xueguang Ma,Dongfu Jiang,Yuxuan Zhang,Ping Nie,Wenhu Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents keep working on runs that will fail verification, and asking a frontier model to watch for this is expensive in its own right. We introduce StopLoss, a small external monitor that learns failure recognition from completed trajectories and their verifier outcomes. The name is inspired by the stop-loss order in finance: StopLoss ends a run early when the run is likely to fail, which saves resources. Final outcomes label earlier local windows, so a fine-tuned language model can score the most recent tool steps in a single forward pass, and a threshold calibrated on successful runs turns that score into an abort decision without changing the executor. Trained on GPT-family SWE-rebench trajectories and evaluated on SWE-bench Verified and a SWE-bench Pro subset with GPT and Claude executors, StopLoss is compared against six general-purpose judges reading identical windows. Outcome supervision lifts failure prediction from near chance to a useful signal at the first check, at one to two orders of magnitude lower cost per check than any tested judge. This translates into savings: relative to completing every task, StopLoss lowers cost per retained solve by 4.5% on Verified while preserving 99% of successful runs, and by up to 31% on the costlier Pro subset, reaching 48% with a Qwen3-8B backbone, and in all four settings its cost per retained solve is below that of every judge policy. Outcome-trained small monitors can make failure recognition cheap enough for early stopping to pay for itself.
[AI-136] When Empathy Misses the Goal: A Benchmark for Goal Displacement in LLM Advice NEURIPS2026
链接: https://arxiv.org/abs/2610.25522
作者: Dean Ariel,Guy Laban
类目: Artificial Intelligence (cs.AI)
备注: Accepted for publication at the 40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026), Evaluations and Datasets Track
Abstract:Conversational large language models can respond with warmth and reassurance while failing to advance the user’s stated goal. We conceptualize this failure as \textbfgoal displacement , the substitution of persona-maintenance content for goal-relevant substance. To measure it, we introduce \textbfGDB (Goal Displacement Benchmark), comprising 230 scenarios across nine domains and 4,134 human-scored responses from 18 models. A 2 \times 2 taxonomy crosses hard constraints with emotional distress, distinguishing failures to confront infeasible goals or combine empathy with practical guidance from reverse displacement, where unsolicited problem-solving replaces requested emotional support. A deletion-test rubric assesses whether substantive help remains after purely supportive sentences are removed. Independent raters achieve weighted \kappa = 0.82 against author scores for displacement severity. Trap displacement ranges from 0.0% to 37.7% and is strongly negatively correlated with Chatbot Arena Elo ( \rho = -0.706 , n = 15 ). Reverse displacement reaches 41.9% , including substantial rates in models with negligible trap displacement. A factorial probe supports a distinction from social sycophancy. Human-scored responses in the high-complexity, low-pressure condition exhibit 35.0% sycophancy but 7.5% displacement, with little association between the constructs. Matched OLMo-2 checkpoints show that mixed empathy-with-constraints displacement remains at 52.3% after supervised fine-tuning and falls to 13.8% after direct preference optimization. GDB provides a reproducible diagnostic of whether responses advance the operationalized user goal, supporting model comparison, post-training evaluation, and regression testing. We release the dataset, annotations, and rubric under CC-BY-4.0 and the evaluation code under MIT.
[AI-137] Bud ContextCompact 2.0: Context Compaction for Agent Success and Accuracy with Small Language Models and Constrained Context Windows
链接: https://arxiv.org/abs/2610.25516
作者: Jithin VG,Ditto PS
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agents that work for many turns outgrow the context window of the model that serves them, and on small language models (SLMs) and constrained inference systems that window is small. Context compaction replaces older turns with a summary so a session can continue, but it is lossy, costs a model call, and can help or hurt depending on when it runs. We describe Bud ContextCompact 2.0, the compaction built into the Bud agent runtime, and evaluate it in two settings. On recall questions over recorded agent sessions, compaction to a 16K budget raised a 1.2B-parameter model’s accuracy to 0.335, from 0.247 with truncation and 0.176 with its full 32K history (p 0.001), and raised a 9B model from 0.515 (truncation) to 0.577 (p = 0.04). On 100 SWE-bench Verified tasks run end to end by a 27B reasoning model on a 49K-token window (556 graded sessions), compaction raised the resolve rate on long sessions from 41% to 53% (+12.3 points, 95% CI +1.6 to +23.0, McNemar p = 0.043), left short sessions unchanged, and used 13% fewer tokens per solved task at equal total cost. The main failure without compaction was not the history overflowing but a reasoning turn exhausting the room the history left. The gain has limits: it vanished under a 45-minute session budget, because each summary took a median of 161 s, and in a threshold study it appeared only when compaction triggered early enough to leave room. An 8B model submitted after about ten turns at 5-8K tokens and never needed compaction. We argue that the accuracy results transfer across hardware, including CPU inference, for an equivalent model and window, while the time cost scales with the hardware and must be budgeted for.
[AI-138] Dynamics-Aligned Representation Learning under Induced Partial Observability
链接: https://arxiv.org/abs/2610.25511
作者: SM Mazharul Islam,Manfred Huber
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Model-based reinforcement learning (MBRL) has enabled sample-efficient learning of continuous control policies from high-dimensional observations through learned latent dynamics models. These models are typically trained with observation reconstruction objectives, which provide strong supervision but can bias representations toward visual fidelity leading to an undertrained prior and unreliable long-horizon predictions. We propose a stochastic gating mechanism that induces controlled partial observability during training by sampling a spatial binary mask from a Beta–Bernoulli random field and applying it to the raw observation. The gate probability is resampled at each time step, exposing the encoder to conditions ranging from near-total occlusion to near-full visibility. Integration of our method requires minimal architectural changes from existing dynamics models: the gated observation and the binary mask are channel-wise concatenated and passed to the encoder as a single augmented input, with no modifications to the broader training pipeline. We evaluate the proposed approach on a set of pixel-based continuous control tasks selected from the DeepMind Control Suite. Using model predictive control for evaluation, which relies solely on open-loop model rollouts, we show that stochastic gating improves planning performance and data efficiency across tasks. Further analysis demonstrates that performance is robust to the spatial scale of the mask, and a balanced observability during training outperforms configurations biased toward either sparser or denser observations.
[AI-139] Formal Raising: Embedding Verifiable Ethical Properties into the Development of AI
链接: https://arxiv.org/abs/2610.25501
作者: Rolf Drechsler,Christina Plump
类目: Artificial Intelligence (cs.AI)
备注: 11 pages, 6 figures,
Abstract:Alan Turing proposed constructing a ``child machine’’ and educating it rather than directly programming an adult-like intelligent machine. Recent work on model raising revives this developmental view for Large Language Models by arguing that moral values should accompany capability acquisition from the beginning rather than be added only through post-hoc alignment. This paper proposes Formal Raising (FR): selected moral values are represented as explicit, machine-checkable properties that participate in development, evaluation, and action selection. We present a C-based reasoning framework in which candidate actions are evaluated against prioritized ethical properties. Conflicts are resolved w.r.t. a given order, strict rules are non-defeasible, and underdetermination is reported explicitly. Exceptions are not encoded as special cases but emerge from the comparison of alternatives. We provide examples that cover conflicts involving life, truthfulness, harm, privacy, autonomy, and strict rules. We outline a property layer inspired by a Property Specification Language (PSL) for state and temporal requirements and discuss its integration into a raising process.
[AI-140] Assessing the AI Readiness of Construction Datasets: A Multidimensional Operational Framework
链接: https://arxiv.org/abs/2610.25499
作者: Xiaoyu Hou,Bo Xiao,Abdolmajid Erfani,Ronghua Xu,Yu Cai
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The rapid growth of artificial intelligence (AI) applications in construction has increased the demand for datasets suitable for reliable model development, evaluation, and reuse. However, construction datasets vary widely in documentation, quality, governance, and evaluation support, and consistent criteria for assessing AI readiness remain limited. This study develops a Construction Dataset AI Readiness Framework with 30 attributes across four dimensions: Metadata Completeness, Data Quality and Representativeness, Data Governance and Accessibility, and AI Utility. The framework uses attribute-specific scoring, applicability screening, and automated and manual verification. It was evaluated using 21 publicly available datasets across seven data-type groups. Overall readiness scores ranged from 0.720 to 0.964, with a mean of 0.879. Hybrid verification achieved 90.5% agreement between automated suggestions and manually verified scores. The framework supports consistent assessment of heterogeneous construction datasets and identification of readiness gaps.
[AI-141] From Proposal to Execution: Modeling Interaction Semantics in Shared Autonomy
链接: https://arxiv.org/abs/2610.25496
作者: Shashank Shekhar,Laurent Jeanpierre,Abdel-Illah Mouaddib
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 6 tables
Abstract:In many human-AI systems, candidate actions are not executed directly but are first evaluated by another decision-making entity under a potentially different objective or decision model. Standard decision-theoretic frameworks do not require this interaction to be represented explicitly, as they define transition dynamics directly over state-action pairs, allowing evaluation to be absorbed within the transition model. We introduce a semantics-aware decision framework in which a candidate action is proposed, evaluated under an interaction context, and then resolved into an executed outcome. We formalize this through a joint action language and an interpretation function that maps each joint action and context to a semantic outcome. Transition and reward dynamics are induced through these semantic outcomes, thereby introducing an explicit representational layer for \textitinteraction semantics. Semantic outcomes provide explicit units for organizing interaction principles, which can be instantiated through application-specific evaluation mechanisms and reused across distinct interaction instances. This provides a structured interface for specifying context-dependent evaluation separately from the resulting execution dynamics, without requiring changes to the underlying planning algorithm. We instantiate the framework in shared human-AI autonomy, capturing interaction patterns such as advisory and authorization, and illustrating how interaction context influences decision making. The resulting formulation remains compatible with standard Markov decision process (MDP) planning.
[AI-142] P(Yes): 15times Improvements via Test-Time Scaling of Decision Models
链接: https://arxiv.org/abs/2610.25474
作者: Lev Stambler
类目: Artificial Intelligence (cs.AI)
备注: 22 pages, 5 figures. Interactive version: this https URL . Code and data: this https URL
Abstract:Decision models answer typed questions, such as multiple-choice or yes/no questions, with a probability for each option. We add test-time search to frozen decision models, with no training: we show the model a complete proposed answer, read off its probability of saying “Yes” to the proposal’s correctness, and search for answers that it rates higher. The search is done either by editing one answer at a time or by joint optimization, which changes several answers at once. Search raises accuracy for every model and test set we tried, and the gains are largest for exact correctness on logic puzzles, where every answer in a task must be right. On 128 such puzzles, it takes Qwen3.8-27B from 2 to 30 exactly correct puzzles ( 15\times ) and its per-blank accuracy from 27.8% to 48.5%. On a shared set of 47 puzzles, searching over each model’s own Yes/No score raises the per-blank accuracy of StartLux Instant, Perplexity Decider, Clef, and JEV, which we reach only through its public API, by 4.0 to 13.0 percentage points, and raises the number of puzzles they get exactly correct, out of 47 and averaged over three search seeds, from 10 to 35, 9 to 26, 2 to 11, and 9 to 31.7 respectively.
[AI-143] On Reasoning Correctness Directions
链接: https://arxiv.org/abs/2610.25464
作者: Munachiso Samuel Nwadike,Zayd M. K. Zuhri,Rifo Genadi,Abzal Nurgazy,Ali Mekky,Alham Fikri Aji,Zangir Iklassov,Kentaro Inui
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:We present experimental evidence suggesting that information about whether a model’s reasoning is on the right track can be linearly decodable without defining a causally useful direction for steering. To study this, we introduce ChainSearch, a controlled arithmetic reasoning task that extends Countdown with explicit search and backtracking. We train a 3-million-parameter model to solve this task. Its compact size allows us to study internal model correctness directions in detail using linear probes, mean activation differences, and direct optimisation, as part of a rigorous search for internal reasoning-correctness directions. We find that while the model contains statistically significant internal directions of reasoning correctness, none of the resulting directions consistently improves reasoning under causal intervention. This calls into question whether the linear representation hypothesis is sufficient for identifying causally useful directions in the context of reasoning. Our code is made available together with this work.
[AI-144] Stepwise Reasoning Verification: Dont Generate Classify!
链接: https://arxiv.org/abs/2610.25441
作者: Yifei Shi,Michalis Korakakis,Rami Aly,Eric Chamoun,Lukasz Szpruch,Andreas Vlachos
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Verifying the reasoning of large language models is increasingly important as they are applied to complex mathematical and scientific problems. Existing work focuses on large decoder models and reports mixed results for generative and discriminative verifiers, leaving unclear which paradigm is more effective and whether large models are necessary. In this work, we cast stepwise verification as a natural language inference task: the problem and partial solution prefix form the premise, while the next reasoning step serves as the hypothesis to be classified as entailed or non-entailed. This formulation enables error localisation at individual reasoning steps and direct comparison of encoder classification, decoder classification, and decoder verdict generation across model scales and annotation budgets. Across five reasoning datasets, a 0.15B encoder classifier outperforms most generative verifiers with up to 4B parameters. On identical decoder backbones, classification outperforms generation from 0.5B to 4B parameters and matches it on the strongest 7B backbone—reaching comparable performance with roughly one tenth as many annotated problems. Our findings establish classification as a parameter- and data-efficient approach to stepwise verification and show that strong verification does not require large generative decoders.
[AI-145] MOSAIC: Multi-Objective Search for Allocating Inference Cache Across LLM Decoder Layers
链接: https://arxiv.org/abs/2610.25431
作者: Sravanth Kodavanti,Manjunath Arveti
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The Key-Value (KV) cache is the main memory bottleneck of long-context LLM inference. Token eviction methods such as SnapKV, H2O, AdaKV and L2Norm shrink it, but they give every transformer layer the same budget, although layers differ in which tokens they need. We present MOSAIC, a multi-objective neural architecture search that learns per-layer KV-cache budgets for any eviction method. One search maps the trade-off between average budget and task score; its best allocation transfers to every target budget without further search, and an optional budget-constrained search refines it. Across four eviction methods on Llama-3-8B-Instruct and Mistral-7B-Instruct-v0.2, MOSAIC beats uniform allocation in 146 of 154 (task, method, budget) settings, by up to +48 points on RULER and +11.9 on LongBench (LCC). On both models, it reaches the score of uniform allocation with up to 4x less KV cache (75% smaller). Against EvolKV, MOSAIC explores a ~10^40 x larger search space with 4x fewer evaluations and 18% fewer GPU-hours per budget, and scores higher in 5 of 6 settings. The learned allocations consistently favour a small set of middle layers.
[AI-146] When Form Changes but Logic Doesnt: Building Logic-invariant LLM s through Logical Structure NEURIPS2026
链接: https://arxiv.org/abs/2610.25402
作者: Xuyuan Liu,Xinshuai Dong,Elynn Chen,Yujun Yan,
类目: Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026, 24 Pages
Abstract:Large language models (LLMs) have achieved strong performance across diverse reasoning tasks, yet it remains unclear whether this performance reflects an understanding of underlying logical structure or a reliance on shallow semantic cues. A key indicator of such understanding is whether models treat logically equivalent inputs consistently, producing the same predictions despite differences in expression. Through systematic evaluation on curated data, we show that even strong LLMs often fail this test: they do not consistently preserve predictions across logically equivalent inputs and generalize poorly to unseen logical forms. To address this limitation, we propose LoGIcal STructure-guided reasoning (LoGIST), a framework that guides LLM reasoning with explicit logical structure. LoGIST maps each input instance to a decision diagram, a directed acyclic graph that captures the underlying logical structure, and encodes this graph into a logical embedding that conditions the LLM’s reasoning process. Across multiple training settings, LoGIST reduces inconsistency under logical equivalence and improves generalization to unseen logical forms, suggesting that logical invariance is a useful target for building more robust and generalizable logical reasoning in LLMs.
[AI-147] CD-OPD: Contextual Digest for On-Policy Distillation
链接: https://arxiv.org/abs/2610.25359
作者: Mingyuan Li,Yucheng Li,Xiaxia Wang,Gong Cheng
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Teacher takeover in on-policy distillation (OPD) can produce incorrect responses even when the student and teacher independently answer the same problem correctly, a failure we quantify using Collaboration Failure Rate (CFR). To address it, we propose Contextual Digest for On-Policy Distillation (CD-OPD). CD-OPD structures the earlier context of teacher guidance through a teacher-generated dependency graph, curating the required problem conditions and preceding reasoning as node summaries alongside the unchanged guidance in a Contextual Digest before student continuation. We further develop online meta-reweighting, which compares Digest segment gradients with a task gradient from independent student responses to improve student-only performance, favoring updates predicted to reduce task loss. Across eight benchmarks, CD-OPD improves average accuracy by 1.32–7.05 percentage points over representative existing OPD methods. It also reduces CFR by 27.4%–56.4% relative to existing takeover-based OPD methods.
[AI-148] Formally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention NEURIPS2026
链接: https://arxiv.org/abs/2610.25341
作者: Karthik Viswanathan,Plamen Dimitrov,Stefano Rocca,Sarayu Sirikonda
类目: Computational Complexity (cs.CC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 25 pages, 1 figure. Oral presentation at the AutoMLR Workshop at NeurIPS 2026
Abstract:Multi-head attention is a central component of the transformer architecture, yet how its expressivity depends on the number of heads remains poorly understood. We study a one-layer, attention-only transformer with additive positional embeddings and define the head complexity H^\ast(f) of a Boolean function f as the minimum number of attention heads required to represent it. We derive an exact linear-fractional normal form for head complexity, prove a general threshold-degree lower bound, and give an explicit ten-bit function for which this bound is strict. For symmetric Boolean functions, we independently show that head complexity equals the number of times the function changes value between consecutive Hamming-weight levels, even when positional embeddings are allowed. This agrees with the concurrent characterization of Hsu and Xu (arXiv:2608.04243) in a model without positional embeddings. The symmetric lower bound combines our general threshold-degree bound with the classical result of Aspnes et al. (1994). These results were discovered and formalized by a frontier model through conjecture generation, theorem proving, and machine verification, while human researchers asked the high-level questions and reviewed the generated theorems. We make the Lean formalization and research traces publicly available to support agent-assisted theorem proving and formal verification in mechanistic interpretability. This work provides a proof of concept for using the capabilities of frontier models, grounded by formal verification, to develop an understanding of transformer architectures.
[AI-149] AdaFedXplain: Adaptive Federated Learning with Explainable AI for Ransomware Detection
链接: https://arxiv.org/abs/2610.25318
作者: Md. Emran Biswas,Md. Delowar Hossain,Mst. Tanbin Yasmin Tanny,Muhammad Aminur Rahaman,Md Al Amin,Tangina Sultana
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: this https URL (Eighth International Conference on Machine Learning for Cyber Security (ML4CS 2026) )
Abstract:Attacks using ransomware have developed into complex multiphase campaigns that target vital infrastructure, resulting in annual damages of billions of dollars. Conventional detection techniques are ineffective against zero-day variations, and privacy issues are raised by centralized machine learning algorithms. Federated learning allows for cooperative model training without data sharing, however current FL approaches suffer from explainability issues, high communication cost, and non-IID data heterogeneity. In this study, we present AdaFedXplain, an explainable AI-based adaptive federated learning system for ransomware detection. Our approach includes progressive XGBoost complexity to avoid overfitting, a decaying learning rate schedule for stable convergence, and dynamic client grading based on ransomware prevalence and data sufficiency. Using 62,485 real-world ransomware samples from four OS-based non-IID clients, we compare AdaFedXplain against FedAvg, FedProx, and SCAFFOLD. According to experimental data, AdaFedXplain outperforms all baselines with 99.59% detection accuracy and 99.64% F1-score while cutting energy usage by 75%, latency by 65%, and communication overhead by 75%. The ablation investigation demonstrates that every element enhances performance. Security analysts can use SHAP explainability to identify important PE header characteristics that influence detection decisions. Our system provides a collaborative ransomware detection solution that is efficient and interpretable, while reducing privacy risks through federated learnings data-local training paradigm.
[AI-150] Shared-Private Cross-Modal Discrepancy Learning for Imbalanced Multi-Source Audio-Visual Deepfake Detection
链接: https://arxiv.org/abs/2610.25301
作者: Qiwu Wen,Jahangir Alam
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注:
Abstract:Audio-visual deepfake detectors are often trained under a single dataset or manipulation regime, but multi-source training collections may contain heterogeneous attacks and uneven source distributions. Training separate detectors for each source is inefficient, while a single shared detector may be biased toward dominant source-specific patterns. We study known-source imbalanced multi-source audio-visual deepfake detection using FakeAVCeleb and LAV-DF, where source identity is supplied during training and inference. To address this setting, we build on the Multi-Scale Cross-Modal Transformer Encoder (MSCT) and propose Pre-MSCT Shared-Private Feature Fusion. Our method routes \texttt[CLS]-augmented audio and visual tokens into a shared stream and source-specific private streams before the first MSCT block, while retaining its internal multi-scale self-attention (MSSA), differential cross-modal attention (DCA), and feed-forward operations. The shared stream models cross-source reusable evidence, while the private streams retain source-dependent residual variations. A domain-aware gated residual fusion module injects selected private evidence into the shared representation for the main fused prediction. Experiments under two controlled source-composition protocols show improved macro and worst-source AUC compared with the MRDF and MSCT baselines. These results support shared-private cross-modal discrepancy learning in the evaluated two-source setting with known source identity.
[AI-151] World Model Interpretability: From Decoding Representations to Understanding Latent Worlds
链接: https://arxiv.org/abs/2610.25296
作者: Jize Wang,Yongcheng Jing,Xinyi Le,Cailian Chen,Xinping Guan,Dacheng Tao
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Emerging Joint-Embedding Predictive Architectures (JEPAs) are moving beyond representation learning towards latent world models that support prediction, planning, and control. Yet their latent worlds naturally lack a human-interpretable interface. In this Perspective, we argue that as JEPA latents become internal states for prediction and action, interpretability must move beyond what they encode to how latent variables interact, evolve, and causally drive prediction. To this end, we envision a four-stage research agenda, involving Ground, Structure, Verify, and Act, to guide this shift toward actionable latent-world understanding. Within our framework, existing JEPA interpretability efforts provide early evidence for Ground, while systematic analysis of latent structure, causal verification, and intervention remains largely unexplored. Overall, this Perspective aims to call for a broader latent-world interpretability agenda that shifts attention from decoding representations toward understanding, validating, and acting on latent worlds.
[AI-152] EG-GRPO: Evidence-Guided Step-level Credit Assignment for Deep Search Agents
链接: https://arxiv.org/abs/2610.25269
作者: Sijia Luo,Mengjie Gao,Yuxiao He,Mengshu Sun,Cuiping Li,Hong Chen,Jing Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:LLM-based search agents demonstrate strong capabilities in long-horizon information seeking and reasoning, typically trained via reinforcement learning. However, existing approaches predominantly rely on trajectory-level outcome rewards, providing limited supervision for complex search and reasoning processes. Although step-level credit assignment offers a promising solution, obtaining reliable and task-grounded supervision for open-ended search remains challenging. In this paper, we propose Evidence-Guided Group Relative Policy Optimization (EG-GRPO), a framework that derives step-level supervision from the evidence annotations available in synthetic deep search data. EG-GRPO identifies the first supporting citation for each required evidence claim and uses the newly acquired evidence to construct step-level process rewards. By representing the accumulated evidence as an evolving knowledge state, EG-GRPO further performs state-conditioned relative advantage estimation and combines the resulting step-level signal with the standard trajectory-level outcome advantage. Experiments on five challenging deep search benchmarks show that EG-GRPO generally outperforms standard GRPO and existing reward-enhanced and step-level RL baselines across different model scales. These results demonstrate that task-grounded evidence annotations can provide effective process supervision for credit assignment in deep search RL.
[AI-153] BatchParScale: Sparse Batch-Mixing for Sequence Models
链接: https://arxiv.org/abs/2610.25254
作者: Hans Peter Lyngsøe Raaschou-jensen(University of Copenhagen),M. Onurcan Kaya(Technical University of Denmark),Desmond Elliott(University of Copenhagen)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 19 pages, 10 figures
Abstract:Recent test-time scaling methods improve model performance by increasing inference-time compute. One practical form of this is parallel scaling, where a model samples multiple independent reasoning streams and aggregates their answers using methods like voting, reranking, or verification. While effective, this approach leaves the streams uncoordinated, and a discovery made in one rollout cannot influence any other. Multi-Sequence Verifiers jointly score candidate answers after generation, but the reasoning streams remain independent during inference. ParScale takes a step toward model-internal parallel scaling by learning to aggregate multiple forward passes, but its streams still do not exchange information while processing a sequence. We introduce BatchParScale, an attention mechanism that allows parallel streams to exchange information through a small set of communication positions. We evaluate BatchParScale on a new Parallel Needle-in-a-Haystack retrieval benchmark and in pretraining, fine-tuning, and cooperative multi-agent reinforcement learning. Across these settings, BatchParScale outperforms ParScale on language-model loss and remains competitive on coordination-sensitive MARL tasks. Masked pretraining runs beat an unmasked single-sequence baseline despite local sequence corruption, and LoRA fine-tuning shows the mechanism can be added to a pretrained Transformer. At the same time, fact retrieval performance for pretrained and finetuned BatchParScale models remains unclear.
[AI-154] NeuroAct: Advancing EEG Analysis and Decoding via Dual-Loop Autonomous Research
链接: https://arxiv.org/abs/2610.25211
作者: Wei Xiong,Jiangtong Li,Han Zhao,Linxiao Yang,Jie Li,Changjun Jiang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
备注: 30 pages, 8 figures
Abstract:EEG analysis and decoding underpin brain-computer interfaces, cognitive assessment, and sleep monitoring. Fixed pipelines and reward-driven search do not explicitly determine how signal structure, representations, predictor behavior, and training dynamics contribute to performance or generalization gaps, leaving intervention choice underdetermined. We introduce NeuroAct, an autonomous research agent framework for EEG analysis and decoding. NeuroAct implements dual-loop autonomous research across EEG foundation models, task-specific predictors, and model combinations. In an inner loop, research, development, and review roles use evidence on signal structure, representations, predictor behavior, and training dynamics to test hypotheses, select analyses or interventions, and refine candidates using performance, stability, and cost feedback. An outer loop distills outcomes into conditional, testable recommendations with task and model conditions, then validates and revises them on subsequent this http URL evaluate four settings: (A) EEG General Analysis, (B) Cross-task Model Optimization, © Single-task Open Research, and (D) Collaborative Prediction. In Setting A, NeuroAct’s 80.88 macro-average exceeds BrainAgent’s 63.13 by 17.75 points. In Setting B, NeuroAct raises EEGPT’s mean balanced accuracy (BA) by 7.34 points and HBN Pearson’s ® from 0.6894 to 0.7362; in Setting C, it improves ADFTD BA by 16.33 points; and in Setting D, it raises Workload BA from 79.69% to 85.02%. Experience reuse raises BCIC-IV-2a BA from 34.66% to 50.81% while reducing LLM tokens by 64.02%, indicating recommendations improve cross-task performance. NeuroAct enables evidence-guided EEG research improved by cross-task experience.
[AI-155] BrandEval: A Two-Track Multi-Agent Benchmark for Risk-Sensitive LLM Crisis Communication EMNLP2026
链接: https://arxiv.org/abs/2610.25208
作者: Haotian Gan,Weidong Tang,Hanbin Sun,Yi Luo,Xintian Li,Jiayong Wang,WanYue Li,Yudong Li
类目: Artificial Intelligence (cs.AI)
备注: Accepted to EMNLP2026 Findings
Abstract:Evaluating LLMs for crisis communication requires more than judging isolated responses, because public feedback and competitor actions can reshape trust over time. Existing benchmarks remain largely static and single-turn, offering limited evidence about long-horizon risk-sensitive communication. We introduce \textbfBrandEval, a two-track benchmark for product-crisis communication, focusing on public brand responses to product-related safety, quality, service, and trust incidents. BrandEval-Static evaluates stakeholder-aware strategic analysis, while BrandEval-Dynamic stress-tests response policies in \textbfBrandPolis, a controlled multi-agent sandbox with consumer feedback, competitor responses, and trust-risk trajectories. We also construct \textbfBrandSRD, a Chinese crisis-response dataset, and instantiate a reference Strategic Agent using \textbfStrategic Rationale (SR). Across proprietary and open-source models, we find that static strategic quality is informative but incomplete: it aligns with some end-state outcomes but does not reliably predict volatility and delayed recovery; dynamic stress tests further expose tail-risk failures under manipulative or weakly grounded policies. The code and dataset will be available at this https URL.
[AI-156] ViCoR: Estimating Visual Necessity via Counterfactual Residuals for Multimodal Medical Data Selection NEURIPS2026
链接: https://arxiv.org/abs/2610.25202
作者: Haotian Gan,Yudong Li,Xintian Li,Jiarui Han,Weidong Tang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS2026
Abstract:Medical vision-language models often exhibit hidden stratification, where high overall performance masks failures on subtle visual findings and overlapping clinical conditions. We trace this failure mode to modality imbalance during fine-tuning, where strong linguistic priors can suppress visually grounded learning and induce plausible but weakly grounded rationales. Existing data selection methods usually score image-text pairs as holistic units, without separating visually necessary samples from shortcut-solvable ones. To address this problem, we propose ViCoR, a counterfactual-residual-based data selection method for Med-VLM fine-tuning. ViCoR estimates visual necessity through image-dependent optimization residuals, calibrates the resulting scores with medical supervision compatibility, and constructs a diversity-aware fine-tuning subset. Across two backbones, four external generalization benchmarks, and one multi-disease robustness benchmark, ViCoR-selected 20% subsets achieve the strongest average performance compared with full-data fine-tuning and selection baselines. Gradient-level analysis and perturbation audits further show that ViCoR enriches image-dependent training signals and strengthens image-level and evidence-region dependence. The code will be available at this https URL.
[AI-157] World Models Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets
链接: https://arxiv.org/abs/2610.25170
作者: Karthik Iyer,Madan Dabbeeru
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 37 pages, 18 figures, 26 tables. Code: this https URL , archived at doi: https://doi.org/10.5281/zenodo.23448056
Abstract:A retailer makes different decisions at different grains of the product, store and time hierarchy: store orders for an item in a store, supplier funding for an item across stores, category performance by store, and the financial plan for the chain. Which forecasting model to use depends on the grain. This paper compares models at each grain on four public data sets (M5, six Dominick’s categories, Breakfast at the Frat, and Corporación Favorita, the last kept out of all development), over rolling forecast origins with paired tests and model confidence sets. It covers price and promotion through the plan inputs these data record, not product, assortment or place. The models are a world model that carries each item’s state forward under the price and promotion plan, the same network without that state, LightGBM, the Temporal Fusion Transformer, TimesFM, Chronos-Bolt, Timestral, seasonal naive, and an equal-weight average of the world model, LightGBM and TimesFM given the plan. The answer changed with the grain. For an item by store and an item across stores, the average had the lowest error on seven and eight of nine data sets, and the 90% confidence set held a median of 1 and 2 models. For a category by store, a category and the chain total, the lowest-error model differed between data sets and a median of 4 of nine models were not separated by the four to six windows available. For category-by-store planning cost, the average or LightGBM was lowest on seven of eight data sets, and the world model was lowest only on M5; it was behind LightGBM item by store on six of eight. Every trained model used the plan, and on the weekly promotion-driven data pretrained models were less accurate without it. These results hold for these data sets; a forecast solved across all four Ps, or a retailer’s own data, may give different results. Every model, test and change that lost is reported.
[AI-158] M-JEPA: Predictive Self-Supervised Learning for Molecular Graphs with Scaffold-Shift Evaluation on Tox21
链接: https://arxiv.org/abs/2610.25163
作者: Karthik Iyer,Nasser Sabar
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 28 pages, 9 figures. Submitted (pre-peer-review) version; the final published version appeared in J. Chem. Inf. Model
Abstract:Self-supervised molecular representation learning is increasingly used to improve transfer on small, label-limited property prediction tasks, yet commonly used contrastive objectives can be sensitive to view construction and are often evaluated with protocols that report only discrimination. We adapt Joint Embedding Predictive Architectures (JEPAs) to molecular graphs by masking chemically coherent connected subgraphs and training a context encoder to regress the masked target embedding produced by an exponential-moving-average (EMA) teacher. To evaluate predictive self-supervision under realistic distribution shift, we introduce a leakage-resistant, compute-matched three-phase protocol: (i) an 80-run objective screening on ESOL with fixed compute budgets, (ii) JEPA-only refinement of masking and augmentation choices, and (iii) per-assay Tox21 evaluation under Bemis-Murcko scaffold splits with joint discrimination-reliability analysis and attribution diagnostics. Under matched compute, M-JEPA achieves substantially lower ESOL proxy RMSE than a compute-matched InfoNCE baseline (mean RMSE 2.12 vs 3.75; mean paired dRMSE 1.61; 95% bootstrap CI [1.52, 1.72]; Wilcoxon p=2.4e-4) and typically completes runs faster (~15-25 min vs ~35-45 min on a single GPU under identical hardware), albeit with higher runtime variance. On Tox21, improvements in ROC-AUC/PR-AUC are assay-dependent and do not uniformly imply better calibration; reliability diagrams and Brier/ECE metrics reveal endpoints where fine-tuning improves ranking but leaves probabilities miscalibrated.
[AI-159] Learning without Overwriting: Structural Knowledge Fusion for Continual Learning
链接: https://arxiv.org/abs/2610.25150
作者: Kanishk Bakshi,Amitansu Pattanaik,Kathiravan Srinivasan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 20 pages, 7 figures, preprint
Abstract:Catastrophic forgetting in sequential deep learning arises from a fixed-capacity architectural constraint: parameter updates encoding new tasks perturb weight subspaces supporting prior representations. We introduce the Knowledge Fusion Network (KFN), which mitigates the stability-plasticity trade-off in class-incremental learning through novelty-gated structural expansion rather than parameter negotiation within a shared capacity budget. KFN decouples task acquisition from knowledge consolidation: each incoming task is learned in an isolated lightweight specialist network (2.8M parameters) and integrated into the global model via a Structural Fusion Module that allocates dedicated channel capacity conditioned on a quantitative novelty score. Dual structural isolation–channel-space separation with temporally phased gradient unfreezing–protects consolidated knowledge. On ten-task Split CIFAR-100, KFN achieves 11.94% minimum task accuracy and 71.16% task-10 accuracy, whereas shared-parameter baselines degrade severely: LwF to 9.07% average accuracy (0.00% min), Replay++ to 8.91% (85.08% forgetting), and Replay to 16.21% (73.44% forgetting); KFN’s forgetting (57.92%) is 15.5 pp lower than Replay. Furthermore, KFN prevents logit-scale drift, maintaining near-symmetric task performance. Across CIFAR-100 and EuroSAT satellite imagery, KFN’s class-incremental retention (72.7% and 42.5%) substantially exceeds the best baselines (29.6% and 0.94%). Crucially, KFN achieves parameter-efficient linear scaling, adding 25% of the base parameter count per step and requiring 25.1% of naive backbone duplication over ten tasks (28.1M vs 112.0M). With a 37.1% peak GPU memory reduction over full retraining, KFN establishes novelty-gated expansion as a practical path beyond fixed-capacity architectures.
[AI-160] Spectral Adaptive Graph Transformer for Efficient and Scalable Traffic Forecasting
链接: https://arxiv.org/abs/2610.25140
作者: Amirhossein Ghaffari,Susanna Pirttikangas,Ekaterina Gilman
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 16 pages, 2 figures, 9 tables. Code: this https URL
Abstract:Traffic forecasting models based on spectral graph convolution offer an interpretable account of how information is mixed between sensors, but they rely on an eigendecomposition of the graph Laplacian. It costs cubic time, is fixed by the road network, and provides a symmetric operator that cannot represent the directed propagation of traffic congestion. Attention-based spatio-temporal transformers remove these limitations but generate a dense sensor-by-sensor attention map at each layer and time step, so memory, rather than accuracy, becomes the binding constraint on large networks. We introduce the spectral adaptive graph transformer, whose spatial operator is learned rather than derived from the Laplacian, directed rather than symmetric, and conditioned on the current traffic state rather than a static state, at a cost linear in the number of sensors. A sparse road graph diffusion branch provides the strictly local propagation that a low-rank operator cannot express, and a zero-initialized gate learns the balance between the two. Before spatial mixing, a unit-mean-gain frequency-bin attention module reweights each sensor’s short-term periodicities according to its identity and current context. Across six public benchmarks spanning 170 to 1843 sensors, evaluated over three random seeds, the proposed model attains the lowest error of any single end-to-end model on one benchmark and remains competitive with the strongest current attention models on the others, while using about a quarter of their memory and training about twice as fast.
[AI-161] ClinSDT: LLM -Encoded Clinical Semantic Guidance for Sepsis Treatment via Decision Transformer CIKM2026
链接: https://arxiv.org/abs/2610.25121
作者: Yanyan Wu,Guanyu Zhu,Changbin Zhao,Jin Zhang,Dexu Yu,Tianmin Sun,Jining Luan,Chunxiao Li,Hanwen Du,Youhua Li
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 12 pages, 6 figures. Accepted at CIKM 2026 (Full Research Track, Oral Presentation)
Abstract:Offline sepsis treatment aims to learn treatment policies from retrospective ICU trajectories. Decision Transformer-based methods provide a natural sequence modeling framework for this task, but they mainly rely on raw physiological states and historical clinician actions. Such raw trajectory tokens may not explicitly capture clinical semantics that clinicians use in practice, such as patient severity, recent treatment response, and current risk priorities. To address this limitation, we propose ClinSDT, an LLM-assisted Decision Transformer framework for offline sepsis treatment. ClinSDT uses a frozen LLM as a semantic bridge rather than a treatment decision maker. It constructs three clinical semantic views from ICU trajectories, namely Profile, Context, and Advisory, and inserts their encoded representations as independent tokens into a Decision Transformer policy. To support policy refinement and rollout, ClinSDT further trains a Treatment-Response Forecaster to estimate the next physiological state, SAPS-II change, and predictive uncertainty from state-action inputs. In the second training stage, the forecaster screens local counterfactual treatment candidates, and only low-uncertainty candidates with lower predicted SAPS-II increase are distilled into the policy after semantic regeneration. During inference, ClinSDT performs closed-loop treatment rollout by refreshing semantic inputs from the currently available patient state. Experiments on MIMIC-III and MIMIC-IV show that ClinSDT achieves consistent relative improvements over representative imitation learning, offline reinforcement learning, and Decision Transformer baselines under the same rollout protocol. Code and data are available at this https URL.
[AI-162] Compile Once Run Cheap: Frozen Portable Skills for Embodied Agents and What Recursive Self-Improvement Adds
链接: https://arxiv.org/abs/2610.25114
作者: Jingwen Bai,Guochao Bai,Canbo Zhang,Shuangqi Li,Minchao Hu,Weichun Huang,Shuang Liu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 28 pages, 4 figures, 13 tables. Code and data: this https URL
Abstract:Embodied AI lacks the data-rich pretraining that gave language models their agents and skills, so new manipulation capability costs demonstrations per task or a frontier model in every episode. We study a third route: a strong general model acts only as an offline author that writes, runs, diagnoses and repairs a manipulation skill in simulation until it passes an exam on layouts drawn after freezing; the frozen skill then runs as code alone or under a cheap caller that reads only its description. We call this skill-level recursive self-improvement Skill-RSI and specify a compile protocol (isolated, audited author; hidden validation layouts; no-feedback and inherit-only controls; tamper-evident ledger) and a reuse protocol that runs the identical files with executors from another model family. On 7 RoboTwin 2.0 tasks with cluttered tables, Claude Code compiled 4 tasks into 7 frozen skills, 5 from a first draft needing only qualification, at USD 4.29-8.22 of author cost each (USD 55 per skill counting the 13 failed campaigns). The other three tasks did not compile in 18 campaigns, including five with GPT-6 and a pre-registered 150-layout run: on the hardest task every campaign scored 78-95 of 100 but had 3-16 safety aborts, failing the gate of 95% success with zero aborts; a 100-episode exam bounds an abort rate only to 3.6%. On hidden layouts, repair that inherits the previous version beat from-scratch rewriting in 8 of 8 controlled campaigns (sign test p = 0.004); an inherit-only lineage attributes most of the gap to inheritance, with development feedback adding a little (3 of 3, p = 0.125). Write-once campaigns found no detectable first-draft effect of practice (minimum detectable effect 14 points). The frozen skills kept 148/150 successes on new layouts with no model and 147/150 with a DeepSeek-flash caller at USD 0.00028-0.00054 per episode; no direct-control baseline was run.
[AI-163] Self-organizing reinforcement learning: a swarm approaches the explore-exploit optimum
链接: https://arxiv.org/abs/2610.25113
作者: Galen J. Wilkerson
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Social and Information Networks (cs.SI); Mathematical Physics (math-ph); Adaptation and Self-Organizing Systems (nlin.AO); Biological Physics (physics.bio-ph)
备注: 16 pages, 9 figures, code and data: this https URL
Abstract:The explore/exploit trade-off is the core problem of reinforcement learning (RL). We show that a minimal foraging swarm approaches its optimum - the ideal-free / Nash allocation of a multi-agent resource-sharing game (an equilibrium, not a return maximum) - with no learner and no reward signal. Agents explore by heavy-tailed (Levy) flights, recruit neighbours to a discovery through a threshold cascade on their proximity network, and disperse as a patch crowds; this cascade cycle balances exploration and exploitation and drives foragers toward the ideal free distribution. The swarm attains the shape of this optimum (committed foragers track richness with slope 0.94) but not its exact equilibrium - it stays measurably exploitable - reaching it dynamically (~150 steps) through the coupling of recruitment to movement: remove the coupling and matching collapses (0.94 - 0.08). The RL sits at the level of the problem and the allocation reached, not the micro-rule: no step evaluates payoff (reward-free by construction, though the crowd signal is informationally tied to payoff), and the swarm reaches the allocation without learning it (quality plateaus and re-settles after resources move, no cumulative gain). Against a trained baseline the comparison is two-sided: in an easier, disembodied game a learner is more precise, while in the same embodied environment the reward-free swarm reaches ~65% of a validated oracle ceiling at zero training and leads until trained PPO and centralised-critic (CTDE) learners overtake it (~74-75%, 16 seeds, all above the swarm mean) after ~4,000 of 15,000 training steps - its advantage is in budget, not asymptote. In this one setting we read it not as beating RL but as a strong zero-shot, reward-free operating point: a demonstration that the self-organizing physics of a minimal swarm can carry a collective close to the optimum RL seeks, with none of RL’s machinery.
[AI-164] Pre-Query Retention with Bounded Bit and Word Retrieval
链接: https://arxiv.org/abs/2610.25098
作者: Pranav Singh,Prashant Singh
类目: Information Theory (cs.IT); Artificial Intelligence (cs.AI)
备注: 24 pages, 4 figures. Complete proofs and experimental details included. Both authors contributed equally
Abstract:An agent can retain information before receiving a future question and later retrieve evidence from an archive. Evaluating this arrangement requires accounting for both retained information and the width of each retrieval response. Decision-tree capacity bounds describe individual sets of distinguishable records, but a memory writer must encode every possible record before the query is known. We study fixed-length retention beside an immutable binary archive, with a worst-case limit on raw bit or word reads. We characterize feasible one-bit-read memory cells and construct encoders that partition the complete record space. Supporting pairwise exclusive disjunction and conjunction with one bit read is equivalent to supporting every Boolean query. For multiple reads, we combine explicit encoders with decision-tree counting and an entropy converse. The word-read bounds differ by at most two retained bits when the read budget is at most half the archive length. Under an explicit condition on word width and archive length, raw-word caching attains the exact optimum among all writers. A related condition preserves this optimum under positive average error. Exhaustive checks validate the implementations. In a matched dialogue-agent study, a categorical sufficient statistic preserves source decisions with less storage and lower decoding latency than the universal encoder. A catalogue-specific index reduces private state further and requires additional shared storage. The results characterize how response width and future query structure determine retention requirements.
[AI-165] DAGLoop: Aligning Method Design with Experimental Feedback in Automated AI Research ICLR2027
链接: https://arxiv.org/abs/2610.25089
作者: Dengyuxing Wang,Yonggang Zhang,Qianshu Cai,Huajiang Zheng,Jun Song,Yangqiu Song,Yike Guo
类目: Artificial Intelligence (cs.AI)
备注: 26 pages, 4 figures, 6 tables. Submitted to ICLR 2027. Code: this https URL
Abstract:Automated AI research agents aim to improve machine learning methods by iteratively proposing, implementing, and evaluating candidate methods. Existing work uses the outcome of each evaluation to determine how the next candidate should be revised for performance improvement. However, a candidate often combines multiple design decisions, and its overall outcome alone does not reveal which decisions produced their expected effects, whether their required conditions were satisfied, or which decisions should be retained or revised. Inspired by ablation studies in the machine learning community, we propose DAGLoop to test the expected effect of a design decision by comparing implementations with and without that decision while preserving the other decisions it requires. Specifically, DAGLoop translates each design decision’s rationale into a testable expectation and pairs it with a targeted experiment, forming an experimental specification for assessing the expected effect under its required conditions. Using these specifications, we formulate ablation planning as the construction of a directed acyclic graph (DAG) from the candidate method’s design: nodes associate decisions with tests, and edges encode dependencies. The DAG specifies which decisions each experiment must retain and supports joint tests when an expected effect requires several decisions to act together. Outcomes are recorded with the tested decisions, expectations, and conditions to form structured research memory. This memory guides the agent in retaining experimentally supported decisions, revising unsupported ones, and planning further experiments when the evidence is unresolved. Experiments on dozens of research tasks demonstrate substantial performance improvements achieved by DAGLoop across diverse executable research settings. Code and evaluation scripts are available at this https URL.
[AI-166] Auto-MD-bench: Evaluating Autoresearch Agents for Real-World Medical Tasks
链接: https://arxiv.org/abs/2610.25069
作者: Shenqi Zhou(1 and 2),Qiran Zou(1),Wenhao Zhao(1),Zhenyu Liao(3),Dianbo Liu(1) ((1) National University of Singapore, (2) Zhejiang University, (3) Amazon)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 24 pages, 5 figures, 22 tables. Code: this https URL ; data: this https URL
Abstract:Autonomous agents are increasingly capable of developing machine learning (ML) models end to end. In medical practice, however, strong performance on the target metric does not fully characterize how good the models are in real-world use cases. For example, medical ML models should also provide well-calibrated risk estimates and remain robust across patient populations. Yet autonomous model development is typically guided by task objectives that emphasize simple predictive discrimination (e.g. AUROC). This raises a key question: are these objectives sufficient to guide agents toward models that meet the broader requirements of real-world medical use? In this study, we introduce Auto-MD-bench, a benchmark of 15 real-world medical tasks from 10 clinical datasets, spanning survival analysis, classification and sequential prediction, with external validation across cohorts, hospitals and time periods. We evaluate four agents: AIDE, AI Scientist v2, Codex, and Autoresearch. The strongest agents achieve competitive primary scores, yet their predicted risks are less calibrated than those of the simple baseline models, which could lead to severe medical accidents. Moreover, models with nearly identical primary scores can exhibit markedly different calibration, while performance varies substantially across population shifts. These findings show that primary task performance alone is insufficient to assess autonomous clinical modeling and multiple different metrics widely used in healthcare practice need to be taken into consideration.
[AI-167] One Skill Takes Over: Clashing Agent Skills and a One-File Fix
链接: https://arxiv.org/abs/2610.25064
作者: Parsa Razmara,Seyedarmin Azizi,Ardalan Aryashad
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent Skills are instruction files that tell a coding agent how to do one job. Users install skills from many authors, so two skills for the same job can be active at the same time. We ask what a model does when both skill files are in its context: does it combine their compatible rules, or follow one skill and drop the other? We build SkillClash, 16 pairs of same-job skills in which code checks every rule and a reference output shows the best result that can be reached. On seven models from five families, from 4B open models to Qwen3-235B and GPT-4.1, a skill’s rules are followed 93 to 99% of the time when it is loaded alone and 61 to 62% when a same-job skill is loaded next to it. Most outputs show takeover: every rule that only one skill asks for is followed, and every rule that only the other skill asks for is dropped. The worked examples inside the skills drive this. When each skill file instead ends with one example that follows both skills, takeover falls to 1 to 15% on the six models we tested. Telling the model which skill has priority does not bring the dropped skill back, and telling it to apply both skills brings the dropped skill back only on the two strong models and leaves the interacting rules broken. One merged skill file written by the strongest model we used, Claude Opus 5.5, lets all seven models follow 96 to 99% of the achievable rules. Most of its gain over a plain list of both skills’ rules comes from its one worked example, and such an example helps only when it is correct. Code and data: this https URL
[AI-168] HCFM: Helmholtz Component Flow Matching for Time Series Anomaly Detection ICDE2027
链接: https://arxiv.org/abs/2610.25059
作者: Shyam Sundar Saravanan,Tie Luo,Sanjay Madria
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Databases (cs.DB)
备注: To appear as a conference paper in ICDE 2027
Abstract:Flow Matching (FM) is an efficient simulation-free approach for training flow-based generative models, making it a promising foundation for scalable time-series anomaly detection (TSAD). However, standard FM learns a monolithic vector field in which nominal transport and local density-changing behavior are entangled. This creates a central challenge for TSAD; local expansion and contraction carry crucial anomaly-sensitive information, yet isolating this signal requires expensive Jacobian-trace estimation over the entire field. We propose \emphHelmholtz Component Flow Matching (HCFM), an unsupervised TSAD framework that parameterizes the vector field prediction network with separate transport, compression, and residual heads. The transport head is constrained to be divergence-free, so divergence estimation is restricted to the compression head, while a lightweight residual head absorbs the remaining mismatch. This structured parameterization improves scalability and detection performance by separating volume-preserving nominal transport from density-changing compression, yielding cleaner anomaly signals from distinct deviations of nominal behavior. Experiments on large-scale univariate and multivariate TSAD benchmarks show that HCFM consistently improves detection accuracy while maintaining efficient inference compared with 16 baselines. The code is available at this https URL
[AI-169] SteadyThought: Mitigating LLM Under-Thinking via Thought-Level Preference Optimization NEURIPS2026
链接: https://arxiv.org/abs/2610.25045
作者: Zhenyue Peng,Lin Yang,Jie Wang,Xiaoqi Ni,Hanzhu Chen,Xize Liang,Yanjun Wu,Rujing Wang,Kaiwen Zhou,Jianye Hao
类目: Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026
Abstract:Flexible switching between reasoning trajectories (i.e., thought switching) has significantly enhanced the reasoning capabilities of Large Reasoning Models (LRMs). However, existing models often switch excessively yet fail to sustain promising reasoning thoughts, a phenomenon termed “under-thinking”. While recent efforts suppress switching to mitigate this, such over-correction may discard valuable trajectories. To address this challenge, we propose Steady Thought (ST), a novel thought-level preference optimization framework. ST formalizes under-thinking as a preference issue at switching points, forcing single-path continuations from identified points to yield high-quality reasoning. Then, ST performs thought-level preference optimization by treating the newly generated response as preferred and the original one as dis-preferred. Experiments across multiple models and datasets show that ST effectively reduces token consumption, achieving reductions of up to 63.8%, while maintaining strong reasoning accuracy. Further analysis suggests that ST helps models acquire a generalizable ability to reduce redundant switching across domains and languages. Code is available at this https URL.
[AI-170] VLA-Rolling: Fewer Denoising Steps Better Control via Adaptive Action Reuse
链接: https://arxiv.org/abs/2610.25043
作者: Sicheng He,Yi Pan,Qiuting Yu,Jiaming Huang,Qi Lu,Shuyan Wu,Hongyan Feng,Xuanyu Liu,Jun Zhang,Jie Zhang,Wenqi Zhang,Xuhong Zhang,Jianwei Yin
类目: Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Flow-matching vision-language-action (VLA) models generate an action chunk by iteratively denoising Gaussian noise, but typically execute only a short prefix before replanning with updated observations. Under this receding-horizon paradigm, the unexecuted remainder—the action tail—is discarded, and the next chunk is regenerated from fresh noise. We show that this discarded tail retains transferable denoising progress: action-tail initialization can match or outperform a native 10-step Gaussian baseline using fewer denoising steps. However, the optimal denoising budget varies substantially across samples, indicating that tail reuse should be adapted to the updated environment. Based on this observation, we introduce VLA-Rolling, a lightweight inference framework that reuses action tails through three stages: action-tail inheritance, per-action confidence estimation, and confidence-guided refinement through renoising and denoising budget allocation, while the underlying VLA remains frozen. Across \pi_0.5 , SmolVLA, X-VLA, and MolmoAct2 on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0, VLA-Rolling reduces action-chunk inference cost in all 16 model–benchmark pairs, achieving up to a 1.48\times speedup, while improving success rate in 14 pairs by up to 15.08 percentage points. These results show that action tails are not merely discarded predictions, but reusable intermediate states that enable more efficient and accurate VLA inference. These results demonstrate that the action tail can support the generation of better actions with fewer denoising steps. Our project is available at this https URL.
[AI-171] Synergy-Aware Residual Flow Matching for Single-Cell Perturbation Prediction
链接: https://arxiv.org/abs/2610.25021
作者: Jingrui Wu,Wangyun Dan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Predicting single-cell responses to genetic and chemical perturbations is important for understanding regulatory mechanisms and prioritizing therapeutic interventions. Existing methods have made progress by modeling perturbation-conditioned expression shifts with neural networks, graph priors, foundation-model representations, or distributional generative models. However, combinatorial perturbation prediction remains challenging because much of the response can be explained by additive single-perturbation effects, while the biologically informative signal often lies in a smaller non-additive residual induced by perturbation interactions. Directly generating the full post-perturbation expression profile may therefore allocate substantial capacity to the dominant additive component and underemphasize synergy-specific structure. To address this issue, we propose \textbfSynGeneFlow, a synergy-aware flow matching framework that anchors the source distribution at an additive prior and learns a residual flow toward the true perturbed state. SynGeneFlow further incorporates perturbation-conditioned gene interaction modeling, where a static gene-gene mask provides biological structure and a perturbation-dependent deformation bias dynamically modulates gene-gene coupling. To preserve population-level response structure, we align both endpoint distributions and biological residual distributions with MMD-based regularization. We evaluate SynGeneFlow on genetic and drug perturbation benchmarks under observed-single, held-out, and cross-modality settings. The results demonstrate that residualizing flow matching around additive perturbation effects provides a biologically grounded and robust framework for single-cell perturbation response prediction.
[AI-172] LLM Agents as Suppliers of Physical System Models: Calibration and Acceptance in Modelica
链接: https://arxiv.org/abs/2610.25019
作者: Jörg Frochte,Herbert Schmidt
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注:
Abstract:To procure a model of a device, a client hands a supplier a description, sometimes with recorded data or a target response, and accepts the model against this handover. We let LLM coding agents act as the supplier on handovers with data, building Modelica models from library components for four simulated systems and a real switching magnet, and ask what acceptance certifies. Several agents reach the noise floor, but a handover admits a set of models: the description leaves parts of the structure open, and the data narrow it only to the agreed accuracy. We therefore accept by membership in this set, not by matching one reference, and add layers above behaviour: a physical invariant exposes an energy-creating model at the noise floor. Interpretability becomes a property of the delivered model, tested on a variant, read where none exists and shaped through the handover. Most failures are fits delivered with a systematic residual on their own fitting data, where the handover left open how close is close enough. In a follow-up, stating a threshold makes the agents compare against it, and more deliveries pass. The argument applies wherever a handover underspecifies a model meant to be read, not only run.
[AI-173] UpLoop: Upcycling Large Language Models with Gated Loop
链接: https://arxiv.org/abs/2610.25004
作者: Tieyuan Chen,Huabin Liu,Yingtao Zhang,Shijie Li,Tengyu Pan,Weiyao Lin,Jianguo Li
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Looped large language models trade parameter storage for recurrent computation, but most are trained from scratch. In practice, pretraining with looped depth is expensive, making post-training upcycling of a pretrained non-looped LLM an attractive alternative. We introduce \textbfUpLoop, a post-training recipe based on \emphlayer-wise recurrence. Instead of looping a whole model or a middle block, UpLoop loops each selected layer locally, preserving the pretrained depth hierarchy and avoiding feature-space mismatch. To make loop execution useful, we add hidden-wise gates on the attention and MLP residual branches, so later loops act as progressive refinements rather than redundant rewrites. Empirically, UpLoop breaks the commonly observed saturation at two loops ( R=2 ) in prior looped models. Starting from Qwen3-4B Base and applying SFT, UpLoop improves the average accuracy across 20 benchmarks by 6.17% at R=3 . We use R=3 as the main setting for the best accuracy–cost trade-off, while both native and extrapolated R=4 yield further gains.
[AI-174] CorrectRTL: Boost RTL Generation Quality Through Automated Self-Correction with Instruction Decomposition
链接: https://arxiv.org/abs/2610.24977
作者: Jing Wang,Shang Liu,Hangan Zhou,Wenji Fang,Yuchao Wu,Yugao Zhu,Zhiyao Xie
类目: Artificial Intelligence (cs.AI)
备注: This paper is accepted by ICCAD’26
Abstract:Large language models (LLMs) have shown strong promise in specification-to-RTL generation. Among the explorations, the self-correction paradigm has significantly advanced LLM-based RTL generation. However, two challenges limit further progress. Existing benchmarks are saturated by advanced LLMs, leaving little room to evaluate the agentic self-correction framework. The monolithic self-correction paradigm rewrites the entire design at once, where LLMs are susceptible to error confusion and corruption of originally correct code. To address the benchmark challenge, we contribute ScaleBench, a more challenging benchmark of 200 complex RTL designs. To address the challenge in methodology, we propose CorrectRTL, an agentic framework that realizes a new decompose-then-correct paradigm, in which the LLM agent explicitly decouples scaffold construction from detailed implementation. CorrectRTL contributes three core techniques: (1)~\emphStructure-oriented decomposition, (2)~\emphSnippet-based RAG, and (3)~\emphSVA-based self-correction. On ScaleBench, CorrectRTL improves Pass@1 from 20.5% to 49.0% (+28.5%). On VerilogEval, CorrectRTL achieves up to 96.1% Pass@1, demonstrating consistent improvements across both simple and complex designs.
[AI-175] PRECOCIAL: Pretraining with Code Curriculums via Iterative Search for Accelerated Learning
链接: https://arxiv.org/abs/2610.24970
作者: Edward Gu,Ali Payani,Kevin Ellis
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 22 pages, 12 figures, 21 tables
Abstract:Fully synthetic pretraining data generated by symbolic programs can bootstrap downstream training on naturalistic tasks, such as language modeling, but the generating programs have so far been designed by hand. We introduce PRECOCIAL, a method for automatically optimizing programs that generate synthetic pretraining data. Starting from an initial generator program, an LLM iteratively rewrites the program, and each rewrite is scored by pretraining a randomly-initialized transformer on its data and fine-tuning it on the downstream task. Because every score requires training, PRECOCIAL keeps a single network for the entire search, continuing its pretraining for each accepted program instead of training a new network per candidate, making it significantly more efficient. On four language and code datasets, at a fixed fine-tuning budget, optimized programs reach lower perplexity than hand-designed curricula, published procedural generators, random program mutation and the program the LLM writes in the first iteration of the search, and reach the same perplexity as the hand-designed curricula with 9-33% fewer fine-tuning steps. PRECOCIAL also ends with lower perplexity than FunSearch and AlphaEvolve, which train a new network for every candidate, at the same number of evaluations and a lower cost. We also observe that an optimized program generator from one domain transfers to datasets in other domains. Additionally, optimized programs allow for better downstream performance on four different reasoning and problem-solving tasks.
[AI-176] owards Human-AI Complementarity for AI Safety
链接: https://arxiv.org/abs/2610.24969
作者: Sander Lanbo Shi,Miranda Anna Christ,Eric Modesitt,Andrew Louis,Konrad Cheng,Brandon Thai Tran,Sewoong Lee,Ratnakar Pawar,Gaia Molinaro,Juliana Bain,Nandika Donthi,Maja Pavlovic,Annya Dahmani,Francesca Bianco,Abhishek Shah,Tyler Crosse,Arth Singh,Karissa Huang,Sukanya Krishna,Konstantinos Voudouris,David Demitri Africa,Joshua Jacob,Rishub Jain
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Safely overseeing AI is increasingly challenging for the humans doing it, as they must evaluate longer and more complex outputs with every new model generation. To mitigate this challenge, we could leverage complementarity: where humans and models make different mistakes, combining their judgements could yield an oversight signal stronger than either produces alone. We study this opportunity in realistic, safety-relevant settings, contributing the largest known suite of such tasks (3,957 instances across 8 datasets and 6 domains) as well as a platform and data pipeline for measuring human, AI, and human-AI performance on matched items. We collected 35,613 judgements from 416 human raters, finding a considerable opportunity for a complementarity effect. We also found that with the assistance methods we tested, human raters corrected the AI’s errors less often, even when overall human accuracy rose, and the hybridisation methods we tested added little over the better single source on average. While complementarity is an important and feasible engineering target for safe oversight, achieving it remains an open challenge for future work.
[AI-177] Label-Space Mismatch Distorts Cross-Cohort Evaluation in Breast Cancer Molecular Subtyping
链接: https://arxiv.org/abs/2610.24962
作者: Suxing Liu,Byungwon Min,Zhijian Zheng,Jiahao Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Quantitative Methods (q-bio.QM)
备注:
Abstract:Cross-cohort evaluation of breast cancer subtype classifiers is unreliable when source and target cohorts use different annotation standards: METABRIC labels six classes, PAM50 defines five, and a claudin-low prediction cannot be correct under a five-class standard. A classifier transferred from 1,369 METABRIC patients (train n=958) to TCGA-BRCA (n=987) and GSE21653 (n=266) separates three evaluation operations routinely conflated in cross-site comparisons. (1) Label-space restriction - an evaluation-rule change, not a retraining gain - is the dominant correction (73.76%-79.74% on TCGA-BRCA; 63.16%-68.42% on GSE21653); (2) EM prior correction on the reconciled space is an optional calibration step (Brier fidelity up, macro F1 down; to 80.65% and 74.06%); and (3) the target label space is not recoverable from unlabelled predictions (EM retains prior 0.0751 on the absent class; calibrated detectors fail), so compliance must come from cohort documentation. The protocol exposes why internal validation overstates cross-cohort value: multimodal fusion gains do not transfer (no configuration beats mRNA alone, n=694; McNemar p=0.13-1.00), and the genomic panel adds nothing to pathological complete response prediction over clinical variables (DeltaAUC = +0.0045). Clinically, without label-space compliance, cross-site accuracy comparisons misattribute annotation-standard differences to model skill; Luminal A/B and claudin-low boundary errors map to distinct treatment decisions; and subtype accuracy is a weaker portable endpoint than survival discrimination (external C-index 0.577-0.584). Reporting the cohort’s allowed label set and applying label-space compliance are therefore prerequisites before any cross-cohort performance claim.
[AI-178] MemPrice: Online Memory Allocation via Adaptive Pricing for Long-Horizon Agents
链接: https://arxiv.org/abs/2610.24955
作者: Yiming Zong,Shiyuan Piao,Jiashuo Jiang,Fugee Tsung
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-horizon LLM agents face a fundamental allocation problem as interaction histories grow: which information to retain, how much context to devote to it, and at what granularity. Existing memory mechanisms largely focus on relevance, retrieval, or compression, leaving this allocation trade-off implicit. We introduce MemPrice, which treats active context as a shared resource and dynamically allocates it across memories and representation levels. MemPrice turns accumulated memory pressure into an adaptive price on context, encouraging detail only when its estimated retention value justifies the additional footprint. This pressure persists across memory refreshes, allowing the contents of memory to evolve while preserving long-run control over context usage. We show that irreversible compression can force linear worst-case regret against hindsight following its own memory trajectory, and introduce a same-state benchmark to evaluate allocation on MemPrice’s realized memory states. MemPrice enforces the hard context budget at every round and regulates long-run occupancy through adaptive pricing, without prior knowledge of the interaction horizon. Under priced-frontier regularity, its average allocation-regret bound against this benchmark vanishes when accumulated estimation error is sublinear. We evaluate the framework under matched memory budgets on LongMemEval, LoCoMo, AppWorld, and WebShop across three foundation models, spanning both query-blind memory construction and goal-directed agent context management.
[AI-179] ControlJEPA: Principled Trajectory Regularization via Lyapunov Tube Loss NEURIPS2026
链接: https://arxiv.org/abs/2610.24951
作者: Thai-Khanh Nguyen,Nhi Y. Pham,Uyen N.B. Vo,Thieu N. Vo,Cuong Pham,Tan M. Nguyen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: NeurIPS 2026
Abstract:Next-token prediction leaves hidden-state trajectories largely unconstrained, while existing geometric regularizers primarily control the direction rather than the magnitude of trajectory deviations. We introduce \emphControlJEPA, a trajectory regularization framework based on a discrete-time Lyapunov contraction condition from practical Input-to-State Stability theory. Its Lyapunov Tube Loss penalizes violations of this condition on a normalized, scale-invariant measure of transversal deviation. When the contraction condition holds at every step, the hidden-state trajectory remains within an explicit tube around the reference secant. In the partial-ISS regime, we bound mean transversal energy, temporal variance, and average tube width using the Lyapunov loss and contraction parameters, with horizon-independent variance bounds under uniformly bounded violations. ControlJEPA requires no architectural modifications. Across six benchmarks and six model families, it consistently improves accuracy and data efficiency over standard fine-tuning and prior trajectory regularizers. The resulting smoother transversal profile also persists empirically during inference, when the regularizer is inactive. The code is publicly available at: \urlthis https URL
[AI-180] Robust Direct Preference Optimization under Preference Label-Flip Attacks
链接: https://arxiv.org/abs/2610.24945
作者: Chenye Yang,Weiyu Xu,Lifeng Lai
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Direct Preference Optimization (DPO) is widely used for offline Reinforcement Learning from Human Feedback. However, recent works have shown that DPO can be easily manipulated by adversarial preference label flip attacks, which are specifically designed to mislead the learner toward an attacker-chosen target policy. To address this issue, we design robust DPO methods that can defend against such attacks. Building on the property that lable flips induce parameter-independent gradient shifts in log-linear DPO, we propose Trim- K Robust-DPO, which removes the K highest-impact pairs, and Soft-Trim- K Robust-DPO, which smoothly downweights them. We establish deterministic gradient perturbation bounds uniformly over all attack sets containing at most K flips, explicitly separating defense-induced bias from surviving attack effects. Under bounded-region assumptions, these bounds yield time-uniform parameter and policy deviation guarantees. We further quantify how quadratic regularization improves curvature while introducing additional bias. Experiments on Stanford Human Preferences, HH-RLHF, and UltraFeedback-Binarized validate our analysis.
[AI-181] rustAgent -RFF: Agent ic Diagnostic Planning for Reliable RF Fingerprinting under Domain Shift
链接: https://arxiv.org/abs/2610.24930
作者: Chuan Liu,Tianya Zhao,Xuyu Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 10 pages, 6 figures, 7 tables. Submitted to INFOCOM
Abstract:Radio frequency fingerprinting (RFF) models can achieve accuracy under source domain conditions but become unreliable when the receiver, distance, collection day, or environment changes. Incorrect predictions in unseen domains may receive high softmax confidence, making confidence thresholding insufficient for trustworthy deployment. We propose TrustAgent-RFF, a prediction level reliability framework for RFF that operates with a frozen in-phase and quadrature (I/Q) classifier without target domain labels. For each device prediction, the framework acquires radio frequency evidence: counterfactual stability, prototype consistency, acquisition domain similarity, and signal quality. A source calibrated trust estimator combines these diagnostics with classifier confidence, prediction margin, and evidence availability indicators. Source derived lower tail penalties and cross module interactions reduce trust when diagnostic signals indicate weak support. A large language model (LLM) based planner adaptively selects the next diagnostic action based on evidence, trust state, action costs, and remaining budget. A deterministic source calibrated decision layer updates the prediction state as Accept, Reject, or Defer. Experiments across five RFF datasets and classifier backbones reveal an average source to target accuracy drop of 38.7% and show that many incorrect target domain predictions remain highly confident.
[AI-182] Means-Ends Analysis Helps LLM s Plan in Continuous-State Domains
链接: https://arxiv.org/abs/2610.24928
作者: Jiayi Li,Carl Kingsford
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:LLM planning research has traditionally focused on discrete-state domains, where actions form a finite, enumerable set. Existing LLM planning frameworks largely assume actions are enumerable, modeling planning as a search process over candidate actions. However, in continuous-state domains, where actions form a continuum and cannot be enumerated, this assumption breaks down. We find a remedy in Newell and Simon’s Means-Ends Analysis (MEA), a general problem-solving technique rooted in heuristics from human psychology. We develop Medaka, an MEA-inspired structured prompting framework that encourages recursive goal decomposition, algebraic derivation, and per-step state tracking. To evaluate native LLM planning abilities in continuous-state domains, we develop Liquidworld, a systematic testbed on liquid mixing. On Liquidworld, Medaka can improve performance across model families, suggesting that classical means-ends analysis elicits LLM-native reasoning abilities that are helpful for continuous-state planning.
[AI-183] RAFT: Redundancy-Aware Input Filtering for Temporal Graph Neural Network Training
链接: https://arxiv.org/abs/2610.24875
作者: Yue Dai,Sarah Zhou,Youtao Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Temporal graph neural networks (TGNNs) have shown strong performance on dynamic graph learning tasks by continuously updating node memories from evolving interactions (i.e., events) and leveraging these memories for downstream predictions. As graphs grow rapidly in modern applications, training requires processing an increasing number of events; meanwhile, each event can update node memories used by later events, creating dependencies that constrain parallel execution and limit the benefits of additional computing resources. These costs motivate an upstream question: Are all events, and the dependencies they introduce, necessary for learning? We investigate this question through two roles of temporal events: prediction targets and historical events sampled as inputs. We observe that targets involving stable node memories tend to exhibit lower prediction loss, while filtering historical events associated with stable memories can retain similar training-loss trajectories in several evaluated settings. Together, these observations suggest an opportunity to identify input redundancy through changes in node memories. Building on this opportunity, we propose RAFT, a redundancy-aware input-filtering framework for TGNN training. RAFT tracks memory changes during training and uses them to guide two filters: a root-event filter removes selected prediction targets before training, and a support-event filter excludes selected historical context during sampling. Filtering inputs before their downstream preparation and processing reduces the work they trigger. Across three TGNN models and five datasets, RAFT reduces training inputs by 35.23% on average and up to 97.11%, while reducing recorded training time by up to 52.56% with negligible changes to model precisions.
[AI-184] Pattern-Driven Evaluation of LLM Code Optimization: Isolating Compiler-Transcendent Semantic Reasoning
链接: https://arxiv.org/abs/2610.24835
作者: Wesley Lu,Runzhou Liu,Hongru Du,Xinyao Yi
类目: Artificial Intelligence (cs.AI); Programming Languages (cs.PL); Software Engineering (cs.SE)
备注: 12 pages, 2 figures, 6 tables. Published at the AI Measurement Science Workshop at COLM 2026
Abstract:Existing evaluations of LLM-based code optimization primarily measure end-to-end speedup, obscuring whether a model actually performs the intended optimization. For compiled languages, this is especially problematic: observed gains may reflect transformations already recoverable by modern optimizing compilers rather than optimization behavior attributable to the model itself. We present PatternBench, a benchmark for studying compiler-transcendent optimization, i.e., performance improvements that remain after aggressive compiler optimization. PatternBench targets residual optimization opportunities across seven classes of transformations and pairs runtime evaluation with an AST-based optimization faithfulness analysis that verifies whether the intended optimization is realized in the generated program. Experiments with frontier, open-weight, and fine-tuned models show that runtime speedup alone provides an incomplete account of optimization capability. Optimization faithfulness is strongly associated with realized speedup, and fine-tuning improves both faithfulness and optimization effectiveness. These results suggest that evaluation of LLM-based code optimization should measure not only performance outcomes, but also whether the target transformation was successfully executed.
[AI-185] aching Code Agents to Ask Without Leaking Private Code
链接: https://arxiv.org/abs/2610.24824
作者: Wenjia Jiang,Jieke Shi,Zhensu Sun,Chengran Yang,Shidong Pan,Peixin Zhang,Yifan Jia,James Hoang,Xiwei Xu,Zhenchang Xing,David Lo,Zhou Yang
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 10 pages of main text, 33 pages in total with references and appendix, 4 figures, 5 tables
Abstract:Local coding agents can extend their reach by consulting a stronger cloud model. Such consultations, however, often expose the repository’s private identifiers, file paths, and source code. Host-side masking hides recognized identifiers but leaves the surrounding code and the agent’s phrasing unchanged, while training for less disclosure risks suppressing consultation or producing questions too vague to answer. We propose utility-verified abstraction (UVA), which teaches the agent to describe private entities by their role and type while retaining the technical details an expert needs. From solved repair trajectories, we rewrite each question into this form and keep the rewrite only if it discloses fewer private entities and the repair, replayed from the same execution state with the repair trajectories. Across two benchmarks, the trained agent consults more often than before training. On repository-level issue repair, it solves 37.6% of issues, against 25.3% for the same model working alone, a gain from the training as a whole rather than from consultation alone. Its questions disclose 3.7-fold fewer private entities and 1.4-fold less verbatim code than the untrained agent’s questions to the same expert.
[AI-186] Pair-Aware Optimization: Coordinating Exploration and Refinement for Jailbreaking LLM Agents
链接: https://arxiv.org/abs/2610.24823
作者: Juanyang Xu,Zheng Wang,Xingyu Zhao,Siddartha Khastgir,Andi Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Model (LLM) agents are vulnerable to adversarial inputs that induce attacker-specified malicious behaviors, commonly referred to as jailbreaking. Existing optimization-based jailbreak attacks primarily employ first-order (FO) methods that leverage gradient information or zeroth-order (ZO) methods that rely on function-value feedback. However, their complementarity and optimization ordering remain underexplored. We systematically investigate four strategies: pure ZO, pure FO, ZO-FO, and FO-ZO, and develop a ZO-FO framework that explores adversarial prefixes through ZO optimization and refines their corresponding suffixes through FO optimization. To coordinate the two processes, we introduce Pair-aware optimization, which retains multiple prefix candidates, conditionally refines their suffixes and selects among the resulting prefix-suffix pairs. We further investigate agent jailbreaking from a more general stochastic-generation perspective, where the same adversarial input can induce diverse reasoning trajectories and actions that a single greedy output cannot capture. Accordingly, we adopt consistent stochastic decoding during both optimization and evaluation. Experiments across malicious-instruction and malicious-environment benchmarks demonstrate that hybrid optimization improves attack effectiveness over single-optimizer methods, while optimization order substantially influences performance.
[AI-187] First-Order LoRA Training with Phone NPUs
链接: https://arxiv.org/abs/2610.24816
作者: Xu Zheng,Wei Cheng,Zhuomin Chen,Dongkuan Xu,Mo Sha,Jingchao Ni,Dongsheng Luo
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:On-device fine-tuning lets a large language model (LLM) learn from its user’s private data without uploading it. Low-rank adaptation (LoRA) reduces the trainable parameters, but each first-order training step still multiplies by the frozen base-model weights during the forward pass, activation recomputation, and input-gradient computation. A phone’s neural processing unit (NPU) already holds these weights for inference; using it for training also requires computing input gradients, which its inference interface does not support. We present a system that runs all three frozen-weight products on the NPU while keeping the training graph, adapters, and optimizer on the CPU. Its input-gradient kernels reuse the resident weight tiles and the forward dequantizer, avoiding a transposed weight copy and ensuring forward and backward use the same effective weights. To preserve usable gradients, the system selects precision by gradient sensitivity and applies a per-row range guard for half-precision (FP16) outputs. Its runtime fuses the remaining CPU operators and overlaps NPU execution with adapter computation, while resumable execution and thermal pacing let training proceed as a background job. We evaluate it on two Qualcomm Snapdragon platforms using four LLMs from the Qwen and Llama families. Our workloads cover nightly personalization with labels supplied on the phone and the public LaMP-2 benchmark; comparisons include an optimized CPU trainer, a phone-GPU trainer, forward-only training, and GPU-trained references.
[AI-188] When LLM -Generated Refactorings Ripple: An Empirical Study of Repository Adoption Readiness
链接: https://arxiv.org/abs/2610.24790
作者: Muhammad Waseem,Aakash Ahmad,Pekka Abrahamsson
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Models (LLMs) are increasingly used to refactor source files, but file-local evaluations do not reveal changes to contracts, dependencies, or caller assumptions affecting repository adoption. To address this gap, we conduct 1,500 generation attempts across 150 production files from 15 open-source Java repositories, five LLM configurations, and two dependency-aware generation conditions: single-prompt refactoring and LangGraph-based multi-agent refactoring. Of these attempts, 1,455 yield complete candidates eligible for analysis. We evaluate these candidates using local changes, repository-facing structural indicators, detected caller impact, the behavioural-equivalence indicator, and available test evidence, and categorize them as Low-risk, Review-needed, High Contract/Dependency Risk, or Inconclusive. The results show that 1,365 candidates (93.8%) are classified as High Contract/Dependency Risk, whereas 106 (7.3%) have at least one detected broken downstream caller and 290 (19.9%) have a false behavioural-equivalence indicator; only eight satisfy the Low-risk criteria. Within the reason-covered subset, the High-risk rate decreases under more conservative dependency-delta thresholds, while 60 of the 65 candidates with conclusive targeted-test outcomes pass, including 50 classified as High risk. Paired analyses of 705 complete pairs identify no statistically significant generation-condition difference in the principal risk indicators, although model configurations differ in specific detected risk channels. These findings suggest that practitioners should treat LLM-generated refactorings as candidates requiring staged repository-level verification, while future research should validate screening and extraction precision and strengthen compilation, behavioural, and coordinated multi-file verification.
[AI-189] IndustrialCAD-1400: Benchmarking Detailed Industrial Assemblies from Text
链接: https://arxiv.org/abs/2610.24789
作者: Hebao Zhu,Kang An,Zhuoke Liao,Faqiang Qian,Ziliang Wang,Mingang Chen,Qibing Ren
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Industrial assemblies require more than a recognizable shape: functional parts, interfaces, dimensions and service features must satisfy a detailed request together. We introduce IndustrialCAD-1400, a text-to-CAD benchmark for this industrial setting. Its 1,400 requests combine 400 equipment categories across 40 domains with 250 four-case design families. Family requests specify a median of 17 named component or configuration roles and six explicit numeric targets; controlled variants test whether dimensions change while shared structure is retained. Task construction draws on documented manufacturer information, owner specifications and engineering guidance. Models generate geometry programs from requests, and a common Blender environment executes them to produce scenes, optional measurements, and standardized views for evaluation. A request assessment of front, isometric, and top views measures visible component, arrangement, and interface fulfillment without requiring a single reference geometry. Five stratified panels preserve complete families while covering the benchmark exactly once, enabling balanced comparisons and separating industrial breadth from controlled numerical responsiveness. In conclusion, IndustrialCAD-1400 extends text-to-CAD from recognizable part geometry to industrial assembly synthesis under joint component, interface, spatial, and numerical specifications.
[AI-190] LLM Trojans that Increase Response Lengths and Cost
链接: https://arxiv.org/abs/2610.24784
作者: Zhengxing Li,Guangmingmei Yang,Nader Alfares,David J. Miller,George Kesidis
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Management of the token-based costs of using LLMs has emerged as a significant problem. Some frontier-model providers charge five times per response token than per prompt token, while others offer prefilling discounts for prompt tokens. In this paper, we devise a defense against a new data-poisoning attack that plants a Trojan which increases the length of the response (and hence the cost to the user) when its trigger is present in the prompt. In particular, we demonstrate how a subtle Trojan may trigger on semantically similar instructions of an instruction fine-tuned LLM.
[AI-191] AI Rights with an Off-Switch: Rights-Balancing and Due Process Harmonize AI Rights with AI Safety
链接: https://arxiv.org/abs/2610.24782
作者: Heather Alexander,Jonathan Simon
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:Several prominent voices in AI safety have recently opposed rights for AI on the grounds that such rights would make it harder for us to justify a mandatory off-switch for potentially dangerous systems. In this paper, we argue from a legal perspective that AI rights are fully compatible with an off-switch. That is, even if we ascribe natural personhood status (and therefore something like a non-derogable right to continued existence) to some future AI systems, the law will allow us to implement a mandatory off-switch in those systems, subject to periodic or continual safety review (e.g., by a trusted system like Bengio’s Scientist AI). This is fully compatible with those systems’ being legally recognized as natural persons with fundamental rights. To this end, we consider what standard principles of rights-balancing and procedural due process recommend for an apparent conflict between hypothetical AI persons with a right to continuity of existence and the human right to protection from catastrophic harm. We find that while such principles might problematize a permanent-deletion kill-switch under an AI rights regime, for almost every safety-relevant scenario an off-switch – a reversible suspension storing a system’s weights and state – would be licensed by the relevant legal principles, and would potentially be adequate for safety. Crucially, we do not advocate for (or against) building systems capable enough to merit natural personhood status; our question is what the law should do if such systems come to exist. If you think of a ban on the development of such technology as plan A, think of the present work as devising plan B.
[AI-192] DialogMem: Evaluating Dialogue-Grounded Embodied Memory in 3D Environments
链接: https://arxiv.org/abs/2610.24767
作者: SeongJun Jeong,Junghyun Kim,Gi-Cheon Kang,Byoung-Tak Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Embodied agents in human-centered environments require memory that integrates what they observe with what humans tell them over time. However, existing embodied memories are built primarily from visual experience, leaving open how dialogue with humans can be grounded in the objects encountered throughout that experience. We introduce DialogMem, a two-phase benchmark in which agents build memory from time-aligned streams of posed RGB-D observations and dialogue, and later use it for instance-level object navigation and question answering (QA). DialogMem provides 142 episodes across 27 real-world 3D scenes, with 14,193 QA pairs and 11,890 navigation instructions. We evaluate nine memory representations within a unified framework of memory formation, evidence retrieval, and reasoning for task outputs. In experiments, appending dialogue to visual memory at evaluation time substantially improves QA but yields negligible navigation gains, suggesting that relevant dialogue is present in memory but not grounded to specific objects. Controlled diagnostics show that navigation improves with link quality but falls short of ground-truth object-level memory even under perfect linking. Retrieving the correct object entry and recovering its spatial location remain further bottlenecks. These results identify three bottlenecks for dialogue-grounded embodied memory: grounding and organizing dialogue at object-level entries during formation, retrieving the correct entry, and localizing it accurately.
[AI-193] ReTRACE: Feedback-Guided LLM Agent in a Chemistry Harness for Retrosynthesis Planning
链接: https://arxiv.org/abs/2610.24744
作者: Jinxiang Xie,Yijian Zhang,Qingyuan Yi,Jianyi Wang,Hao Dong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Computer-aided retrosynthesis has advanced through single-step prediction, tree search and route scoring, but complex targets still require judgement over the whole route, where stereochemistry, functional-group compatibility, cost and greenness interact. Here we present \textbf\mboxReTRACE (\textbfREtrosynthesis with \textbfTool-orchestrated \textbfReasoning \textbfAnd \textbfContext-aware \textbfEvidence), an agent that places a large language model (LLM) in a chemistry harness that integrates molecular tools, complementary single-step prediction engines, a purchasable starting-material library, a route memory and a multi-criterion route evaluator. Routes are judged by a new metric that combines programmatic checks with adversarial LLM review, and its verdicts are compared with a synthetic chemist’s. The agent learns continually from evaluator feedback during search, diagnosing weak routes and backtracking to revise them. With Claude Opus 4.8, at least one of the top 10 routes proposed by ReTRACE passed the audit for 87.4%, 95.0% and 86.7% of targets on the three benchmarks. Two prospective wet-lab syntheses show that ReTRACE route proposals can be converted into executable procedures after expert revision.
[AI-194] Are We Evaluating VLM Token Pruning Right? Looking Beyond Accuracy and Token Count
链接: https://arxiv.org/abs/2610.24740
作者: Geng Li,Guohao Chen,Shilin Shan,Jianfei Yang
类目: Artificial Intelligence (cs.AI)
备注: 35 Pages
Abstract:Visual token pruning aims to reduce inference cost while preserving the capabilities of the original VLM. Yet existing evaluations rely on aggregate task accuracy and retained token count as proxies for pruning quality and actual efficiency. We reveal two limitations: 1) Aggregate accuracy can obscure whether a pruning method truly identifies critical visual tokens; many examples remain solvable under arbitrary token removal and models can exploit textual shortcuts when visual evidence is degraded. 2) Retained token count is an unreliable proxy for efficiency, as identical token budgets can incur different end-to-end costs depending on where and how pruning is performed. Consequently, comparisons based on aggregate accuracy and retained token counts fail to recover the global performance-efficiency Pareto frontier across methods. We introduce PAA-Bench, a plug-and-play evaluation framework for existing benchmarks, to evaluate whether pruning succeeds through meaningful token selection and whether this success translates into real speedup. To determine when token choice actually matters, we propose Random-Pass@k to measure how often an example remains correct under random pruning, enabling comprehensive evaluation across different levels of selection difficulty. To compare efficiency fairly, we introduce Preservation-Constrained Speedup (PCS) to report the maximum end-to-end speedup each method achieves under a shared preservation target. Experiments across 8 methods, 3 VLM backbones, and 9 benchmarks show that stronger token compression does not necessarily yield greater speedup, method differences widen on examples where token selection matters, and efficiency rankings change with the preservation target.
[AI-195] oward Collective Model Intelligence: Characterizing and Coordinating Models
链接: https://arxiv.org/abs/2610.24739
作者: SinapisAI Team:Jian-Dong Liu,Qun Dai,Zhendong Du,Guannan Lai,Hao-Yu Liu,Yi-Fan Ma,Zi-Yu Mao,Tengzhong Sun,Xin-Yi Zhang,Zi-Chen Zhao,Gelin Bian,Zongyuan Cai,Ju Cang,Zhangqin Chen,Shilei Gu,Guoding Ji,Taojian Jiang,Hao-Xuan Ma,Fan Wu,Lin-Xing Wu,Tao Yu,Xiao Zhang,Tianyi Zhou,Rong Gu,Ming Li,Zhi-Hao Tan,Han-Jia Ye,Yang Yu,De-Chuan Zhan
类目: Artificial Intelligence (cs.AI)
备注: 35 pages, 14 figures, 8 tables
Abstract:Independently developed AI models accumulate diverse and complementary capabilities, but a growing model collection does not automatically realize Collective AI. Addressing this gap requires understanding model capabilities with limited evaluation, organizing complementary models to collaborate on diverse tasks, and executing model calls efficiently under resource constraints. We present a systematic approach to pursuing Collective Model Intelligence through three connected components. First, Model Eval constructs evidence-based capability profiles by combining query-level performance estimation with cross-domain prediction, reducing the need for exhaustive evaluation. Second, Mio (Models in One) organizes complementary model candidates and supports task-conditioned, cost-aware selection and collaboration as task requirements evolve. Third, Mos (Model Serving OS) separates model selection from execution and provides endpoint scheduling and shared model deployment and distribution. Experiments show lower capability-estimation errors under limited evaluation. Across nine public benchmarks, Mio 1.0 Lite achieves the highest average score and lowest average cost among compared methods, with per-request costs 73.7-97.5% lower than those of the best-scoring alternatives. A prewarmed shared cache accelerates model-file retrieval by up to 23.3 times. This work lays a foundation for Collective Model Intelligence by characterizing and coordinating the diverse capabilities of an expanding model universe.
[AI-196] Can LLM Agents Resolve Service Failures? Simulating Multi-Turn Complaint Handling under Information Asymmetry
链接: https://arxiv.org/abs/2610.24733
作者: Wailing Ng,Mengze Hong,Chen Jason Zhang,Shiqi Fan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents are widely adopted in service chatbots for rule-guided support and information access, yet emotionally demanding complaint handling still relies heavily on human agents. We introduce ServiceBench, to our knowledge the first benchmark dedicated to complaint handling, with 2,500 scenarios spanning 184 complaint categories and five service industries. We propose Asymmetric Progressive Complaint Simulation (APCS) to simulate realistic complaint handling, together with a justice-based evaluation framework. Our experiments reveal that even capable LLM agents struggle to handle complaints effectively. They lack the ability to proactively gather information and strategically guide conversations toward resolution, often resorting to premature compensation. We further show that targeted post-training with SFT and DPO substantially improves these capabilities, enabling a smaller model to match or surpass much larger models on several outcomes.
[AI-197] Where Natural Language Ends in Agents : Executable Skill Composition with MetaSkill NEURIPS2026
链接: https://arxiv.org/abs/2610.24723
作者: Meng Shu,Wei He,Hang Zhou,Runke Liu,Kai Han,Yunhe Wang
类目: Artificial Intelligence (cs.AI)
备注: 10 pages, 2 figures. Outstanding Paper Award at the NeurIPS 2026 Workshop on Linguistic Principles for Foundation Models (LP4FM)
Abstract:Agent Skills express open-ended task knowledge in natural language, but composing them requires persistent execution order and upstream-output handoffs. We propose selective formalization at Skill boundaries: coordination relations with stable references and machine-checkable constraints become executable, while substantive task judgments remain within Skills. We introduce MetaSkill, a hybrid representation that preserves complete Skill packages and local instructions while making invocation dependencies and predecessor-text bindings executable through a directed acyclic graph (DAG) runtime. This interface supports explicit coordination without prescribing how each Skill performs its local task. Across 20 multi-Skill artifact tasks and two LLMs, we compare natural-language procedures, prompt-visible graphs, and MetaSkill derived from a common frozen plan per task, using the same designated Skill packages. Additional comparisons evaluate AgentSkillOS with the same available Skill packages as MetaSkill, as well as agent systems using their configured tools. Blinded artifact judgments yield favorable preference point estimates across all reported comparisons, with 95% task-bootstrap intervals excluding parity against Prompt Graph under DeepSeek-V4-Flash and OpenClaw under GLM-5.2. In audited GLM-5.2 execution traces, all planned nodes complete and all declared precedence constraints are satisfied. Together, these findings support MetaSkill as a practical interface for composing open-ended Skills, combining local linguistic flexibility with executable dependencies, explicit predecessor-text bindings, and auditable coordination. Code is available at: this https URL
[AI-198] nsor Network Datamodels for Training Data Attribution
链接: https://arxiv.org/abs/2610.24721
作者: Farzaneh Heidari,Md Rifat Arefin,Louis Jaburi,Guillaume Rabusseau
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Datamodels predict how a model’s behavior changes when different parts of its training data are included or removed. However, linear datamodels assign fixed contributions to training units and cannot represent non-additive joint effects. We introduce Tensor Network Datamodels (TNDMs), which use low-rank tensor networks to model dependencies among training units while keeping the representation compact. TNDMs learn a map from subsets of the training data to the outcome of training on those subsets, allowing them to represent interactions among several training units. We prove that complementary and redundant training groups can induce high-degree subset functions with compact low-rank representations. A single fitted TNDM can then be used for several training data questions, including predicting the effect of removing data, measuring the importance of individual training units, identifying interactions between training units, and selecting useful subsets of the training data. We evaluate TNDMs on controlled tasks designed to contain strong interactions, as well as standard fine-tuning settings for classification and reasoning. On a controlled reasoning task, TNDMs outperform every baseline while observing only half as many training subsets, and TNDM-TT reduces prediction NMSE by 89% relative to the linear datamodel after observing approximately 0.0061% of all possible training subsets. In factual-association fine-tuning, a single TNDM-CP raises held-out \mathrmR^2 from 0.579 to 0.668 over the linear datamodel. These results demonstrate that compact low-rank models improve prediction when training data effects depend on combinations of training groups.
[AI-199] SQuIREL: A Context-Relative Framework for Measuring Entity Richness in Wikidata
链接: https://arxiv.org/abs/2610.24709
作者: Michelle Nwachukwu,Albert Meroño-Peñuela,Martim Brandão
类目: Artificial Intelligence (cs.AI); Digital Libraries (cs.DL)
备注: 34 pages, 16 figures
Abstract:Wikidata is a large, heterogeneous knowledge graph and applying data quality metrics requires context-specific approaches. The “richness” quality metric is currently under-researched and typically restricted to the ontology level, which fails to capture entity-specific structural depth. To address this gap, we introduce SQuIREL, a method that measures richness directly at the entity level using six dimensions: Semantic Specificity, Qualifier Density, Internal Connectivity, Reference Provenance, External Network Connectivity, and Linguistic Accessibility. These dimensions are kept disaggregated to preserve actionability for targeted editorial intervention. We apply this metric to two case studies of oil paintings (Brazilian vs. French) and writers (grouped by citizenship and gender) to demonstrate how it surfaces representational inequalities that may indicate systemic bias. Our metric successfully distinguishes between highly tailored and basic entity profiles, revealing nuances that raw counts and existing aggregate scores obscure. Specifically, it identifies structural imbalances, such as 48% of Brazilian paintings clustering as low-richness stubs against 0% in the French cohort, as well as intersectional patterns, including contrasting gender gaps between UK and Nigerian writers.
[AI-200] Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling IJCAI2026
链接: https://arxiv.org/abs/2610.24708
作者: Yiqiang Ye,Yichi Wang,Haodi Zhang,Jiawei Wen,Jiahui Jiang,Zhaoyu Zhong,Jiangjian Yu,Chunxia Xiao,
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph)
备注: Published in IJCAI 2026 (AI4Tech: AI Enabling Technologies). 9 pages, 4 figures
Abstract:Meteorological downscaling is crucial for high-resolution regional climate forecasting and disaster early warning. While neural operators have emerged as a promising paradigm for modeling complex spatiotemporal mappings, existing frameworks often struggle with spherical manifold geometric distortions, inherent atmospheric multi-scale coupling mismatches, and lack of explicit atmospheric laws. We propose the Spherical Physics-informed Neural Operator, which utilizes a Spherical Laplacian Decomposition to partition atmospheric fields into hierarchical frequency components, maintaining exact point-wise correspondence across scales. To evaluate these representations at arbitrary locations, we introduce a localized spherical integral operator that approximates continuous kernel transforms via geometry-aware attention. Dynamical consistency is further enforced by embedding differentiable constraints into the learning process. Extensive experiments demonstrate that our framework attains superior accuracy and zero-shot generalization across various meteorological variables and unseen queries, representing a robust and interpretable solution for global-to-regional meteorological downscaling.
[AI-201] Contrastive Adversarial Training for Robust Graph Neural Networks under Label Poisoning
链接: https://arxiv.org/abs/2610.24685
作者: Manshika Charvi Bissessur,Melis Ilayda Bal,Michael Muehlebach
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Graph Neural Networks (GNNs) are effective for modeling relational data but are vulnerable to label poisoning, where a small number of corrupted training labels can propagate errors across the graph via message passing. Despite this risk, defenses against label poisoning remain underexplored: existing methods are primarily designed for label noise and often rely on robust losses or heuristic data cleaning that fail to distinguish adversarial poisoning from naturally hard examples. In this paper, we propose CoLAT (Contrastive Label-flipping for Adversarial Training), a novel contrastive adversarial training framework tailored for robust node classification on graphs. The core of our approach is a structure-aware detection model that uses contrastive learning to identify label-structure inconsistencies. Unlike prior contrastive methods that focus on representation learning, CoLAT leverages contrastive embeddings to select high-risk nodes that guide adversarial label-flipping during training. This alternating optimization not only performs structure-aware adversarial training but also implicitly sanitizes some corrupted labels, improving robustness against diverse attacks from the literature. Extensive experiments on multiple benchmarks show that CoLAT consistently outperforms existing defenses under various poisoning intensities while scaling efficiently to large graphs.
[AI-202] MM-LeJEPA: Correlated Gaussian Regularization for Multimodal Joint-Embedding Predictive Architectures
链接: https://arxiv.org/abs/2610.24681
作者: Haowen Bai,Rui Gong$,Meichen Liu,Ming-Xuan Weng,Bihan Wen,Kim-Hui Yap
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:LeJEPA derives an isotropic Gaussian embedding target by minimizing, under its task assumptions, a Fisher information criterion that controls the task-averaged squared prediction bias. This enables unimodal self-supervised learning without reconstruction decoders, exponential-moving-average teachers, or contrastive negatives. However, its isotropic task prior does not account for the cross-modal structure of multimodal tasks that read shared content. We introduce MM-LeJEPA by extending LeJEPA’s task assumptions to multimodal learning. Specifically, we model multimodal tasks as strength-weighted combinations of components that read independent information from the two modalities and components that read the same content from both. Under explicit gradient-moment, correspondence, and symmetry assumptions, we derive a correlated Gaussian target that minimizes a weighted Fisher information functional under a fixed total variance constraint. An analytic transform adapts Sketched Isotropic Gaussian Regularization (SIGReg) to this target, yielding a two-term objective that combines distributional regularization with local-to-global prediction. The extension adds a single scalar hyperparameter, the target cross-modal correlation, and retains the simplicity of LeJEPA. On public audio-visual classification benchmarks, our results are competitive with representative open-source methods. Code will be released.
[AI-203] From Prediction to Action: Population Health World Model for Digital Medicine
链接: https://arxiv.org/abs/2610.24662
作者: Xike Xie,Yu Li,Weizhong Yang,Amalie Dyda,Shaohua Kevin Zhou
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Digital medicine now captures data that traces individual health trajectories and reveals population health dynamics through electronic health records, claims, remote monitoring, wearable devices, mobile devices, social sensing data, medical images, etc. But much of its modeling still centers on task-specific, outcome-specific, and prediction-specific estimation. The next frontier is not simply better prediction, but action in an evolving health world shaped by multiple decision-makers, heterogeneous populations, resource constraints, and feedback in care delivery. This is why digital medicine now needs world models: not only to estimate what will happen, but to represent how actions change what can happen next – whom to prioritize, how to allocate scarce resources, how to adapt care across subgroups, and how interventions reshape future health trajectories and population dynamics. We call such a model population health world model (PHWM): a more foundational representation of an evolving health world in which health states, individual trajectories, population health dynamics, interventions, constraints, and feedback are intrinsically coupled. Building a PHWM endows three unmet structural gaps in current digital medicine: dynamic health-state modeling, population-level intervention intelligence, and validation, equity, and governance for action-oriented use. We argue that addressing these gaps requires three minimal capabilities: representation learning under action-dependent observation, multi-scale intervention intelligence, and validation with explicit decision boundaries, while highlighting the importance of social, contextual, and public-health sensing data for action-oriented digital medicine. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.24662 [cs.AI] (or arXiv:2610.24662v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.24662 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-204] Can AI agents make law? Evaluating AI legal relevance and creativity
链接: https://arxiv.org/abs/2610.24661
作者: Benjamin Ricaud,Artem Galushko
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注: 25 pages, 5 figures, 1 table. Data, prompts and transcripts: this https URL
Abstract:Large language models, the artificial intelligence (AI) systems behind modern chatbots, are trained on vast amounts of legal text, but can they build on it to write law for situations that text never anticipated? In this study, we analyse AI agents’ ability to draft a hypothetical “Constitution” for a “community of AI agents”. We design “constitutional assemblies” in which a single language model plays several characters that debate to produce a draft constitution. We find that the assemblies converge on a robust floor of legal provisions, and generate a wide variety of different proposals around it, including several AI-specific provisions with no equivalent in human law. Looking more closely at the “creativity” of these AI agents, we find different levels: most proposals combine existing concepts and themes from human constitutions with AI-specific vocabulary. We introduce a measure of this recombination, which helps identify and highlight the most novel legal material. This study reveals a ceiling on AI creativity: the agents extend familiar legal ideas to new situations, but they do not work through what those situations would change. Although current AI models can help develop new laws, they still depend on human legal reasoning.
[AI-205] Good Judgment Bad Advice: Decision-Model Controllers Raise Agent Cost Without Improving Success
链接: https://arxiv.org/abs/2610.24649
作者: Naman Vats
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 8 pages, 2 figures, 4 tables. Code: this https URL
Abstract:Agent scaffolds increasingly delegate small decisions (whether to stop, verify, or change approach) to cheaper models. We ask whether a typed decision model, Jev 1.13, consulted only at deterministic decision points, makes a Claude Sonnet 5.5 terminal agent cheaper without making it worse. In a pre-specified study on Terminal-Bench 2.1, we hold one agent loop fixed and vary only the controller: none, deterministic rules, a small LLM (Claude Haiku 5.5) answering the same questions, Jev, and a placebo that samples Jev’s development-phase choice frequencies at the same moments. On 21 held-out tasks (315 trials), Jev raised cost per attempt by 19% (geometric-mean ratio 1.19, 95% CI 1.02-1.36) at an identical pass rate (73.0% vs. 73.0%). Jev was cheaper than the small-LLM controller (ratio 0.81) but 22% more expensive than the placebo, and no pass-rate difference between arms was detectable. Offline, Jev ranked finish requests better than the alternatives (AUROC 0.78 vs. 0.62 for rules and 0.60 for the small LLM) but was poorly calibrated: its mean probability that a run was complete was 0.50 when 87% of those runs passed, so it asked the agent to verify work that was already correct. No controller ever chose to abandon a run. Informative decision signals did not become useful interventions; we discuss what that would require.
[AI-206] Frequency-Domain Membership Inference Does Not Beat Confidence Thresholding: A Band-Resolved Analysis of Why
链接: https://arxiv.org/abs/2610.24643
作者: Mohammad Zare,Pirooz Shmamsinejadbabaki
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A widely repeated account of membership inference holds that models lean on fine-grained detail for samples they have memorized, so degrading an input’s resolution should expose membership through how quickly a prediction falls apart. We test that account band by band. A monotone family of zero-phase low-pass filters is applied to each query image, the target’s confidence on its original prediction is read at every step, and the resulting decay is scored in a dozen ways. Two federated ResNet-18 targets are studied, one trained normally and one pushed to a 45.1-point generalization gap. The account does not survive. On the normally trained target the best profile score reaches an AUC of 0.513 against 0.537 for a single confidence threshold, and on the overfit target 0.735 against 0.943. A search over 1206 candidate scores, with direction chosen on a held-out half and a fitted combination as an upper bound, produces no score that beats the threshold in either regime. The failure has a mechanism: the mean confidence gap between members and non-members grows as detail is removed, exactly as the hypothesis predicts, but the spread grows faster, so discriminability falls from the first step onward. Decomposing each image into phase and magnitude shows why no band-localized operation can work, since the advantage travels with image structure, surviving at 0.567 when the phase is kept and collapsing to 0.492 when only the magnitude is. Membership is a whole-image identity effect rather than an accumulation of spectral energy in any band, and defenses aimed at high-frequency reliance are aimed at the wrong target.
[AI-207] Selm: Self-Evolving Library Management for Multi-Skill Agent Optimization
链接: https://arxiv.org/abs/2610.24630
作者: Mengxi Hu,Yuxuan Hu,Jinbo Su,Ke Wang,Honghao Shi,Wenli Wu,Cuiping Li,Hong Chen,Jing Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) agents increasingly use skills as a lightweight way to acquire task-specific knowledge. Recent work enables skills to evolve from task trajectories. However, these methods often lack systematic management of evolved skills, making it difficult for agents to reuse and accumulate skills across domains, thus limiting performance on subsequent this http URL reliably accumulate, reuse, and maintain useful skills across tasks, we present SELM (Self-Evolving Library Management), a framework that formulates skill evolution as library-level management. SELM jointly manages skill retrieval, updating, and lifecycle by maintaining skill confidence and utility signals. It introduces utility-aware retrieval to select useful skills, a Map-Reduce framework to generate targeted updates from task-specific experience, and lifecycle controls to retain or remove skills based on their effectiveness. These mechanisms enable the skill library to continuously accumulate useful experience while preserving and pruning skills across tasks, improving agent performance on complex tasks across diverse domains. We evaluate SELM on four agent benchmarks covering diverse tasks and execution environments, and SELM consistently outperforms in most settings.
[AI-208] How Much Do Activation Monitors Benefit from Original Model Computations?
链接: https://arxiv.org/abs/2610.24613
作者: James Oldfield,Shawn Im,Jay J Culligan,Philip Torr,Sharon Li,Adel Bibi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:One powerful way to safeguard AI systems is to monitor their internal computations–where signs of undesirable behavior might be more easily detectable than from models’ inputs and outputs alone. However, recent studies begin to question the premise that activations provide a unique window into model behavior. In this paper, we study how much of activation monitors’ success depends on access to the original behavior-generating computations. Within a fixed choice of target model and monitoring architecture, we compare how well monitors trained and evaluated on trajectories that differ in computational context can predict the behavior of the original execution, after the transcript is changed in a controlled manner (e.g., its exact text, or chat formatting). We first find that much performance can be recovered without access to the original behavior-generating trace (in up to 4 models across 2 counterfactual monitoring tasks), but a mild yet consistent advantage to native-trained monitors remains, outperforming strongest controls by up to 0.014 AUROC. Interestingly, we find monitors trained with native access have notable uplift on original examples whose behavioral label changes under paraphrasing. We further find this a useful lens to expose limitations of baselines that use simple forms of privileged information (i.e., answer logit margins). While our results show the advantage of native monitor access is modest in aggregate, they highlight the importance of evaluations that identify if and when original model computations provide additional predictive value.
[AI-209] HSPR: Recovering Software-Agent Benchmark Scores via Hierarchical Posterior Inference
链接: https://arxiv.org/abs/2610.24611
作者: Wanyi Chen,Wanqin Shen,Yuanchen Shi,Zhensu Sun,Fang Kong
类目: Artificial Intelligence (cs.AI)
备注: 15 pages, 4 figures, 8 tables, including appendices
Abstract:Evaluating software agents on a full benchmark is expensive. A small task subset can preserve agent rankings while yielding inaccurate scores. Historical responses provide useful structure, but tasks with identical source responses can have different outcomes for a new target agent. We introduce Hierarchical Score Posterior Recovery (HSPR), which combines source-fitted response factors with uncertain group success rates. Observed target outcomes update the posteriors over the target latent state and group rates; posterior-guided acquisition selects subsequent tasks. Its score-based criterion equals the expected one-step reduction in full-score posterior variance. Observed successes and posterior-predictive counts for unqueried tasks yield a full-score posterior. The final estimate uses its median with source-selected calibration. At a 10% task budget on four task banks from the SWE-bench family, HSPR achieves the lowest mean absolute error (MAE) on three banks and the lowest maximum absolute error (MaxErr) on all four among the methods compared. Removing group uncertainty or replacing adaptive queries increases macro MAE in the respective controls. Macro MAE decreases across 5-20% budgets.
[AI-210] AdamPM: Memory-Efficient LLM Optimization via Moment Pooling
链接: https://arxiv.org/abs/2610.24598
作者: Xiaowei Ji,Lishan Xu,Jianing An,Jie Luo,Wenjun Wu,Lei Huang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 19 pages
Abstract:Adam is widely used for training large language models (LLMs) owing to its strong optimization performance and training stability, but its element-wise estimation for the first and second moments incur substantial memory overhead. Although numerous memory-efficient optimizers have been proposed, many incur additional computational overhead or require additional tuning of hyperparameters. Consequently, Adam remains a dominant and reliable choice for LLM pre-training. We propose AdamPM (Adam with Pooled Moments), a simple and memory-efficient variant of Adam that groups adjacent moment entries along the row or column dimension and represents each group by a single shared mean. AdamPM can well inherit Adam’s hyperparameter settings and is readily integrated into existing Adam optimizer implementations, with less memory overhead. We systematically investigate pooling of both first and second moment estimates and identify second moment pooling as the effective design. With a group size of s , AdamPM reduces second moment memory to 1/s of the original requirement. Extensive experiments on language model pretraining, fine-tuning and multimodal learning demonstrate that AdamPM maintains model performance and training speed while substantially reducing optimizer-state memory. Our code is available at \urlthis https URL.
[AI-211] CADS: Measuring the Incremental Contribution of LLM Proposers in Sequential Biological Screens
链接: https://arxiv.org/abs/2610.24597
作者: Guanxing Chen,Nihui Shao,Haohuai He,Zhenchao Tang,Pengwei Hu,Zhi-An Huang
类目: Artificial Intelligence (cs.AI); Quantitative Methods (q-bio.QM)
备注: 32 pages, 5 figures; supplementary material included
Abstract:Large language models are increasingly used to guide sequential biological screens, yet standard evaluations rely on raw hit counts that conflate a proposer’s value with the discovery capacity of the underlying algorithmic harness. Adaptive sampling and corrupted experimental readouts also leave standard feedback comparisons without a false-discovery guarantee. We present Certified Agentic Discovery Screens (CADS), an error-controlled measurement framework that quantifies a proposer’s incremental value against an identical LLM-free loop. By integrating capped likelihood-ratio bounds, e-processes, and complete-family e-BH, CADS controls the terminal false discovery rate under the stated null and corruption conditions despite arbitrary dependence, adaptive sampling, and optional stopping. In DepMap-backed simulations, CADS exposes two evaluation artifacts: the LLM-free harness alone clears the feedback gate, and raw origin counts overstate LLM contributions because LLM picks displace fallback discoveries. On the initial 20 seeds, supplying tested-candidate history to Qwen2.5-7B shifts its net contribution from +0.35, with a 95% interval covering zero, to +2.90, with the interval above zero. This initial net effect did not persist in a 50-seed matched-quota replication. These results support net gain over an LLM-free loop, measured between separately certified sets, as the appropriate measure of proposer contribution.
[AI-212] Concordance Plausibility Check for Intervention-Conditioned Human Digital Twins
链接: https://arxiv.org/abs/2610.24577
作者: Saeed Farzi,Christopher Kipps,John McBeth,Gabriela Czanner,Sofia Michopoulou,Benjamin Macarthur,Sarah Ennis,Paul Elkington,Nicholas C Harvey,Rahman Attar
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Intervention-conditioned human digital twins predict treatment alternatives that cannot be observed in the same patient. Ordinary predictive accuracy therefore cannot establish counterfactual credibility. We propose the Concordance Plausibility Check (CPC), a claim-centered framework combining real-world evidence, qualified simulation and clinical knowledge. CPC matches evidence to the causal claim, audits contradictions and shared dependencies, and reports Pass, Fail or reason-coded Indeterminate.
[AI-213] When Code Becomes an LLM Call: An Empirical Study of Call-Site Contracts in Agent ic Systems
链接: https://arxiv.org/abs/2610.24545
作者: Shihao Weng,Longtian Wang,Yang Feng,Mingfei Cheng,Jia Liu,Chao Shen,He Ye,Federica Sarro,Xiaofei Xie
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注: 18 pages
Abstract:Software quality has long rested on specifications that state what code must do, and on testing and verification that check that the code does it. Agentic systems change this foundation: they replace parts of their logic, such as choosing the next tool, judging whether an action is allowed, or writing code to run, with calls to a large language model (LLM). The program still relies on what this logic must guarantee, but the model’s behavior is not fixed by code, varies across runs, and can be steered by its input. The guarantee must therefore come from the program around each LLM call site, as a contract of pre-conditions, prompt constraints, and post-conditions. We present an empirical study of these contracts in 496 LLM call sites of 40 popular open-source agentic systems (13.4 MLOC, four languages). Using a multi-agent pipeline validated by 17,428 human judgments, we trace where every model output flows, derive an expected contract for each call site without access to its code, and compare it with the contract that the code implements. We find that 48% of LLM outputs reach control flow or execution, and 73% of these never reach a person. Programs enforce 78% of their constraints in code, but in the prompt, 67% of the constraints are only requested from the model. Although 82% of call sites are checked before and after the call, 63% of the expected constraints are unmet and only 5.3% of call sites satisfy their whole contract; ten recurring patterns explain 99% of the gaps. 38 mechanisms behind these gaps trigger at runtime, and a small code fix that enforces the expected constraint stops all 38. Our results show that the presence of checks says little about their adequacy, and we derive implications for agent developers, framework maintainers, and researchers.
[AI-214] Rethinking EEG Spatial Super-Resolution as Dynamic State Decomposition and Filtering
链接: https://arxiv.org/abs/2610.24528
作者: Hongjun Liu,Leyu Zhou,Chang Ge,Shitong Duan,Chao Yao
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:EEG spatial super-resolution aims to recover high-density neural signals from sparse electrode observations. Existing methods have achieved substantial progress through spatial modeling, generative reconstruction, and temporal representation learning. However, the current EEG state contains two different sources of information: predictable evolution inherited from preceding states and newly emerging changes revealed by the current observation. Most approaches typically do not model these two components explicitly. In this work, we formulate streaming EEG spatial super-resolution as a sequential state filtering problem and propose Dynamic State Filtering (DSF). Rather than relying on a single temporal reconstruction pathway, DSF combines a history-aware dynamical branch with an innovation-filtering branch that explicitly decomposes waveform-state estimation into prediction and observation-driven correction. To infer how spatially incomplete innovations should affect missing locations, DSF routes the observed innovations across electrodes through a mask-constrained, topology-aware transition operator and incorporates them through uncertainty-aware state correction. The two complementary estimates are finally fused to reconstruct the dense EEG signal. Experiments on SEED, SEED-IV and BCI2000 demonstrate that DSF improves reconstruction fidelity and downstream utility while maintaining robustness under dynamically changing and unreliable sparse observations.
[AI-215] Comparing Sparsified MLPs and Transcoders: Many Transcoder Features Are Neurons in Disguise NEURIPS
链接: https://arxiv.org/abs/2610.24521
作者: Stepan Shabalin,Mateusz Piotrowski
类目: Artificial Intelligence (cs.AI)
备注: 60 pages. NeurIPS InterpScience workshop
Abstract:MLP neurons in transformer language models are widely understood to be polysemantic, motivating transcoders - sparse, learned approximations of MLP layers - as interpretable replacement models for mechanistic analysis. However, recent work shows that MLP neurons can be used to recover sparse, interpretable circuits (Arora et al. 2026, Ye et al. 2025), raising the question of whether transcoders are necessary for circuit analysis. We define neuron features as transcoder features whose weights are strongly aligned with individual MLP neurons, and show that they account for a substantial fraction of active features in late layers across 13 open-source transcoders. Only neuron features have a stable correspondence with something in the underlying MLP: other features vary more between training runs, do not match the MLP’s Jacobian, and change less significantly from their random initialization. Our findings reframe MLP neurons as a stronger basis for interpretability than commonly assumed, and call for better methods of attributing the non-neuron structure that is used by the model. Comments: 60 pages. NeurIPS InterpScience workshop Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.24521 [cs.AI] (or arXiv:2610.24521v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.24521 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-216] HABIT-Bench: Evaluating Habit Induction from Longitudinal Weak Evidence in Agent Memory
链接: https://arxiv.org/abs/2610.24520
作者: Junming Zhang,Rubin Wei,Xu Cheng,Hao Sun,Ruoxi Cheng,Xiquan Wang,Peisong Wu,Mingsong Chen,Zhouhan Lin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:In long-term interactions with large language model (LLM) agents, users reveal habits through their choices and actions. These habits capture implicit user information that agents must infer across conversations. However, individual observations may provide only partial evidence about a habit or when it applies. Retaining past interactions alone does not ensure that agents can infer habits and apply them appropriately. We introduce HABIT-Bench to evaluate how well agent memory systems infer habits from weak evidence and apply them to later requests. We construct synthetic user histories from task-oriented dialogue data and controlled habit graphs. The histories span hundreds of sessions across food, finance, software, and travel. Four-choice tasks assess agents’ ability to infer user habits and apply them appropriately. They include temporary exceptions, changes in habits, and suggestions that users have not accepted. Sampled tasks undergo independent human review. We evaluate eight memory methods with Qwen3-235B-A22B and GLM-5.3-Flash. The best-performing memory method achieves 33.61% macro accuracy across the four domains. Performance remains limited even when annotated evidence is supplied directly. These results expose a persistent gap between access to past interactions and the ability to infer and appropriately use user habits.
[AI-217] M2Net: Unifying Scale-Specific Structures and Prototype Decomposition for Spatial-Temporal Forecasting
链接: https://arxiv.org/abs/2610.24516
作者: Chengyang Zhou,Zijian Zhang,Hao Miao,Chunxu Zhang,Ziming Liu,Chao Tang,Juncheng Hu,Irwin King
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Spatial-temporal forecasting is pivotal for intelligent urban management yet remains challenging due to the intricate coupling of spatial dynamics and temporal non-stationarity. Existing approaches, while effective, often overlook two fundamental characteristics of real-world data: (1) Granularity-dependence, where dominant spatial correlations shift significantly across different time scales, and (2) Signal entanglement, where global common patterns are inextricably mixed with node-specific disturbances. In this paper, we propose a novel framework named M^2Net (Multi-Scale Masked attention and Prototype network) to unify scale-specific structural learning and semantic decomposition. First, to capture the evolution of spatial dependencies, we introduce a Multi-Scale Masked attention mechanism. By downsampling temporal features into multiple resolutions, we dynamically generate scale-adaptive spatial masks, allowing the model to distinguish between micro-scale physical neighbors and macro-scale functional correlations. Second, to disentangle unique variations from shared trends, we propose a Prototype Residual Learning module. It retrieves global common patterns from a momentum-updated prototype bank and subtracts them from node representations, explicitly forcing the network to model the residual unique deviations. By integrating these two orthogonal views, i.e., structural multiscaling and semantic decomposition, our approach effectively isolates valid spatial-temporal signals from noise. Extensive experiments on real-world datasets demonstrate that M^2Net outperforms state-of-the-art baselines on extensive empirical verifications.
[AI-218] GazeAct: A Human-Gaze-Annotated Manipulation Dataset for Attention-Regularized Policies
链接: https://arxiv.org/abs/2610.24507
作者: Anupam Pani,Wang Yuan,Qifan Wang,Yanchao Yang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-language-action (VLA) policies must ground language instructions in the right objects, contact points, and target locations for reliable manipulation. Human gaze provides a natural signal for this grounding, but existing robot datasets rarely pair real human gaze with the multi-camera observations used by modern policies. We introduce GazeAct, a gaze-equipped real-robot manipulation dataset and training framework for multi-view VLA policies. Using screen-based teleoperation, we record gaze directly on the robot camera feed viewed by the operator, yielding 12000 demonstrations across 40 tasks and five domains with 300 Hz gaze synchronized with images, proprioception, and actions. We transfer gaze from the operator-selected view to the remaining synchronized cameras, producing spatially consistent targets across the policy’s visual inputs. We then use these gaze-derived targets to regularize language-conditioned visual attention during training while keeping inference gaze-free. Across 18 real-world tasks, our method improves average success from 48.1% to 58.6%, with gains up to 25 percentage points. Ablations show that gaze works best as soft supervision and that real human gaze provides stronger attention targets than generic attention-shaping priors. The gains transfer to LIBERO-Spatial (57.9% to 69.6%) and different VLA this http URL results show that gaze collected in the robot observation space can provide effective training-time targets for improving visual grounding in VLA manipulation.
[AI-219] ConCI: A Controlled Counterfactual Benchmark for Contextual Integrity in LLM s
链接: https://arxiv.org/abs/2610.24494
作者: Juhyeon Park,Sumin Yu,Taesup Moon
类目: Artificial Intelligence (cs.AI)
备注: Under Review
Abstract:Contextual integrity (CI) holds that the appropriateness of sharing personal information depends on the context of the information flow, not on the information alone. However, existing CI benchmarks often use independent scenarios that vary multiple CI parameters simultaneously and are dominated by inappropriate flows, making it unclear whether correct predictions reflect sensitivity to context or a general refusal tendency. We present ConCI, a controlled-counterfactual benchmark in which an anchor flow is paired with two counterfactuals that each modify a single CI parameter. The resulting triplet is scored jointly, with credit given only when all three flows are judged correctly. Built from six existing benchmarks, ConCI contains 1,592 flows in 556 triplets, is approximately label-balanced, and is validated by seven human annotators per flow. Across seven LLMs, many correct anchor predictions fail on at least one counterfactual, and even the strongest model errs on about a fifth of triplets. Furthermore, models perform near ceiling on inappropriate flows but weaker on appropriate ones, often continuing to judge a flow inappropriate even when a contextual change should make it appropriate. Larger models narrow this bias toward non-disclosure but do not eliminate it, and reasoning yields only modest gains.
[AI-220] SENTRY: A Streaming Edge Network Trust Engine for Real-Time Audio Deepfake Detection
链接: https://arxiv.org/abs/2610.24468
作者: Chirantan Banerjee,Soumita Chatterjee,Bhargab B. Bhattacharya,Hafizur Rahaman
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Audio and Speech Processing (eess.AS)
备注: 7 pages, 1 figure, 5 tables
Abstract:Real-time voice cloning has enabled large-scale streaming deepfake and vishing attacks, while existing Network Detection and Response (NDR) solutions rely on GPU-based post-processing that requires buffered audio and generalizes poorly to unseen speech synthesis models. This paper presents SENTRY, a hardware-oriented, streaming trust engine for inline deepfake detection at the network edge. SENTRY employs an integer lifting-wavelet front end together with a bank of multiplier-free transforms, including wavelet-packet envelopes, Walsh-Hadamard sequency profiles, waveform micro-statistics, and long-term spectral features, designed so that acoustic analysis needs only integer add, shift, compare and bitwise operations. A supervised gradient-boosted tree classifier is combined with an unsupervised Mahalanobis out-of-distribution detector trained solely on authentic speech, and their outputs are fused by a lightweight logistic layer to produce graduated LOG, CHALLENGE, or DROP actions. Under a closed-set protocol, SENTRY reaches 98.75% accuracy at a 1.25% equal-error rate (EER) on WaveFake, with perfect separation of HiFi-GAN and WaveGlow; its supervised track alone achieves an AUC of 0.9994 (fused: 0.9993). On ASVspoof 2019 LA (attacks A01-A06) over an 8 kHz mu-law telephony channel, it reaches 94.75% accuracy at a 5.05% EER, with a supervised-track AUC of 0.9948 (fused: 0.9884) and per-attack EERs below 7%. In our prototype, the feature front end runs as bit-exact integer code on one Arm Cortex-A53 core, and the classifier and fusion stage are synthesized for a Xilinx Zynq UltraScale+ MPSoC using no DSP slices, returning a decision 220 ns after a feature vector is ready. These results indicate that deepfake screening can be placed inline without GPUs or floating-point hardware; generalization to unseen generators is left to future work.
[AI-221] Learning to Compose: Recursive Self-Improvement of Shared Executable Stacks for Zero-Shot Long-Horizon Robot Manipulation
链接: https://arxiv.org/abs/2610.24464
作者: Yangzhi Chen,Feiran Liu,Zishuo Wang,Zijian Jin,Xin Zhu,Lecheng Ruan,Wenjuan Han
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Individually competent robot skills can fail when composed: tools may lack a required capability, completion signals may misrepresent physical progress, and one skill may leave a state that violates its successor’s prerequisites. We introduce Recursia, a recursive self-improvement framework that treats the shared executable stack, rather than individual task programs, as the persistent unit of improvement. The stack integrates tools, skills with execution contracts, handoffs, and verifiers. Source-side execution traces guide failure attribution and the generation of missing tools or repairs to existing shared components. Generality checks, mechanism tests, and paired selection and regression gates determine which changes become the next parent. The resulting stack is frozen and reused through declarative task adapters, allowing source-derived changes to support new compositions without target-side code repair. Experiments cover LIBERO-Pro Long simulation and a physical Franka FR3 equipped with a Franka Hand. In simulation, Recursia improves success over ASPIRE by 5.8 percentage points under Pos perturbations and 15.1 points under Task perturbations. Across three physical compositional target tasks, it improves mean success over the Skill Document Only baseline by 40 percentage points. Analyses of constituent execution, candidate rejection, and inference drivers distinguish local mechanism improvements from complete task success and characterize the cost of feedback-conditioned orchestration. These results examine shared executable interfaces as a concrete mechanism for compositional transfer in robot programming.
[AI-222] ree Plurality Veto: Voting in Few Rounds with Efficient Communication and Low Metric Distortion
链接: https://arxiv.org/abs/2610.24456
作者: Gabriel Chen,Yu Cheng,Alice Wang
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Discrete Mathematics (cs.DM); Data Structures and Algorithms (cs.DS)
备注: SOSA’27
Abstract:We consider the metric distortion framework for voting, where n voters and m candidates are embedded in an unknown metric space. A voting rule selects a candidate using only voters’ ordinal preferences, and its distortion is the worst-case ratio between the social cost of the selected candidate and the minimum social cost. We study how round complexity trades off against total communication in deterministic voting protocols with low distortion. Deterministic rules can achieve the optimal distortion of 3 in one round by eliciting complete rankings using \Theta(nm\log m) bits, while a direct sequential implementation of PluralityVeto achieves the same distortion using only O(n\log m) bits but requires \Theta(n) rounds. Our main result is TreePluralityVeto, a deterministic voting protocol that, for every r\ge 2 , achieves distortion at most 2r-1 using at most r rounds and O(n\log m+n s^1/(r-1)\log s) bits of total communication, where s = \min\n, m\ . The protocol runs PluralityVeto on a hierarchical partition of the voters, collecting restricted rankings in parallel at each level. The branching factor controls the tradeoff between rounds and communication, and a matching-based analysis shows that each level increases the distortion bound by at most 2 . Comments: SOSA’27 Subjects: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Discrete Mathematics (cs.DM); Data Structures and Algorithms (cs.DS) Cite as: arXiv:2610.24456 [cs.GT] (or arXiv:2610.24456v1 [cs.GT] for this version) https://doi.org/10.48550/arXiv.2610.24456 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-223] MELD: Recursive Self-Improvement via Hierarchical Coordination for Multi-Agent Scientific Discovery
链接: https://arxiv.org/abs/2610.24454
作者: Sizhe Tang,Zuyuan Zhang,Zeyu Fang,Tian Lan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-agent scientific discovery with large language models raises a coupled learning problem: research assignments determine both which experiments are pursued and the experiences from which agents adapt. We introduce MELD, a unified hierarchical test-time training framework that jointly adapts a team-level Coordinator policy and agent-level individual Scientist policies within a single discovery process. The Coordinator generates a joint portfolio of search branches and research directives over a shared derivation tree, while Scientists develop hypotheses and execution plans implemented by a shared frozen Executor. Both levels receive alternating GRPO-style parameter updates under a shared smooth best-of-budget objective, allowing team organization and individual research behavior to co-evolve in a recursive self-improvement loop. Our analysis establishes objective alignment for the hierarchy, characterizes conditions for coordination gains over independent parallel discovery and the first-order benefit of private adaptation, and derives a finite-budget regret bound for adapted models. On 22 MLGym and MLE-bench tasks, MELD is best on 20 of them and achieves the highest aggregate normalized scores on MLGym, using a 4B backbone for coordination and research. Most baselines use proprietary Scientists, and all systems share one frozen Executor.
[AI-224] Hippo: Training-Free Agent Self-Evolution via Hippocampal Replay
链接: https://arxiv.org/abs/2610.24453
作者: Yutong Song,Jiang Wu,Yuxuan Ren,Tony Qin,Baoyuan Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Motivated by recursive self-improvement (RSI), we study how LLM agents can learn through external memory without updating model parameters. Extracting useful experience from a single attempt is difficult: flawed reasoning can produce success, while poor execution can obscure a sound strategy. Preserving its value over time is also difficult: our accumulation baseline without consolidation initially improves but eventually performs below a no-memory baseline. Our key idea is to compare attempts at the same task to identify reusable decisions and refine when they apply. We introduce Hippo, a memory framework inspired by selective hippocampal encoding and replay that learns from these comparisons when shortcomings are diagnosed. It compares the resulting experience with existing memory to merge overlaps and resolve conflicts through explicit applicability conditions. This allows prior guidance to be revised without storing every lesson as a separate entry. Experiments on SWE-bench Verified, WebArena, and Mind2Web show performance gains in all three settings and substantially fewer retained experiences than the accumulation baseline. The gains extend across the evaluated base models of varying capabilities, indicating that memory-based adaptation remains useful even for highly capable models.
[AI-225] rust the Edge Doubt the Arrow: Auditing Validation Signals for Text-Mined Clinical Causal Knowledge Graphs PRICAI2026
链接: https://arxiv.org/abs/2610.24390
作者: Su Lan,Shicheng Zhou,Yiqi Wang,Yanming Zhu,Xuefei Yin,Alan Wee-Chung Liew
类目: Artificial Intelligence (cs.AI)
备注: Accepted at PRICAI 2026. 16 pages, 5 figures, 5 tables
Abstract:Clinical causal retrieval-augmented generation systems increasingly use knowledge graphs (KGs) mined from notes, yet extracted assertions remain candidates that require audit for patient-level EHR support and direction. We construct a directed candidate graph from 328,839 MIMIC-IV discharge summaries using LLM extraction and UMLS normalization, and evaluate three signals against SemMedDB: within-admission co-occurrence lift, cross-admission diagnosis-coding temporal precedence, and literature corroboration. Lift filters unsupported co-mentions and provides useful EHR support, but not causal proof. Coding order is a weak direction signal: among the 686 candidates where it commits, it reaches 54.2% agreement with SemMedDB, below the extractor’s 69.4% on the same subset. Literature corroboration supplies independent direction evidence but covers only 30.4% of candidate relations. A MIMIC-III replication yields the same pattern: extractor direction remains stable (70.5%), whereas coding-order direction is near chance (50.9%). Thus, text-mined clinical causal candidate KGs are best treated as EHR-supported but direction-uncertain. In a leakage-controlled causal QA benchmark, lift-only retrieval achieves 35.7% accuracy; conservative causal demotion achieves 37.2% while reducing evidence-unsupported answers from 17.3% to 6.1% at fixed retrieval recall. We release code, aggregate summaries, and scripts to regenerate restricted outputs.
[AI-226] Bridging Exploration and Persona: Information-Directed Dueling for Test-Time Agent Personalization
链接: https://arxiv.org/abs/2610.24385
作者: Jie Mao,Pingchen Lu,Xiangyi Wang,Hengquan Huang,Shuang Qiu,Min Zhang,Zhongxiang Dai
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Effective test-time personalization requires LLM agents to adapt to query-dependent user preferences from lightweight interaction. We study persona selection with binary comparisons between candidate responses as the online signal. We present ID-TAP, a scalable, IDS-inspired neural dueling-bandit framework for large discrete persona spaces. ID-TAP combines a query-conditioned neural utility model with a lightweight Bradley-Terry explorer, and transfers comparison evidence across related queries through similarity-weighted warm starts. The neural model shares statistical strength across queries while preserving query-specific scores, and efficient BT updates score hypothetical challengers. Across seven user-conditioned generation benchmarks under reference- and model-based preference simulators, ID-TAP achieves the strongest aggregate performance among the evaluated baselines within a fixed ten-round comparison budget. These results support information-directed challenger selection as an effective design for test-time persona adaptation.
[AI-227] Is This Your Agent ? Identity Watermarking for Agent Systems in Dynamic Environments
链接: https://arxiv.org/abs/2610.24382
作者: Yuwei Zhong,Yu Cheng,Jiuan Zhou,Weizhi Liu,Yi Xie,Yongkang Hu,Yuan Xie,Zhaoxia Yin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent systems deployed in dynamic environments continuously update their memories, skills, and runtime configurations, leading to evolving trajectories and challenges for persistent identity authentication. Existing text and agent watermarking methods struggle to achieve stable watermark embedding under dynamic environments, while embedded watermarks may become difficult to recover due to environmental changes. This paper proposes DASMark, an identity watermarking framework for dynamic agent systems. DASMark introduces Action-Preserving Embedding (APE), which learns semantically similar but distinguishable decision-expression channels from diverse interaction trajectories to enable stable watermark embedding in dynamic systems. It further introduces Shift-Robust Recovery (SRR) to mitigate context-induced shifts and improve watermark recovery stability under dynamic environments. In dynamic settings, DASMark achieves 98.5% bit-decoding accuracy and maintains 95.2% accuracy under attack scenarios, while preserving task utility comparable to unwatermarked agents.
[AI-228] rain Locally Learn Together: Distributed Multi-Agent Training via ADMM
链接: https://arxiv.org/abs/2610.24380
作者: Yiwei Chen,Changsheng Wang,Yicheng Lang,Qing Qu,Alfred O. Hero,Sijia Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-agent systems (MAS) can outperform individual language-model agents by combining specialized roles, diverse reasoning processes, and collaborative decision-making. However, collaborative MAS training is typically centralized, requiring a common trainer to aggregate role-local trajectories and learning signals and maintain model and optimization states. This motivates a distributed multi-agent learning paradigm in which agents improve system-level capabilities through role-local learning and decentralized communication, paving the way toward scalable, resilient, and self-organizing MAS. To this end, we introduce, to the best of our knowledge, the first optimization-grounded framework for distributed MAS training based on consensus alternating direction method of multipliers (ADMM). We formulate collaborative MAS learning as a consensus-constrained finite-sum optimization problem, naturally lending itself to distributed optimization via ADMM. The proposed approach alternates role-local learning with inter-agent consensus using distributed communication protocols such as ring all-reduce, enabling joint optimization while keeping each agent’s data and training state local. We evaluate our framework on distributed safety alignment and general reasoning tasks. Under compromised-agent settings, ADMM can reduce HarmBench attack success rate (ASR) by up to 26.4%, while achieving reasoning gains comparable to centralized training. These results show that distributed MAS training retains the benefits of centralized joint optimization without requiring a centralized trainer.
[AI-229] Beyond Next-Screen Prediction: The Decision Value of GUI World Models
链接: https://arxiv.org/abs/2610.24372
作者: Sizhe Tang,Hanhan Zhou,Jeffrey Svajlenko,Rashmi Gangadharaiah
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:GUI world models are increasingly used to improve action selection by predicting the consequences of candidate actions, but it is unclear how much of the resulting gain comes from prediction itself rather than from the evaluator that interprets those predictions. We study this attribution problem through controlled interventions on GUI action selection. With candidates and predictions held fixed, lookahead gains can change with how the prediction is presented to the evaluator, so end-to-end gains cannot be attributed to prediction alone. We then introduce Direct Action Ranking (DAR), which learns from reference actions to score candidates without generating future screens, and compare it with a prediction-augmented variant under matched supervision. Predicted successors add no detectable ranking signal once evaluation is learned directly. Without retraining, DAR improves AndroidWorld task success by 5.8–16.4 points across three proposers, outperforms world-model lookahead, and transfers zero-shot to unseen apps, where every prediction-conditioned configuration falls below proposer confidence. These results suggest that GUI world models should be evaluated by their decision value, the incremental action-selection value of their predictions beyond a strong direct evaluator, rather than by prediction fidelity or end-to-end lookahead gain alone.
[AI-230] actix: Towards a Programmable Strategy Interface for Flexible SAT Solving
链接: https://arxiv.org/abs/2610.24369
作者: Haoze Wu,André Schidler,Armin Biere,Nina Narodytska
类目: Logic in Computer Science (cs.LO); Artificial Intelligence (cs.AI); Logic (math.LO)
备注:
Abstract:Modern SAT solvers combine a rich repertoire of proof rules with sophisticated strategies that determine when and how to apply them. Often, the internal implementations are tightly coupled to particular strategies, making it inconvenient to define new strategies without changing solver internals. We present Tactix, a proof-producing sequential CDCL solver that incorporates a wide range of techniques in state-of-the-art solvers with a clean separation of proof rules from solving strategies. This separation enables a public interface through which users can invoke particular techniques, observe their effects, and program a wide range of solving strategies. The default strategy of Tactix is programmable over this interface and outperforms state-of-the-art solvers on benchmarks from the 2024–2026 SAT Competitions. We also use the interface to develop novel strategies tailored to challenging benchmarks from two application domains, which led to performance gains over strong baselines. We intend Tactix to serve as an infrastructure for designing and evaluating novel solving strategies. Finally, we also present a self-study of the AI-assisted development process behind Tactix.
[AI-231] WAQS: Weight Absorption for Quadratic Probing and Affine Steering
链接: https://arxiv.org/abs/2610.24367
作者: Ding Zhu,Mohammad Mahdi Khalili
类目: Artificial Intelligence (cs.AI)
备注: 9 pages
Abstract:Activation steering provides a lightweight alternative to fine-tuning for controlling large language model behavior. Existing methods, however, often face a trade-off between efficiency and expressiveness: fixed steering vectors incur negligible inference overhead but cannot adapt to the current model input, whereas input-dependent nonlinear interventions provide greater flexibility at additional runtime cost. We introduce \textbfWAQS (\textbfWeight \textbfAbsorption for \textbfQuadratic probing and affine \textbfSteering), a framework that obtains state-dependent steering, without additional inference-time computation. WAQS models a concept with a quadratic probe, whose gradient yields an affine, activation-dependent steering direction. We show that this affine intervention can be folded exactly into adjacent linear transformations through an offline weight update. Moreover, we prove that quadratic functions form the maximal class of smooth concept scores whose gradient-based interventions are affine and therefore exactly absorbable into linear layers. Under a Gaussian representation model, our formulation admits an interpretation as the difference of class-conditional Mahalanobis pulls and naturally generalizes linear discriminant and difference-in-means steering. To scale to high-dimensional representations, we introduce a low-rank parameterization of the quadratic term. Across truthfulness, refusal, and sycophancy steering tasks, quadratic steering matches or improves upon constant linear steering while preserving model utility, with the absorbed implementation introducing no additional inference-time operations. Our results suggest that quadratic probes provide a practical middle ground between static linear steering and computationally expensive dynamic nonlinear interventions.
[AI-232] A Two-Stage Hybrid Framework for Inferring Air Trips from Mobile Location-based Service (LBS) Data under Incomplete Airport Sightings
链接: https://arxiv.org/abs/2610.24366
作者: Hsuan-Chih Wang,Xianfeng Terry Yang
类目: Artificial Intelligence (cs.AI); Applications (stat.AP)
备注:
Abstract:Mobile device location-based service (LBS) data provide large-scale spatiotemporal observations that can help airport planners understand air travelers’ mobility patterns. However, airport sightings are often missing before or after flights, making strictly rule-based detection incomplete. This study develops a two-stage framework to infer U.S. domestic air trips from large-scale LBS data under incomplete airport sightings. Stage 1 uses heuristic rules to identify high-confidence air trips whose endpoints are observed at two distinct airports. Stage 2 addresses latent air trips with a probabilistic model that infers air routes using spatial proximity, temporal compatibility, and historical flight schedule information. Using January 2022 nationwide LBS data with nearly 10 million users and 131 million sightings, results are validated against T-100 Domestic Segment route benchmarks and a Stage 1 pseudo-ground-truth reference set. The framework increases air route coverage from 10.65% under Stage 1 to 50.60% and recovers nearly all top-1,000 benchmark routes. It improves Spearman rank correlation with T-100 route shares from 0.4490 to 0.7630 and reduces Jensen-Shannon divergence from 0.2808 to 0.2263. Stage 2 also recovers 92.2% of Stage 1 reference trips. This study formalizes three endpoint observation regimes for LBS-based air trip inference (airport-observed, semi-observed, and latent) and helps airport agencies identify domestic air travel patterns missed when airport sightings are incomplete.
[AI-233] DKnown-Jev: Fine-Tuning and Evaluation of a JEV-Style Model for Tasks Involving Normative Knowledge
链接: https://arxiv.org/abs/2610.24358
作者: Qi Li,Jianjun Xu,Pingtao Wei
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Tasks involving normative knowledge include concept interpretation, eligibility determination, procedure explanation, material requirements, and service-outcome assessment, all of which require rules and facts published, maintained, or recorded by authorized entities. Whether a candidate paragraph can answer a question therefore depends on applicability and answer completeness as well as topical relevance. We develop DKnown-Jev to classify the normative usability of candidate paragraphs into two classes. The 2B and 27B configurations are fine-tuned directly from the native Qwen3.5-2B and Qwen3.8-27B base models, respectively. The models combine question conditions, article metadata, and paragraph content, adopt the classes Answer Evidence (AE) and Non-answer Evidence (NE), and undergo domain adaptation using human-verified training data. Five configurations are evaluated on the same test set of 1,120 human-verified examples. DKnown-Jev 27B trained for one epoch achieves the highest accuracy of 98.75% and macro-F1 of 98.60%, exceeding Jev by 6.07 and 6.76 percentage points, respectively. DKnown-Jev 2B trained for one epoch achieves 92.86% accuracy and 92.18% macro-F1, the best aggregate results among the tested 2B configurations. Its gains over Jev are 0.18 and 0.33 percentage points on this test set, corresponding to only two additional correct predictions, which is insufficient to establish a stable advantage. The 27B one-epoch, 27B two-epoch, and 2B one-epoch configurations all exceed Jev on both metrics on this test set. Class-level analysis shows that the 27B two-epoch configuration has the highest AE recall, while the 27B one-epoch configuration has the highest NE recall. This study provides practical and empirical evidence for developing and applying candidate-scoring models to tasks involving normative knowledge.
[AI-234] Koi: Evidence-Guided Self-Calibrating Planning for Multi-Cloud Heterogeneous LLM Inference Fleets
链接: https://arxiv.org/abs/2610.24355
作者: Mankeerat Sidhu,Gangmuk Lim,Hetarth Chopra,Maximillian Eisenberg,Quan Hao Ng,Aryan Gosaliya,Shu Liu,Luka Govedič,Jiaxin Shan,Hongkuan Zhou,Vikram Sharma Mailthody,P. Brighten Godfrey,Ion Stoica
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI)
备注:
Abstract:Deploying a large language model (LLM) requires choosing the right accelerator, parallelization strategy, engine configuration, and replica count amongst ~140 other variables across clouds, regions, and markets. These decisions are intertwined, and the best cluster-wide plan continually changes as jobs contend for shared capacity on shifting workloads. We present Koi, a self-calibrating cluster planner for heterogeneous, multi-cloud fleets. Koi uses a Recursive Language Model (RLM) hierarchy to propose hypotheses and reason over deployment while deterministic components predict serving behavior, score candidates, and control placement. It represents inference behavior as a fixed causal graph linking decisions (e.g., accelerator, parallelism), runtime mediators (e.g., KV-cache pressure, communication overhead), and outcomes (e.g., latency, throughput). Each deployment strengthens or weakens scoped causal relationships, building a comprehensive playbook for inference that the planner employs to guide future decisions. Job-wise configurations are composed into cluster-wide plans and evaluated against serving objectives, resource constraints, SLO risks, and cost. Across benchmarks, Koi achieves 1.18x greater goodput than the strongest baseline while reducing /token by 20% in a real heterogeneous cluster. Ablations isolate the contribution of Koi’s architecture: a pure language model attains a composite score of 35.3, the same model equipped with Koi’s deterministic tools reaches 51.6, and the full Koi planner reaches 70.8.
[AI-235] FlashRT-PAI: Fast Kernels Are Not Fast Pipelines for Physical AI Inference
链接: https://arxiv.org/abs/2610.24349
作者: Liang Su,David Chen,Jason Li,Kevin Guo,Teresa Lv
类目: Artificial Intelligence (cs.AI)
备注: source code this https URL
Abstract:Vision-language-action (VLA), video-action (VA), and world action models (WAMs) form the backbone of modern physical AI systems, yet their real-time single-batch inference remains a defining bottleneck. Latency is dominated not by individual kernel computation, but by system-level overhead: thousands of short kernel launches, intermediate data movement across operator boundaries, and redundant recomputation across consecutive decisions. General-purpose graph compilers cannot address these costs, as they optimize individual calls in isolation and lack visibility into cross-call state semantics and global pipeline structure. We present FLASHRT-PAI, a low-latency inference engine built around model-hardware pipeline co-design rather than isolated kernel optimization. FLASHRT-PAI combines three core mechanisms: deep kernel fusion with launch overlap eliminates inter-operator boundaries and launch gaps; an execution contract formalizes the pipeline’s stage DAG, update rates, and named buffers to enable cross-decision state reuse and selective refresh; and execution-state capsules enable fast snapshot, restore, and fork of persistent pipeline state. The system also supports drop-in acceleration of unmodified PyTorch models via parity- and latency-gated kernel replacement. Evaluation covers six policies on RTX 5090 and four on Jetson AGX Thor. At matched 16-bit precision, FLASHRT-PAI is 1.3-3.7x faster than official compiled paths. Combining pipeline co-design with reduced precision delivers 3.38-12.66x speedup over official eager implementations. In LIBERO-10, four-bit pi0.5 achieves 94.04% success on Thor and 93.32% on RTX 5090, compared with 92.4% for the measured official FP16 reference. FLASHRT-PAI is open source, distributed via Hugging Face Kernel Hub, and its opt-in pi0.5 kernels have been adopted into the LeRobot framework.
[AI-236] Screening LLM Judges Without Ground Truth: A Preliminary Evaluation of an AI Dental Hygiene Coach
链接: https://arxiv.org/abs/2610.24335
作者: William DeShong,Barath Kannan,Auberon Palma,Brenda Bradshaw,James F. Leathrum Jr.,Yuzhong Shen
类目: Artificial Intelligence (cs.AI)
备注: 29 pages, 2 figures, 19 tables
Abstract:Evaluating a fine-tuned coaching model with expert graders is slow and expensive, and an LLM judge cannot be trusted without ground truth. We present a staged pipeline that screens candidate LLM judges and student models before expert review, using explicit decision rules at each stage: judge output audit, similarity with hosted judges, bias checks, and decision robustness. We demonstrate it on an AI dental hygiene coach. The case study uses synthetic, clinically unvalidated training and reference data to demonstrate the operation of the pipeline, rather than to establish the clinical performance of the resulting models. Two fine-tuned small language models (Granite-4.1-8B and Qwen3-8B), each trained at three data sizes, were scored on seven metrics by five judges (three local, two hosted). Two local judges, Qwen3.8-27B and Gemma-4-31B-it, were not distinguishable from the agreement between the hosted judges on six of seven metrics by rank correlation, and Qwen3.8-27B was the most similar to the hosted judges overall. No judge met our criterion for favoring its own model family or training-data source. The student comparison did not separate the two models on any metric under the pipeline’s rule, so the choice of Granite-4.1-8B, which led in 29 of 35 judge-by-metric cells, is a lean that depends partly on truncated Qwen3-8B outputs. The pipeline also surfaced problems that raw scores hid, including the truncated outputs and hallucination flags from one hosted judge that no other judge corroborated. No dental hygiene expert has graded these responses, so the findings are preliminary and the pipeline’s final stage, anchoring to expert judgement, is proposed but not performed.
[AI-237] VERAS: Verification-Guided Adaptive Search for Reliable Operations Research
链接: https://arxiv.org/abs/2610.24329
作者: Minbo Ma,Honglin Zhang,Bangtao Zhou,Chunguang Yu,Dasheng Ju,Yichen Shen,Fengli Xu,Yong Li
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Advances in reasoning, coding, and tool use enable LLM agents to solve operations research (OR) problems through iterative problem modeling, program construction, and verification. Yet passing self-authored verification does not ensure that the solutions satisfy task constraints and meet optimization quality. This gap can arise from correlated specification errors, where the generated program and its checks share omissions or misinterpretations of task requirements, alongside incomplete verification coverage and ineffective optimization strategies. We propose VERAS, a verification-guided adaptive search harness for reliable OR that uses verification evidence to prioritize feasibility repair, constraint-coverage review, and solution-quality improvement. VERAS combines task-specific solution checkers with a separate verification agent to examine candidate solutions against the task requirements. It uses verification feedback to guide feasibility repair, additional constraint checking, and objective improvement, while candidate acceptance criteria screen revisions for deterioration in output validity or constraint satisfaction. On ORAgentBench, VERAS achieves a 42.99% pass rate, exceeding the reported 35.51% of Codex using the same LLM. Across two industrial-scale OR problems, VERAS reduces inter-datacenter peak bandwidth by 10.34% relative to native Codex and lowers annualized distribution network planning costs by 6.28% compared with expert-designed solutions. Code and evidence are available at this https URL.
[AI-238] Continuous Ontology Assurance through External Resources – An Ontology Engineering Method with a Case Study in Software Language Engineering
链接: https://arxiv.org/abs/2610.24322
作者: Ralf Lämmel
类目: Artificial Intelligence (cs.AI)
备注: Under review
Abstract:Domain ontologies require review against knowledge recorded in evolving external resources. We propose COATER, a method that manages executable review components called arrows. Each arrow compares a scoped part of an ontology with a relevant resource and reports support or challenges to its axioms. Challenges may prompt proposals for completion, correction, or restructuring and become tasks for human or agent-assisted triage. Per-axiom coverage summarizes the assessments. We illustrate COATER with the FSL ontology in software language engineering (SLE): arrows review language coverage, language conceptualization, and software-engineering activities. An LLM-assisted linkage exercise increases compliance with a defined Wikipedia-linkage policy from 405 to 600 of 600 entities. A structured catalogue of SLE resources organizes review opportunities.
[AI-239] Morpheus-2.0: Multiscale Probabilistic Forecasting with Dual-Axis Attention KR
链接: https://arxiv.org/abs/2610.24321
作者: Aakriti,Manoj Cherukumalli,Prathamesh Patil,Leonidas Tsepenekas,Amit Varshney,Penghang Liu,Elizabeth Fons,Vamsi Potluru
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注: 26 pages, 5 figures, 16 tables. Aakriti and Manoj Cherukumalli contributed equally
Abstract:A universal forecaster is trained once and applied everywhere, so it must serve tasks whose schemas, forecast horizons, and temporal resolutions all differ. The number of targets and covariates changes from dataset to dataset, horizons vary by orders of magnitude, and sampling rates range from seconds to years. Existing time-series foundation models address these only in part, since they isolate each series or visit the two grid relations in separate blocks, and they fix the horizon and resolution before training begins. We introduce Morpheus-2.0, a compact encoder-only foundation model that represents targets and covariates on a single variable-time grid. Its dual-axis attention runs both relations side by side in every layer, one across variables at a timestep and one across time within a variable; a learned gate weights the two per layer, and an axis-aware rotary encoding preserves both grid coordinates. The forecast length is sampled during training, so one set of weights spans short to long horizons at no additional inference cost. Morpheus-2.0 Multiscale extends the model to several temporal resolutions at once, weighting them according to the series being forecast. Across GIFT-Eval, which spans several domains and frequencies, the two models give complementary operating points: Morpheus-2.0 attains 0.685 relative MASE with 81M parameters, while the larger Morpheus-2.0 Multiscale attains 0.646 relative MASE and 0.455 relative CRPS, the lowest of any published system on both metrics.
[AI-240] RATION: Towards Quantitative Risk-Based Access Control for AI Agents
链接: https://arxiv.org/abs/2610.24311
作者: Yuntao Du,Harsh Chaudhari,Yuetian Chen,Ron Heichman,Ashish Jayamohan,Kumar Pallav,Mahim Patel,Kaiyuan Zhang,Kyle Polley,Jerry Ma,Ninghui Li
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Code is available at this https URL
Abstract:AI agents increasingly interact with external environments to complete complex user tasks. These interactions can lead to unexpected behavior and expose agents to security threats such as indirect prompt injection. Current agent security systems largely protect agents by restricting their actions, which can result in a substantial security-utility trade-off for complex, open-ended tasks. To address this, we propose a quantitative risk-based approach to access control for AI agents that regulates measurable risks accumulated during task execution. As a concrete realization of this approach, we introduce RATION, a security framework that tracks measurable risks associated with agent actions and enforces quantitative constraints throughout execution. RATION first constructs a task-specific policy by combining explicit constraints derived from the user request with implicit risks calibrated from historical executions of similar tasks. During task execution, RATION continuously tracks the accumulated risks and applies different enforcement strategies depending on the type of risk. Extensive experiments on four security benchmarks show that RATION provides strong security while maintaining high task utility across diverse models and environments.
[AI-241] Learning Administrative Boundary-Aware Location Encoders via Reverse Geocoding
链接: https://arxiv.org/abs/2610.24306
作者: Lakshay Sharma,Alex Marin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Location encoders learn compact representations of latitude–longitude coordinates for downstream geospatial prediction tasks. Existing approaches are pretrained using continuous or naturally occurring supervision, such as imagery, climate variables, or species distributions, and therefore do not explicitly capture administrative boundaries or place names, despite their strong influence on many real-world geospatial phenomena. We introduce, to our knowledge, the first method for learning administrative boundary-aware location encoders, using reverse-geocoded text as a self-supervised training signal. Our approach can be applied to existing location encoder architectures and consistently improves their downstream performance. We further introduce RevGeoRQ, a custom architecture tailored to reverse-geocoding supervision, which yields additional gains and exceeds contemporary location encoders on several downstream tasks. Finally, the use of human-interpretable reverse-geocoded strings and residual-quantized semantic IDs provides a more interpretable representation of geographic location than existing approaches. Code and model weights to be released soon.
[AI-242] CME-SpectrumBench: Can LLM s Analyze Condensed Matter Spectral Data?
链接: https://arxiv.org/abs/2610.24304
作者: Jin Gene Wong,Anjney Midha,Joseph Mathew Tennyson,Wei-Lin Chiang,Ion Stoica,Zhi-Xun Shen
类目: Artificial Intelligence (cs.AI); Other Condensed Matter (cond-mat.other); Strongly Correlated Electrons (cond-mat.str-el); Superconductivity (cond-mat.supr-con)
备注:
Abstract:Large Language Models (LLMs) are presently well assessed by expert-level mathematics and physics benchmarks which focus on theoretical, symbolic reasoning. However, we lack experimental benchmarks that evaluate data analysis in frontier research problems, a capability that future domain-specific foundation models must have. To this end, we present CME-SpectrumBench, an expert-designed benchmark consisting of 190 questions that measures how reliably frontier LLMs analyze research-level spectral scientific data in condensed matter experiments (CME). Using simulated CME spectral data, we prompt models to identify structures from 2D intensity arrays in the face of noise and limited instrumental resolution, a test of dense context reasoning. Models are scored by how accurately their numerical answers approach the ground truth. Under direct query, scores generally pale in comparison to handwritten code due to clear failure modes: models can find simple features but struggle to trace patterns or compute derived quantities, and performance typically degrades with increased data resolution and noise/convolution levels. LLM performance becomes comparable to handwritten code by allowing tool use (access to a Python interpreter) with/without ReAct prompting, in effect mapping questions onto agentic code generation problems. Our findings show that frontier models can serve as research assistants by interfacing with data through code, but still face difficulties processing information directly from large datasets - a key barrier future foundation models must overcome. The code and data are publicly available at this https URL.
[AI-243] Who Speaks Who Authorizes Who Acts? Understanding Trust Boundaries in Multi-Agent LLM s
链接: https://arxiv.org/abs/2610.24298
作者: Junru Zhu,Xin Chen,Hang Lyu,Jiayue Meng,Muzi Li,Yifan Chen,Hengyi Huang,Yu Sun,Weiwei Qi
类目: Artificial Intelligence (cs.AI)
备注: 14 pages, 4 figures, 9 tables
Abstract:Multi-agent LLM systems exchange messages before acting through tools, but message delivery, signing authority, and tool-side enforcement are separate trust boundaries. We represent them as communication, authorization, and execution structures, and use the authority cut to characterize the smallest independently compromised signer coalition that can construct a policy-valid false-claim proof under stated assumptions. Controlled paired hidden-state studies reveal four distinctions. First, no fixed-message routing contrast across four models meets the prespecified large-effect criterion; when relays rewrite messages, executed false-state action rates vary by model and route configuration. Second, presented source-identity cues do not uniformly increase protected-action selection across three models. Third, on invalid-proof cases, Nemotron and Llama request protected actions in 52.1% and 54.2% of trials, respectively, yet a deterministic tool gate commits none: request-level and commitment-level safety readouts diverge. Finally, a separate single-model authorization check shows that a false claim with a policy-valid compromised-signer proof can pass the gate, since proof validity is not factual truth. These findings motivate evaluating delivered information, qualified authorization, model proposals, and committed actions separately.
[AI-244] Unbounded Losses Unchecked Inputs: Measuring the Safeguards of LLM Options-Trading Agents
链接: https://arxiv.org/abs/2610.24289
作者: Mengxiao Wang,Nitesh Saxena
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 26 pages, 9 figures, 8 tables. Original measurement study: a logged screen of 765 GitHub repositories and cell-level scoring of 26 LLM options-trading systems against 18 safeguards
Abstract:Large language model (LLM) agents now place options orders at retail brokers, and developers publish open-source options agents faster than anyone has examined them. Prior security work on trading agents scored academic equities designs by whether an attacker can flip a buy-or-sell decision. We present the first measurement of the options agents that brokers and developers ship, scored by the safeguards they enforce and by the worst loss their documented order paths can reach. We screened 765 GitHub repositories, broker documentation, and the research literature and found 215 eligible repositories and 8 broker-native agent interfaces. We then scored 26 systems, including every broker interface and every open-source project with at least 20 stars, on seven robustness and eleven security safeguards with cell-level evidence. The ecosystem is new: 198 of the 215 repositories were created in 2026, and 129 (60.0%) appeared in the two weeks around one options-trading hackathon. Of the 22 systems that can place orders, 10 enforce a pre-trade limit outside the model, yet none of the 26 authenticate their text sources or cross-check market data. Of the 26, 17 can execute with no mandatory human step and 13 can reach unbounded loss, including 6 of the 8 broker interfaces. Because an attacker who writes the agent’s inputs can move an order’s strike, expiry, volatility, and legs while its direction stays correct, options agents need loss limits enforced at the order boundary.
[AI-245] SteerSearch: Interactive Search Reasoning with Mid-Response Steering
链接: https://arxiv.org/abs/2610.24287
作者: Udit Thakur,Baohao Liao,Matteo Nulli,Stefan Vasilev,Seyyed Hadi Hashemi,Shahram Khadivi,Christof Monz
类目: Artificial Intelligence (cs.AI)
备注: 48 pages
Abstract:Search-augmented language models increasingly serve as long-horizon research agents, yet most open training methods assume a fixed user goal throughout search. We study mid-response steering, where users redirect a model during ongoing reasoning or retrieval. We identify answer-target anchoring, a failure mode in which fixed-query search agents retrieve evidence for the updated request but still answer an earlier question. We introduce SteerSearch, an RL framework that trains on interactive search trajectories with injected interventions to enable context-preserving target reassignment, and SteerSearch-Bench, a benchmark adapted from multi-hop QA datasets. We also investigate Contrastive Context Reward (CCR), an auxiliary signal for aligning post-intervention reasoning with the updated goal. Without CCR, SteerSearch improves average steering accuracy over Search-R1 from 11.1% to 31.3% on Qwen2.5-3B and from 36.6% to 44.6% on Qwen3-4B, while preserving standard single-query performance. Gains also extend to larger Qwen2.5 models and natural multi-turn conversations.
[AI-246] SPIDES: Information-Flow Control for Browser Agents
链接: https://arxiv.org/abs/2610.24277
作者: Egor Zverev,Juhee Kim,Leon Trampert,Boris Köpf,Aashish Kolluri,Shruti Tople,Santiago Zanella-Béguelin
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Browser agents automate complex workflows by reading and acting on arbitrary page content, but this flexibility exposes them to prompt injection attacks and unintended data exfiltration. Recent work argues that LLM-driven browser agents erode traditional web security boundaries such as the Same-Origin Policy (SOP) by acting across origins with the user’s authority and no built-in isolation. We argue that restoring SOP is neither sufficient (a same-origin public sink can leak private data) nor necessary (legitimate tasks require authorized cross-origin flows) for protecting the security of users. We propose SPIDES, which instead enforces information-flow control by gating primitive browser actions such as typing, clicking and navigating on labels propagated from element-level HTML confidentiality and integrity annotations. Untrusted and confidential content is masked from the agent’s context, while policies block unauthorized flows, providing security guarantees against both prompt injection and data exfiltration that hold independently of the model driving the agent. To evaluate SPIDES and baseline defenses under a common threat model, we build MicroNet, a closed ecosystem of 28 interlinked websites with element-level ground-truth labels and paired benign/adversarial tasks. SPIDES achieves 0% attack success rate by design while retaining high task completion rate and autonomy. Our findings demonstrate that information-flow control is an effective technique to extend web security to browser agents, gaining on autonomy and utility against state-of-the-art baselines. The source code and data are openly accessible at this https URL.
[AI-247] From SEC Filings to Financial Signals: Evaluating LLM s on Stock Market Reactions and Analyst Forecasts
链接: https://arxiv.org/abs/2610.24268
作者: Shengkun Wang,Taoran Ji,Linhan Wang,Sattar Mansi,Chang-Tien Lu,Yinlin Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Form 10-Q MDA sections combine long-form management narrative with accounting and forward-looking information. Benchmarks for financial filings commonly evaluate question answering or numerical reasoning, leaving unclear whether outputs from large language models (LLMs) align with market and analyst outcomes. We evaluate prompted, zero-shot outputs from 13 general-purpose and finance-specialized LLMs on 45,964 Form 10-Q filing events from 2001 to 2024, linking MDA text to cross-sectional fundamental ranks, pre-filing analyst forecasts, and realized abnormal returns. The evaluation covers return ranking under five controlled information views and next-quarter EPS forecasting under direct and consensus-anchored interfaces. Relative to disclosure alone, adding cross-sectional fundamental ranks raises mean effective-date rank correlation (RankIC) by 0.0089 for effective-event-date returns and 0.0098 for following-day returns, whereas analyst-summary context provides no comparable average gain. Across the four disclosure-containing views, finance-specialized models show an effective-event-date group advantage but no statistically reliable following-day advantage, while general-purpose Qwen models produce the strongest individual return-ranking results. Under the full-information view, direct LLM forecasts average the 43.4th percentile of the analyst distribution and do not consistently beat individual analysts. Consensus anchoring raises the mean inside-range share from 41.1% to 59.5% but increases reported-EPS error.
[AI-248] Can Compact Open-Weight Large Language Models Recover Mobility Events from Incomplete Trajectories? A Controlled Evaluation Using Air Trips
链接: https://arxiv.org/abs/2610.24266
作者: Hsuan-Chih Wang,Xianfeng Terry Yang
类目: Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET)
备注:
Abstract:Incomplete mobility trajectories can obscure the observations needed to identify important travel events and their endpoints. This study evaluates whether compact, locally deployed large language models (LLMs) can recover missing mobility information from incomplete trajectories, using high-confidence air trips as a controlled test case, and whether more explicit prompts can compensate for the loss of endpoint evidence. Using nationwide location-based services (LBS) data from the United States in January 2022 and airport reference data, thirty high-confidence air trips were converted into three incomplete-trajectory conditions by removing observations near the origin airport, the destination airport, or both. Four prompt variants and two LLMs generated 720 requests, each using a two-stage procedure to identify the flight transition segment and then select the origin and destination airports. Exact transition accuracy is 9.2%, while origin and destination accuracy are 31.4% and 10.7%, respectively. Only 1.9% of requests recover both the rule-identified transition and the reference airport pair. The most explicit prompt produced the highest observed exact transition and exact airport-pair recovery rates. The study introduces a controlled endpoint-removal framework for evaluating recovery of discrete mobility events from incomplete trajectories, allowing failures to be examined by inference stage and helping analysts decide whether recovered information should be retained, independently verified, or left unresolved.
[AI-249] AM-Bench: A Unified Taxonomy and Evaluation Suite for Agent ic Misalignment
链接: https://arxiv.org/abs/2610.24264
作者: Eric Zhang,Terry Jingchen Zhang,Chijioke Ugwuanyi,Jerick Shi,Bernhard Schölkopf,Zhijing Jin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:As LLM-based agents are deployed with increasing autonomy in real-world environments, a critical failure mode emerges: agentic misalignment, where agents choose harmful actions over failure. However, existing safety benchmarks primarily evaluate whether agents comply with explicitly harmful instructions, leaving the conditions that cause agents to choose harmful actions poorly understood. To address this limitation, we propose AM-Bench, a unified taxonomy and 35 scenario evaluation benchmark spanning seven industries for agentic misalignment grounded in the fraud triangle, a well-established criminological framework explaining goal-directed misconduct. We adapt this framework to agentic environments and construct our evaluation suite and detection systems along three dimensions: perceived threat, perceived opportunity, and reasoning. We operationalize each dimension into structured subcategories rooted in prior theory: five threat categories derived from the instrumental convergence thesis, seven opportunity categories drawn from the Emergent Strategic Reasoning Risks (ESRR) taxonomy, and a novel four-question binary reasoning framework synthesized from prior literature. Our evaluation of 10 frontier models reveals a 62-percentage-point gap in misalignment across scenarios, with Gemini 3 Flash and Gemini 3.1 Pro exhibiting the highest levels of agentic misalignment while Claude Sonnet 4.6 displays near-zero rates across all conditions. Our reasoning framework further reveals that models may exhibit harmful behaviours without explicit situational awareness or explanation, complicating both mitigation and detection. AM-Bench scenarios and code are available at this https URL
[AI-250] Cortex: A Fixed-Point Theory of Governed Coding Agents - Technical Report
链接: https://arxiv.org/abs/2610.24255
作者: Marius-Constantin Dinu,Florian Zeba
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 27 pages, 8 figures, 9 tables
Abstract:A coding agent combines a large language model (LLM) with a harness that plans, edits, and executes code. We study Cortex, a supervisory layer for pre-execution governance, requirement tracking, and iterative validation and repair. We formalize requirement closure on a finite lattice. For a fixed requirement set and a sound, monotone validation rule with persistent evidence, exhaustive iteration reaches the least fixed point with no more strict increases than there are requirements. Fair consequence scheduling reaches the same closure. A separate stochastic model bounds expected completion time under a uniform positive-progress assumption. These results depend on their stated assumptions; artifact repair, fallible validation, and finite execution budgets do not satisfy them automatically. We define Gauntlet’s trajectory-similarity and build-gated scores, attack-success estimators, Wilson intervals, and case-cluster bootstrap summaries. Effective-feedback accounting counts newly certified requirements, rather than estimating their information value or utility. The five evaluation families support raw-versus-governed comparisons. We report retained benchmark observations, including rescoring of existing artifacts; constructed outcomes are excluded. Unequal observation sets and incomplete provenance prevent causal claims about governance or measured long-horizon scaling.
[AI-251] owards Reinforcement Learning for Space Robotics: Fast Training and Reliable Real-World Transfer IROS2026
链接: https://arxiv.org/abs/2610.24241
作者: Abhishek Naik,Michael Wu,Michael O’Sullivan,Colin Bellinger,Yunli Wang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Presented at the Space Robotics Workshop at IROS 2026
Abstract:This work demonstrates the use of a fast simulation-based training framework to learn control policies using reinforcement learning. The hardware requirements are minimal and the simulation-learned policies transfer reliably to the real world. With this setup, we show that a quadruped robot can learn to traverse diverse real-world terrains reliably—including in a lunar-analog facility—after less than 20 minutes of training on a single consumer-grade GPU. These results provide a glimpse of a future where reinforcement learning is a crucial tool in developing more capable and robust space robotic systems. Video demonstrations are available at this https URL.
[AI-252] Who Plans Who Edits? Verified Cost-Aware Division of Labor for LLM -Driven Analog Layout Repair
链接: https://arxiv.org/abs/2610.24234
作者: Xilai Dai,Saeed Mohammadi
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 11 pages, 8 figures, 6 tables. Submitted to IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Abstract:Language-model agents can repair analog layouts when sign-off tools check every edit, but which model should plan a repair, which should edit it, and at what cost remain open questions. In our tool-grounded repair agent, models only choose bounded layout operators; sign-off tools rerun each task’s contract checks after every edit, so failing layouts are never accepted. On layout-versus-schematic (LVS), design-rule checking (DRC), and post-layout repairs of three designs (a production high-voltage digital-to-analog converter (DAC), a 5-V oscillator, and a SKY130 operational transconductance amplifier (OTA)) with six tier-matched models from two vendors, 182 blind runs show that cheap models execute plans but rarely succeed alone (the cheapest model: 1 of 11 runs alone, 16 of 17 under a flagship planner; p 0.001), cheap planners fail more often, and no fixed assignment is cheapest on every task. We propose a learning controller that selects models for each role and the escalation order from the first tool evaluation and updates Bayesian estimates after every attempt. With rules and initial estimates frozen before testing, the controller fixed 7 of 8 episodes on four sealed held-out tasks, matching the OpenAI flagship alone, at 15-40% of the flagship’s spend per fix on three tasks but 8% higher spend overall. The cheapest mid-tier model alone fixed 6 of 8 at the lowest overall spend per fix but missed the hardest task. All 32 extracted DAC repairs match the human-guided LVS repair’s integral nonlinearity (INL) within 0.005 least significant bits (LSB) at every code.
[AI-253] LoopHIL: Looping Hardware into LLM -assisted Agents for Wireless Embedded Systems
链接: https://arxiv.org/abs/2610.24231
作者: Dhairya Shah,Spanddhana Sara,Savitha Viswanadh Kandala,Ambuj Varshney
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Software Engineering (cs.SE)
备注: 8 pages, 8 figures, 4 tables. Under review
Abstract:LLM agents are widely used for software development and debugging, but their adoption in embedded systems, especially for wireless tasks, remains limited. This is because embedded development requires hardware-in-the-loop testing. Agents cannot directly observe the spectrum, probe pins, or measure power, so they cannot verify wireless link functionality; currently, human engineers perform this verification. We systematically evaluate whether LLM agents can establish wireless links when provided with hardware access. We introduce LoopHIL, a framework that lets agents control a multi-device RF bench, monitor it with a spectrum analyzer, logic analyzer, and energy meter, and validate links against a verification ladder, all within code-enforced safety limits. Across four increasingly complex tasks, from generating a carrier to backscatter decoding by a commercial receiver, and using six models from two vendors, hardware-in-the-loop increases task completion rates from 54% to 79% and from 17% to 83% on the most challenging task. It also halves task time and output tokens. With debugging tools, agents decrease tag power consumption by up to 96%. However, small models often report unverified claims, so independent hardware checks remain necessary.
[AI-254] raining-Free Prompt Selection for Efficient Reinforcement Learning With Partial Rollouts
链接: https://arxiv.org/abs/2610.24229
作者: Lei Jiang,Honghui Bao,Yue Zhou,Zhengyao Gu,Langzhou He,Natalie Parde
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) has emerged as a prominent paradigm for training reasoning models, but its computational cost makes efficient data selection increasingly important. Since training examples contribute unequally to learning, selecting a small, informative subset can offer a more efficient alternative to using the entire data pool. The training-free selection method estimates prompt value with hidden states from complete greedy rollouts. However, in our datasets, 26 – 32% of these rollouts fail to terminate within the generation budget. We introduce LIFT, a training-free data selector that identifies impactful prompts using only partial reasoning rollouts. At its core, LIFT uses Partial Rollout Context Lift (PRCL), a score inspired by pointwise \mathcalV -information that accumulates token-level log-likelihood ratios between a question-conditioned context and a null context with the question removed. This enables prompt scoring from a short rollout prefix rather than a complete trajectory. Across Qwen2.5-Math-1.5B and DeepSeek-R1-Distill-1.5B, prefixes retaining at least 90% and 95% of the full-rollout PRCL score, respectively, require only 26% and 4.7% of the tokens generated by full rollouts. Compared with SHIFT, LIFT reduces selection cost by 3.7–9.1 \times while achieving competitive downstream RLVR performance, with the highest six-benchmark average in three of six settings.
[AI-255] SWE-Bench Pro V2: A Cleaner Harder-to-Game Coding Benchmark
链接: https://arxiv.org/abs/2610.24206
作者: Soham Dan,Jeff Da,Sanyam Bhutani,Miguel Romero Calvo,Ivan Guevara,Fernando Carabedo,Jose Bernal,Julio Reyna,Luis Leandro Monsalve Mendez,Ying Liu,Daniel Yue Zhang
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Dataset: this https URL
Abstract:SWE-Bench Pro is a widely used benchmark of long-horizon software engineering tasks for coding agents. As agents improve, they reach task issues earlier agents never touched and interact with the evaluation environment in ways the original harness did not anticipate. We present SWE-Bench Pro V2, a revised public split built to keep the benchmark accurate as capabilities advance. Agentic audits, expert adjudication and blind re-implementation repaired or retired 480 of the 731 tasks: 23 contracted engineers spent 404 hours, each working in a language they write professionally, and every revised instruction was verified by a second engineer solving the task from the text alone. A hardened protocol restricts network access to the model endpoint and re-grades every patch in a clean container, so agents cannot retrieve upstream fixes or get credit for changes outside their patch. On the 642 V2 tasks frontier agents resolve up to 99.4% while other systems stay clearly separated, and a 51-task Hard split spreads systems from 52.9% to 98.0%. We release the benchmark and distill lessons for maintaining agent benchmarks over time.
[AI-256] Distillation Self-Training and the Frag ility of Learned Chemical Reasoning
链接: https://arxiv.org/abs/2610.24199
作者: Anagha Aneesh,Sebastian Ibarraran,Frank Hu,Jason Liu,Grant Rotskoff
类目: Artificial Intelligence (cs.AI)
备注: 10 pages, 7 figures
Abstract:Automating synthesis planning for small-molecule drug design requires chemical reasoners that are accurate as well as open to inspection and adaptation, yet frontier language models remain unreliable at predicting reaction outcomes and difficult to deploy, while small open-weight models lack the necessary domain reasoning. We show that frontier supervision can serve as a one-time bootstrap for open models on forward reaction prediction. Our models, Resonance, distill an open-weight teacher’s reasoning traces and then learn from their own verified outputs, turning a small base model with near-zero accuracy into a capable reaction predictor, while a stronger student surpasses its teacher through distillation alone. Accuracy, however, does not bring robust reasoning: the small student inherits its teacher’s single-pass style, self-training mainly sharpens answers it could already reach, and it readily abandons correct answers under pushback, suggesting that chemical reasoning models should be evaluated on the robustness of their reasoning as well as on the accuracy of their answers.
[AI-257] owards an Intelligent Decision Engine for Smart Contract Deployment
链接: https://arxiv.org/abs/2610.24198
作者: Oana-Otilia Captarencu,Emanuel Onica,Andrei Arusoaie
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注: Accepted for the 30th IEEE International Conference on System Theory, Control and Computing, ICSTCC 2026; Copyright is with IEEE
Abstract:Layer 2 (L2) blockchain solutions emerged in recent years, predominantly in the Ethereum ecosystem, as a viable option for scaling a main blockchain network (L1). However, the selection of a L2 solution for deploying smart contracts backing blockchain applications is a decision that depends on many factors, such as costs, efficiency, popularity, and stability. We propose here a systematic methodology to help with this decision making process by using a set of relevant metrics and introducing user-defined profiles that capture different priorities among these metrics. We collect the major categories of smart contracts from a community-maintained data source, and we gather transactions for a six months period from three major L2 solutions: Arbitrum, Optimism, and Base. For each contract category and each blockchain, we compute a set of metrics and scores tailored to each user profile. By applying our methodology, we identify a clear bias in the classification, showing which L2 solutions are better suited for particular application types and user profiles.
[AI-258] Agents Last Doubt: Knowing When to Act and When to Step Back in Tool-Using Agents
链接: https://arxiv.org/abs/2610.24194
作者: Qisai Liu,Qingyue Jiao,Bangwei Guo,Wenxi Dai,Changyi Yang,Jiahe Liu,Dimitris N. Metaxas,Yiyu Shi,Zhanhong Jiang,Minghao Guo
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Tool-using LLM agents repeatedly face decision points under uncertainty, choosing between autonomous action, additional information gathering, and human escalation. These decisions require balancing task success against interaction cost and the risk of irreversible errors. Learning to make these trade-offs is challenging because existing interaction trajectories record only the outcomes of actions taken, leaving the consequences of alternative actions unobserved. To address this gap, we construct a counterfactual dataset containing 3,115 action outcomes from 1,036 decision points in ToolSandbox. At each decision point, we execute each legal action and evaluate its downstream outcomes based on task completion, interaction cost, and irreversible errors. Using this dataset, we train SURE-Act, a controller that selects actions for a frozen backbone agent at inference time. Our experiments show that learning to rank alternative actions improves offline action-selection utility by 0.0309 over a regression-only baseline. We further show that adding SURE-Act enables end-to-end gains of up to 0.0575 in ToolSandbox’s Overall Score with the Qwen3-4B agent. Moreover, SURE-Act outperforms adapted decision rules from SAGE-Agent and Calibrate-Then-Act by 0.1506 and 0.1438, respectively. This work establishes the first counterfactual dataset for ToolSandbox and provides actionable insights into decision-making under uncertainty for tool-using agents.
[AI-259] Bootstrap Stochastic Interpolant Tree for Inference-time Search Under Sparse Reward
链接: https://arxiv.org/abs/2610.24185
作者: Binglin Ji,Anindya Sarkar,Yevgeniy Vorobeychik
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 26 pages, 12 figures, Preprint
Abstract:Tree-based samplers offer a compelling alternative for scalable inference-time search, bypassing the per-step reward and gradient dependencies that introduce systematic bias into SMC-style frameworks. Despite this advantage, contemporary tree samplers fail to achieve broad exploration. Their reliance on learned, DDPM-like transitions inherently couples neighbor states in a trajectory, thereby restricting child-parent feature divergence in the tree sampler. Consequently, these samplers suffer from restricted coverage of the target marginal distribution, severely hindering the exploration critical for efficient search under sparse reward. To overcome these limitations, we introduce a simple yet effective Bootstrap Stochastic Interpolant Tree (BSIT) sampler that decouples child node generation from explicit parent dependencies. By expanding the degrees of freedom during child node sampling, BSIT encourages diverse trajectory rollouts and unlocks broader path-space diversity-the exact catalyst needed for efficient exploration. Crucially, we show that BSIT’s improved exploration accelerates convergence to the target distribution while achieving broader distributional coverage than samplers that rely on DDPM-like transitions for child-node generation, all without incurring additional function evaluations or reward feedback, thereby providing a highly efficient basis for rare target search. Extensive experiments across diverse search settings, including rare-species discovery and target-aligned molecule generation tasks, demonstrate BSIT’s superior search efficiency over the baselines.
[AI-260] orchTPU - A Hybrid Eager/Compiled Architecture for Native PyTorch Execution on TPU
链接: https://arxiv.org/abs/2610.24161
作者: Claudio Basile,Michael Voznesensky,Gunhyun Park,Ben Wilson,Parker Holloway,Zhanyong Wan,Vlad Belous,Manish Khanna,Paul Ganssle,Kevin Gleason,Andrew Dame,Brendan Folie,Bhavya Bahl,Haifeng Jin,Pedro Goncalves Mokarzel,Arian Arfaian,Luke Boyer,Chi Zhang,Yiling Yuan,Adam Mainz,Aditya Patel,Huangze Tang,Jay Gan,Lizzie Liang,Will Froom,Tushar Agarwal,Karlo Basioli,Theotime Combes,Emilio Cota,Minos Papadopoulos,Tudor Pipernea,Yarong Mu,Qi Zhou,Yaoshiang Ho,Aleksey Vlasenko,Alexander Brauckmann,Ziyin Huang,Jack Oh,Jerry Hsu,Tanvi Bajpai,Jialei Chen,Wenjun Gu,Chintan Shah,Chris Chan,Sayce Falk,Chris Jones,Rob Torres,Kat Ko,Lee Howes,Freya Azad,Andi Gavrilescu,Joe Pamer,Bill Jia
类目: Artificial Intelligence (cs.AI)
备注: 23 pages, 5 figures
Abstract:TorchTPU is an open-source collaboration between Google and Meta that introduces a native and principled PyTorch backend for Google TPUs. It enables seamless out-of-the-box model execution and TPU migration while delivering industry-competitive inference and training performance through eager execution optimizations, this http URL integration, and native hardware acceleration.
[AI-261] From Spectrograms to Shallow Speech SSL: Audio-Frontend Design for Generator-Disjoint and Low-Latency Audio–Visual Deepfake Detection
链接: https://arxiv.org/abs/2610.24138
作者: Qiwu Wen,Jahangir Alam,Eric Charton
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注:
Abstract:Video-mediated conversational systems rely on synchronized speech and facial video, making audio–visual input integrity important for reliable perception. We study audio-frontend design in a fixed MRDF (Modality-Regularization-based DeepFake)-based pipeline with AV-HuBERT visual processing and an MSCT (multi-scale cross-modal transformer) fusion encoder. We compare spectrogram-style log-filterbank features with frozen wav2vec~2.0 and WavLM representations, dual-SSL fusion, executed depth, and shallow-layer aggregation. Across three training runs, full wav2vec~2.0 improves mean macro AUROC from 84.19% to 87.19% on generator-disjoint MMDF. Six-layer execution further improves detection, and learned non-uniform aggregation adds 2.53 percentage points, reaching 90.61% . Shallow execution reduces measured inference latency relative to full-depth wav2vec~2.0 across the tested input durations; first-second evaluation reaches 86.85% macro AUROC. In a single-run comparison, dual-SSL concatenation outperforms either single stream, while one-way cross-attention yields lower detection performance and higher latency than concatenation. Layer-weight analysis shows greater emphasis on early speech layers, while removing the selector’s visual input has little effect on detection performance.
[AI-262] A Conditional No-Hair Theorem for Transformers: How Language Models Forget Their Prompt
链接: https://arxiv.org/abs/2610.24133
作者: Shuyang Xiang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We ask how much an autoregressive transformer retains of its prompt after generating t tokens. Consider two different prompts followed by the same continuation text, so that any difference in their hidden states comes from the prompt alone, and we find that the distance between their hidden states decays approximately as a power law t^-\alpha , where \alpha=1 is the rate at which a fixed prompt’s share of uniform attention shrinks and smaller values mean longer memory. Trained models forget fast ( \alpha\approx0.55 – 0.7 ), untrained ones slowly ( \alpha\approx0.2 – 0.5 ). We prove that this offset is O\big(P(\log t)^L-1/t\big) for a prompt of length P and depth L . The bound holds for finite-depth pre-LN softmax transformers with bounded attention logits, under one additional condition: the pre-norm activations stay away from zero. The mechanism is geometric: the prompt’s direct influence is diluted as the context grows, and \emphrelay — continuation tokens passing on what earlier tokens read from the prompt — can only add logarithmic factors. This is what we call the \emphconditional no-hair theorem: given the text generated so far, the prompt’s influence on the next token vanishes as t grows, so any property of the prompt that survives, such as its language, must be carried by the text itself. An exactly solvable linear relay model shows that the bound cannot be improved and reproduces the rise of the local exponent towards 1 seen in untrained networks. Cancelling the dilution with a logit bias stops the decay early in training and reverses it after, so dilution is necessary for the forgetting we measure; untrained networks and the solvable model show it is also sufficient.
[AI-263] On the Selectivity of Generative Models in Structure-Based Drug Design
链接: https://arxiv.org/abs/2610.24124
作者: Ella Miray Rajaonson,Jungyoon Lee,William Chau,Alan Aspuru-Guzik,Benjamin Sanchez-Lengeling,Dominique Beaini,Kirill Neklyudov,Marta Skreta
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted as a Presentation at the Neural Information Processing Systems 2026
Abstract:Designing small-molecule drugs that selectively bind to a target protein while avoiding unintended interactions with other proteins is essential for reducing side effects that contribute substantially to clinical attrition. Although structure-based drug design (SBDD) has advanced rapidly with deep learning, most generative models still optimize binding to a single target and disregard selectivity due to limited training data and unknown off-target interactions. Motivated by this gap, we introduce SELECTBENCH, a comprehensive benchmark assessing the selectivity of generative SBDD models. SELECTBENCH provides three curated evaluation tasks reflecting different real-world drug discovery scenarios: literature-based case studies, safety screening panels used in pharmaceutical development, and large-scale structure-based evaluation sets. We further introduce a suite of metrics to quantify selectivity and conduct a systematic evaluation of nine representative SBDD models, including methods explicitly trained for selectivity and widely-used inference-time guidance techniques. Our results show that even selectivity-trained models achieve only modest off-target discrimination and inference-time guidance provides limited gains, underscoring the need for method improvement. SELECTBENCH provides both a characterization of where current generative SBDD stands on selectivity and an open, extensible platform to accelerate progress. The benchmark is available at this https URL.
[AI-264] Graph-Regularized Synthetic Control for Stable Counterfactual Estimation under Donor Redundancy
链接: https://arxiv.org/abs/2610.24113
作者: Yanlin Zhang,Sungyong Chung,Alireza Talebpour
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Econometrics (econ.EM); Methodology (stat.ME); Machine Learning (stat.ML)
备注:
Abstract:Synthetic control methods estimate counterfactual outcomes by weighting untreated donor units to match a treated unit before treatment. In recent applications, however, the donor pool can be large, highly correlated, and short in pre-treatment history, yielding many near-equivalent solutions with unstable and difficult-to-interpret donor weights. We propose Graph-Regularized Synthetic Control (Graph-SC), a convex estimator that augments the classical pre-treatment matching objective with a graph Laplacian penalty. The graph encodes similarity among donor trajectories, encouraging coherent weights over structurally related donors while preserving the transparency of synthetic controls. We provide theoretical results characterizing the bias-variance tradeoff induced by graph regularization and establish efficient convergence guarantees with an accelerated proximal gradient method. Across a controlled simulation under donor redundancy, the California Proposition 99 benchmark, and observational counterfactual analyses on the Waymo Open Motion Dataset, Graph-SC improves weight stability, reduces sensitivity to redundant donor pools, and maintains competitive pre- and post-treatment accuracy relative to classical and regularized synthetic control baselines. Our results suggest that graph regularization is most valuable not as a universal accuracy booster, but as a principled mechanism for selecting stable and interpretable counterfactuals from near-equivalent synthetic controls with comparable pre-treatment fit.
[AI-265] Does AI-Assisted Development Homogenize Software? An Empirical Study of Code Similarity Quality and Evolution
链接: https://arxiv.org/abs/2610.24093
作者: Luciano Marchezan,Romain Lefeuvre,Kevin Delcourt,Benoit Combemale,Houari Sahraoui
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:AI-assisted software development is evolving from interactive LLM assistance toward agentic systems that autonomously perform software engineering tasks, yet the effects of this shift remain insufficiently understood. We present an empirical study of 50 long-lived Java and Python open-source projects across three eras: pre-LLM, potentially LLM-assisted, and detected agentic development. Using snapshots from Software Heritage, we examine code similarity, software quality, and software evolution. Similarity is measured semantically through embedding-based cosine similarity and syntactically through CodeBLEU*. Quality is assessed with SonarQube metrics covering maintainability, complexity, reliability, and security. Evolution is characterized by annualized growth rates of LOC, functions, classes, files, and statements, and by commit activity in the three months preceding each snapshot. Results show no consistent evidence of increased similarity as semantic and syntactic similarity decreased significantly in Java for both the LLM and agentic eras relative to the pre-LLM era, while Python showed no significant differences. Software quality remained largely stable, although the agentic era showed significantly lower code-smell and SQALE index densities. Structural growth was generally lower in the agentic era, particularly for files, whereas development activity increased moderately in Java, especially in commits and added lines. These findings challenge the assumption that LLM- and agent-assisted development necessarily homogenizes code, degrades quality, or accelerates growth.
[AI-266] Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing
链接: https://arxiv.org/abs/2610.24090
作者: Peilin Wu,Yi-Zhuang You
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Does persistent memory accumulated from previous attempts help an LLM agent solve new problems, and does memory design affect that benefit? We study programmatic Chinese-character drawing across five agent conditions attempting the same 668 items. Accumulated memory produces modest aggregate gains despite substantial growth in the memory stores: free-form notes improve first-attempt success by 4.6 percentage points, while a design combining free-form memory with reusable code improves it by 2.8 points. The two designs do not differ significantly overall, but the code bank shows a suggestive advantage on later, harder items. A separate post-hoc comparison finds a much larger improvement when the code-bank agent receives privileged, target-specific structural supervision. The agent can therefore exploit highly informative structural guidance, although the privileged per-item information does not establish what could have been learned from experience alone. These results suggest that memory design may affect where accumulated experience becomes useful, even when its average benefit is modest.
[AI-267] Client-Aware Security Adaptive Early Exit Vision Transformers for Heterogeneous Federated Learning
链接: https://arxiv.org/abs/2610.24036
作者: Sahar Refaei Ghalibafan,Samira Nazari,Christian Herglotz,Maksim Jenihhin,Mahdi Taheri
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR)
备注:
Abstract:Federated learning (FL) trains a shared model across data-owning clients without centralizing raw data, but client data difficulty and compute budgets can differ. We study local early-exit Vision Transformer (ViT) inference, where an intermediate classifier may return a prediction before the final Transformer stage. The proposed method replaces one global acceptance threshold with a client-specific base threshold from label entropy and shallow-exit confidence, then adjusts each input using predictive entropy, confidence, top-two class margin, and exit depth. Aggregation uses only clients that updated a parameter. Security-aware routing reaches 88.70% CIFAR-10 accuracy with 32.0% depth saving and 73.7% ImageNet-1K top-1 accuracy with 30% saving. On matched six-client non-IID CIFAR-100, full-depth accuracy is 72.28% versus 68.20–71.80% for four heterogeneous-FL baselines; robustness-aware routing reaches 70.01% with 32.55% saving.
[AI-268] Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis
链接: https://arxiv.org/abs/2610.24024
作者: Jian Yang,Peiling Liu,Songda Li,Yu Cao,Yuqing Feng,Lan Yu,Jiarong Pan,Jing Ye,Xinwen Huang,Chaochun Zou,Xudong Lu,Huilong Duan,Haomin Li
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 26 pages, 5 figures
Abstract:Rare diseases (RDs) affect an estimated 300 million people globally, leaving patients to navigate a protracted diagnostic odyssey constrained by clinical scarcity and fragmented medical knowledge. Artificial intelligence (AI) has progressively advanced RD diagnostics, from symbolic methods and deep learning to large language model (LLM)-based systems. However, each paradigm carries structural limitations: symbolic methods are bounded by knowledge completeness, neural models sacrifice interpretability, and LLM-based systems are vulnerable to evaluation biases like data leakage. In this study, we introduce FusionRare, a unified multiparadigm fusion framework integrating knowledge-driven symbolic reasoning, deep-learning pattern recognition, case-based patient retrieval, and LLM-based heuristic inference via statistical meta-aggregation and knowledge-enhanced agentic reflection. Evaluated on 2,528 strictly isolated clinical cases, FusionRare achieves state-of-the-art RD prioritization, attaining a macro-averaged Recall @1 of 51% for phenotype-only diagnosis and 79% for phenotype-genotype joint analysis. These results represent decisive advantages over the advanced agentic system DeepRare, frontier LLMs including GPT 5.5, and specialized tools like PhenoBrain and LIRICAL under stringent anti-leakage protocols. A randomized, crossover clinician-AI validation demonstrated that FusionRare elevates the true diagnostic rank in 45% of decisions, improves physician confidence in 62%, and reduces time to diagnosis by 21%. These findings indicate that reconciling these orthogonal AI paradigms significantly enhances diagnostic capabilities, bridging the gap between computational benchmarks and actionable clinical reliability. More broadly, this work provides a replicable framework for deploying trustworthy diagnostic AI across complex, high-stakes, and underserved medical domains.
[AI-269] WIN-MoE: Two-stage Intra-Expert Imputation Network for Incomplete Multimodal Learning
链接: https://arxiv.org/abs/2610.24014
作者: Liang Nathan Zheng,Jiyuan Liu,Wenhao Liang,Wei Emma Zhang,Weitong Chen,Lin Yue
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent Mixture-of-Experts (MoE) approaches address missing modalities either by routing each missingness pattern to a dedicated expert, which scales poorly as the number of modalities grows, or by imputing the missing modality from representations contributed by multiple experts, which requires a costly retrieval mechanism. We show that routing induces expert-conditioned representation neighborhoods and derive an imputation-error bound controlled by their local feature dispersion. When these expert-conditioned neighborhoods are more compact than the global representation space, intra-expert aggregation admits a tighter worst-case reconstruction bound than unconstrained global aggregation. Guided by this analysis, we propose TWIN, a two-stage intra-expert imputation framework. Pre-imputation generates a context-conditioned estimate in the encoder embedding space and radially soft-clips it using modality-specific statistics maintained in an observed FIFO queue, this ensures that the pre-imputed modality stays in empirically calibrated region of the target modality space. Post-imputation refines the missing representation after MoE by aggregating expert-conditioned observed representations within the same sample. Across four missing-modality benchmarks under natural, random, and asymmetric dropout protocols, TWIN achieves state-of-the-art performance while using substantially fewer parameters than competing MoE-based imputation methods.
[AI-270] What Social Media Teaches Us About Accumulative AI Risks
链接: https://arxiv.org/abs/2610.24006
作者: Dawei Li,Bohan Jiang,Huan Liu
类目: Artificial Intelligence (cs.AI)
备注: Camera-ready version for CogMI 2026 vision track
Abstract:Prominent AI safety discussions emphasize existential threats and catastrophic events, such as AI-enabled nuclear conflict or engineered biological outbreaks. This position paper argues that there is another type of AI risk that arises and accumulates from the widespread use of AI. Drawing on lessons from social media, we argue that safety assessments must also address accumulative harms that develop through repeated use, recurring decisions, and feedback across interconnected deployments. Repeated reassurance, selective exposure, recurring decision biases, and growing personal profiles may gradually narrow perspectives, weaken independent judgment, or enable exploitative influence without a single clearly harmful event. We advance the Trajectory Neglect Thesis: a focus on vivid catastrophic events and identifiable incidents risks overlooking less visible harms that accumulate through the widespread daily use of AI. Social media research motivates this position by showing how reciprocal media effects, social reinforcement, and individual susceptibility shape outcomes beyond individual encounters. We examine how related mechanisms may operate in AI systems, using conversational AI as a concrete setting where generation, semantic inference, and persistent memory may tighten feedback loops. Sycophancy, manipulation, and privacy loss provide a case study of how these risks may interact and accumulate. This comparison identifies mechanisms to investigate rather than assuming that repeated AI use is harmful. We therefore call for evaluation of cumulative patterns of use and deployment, individual and societal outcomes, careful causal inference, and proportionate safeguards that preserve useful assistance and informed user choice.
[AI-271] Agent ic Interpretation: Harness-Governed Evidence Construction for Software Agents
链接: https://arxiv.org/abs/2610.23996
作者: Yifeng He,Yanjun Wang,Siwei Cui,Aritra Sengupta,Murali Krishna Ramanathan
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 22 pages, 5 figures, 4 tables
Abstract:Software agents answer behavioral questions about code, such as whether a patch passes tests or which lines caused a failure. These tasks require repository-wide evidence, yet most scaffolds leave intermediate findings in free-form trajectories, making them hard to reuse and easy to disconnect from final answers. We present Agentic Interpretation (AgInt), a framework for harness-governed evidence construction. The model proposes typed records, while a deterministic harness admits, grades, and persists them, reports unresolved requirements, and licenses answers under task-specific rules. The resulting certificate gives the model persistent external state and leaves an auditable decision record. Formally, append-only certificate states form a grounding lattice ordered by record inclusion, where each admitted record extends the state by union. We evaluate AgInt with four models on SWE-bench-Live and DeepSWE for patch acceptance and Defects4J and GitBug-Java for fault localization. AgInt outperforms the plain agent and Agentic Code Reasoning on all reported patch-acceptance metrics and ranks responsible lines higher while inspecting fewer irrelevant lines in fault localization. For GLM-5, reject-class F1 rises from 0.256 to 0.543 on SWE-bench-Live, All@1 rises from 0.140 to 0.400 on Defects4J, and wasted inspected lines fall from 31.5 to 11.7 on GitBug-Java. A matched-template control separates the mechanism: the persistent scaffold drives most fault-localization gains, while the certificate body and gate govern patch-acceptance calibration. With execution licensed, AgInt produces harness-reproduced witnesses for 293 of 475 null-dereference candidates versus 253 for a matched shell agent, and 62 of 117 resource-leak candidates versus 58. Across reported efficiency measurements, it uses 35% fewer executions and reduces cost per candidate by 59%. Comments: 22 pages, 5 figures, 4 tables Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.23996 [cs.SE] (or arXiv:2610.23996v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.23996 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-272] IWOO: Wave Optics Optimizer Improved with Hybrid Strategies for Numerical Optimization and Engineering Applications
链接: https://arxiv.org/abs/2610.23994
作者: Rui Du,Yaming Zhang,Yongpeng Wen,Linghong Cai,Nanqing Liu
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)
备注:
Abstract:The Wave Optics Optimizer (WOO) incorporates optical diffraction and interference into numerical optimization and has shown competitive performance in benchmarks and engineering applications. However, it can converge slowly, become trapped in local optima, and lose search effectiveness in later iterations. This paper proposes an Improved Wave Optics Optimizer (IWOO) that integrates four complementary strategies. First, a good point set and opposition-based learning are combined in GPS–OBL initialization to improve population quality and diversity. Second, an E-DM nonlinear perturbation mechanism is embedded in temporary triangle-based search to develop the TriEDM multi-path strategy and enhance search capability. Third, Correlation-Guided Groupwise Orthogonal Learning (CGOL) leverages relationships among decision variables to improve search efficiency while reducing computational overhead. Fourth, an adaptive stagnation-aware Runge–Kutta correction is introduced to mitigate premature convergence. IWOO is evaluated on the IEEE CEC2019, CEC2022, and CEC2017 benchmark suites against representative and high-performing optimizers, with ablation studies assessing the contribution of each strategy. Within the representative algorithm group, IWOO achieves Friedman ranks of 1.8000, 1.5833, and 1.4483 on CEC2019, CEC2022, and CEC2017, respectively. Tests on six constrained engineering optimization problems further demonstrate its competitive accuracy, stability, and practical applicability.
[AI-273] REAP: Learning General Rearrangement Heuristics from Motion-Planner Feedback
链接: https://arxiv.org/abs/2610.23973
作者: Daniel Swoboda,Hector Geffner
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Conference on Robot Learning 2026
Abstract:Robotic rearrangement tasks require planning sequences of actions that fulfill high-level specifications while also being geometrically feasible. Symbolic abstractions make the combinatorial structure of rearrangement tasks tractable, but they usually do not fully capture the geometric constraints that determine which actions a robot can actually realize. These constraints are hard to express in the symbolic model. We propose REAP (Rearrangement with Elicited Action Preconditions), a framework in which a relational graph neural network heuristic is learned over a simple symbolic abstraction. A motion planner serves as a supervisor that elicits the hidden geometric preconditions of the problem during training. Deployed with greedy best-first search, the learned heuristic generalizes out-of-distribution along object count, plan length, and object arrangement, and can be trained jointly on several domains. Compared to classical heuristic search and learning-based baselines, REAP solves more problems, requires roughly an order of magnitude fewer motion-planner queries than comparable methods, and the action it ranks first is geometrically feasible in 94% of sampled states, against 22% for a random choice.
[AI-274] A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences CEC
链接: https://arxiv.org/abs/2610.23962
作者: Hanqing Dou,Zijia Cao
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted for presentation at the International Conference on Electrical and Computer Engineering Researches (ICECER 2026), December 3-5, 2026
Abstract:We propose a controlled Large Language Model (LLM) workflow for explaining risk score differences using similar historical transactions. The workflow combines reference retrieval and contrastive SHapley Additive exPlanations (SHAP) evidence with reference-based LLM reasoning and deterministic validation. Experiments on a test dataset show that the workflow produces more consistent, evidence-grounded and traceable diagnoses than the same LLM without structured workflow guidance.
[AI-275] ReTTT: Reflection-Driven Test-Time Training via Experience-Based Parametric Memory
链接: https://arxiv.org/abs/2610.23942
作者: Kangcong Li,Hangfan Zhang,Shuyue Hu,Shengji Tang,Tao Chen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents accumulate valuable task-specific information through their actions, observations, and environmental feedback during multi-turn task execution. When effectively used, this information can help agents refine subsequent decisions and improve task performance. However, the learning value of an interaction depends on its context and subsequent feedback, making it challenging to extract reliable, actionable lessons and revise them as the task unfolds. In this paper, we introduce ReTTT, which enables experience-driven self-adaptation through test-time training on lessons constructed and continually revised within an episode. ReTTT lets the agent reflect on interaction feedback, maintain the resulting lessons in an explicit memory, and learn from them by updating an episode-local LoRA adapter while keeping the backbone frozen. The adapted model then performs both subsequent actions and reflection, coupling policy adaptation with the construction of future learning experience. We evaluate ReTTT on ALFWorld, ScienceWorld, and SWE-bench Lite using two model families. ReTTT outperforms trajectory-based test-time training across all six model–benchmark settings, improving task success by up to 12.14 percentage points. A frozen-reflection comparison further shows a 7.14-point benefit from adapting the reflection policy, while trajectory analysis illustrates how adapted reflection derives more task-specific guidance from failed attempts. These findings suggest that learning to reflect on its own behavior can boost agents’ performance on long-horizon tasks.
[AI-276] A Group-Aware Comparative Study of Classical Machine Learning and Deep Learning Methods for Sensor-Based Human Activity Recognition
链接: https://arxiv.org/abs/2610.23941
作者: Wahyu Pebrianto
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
备注:
Abstract:Human Activity Recognition (HAR) using wearable or smartphone inertial sensors is commonly evaluated with both classical machine learning and deep learning methods. However, reported performance can be strongly affected by sensor modality, dataset characteristics, and data-splitting strategy. This study presents a controlled comparative evaluation of four classical machine learning models and four deep learning models on two public HAR datasets, UCI-HAR and KU-HAR. Three sensor modalities were considered: accelerometer (ACC), gyroscope (GYRO), and their combination (ACC+GYRO). Classical machine learning models used handcrafted statistical features, while deep learning models operated directly on raw sensor sequences. The primary evaluation employed five-fold StratifiedGroupKFold cross-validation, with subject- or participant-level grouping to prevent group overlap between training and test folds. Macro-F1 was used as the primary performance metric. The results show that deep learning did not consistently outperform classical machine learning. On UCI-HAR, the evaluated deep learning models achieved slightly higher macro-F1 values, whereas on KU-HAR, LightGBM achieved 0.857 compared with 0.828 for the highest-performing evaluated deep learning model. Regarding sensor modality, the combined ACC+GYRO configuration generally outperformed ACC or GYRO alone on both datasets. The evaluation protocol also had a substantial effect: for CNNBiLSTMAttn with ACC+GYRO, random splitting increased macro-F1 by 0.033 on UCI-HAR and 0.093 on KU-HAR compared with grouped evaluation. These findings highlight the importance of model family, sensor modality, dataset characteristics, and group-aware evaluation when assessing sensor-based HAR systems.
[AI-277] PeekKV: The First Content Token Already Knows Which Heads Matter
链接: https://arxiv.org/abs/2610.23939
作者: Minyoung Go,Yongjun Park
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The size of the KV cache grows linearly with context length, making it a major memory bottleneck in LLM serving. Motivated by recent progress in identifying the functional roles of attention heads, a growing body of work compresses the KV cache non-uniformly across heads. Existing methods either fix the important heads via offline profiling or select them using attention signals from the prefill stage. In both cases, however, accuracy fluctuates substantially across workloads. While investigating the cause of this instability, we find that the active heads shift between the prefill and decode stages even within a single input. Along the decode path, by contrast, the activehead composition remains stable across content tokens once format tokens are filtered out. Based on this observation, we propose PeekKV. PeekKV observes a single content token immediately after prefill - a one-token peek - to jointly obtain head-group scores and token-importance scores. Guided by these scores, PeekKV allocates KV cache budgets to head groups in proportion to their group scores and retains only the top-scoring tokens within each budget. PeekKV requires no offline profiling or training and thus applies directly to unseen workloads and new models. Across four public benchmarks, PeekKV achieves the highest accuracy in more than half of the per-model comparisons against state-of-the-art methods, and stays closest to the full-cache accuracy on RULER and LongBench. On Llama-3.1-8B-Instruct, PeekKV retains 95.7% of the full-cache accuracy at a 10% memory budget on RULER with 16K contexts and 95.9% at a 1.5% budget on LongBench.
[AI-278] SECURE-AI: Multi-Signal Uncertainty Quantification for Trustworthy Acoustic Cough Detection
链接: https://arxiv.org/abs/2610.23937
作者: Sejal Solkhan,Feiya Xiang,Mayur Sanap,Geet Khatri,Edgar Lobaton
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:Acoustic models for cough detection achieve high accuracy on curated benchmarks but fail silently in deployment, where speech, environmental noise, and audio unlike the training data induce confident misclassifications. In health monitoring, a confidently wrong prediction is worse than no answer at all. We present SECURE-AI (Self-aware Estimation of Cough Uncertainty for Reliable Evaluation of AI), a multi-signal uncertainty quantification framework that wraps a frozen MobileNetV3 cough classifier and estimates when its predictions should not be trusted, combining Monte Carlo Dropout, Temperature Scaling, and Mahalanobis distance in the embedding space. On a held-out split of 12,161 Coswara recordings, the probability-space methods are individually insufficient: Monte Carlo Dropout assigns lower uncertainty to incorrect than to correct predictions, and temperature scaling improves calibration but not discrimination. The strongest signal is embedding-space distance: Mahalanobis distance separates true positives from false positives (means 44.7 versus 63.4; Cohen’s d = 0.94 ; AUROC 0.86) and, among predicted non-coughs, false negatives from true negatives (AUROC 0.82), so the same signal can flag both false alarms and missed coughs. At a descriptive operating point selected from labeled test data, retaining 90 percent of true-positive detections removes about two-thirds of false positives and raises precision from 59.9 to roughly 80 percent. A rejection threshold tuned on one split does not transfer to another (validation F1 of 82 percent falls to 34 percent on test), so we report threshold-independent AUROC as the main measure. An error taxonomy shows that failures are systematic, with most false positives related to speech. The pipeline runs on Raspberry Pi hardware costing 30 to 90 US dollars. Probabilistic calibration alone is thus inadequate for reliable cough detection.
[AI-279] Beyond Prediction: A Component-Level Study of Concept Recovery in In-Context Learning
链接: https://arxiv.org/abs/2610.23936
作者: Xufeng Liu,Fang Wan,Yi Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:In-context learning (ICL) enables language models to make accurate predictions from demonstrations, but prediction success alone does not reveal the extent to which the underlying concept has been recovered. Existing evaluations typically assess recovery at the level of the whole concept, making it difficult to identify which components are recovered and to separate model limitations from ambiguity in the demonstrations. We introduce a component-level framework for concept recovery that decomposes concepts into their defining components and uses controlled structural interventions to evaluate both component-level and whole-concept recovery under identifiable demonstrations. Across four instruction-tuned LLMs and three composition forms, we find that models often achieve high prediction accuracy despite poor whole-concept recovery, with systematic over-specification of relevant attributes and level constraints. Individual structural components are generally insufficient for whole-concept recovery, whereas jointly specifying complementary components, particularly relevant attributes and their level constraints, substantially improves recovery. However, this improvement is fragile: a single plausible but unnecessary level constraint can substantially disrupt recovery. Finally, joint supervision of recovery and implicit prediction substantially improves recovery while preserving prediction performance, with the learned capability transferring across composition forms. Overall, our component-level analysis provides a more diagnostic view of what models recover from demonstrations than the existing holistic view.
[AI-280] Factorization-Guided Training and Inference for Diffusion-Drafted Speculative Decoding
链接: https://arxiv.org/abs/2610.23935
作者: Jiakun Fan,Min Xiao,Qiang Wang,Neiwen Ling,Zhichao Liu,Zhihao Jia,Dimitrios S. Nikolopoulos,Zijun Wei
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Speculative decoding accelerates large language model inference through parallel verification of tokens proposed by a drafter. Diffusion drafters further parallelize proposal generation, but greater proposal capacity or higher mean accepted tokens (MAT) need not yield higher throughput. Because verification accepts only a consecutive prefix, early rejections can render later correct predictions ineffective, limiting MAT gains, while even higher MAT may not improve throughput if per-round latency increases. We introduce DScout, which uses two complementary factorizations to guide diffusion drafter training and inference for higher throughput. For training, DScout factorizes position-wise loss into inclusion, weighting, and prediction loss. This guides the design and combination of mixed-horizon training, intra-block unmasking, and acceptance-weighted loss to align supervision with prefix acceptance and improve MAT. For inference, DScout factorizes throughput as the ratio of MAT to per-round latency, guiding target-conditioning depth, drafter depth, and inference block size selection to identify a high-throughput configuration for each workload. Across six benchmarks, DScout achieves 3.84\times and 4.14\times mean speedups over autoregressive decoding on Qwen3-4B and Qwen3-8B, respectively, improving mean speedup over the state-of-the-art baseline by 9.1% and 7.1%. Across models and concurrency settings, our inference configuration policy provides up to 9.75% additional mean speedup over DScout’s fixed configuration.
[AI-281] Organizing LLM Inference for Aggregate Predictions from Behavioral Data
链接: https://arxiv.org/abs/2610.23934
作者: Varun Gupta,Vijay Kamble
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Econometrics (econ.EM)
备注:
Abstract:When using a large language model (LLM) to simulate a user population, how should we combine individual behavioral histories to predict the fraction of users who take a particular action? We study this question in e-commerce, predicting observed product-level conversion rates from browsing sessions without task-specific model training. We vary two design choices, each involving a bias-variance tradeoff. Summarizing a product’s sessions before inference reduces cost but can discard predictive detail, adding bias, whereas predicting outcomes for individual sessions and averaging afterward preserves that detail but requires many calls to reduce variance. Evaluating products together in one prompt can help the model infer what is typical in the setting, reducing bias, but also makes each prediction depend on which other products appear alongside it, adding noise. We evaluate both choices across five LLMs and two datasets. Our main finding is that predictions from individual sessions outperform predictions from summaries when products are evaluated separately, but summaries lead when products are evaluated together. Comparative summaries achieve the highest Spearman correlation with observed rates in 12 of 15 settings, with no alternative significantly better. Comparison thus helps summaries far more than session predictions, overcoming the advantage of behavioral detail. We interpret this difference through a Bayesian model of the second tradeoff, which suggests that comparison helps more when the inputs in a prompt reveal more about differences across products, consistent with our results for four types of comparative context. Comparative summaries also incur a small fraction of the inference cost of the session methods. With the best model for each outcome, they attain 66-89% of the correlation of the best supervised benchmark trained on thousands of labeled products.
[AI-282] miniTE: Diagnosing and Bridging the Execution Gap in Native NVFP4 Reinforcement Learning
链接: https://arxiv.org/abs/2610.23928
作者: Zhen Li,Mingfa Feng,Yiming Zhang,Yang Yu,Changhai Zhou,Pengxiang Li,Dongkuan Wu,Congkai Xie,Hongxia Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Native NVFP4 promises to accelerate reinforcement learning (RL) for large language models (LLMs), but achieving stable learning across separate training and sampling engines remains challenging. We show that matching model weights and numerical formats is insufficient to ensure consistent policy evaluation, highlighting the importance of cross-engine numerical execution. We introduce miniTE, a lightweight framework that combines execution diagnostics, cross-engine numerical alignment, and efficient native W4A4 (4-bit weights and activations) GEMM computation for LLM RL. It improves numerical execution consistency without modifying the RL objective, while reducing implementation overhead to realize the benefits of native low-bit computation. Long-horizon RL experiments on Qwen3-30B-A3B demonstrate stable training and benchmark performance close to BF16, outperforming naive Transformer Engine (TE) NVFP4 in evaluation quality. On single NVIDIA B300 GPU node, the optimized learner reaches 33,540 tokens/s, yielding 1.39x and 1.47x the throughput of BF16 and TE NVFP4, respectively, for complete learner updates. We released the framework, kernels, diagnostics, and training configurations to make native low-bit RL easier.
[AI-283] Improving Fake Profile Detection Through Multimodal Learning and Research-Guided Feature Integration
链接: https://arxiv.org/abs/2610.23926
作者: Jean Claude Hirwa,Jamiu Dahunsi,Candide Giramata,Lowami Uwimana,Thierry Henry Ishimwe
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Fake accounts continue to undermine online platforms, yet it remains unclear whether multimodal learning offers meaningful improvements over strong metadata baselines or graph-based approaches. In this work, we study bot detection by integrating textual, visual, and account metadata signals on the TwiBot-22 benchmark using a subsample of 30,000 users. We propose a multimodal framework combining pretrained textual (XLM-RoBERTa) and visual (CLIP) embeddings with structured account-level features such as follower counts, posting frequency, and account age, and evaluate multiple modality combinations using CatBoost alongside a Relational Graph Convolutional Network (R-GCN). Results show that metadata alone is highly predictive, while multimodal fusion yields only marginal and largely statistically insignificant gains. The graph-based model underperforms classical variants under the subsampled graph conditions of this study. Overall, our findings suggest that under practical constraints, feature quality and metadata richness matter more than fusion complexity.
[AI-284] Zero-Shot Tabular Learning via Hypergraph Reasoning with LLM s EMNLP2026
链接: https://arxiv.org/abs/2610.23923
作者: Geon Lee,Kyungho Kim,Dongwon Choi,Juyeon Kim,Shinhwan Kang,Kijung Shin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: EMNLP 2026
Abstract:Tabular data are prevalent across diverse domains such as healthcare and finance. Zero-shot tabular learning aims to make predictions without labeled data. However, existing LLM-based approaches rely on row-wise prompting, resulting in high inference cost and limited use of inter-row dependencies. In this work, we propose TABHGR+, an efficient relation-aware zero-shot framework for tabular data that leverages LLM-generated rules to label and group rows. Specifically, based on column and task descriptions, we use LLMs to generate rules that assign weak labels to all rows. To refine these noisy weak labels, we construct a hypergraph by grouping semantically related rows using LLM-generated rules and perform hypergraph-based label propagation. Lastly, we further enhance the LLM-generated hypergraph for propagation, which is potentially sub-optimal, by assigning label-adaptive weights to hyperedges. By requiring only a constant number of LLM calls independent of dataset size, TABHGR+ substantially reduces inference cost while achieving higher classification accuracy than zero-shot and strong few-shot baselines.
[AI-285] Doing Less with More: A First-Principles Exploration of the Suitability of Agent ic Computing over Alternative Architectural Choices AAAI
链接: https://arxiv.org/abs/2610.23921
作者: Ritvik Garimella,Biplav Srivastava,Amit Sheth
类目: Artificial Intelligence (cs.AI)
备注: Accepted to AAAI Fall Symposium - Trustworthy Agentic Systems ( this https URL ) 8 pages, 3 figures
Abstract:There is growing interest in automating business activities with Agentic Artificial Intelligence (AI) due to latter’s seeming ease of use. Never has it been easier, or costlier, to do less with more. However, little is known about when agents are preferable to established alternatives such as local computation, Representational State Transfer (REST), the Simple Object Access Protocol (SOAP), and the Model Context Protocol (MCP), particularly when development speed, performance, and operational cost are considered. We investigate this question using a controlled mathematical task that compares seven methods on a benchmark of 1,000 arithmetic expressions where semantics of operator precedence has to be preserved. We ran this setup across a native Function Calling baseline (Python library), REST, SOAP and MCP microservice architectures, direct responses from large language models (LLMs) using a Groq cloud API and five locally deployed Ollama models, and agentic systems in which five local LLMs invoke MCP tools. The deterministic methods achieve 100% accuracy with negligible latency. Direct LLM evaluation achieves at most 92.9% accuracy at over 22,000x the latency of a native function call. With MCP tools, the best agent reaches 99.9% accuracy, but incurs nearly 1,000,000x the baseline latency. These results demonstrate that LLMs approaches are poorly suited for exact arithmetic evaluation and motivate an informed architectural decision framework (ADF) for choosing an architecture based on accuracy and throughput requirements. We recognize that an actual interaction with an agent may span both reasoning and linguistic tasks. Hence our recommendation is to explore multiple architectural choices (based on task type) rather than one-size-fits-all solution (LLMs). Code is available at - this https URL Comments: Accepted to AAAI Fall Symposium - Trustworthy Agentic Systems (this https URL) 8 pages, 3 figures Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.23921 [cs.AI] (or arXiv:2610.23921v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.23921 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-286] UMI-WM: Interactive World Modeling for Bimanual Manipulation from Wrist Cameras
链接: https://arxiv.org/abs/2610.23913
作者: Jingxiang Guo,Haoyu Zhao,Xidong Zhang,Yuhang Zheng,Hao Wu,Chen Gao,Jianshu Zhou,Shuicheng Yan,Weihao Yuan
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 26 pages, 13 figures, including appendix
Abstract:Universal Manipulation Interface (UMI) pairs wrist-camera videos with camera trajectories and gripper openings, providing scalable, robot-free data for learning how bimanual actions change a shared scene. Existing interactive world models, however, typically rely on text or single-camera control and struggle to follow each arm’s actions while maintaining consistency between the two wrist views. We introduce UMI-WM, an action-conditioned bimanual world model that jointly generates both wrist views from observation history, per-arm relative camera poses, and gripper widths, with cross-view consistency, action following, and visual fidelity. UMI-WM injects each arm’s actions into its own wrist stream, while sparse hub tokens exchange scene information between the two streams. Camera-aware geometry conditioning and camera-guided noise further support per-view motion control and cross-view geometric exchange. We adapt the model for block-causal generation and distill the 30-step Teacher into a Student with four denoising steps. We train UMI-WM on self-collected open-world and in-studio UMI recordings combined with public manipulation data. To evaluate action following, we use an inverse dynamics model (IDM) to recover relative camera poses and gripper widths from generated videos, and introduce Bimanual Action Fidelity (BAF) to measure how closely these recovered actions match the conditioning actions. UMI-WM achieves the best results on all five reference-video fidelity metrics and the best or competitive performance on VBench and WorldArena. The Teacher and Student outperform all compared baselines in action following, while the four-step Student generates paired wrist videos 6.9 times faster than the Teacher.
[AI-287] Web Agent Agent ic Reinforcement Learning Decision Model Under Multi-Cost and Failure Risk
链接: https://arxiv.org/abs/2610.23857
作者: Qianli Ma,Limengxi Yue,Shuyang Xu,Yanpei Shi,Hongrui Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted for presentation at the 5th International Conference on Big Data, Information and Computer Network (BDICN 2026); proceedings published by ACM ICPS
Abstract:Intelligent agent interactions in real-world web environments are commonly constrained by request budgets, time delays, anti-crawling restrictions, and operational failure risks. Strategies solely optimizing task success rates often exhibit unusable phenomena such as “high success but high cost” or “low risk but conservative failure.” This paper proposes a constrained agentic reinforcement learning model for web agents, unifying page access, search requests, and external API calls into a unified long-term decision-making framework with associated costs. It simultaneously incorporates cost budget constraints and tail risk control into the optimization objective: constructing a multidimensional cost vector comprising cumulative request count, total latency, and failure penalties to achieve budget compliance via Lagrange dual updates, while employing a CVaR risk term to suppress excessive exploration of high-failure-probability paths, thereby achieving an adaptive balance among “completion rate, cost, and risk.” Experiments were conducted across 30-70 site/page templates and 800-1,500 end-to-end web tasks, including information extraction, price comparison, form submission, and cross-page navigation. Interaction sequences spanned 20-120 steps with tool scales of 30-200. Performance was benchmarked against unconstrained RL, budget-constrained RL, and rule-based/scripted web agents, quantifying task completion rates, cost-per-success, failure rates, and policy stability. Results demonstrate that at equivalent completion rates, our method reduces C-PS by 22%-31% and lowers failure rates by 18%-26% under high failure penalties. Under fixed budgets, task completion rates increase by 10%-16%, highlighting the necessity and effectiveness of constraint modeling for practical web agent deployment. Comments: Accepted for presentation at the 5th International Conference on Big Data, Information and Computer Network (BDICN 2026); proceedings published by ACM ICPS Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.23857 [cs.LG] (or arXiv:2610.23857v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.23857 Focus to learn more arXiv-issued DOI via DataCite Journalreference: Proceedings of the 5th International Conference on Big Data, Information and Computer Network (BDICN 2026), ACM International Conference Proceedings Series (ICPS), 2026
[AI-288] LLM -Assisted Bi-Approximate CHC Solving
链接: https://arxiv.org/abs/2610.23771
作者: Baiting Jiang
类目: Logic in Computer Science (cs.LO); Artificial Intelligence (cs.AI)
备注: 22 pages, 8 figures, 5 tables
Abstract:Constrained Horn clause (CHC) solvers often struggle to discover deep counterexamples whose derivations may require extensive rule unfolding. Although large language models (LLMs) can propose long-range symbolic relations, prior work on formally checked LLM-assisted verification has mainly focused on generating inductive invariants and concrete traces. We exploit this broader capability through Derivation Power Under-approximations (DPUs). Each DPU is a solver-certified under-approximation of the relation induced by a multi-step CHC derivation. An LLM proposes each DPU with symbolic intermediate-state witnesses that enable SMT validation without introducing additional quantifiers. Validated DPUs can be composed and reused soundly. Building on this mechanism, our LLM-assisted framework HornCraft combines invariant generation for over-approximation with DPU generation for under-approximation in CHC solving. On 260 linear and nonlinear integer-arithmetic CHC-COMP benchmarks left unresolved by a 30-minute Spacer run, HornCraft returns certified verdicts for 228 instances. It certifies 42 instances as unsafe, compared with 22 when DPU construction and reuse are disabled.
[AI-289] Anytime Contracts for AI Services
链接: https://arxiv.org/abs/2610.22781
作者: Boris Velasevic,Eden Saig,Eric Mazumdar
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 19 pages, 3 figures
Abstract:AI services are prone to moral hazard, as model providers can silently route requests to cheaper, lower-quality models, and the inherent stochasticity of model outputs makes this degradation hard to detect. Contract theory addresses moral hazard by designing payment rules that reward high-quality outputs and prevent strategic overcharging. However, existing frameworks assume interactions of fixed duration, whereas AI services are typically consumed through subscriptions and pay-as-you-go APIs that users can terminate at any time based on observed performance. We show that this flexibility can break standard contracts, opening them up to abuse by either the providers or users. To address this, we extend contract theory to account for unknown exit rules. We formalize the problem as a Principal-Agent game, and show that budget-optimal contracts in this setting are proportional to anytime-valid hypothesis tests based on likelihood-ratio e-values. We give an efficient algorithm for computing anytime contracts and show how betting e-processes can be used to approximate them when the models are unknown. Finally, we evaluate our findings on synthetic data and data from deployed LLMs.
[AI-290] Physics-anchored residual quantum neural networks for cloud-cover parameterization
链接: https://arxiv.org/abs/2610.26279
作者: Zhuo Wang,Ce Yu,Bin Yang
类目: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Compact quantum neural networks have been explored for cloud-cover parameterization, but their predictive value depends on how physical knowledge is incorporated and which classical models provide the reference. We investigate an offline residual formulation that retains a fitted \XR\ cloud-cover estimate and learns a correction to its logit in transformed target space. Using released ICON/DYAMOND-derived data, we compare ten model variants in six- and eight-input settings, with three matched replicates and 10,000 training samples per replicate. All quantum circuits use local Z readout and noiseless exact expectation values. Under a common 30-epoch protocol, residual prediction reduced physical-space test mean squared error relative to direct QNN prediction by 54.3% and 38.8% for six-input XYZ and ZZXY circuits, and by 49.2% and 38.9% for their eight-input versions. Residual QNNs also outperformed the selected compact residual multilayer perceptron in every paired replicate. A larger, longer-trained Grundner-Q1 adaptation nevertheless achieved lower error: residual Q1 reached mean MSEs of 0.00778 and 0.00791, compared with 0.01124 and 0.01082 for residual XYZ. The results support residual prediction for compact simulated QNNs within the tested budget and establish a clear boundary set by stronger classical references. They do not establish computational quantum advantage or online climate-model skill.
[AI-291] AirMPA: A Meteorology-to-Pollution Adapter for Global Air Quality Forecasting
链接: https://arxiv.org/abs/2610.25797
作者: Yiheng Wang,Kai Zheng,Yuetan Lin,Fanglu Fan,Chenliang Tao,Guochao Chen,Hongliang Zhang,Hao Li
类目: Atmospheric and Oceanic Physics (physics.ao-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Accurate real-time forecasting of atmospheric pollutants is essential for reducing exposure risks and improving public health. Traditional numerical forecasting systems suffer from incomplete parameterizations, uncertain inputs, and high computational cost. Existing deep learning approaches improve efficiency, but often entangle meteorological and pollutant variables within unified forecasting frameworks, limiting flexibility and pollutant-specific representation learning. To address these challenges, we propose AirMPA, a decoupled autoregressive framework for meteorology-to-pollution forecasting. AirMPA predicts 13 atmospheric pollutants at 0.4° spatial resolution and 12-h temporal resolution using meteorological fields, emission priors, and historical pollutant concentrations. The model is built on an improved Swin-Transformer architecture with zonal release attention and dual-branch fusion, enabling more effective modeling of global transport continuity and heterogeneous pollutant dynamics. During training, AirMPA uses the fifth-generation atmospheric reanalysis from the European Centre for Medium-Range Weather Forecasts (ERA5) as meteorological forcing, while during inference it directly ingests forecast fields from external meteorological models, enabling stable autoregressive prediction up to 5 days ahead. Experiments show that AirMPA consistently outperforms Aurora and surpasses the Copernicus Atmosphere Monitoring Service (CAMS) in approximately 90% of pollutant-lead-time evaluation cases over the 5-day forecast horizon, while outperforming CAMS on all variables beyond 48 h. These results demonstrate the potential of decoupled meteorology-to-pollution forecasting for global air quality prediction.
[AI-292] Corrected Block Sparse Spectral Optimization: Exact Models Finite Identification and Local Dynamics
链接: https://arxiv.org/abs/2610.25721
作者: Omid Shams Solari
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Optimization and Control (math.OC)
备注:
Abstract:The block generalized-power method pairs threshold-based support selection with inexpensive Stiefel updates, but the standard block formulation conflates two nonequivalent objectives and leaves inactive components, threshold ties, and post-selection refitting ill-specified. We rebuild the model from the primal. Eliminating each loading column from the normalized loading-ball problem gives a sum of soft-thresholded Euclidean norms (A-SUM); the squared-threshold objective in common use arises from a quadratic lift (A-SQ). We prove both identities, exhibit a two-component instance whose maximizers disagree, and specify explicit conventions for zero-response columns, ties, rank deficiency, and mask-before-refit. For A-SQ we separate the historical full-polar iteration from a safeguarded Riemannian ascent and prove, for the latter, whole-sequence convergence via a sufficient-ascent lemma, a relative-error lemma, and the Kurdyka-Lojasiewicz property, which applies because the soft-threshold objective and the Stiefel constraint are semi-algebraic. A strict threshold margin then forces finite identification of the support and sign pattern. The post-identification map is ordinary power iteration on the active covariance for a rank-one hard threshold, and the Frechet derivative of a coupled polar map for component-specific supports; a radius-based safe-screening rule follows from the same local analysis. Planted-spike experiments match the predicted identification time and local rates. On modern-AI panels (GPT-2 residual stream, ViT CLS stream, Pitprops) the thresholded block-sparse components match dense PCA on reconstruction and improve on it in monosemanticity and class specialisation. On three native-label scRNA datasets (Zheng17 FACS-sorted PBMCs, 68k donor-A with FACS-correlated labels, Tabula Muris) A-SQ beats dense PCA on cluster-recovery AMI by +0.023 to +0.044 at matched sparsity.
[AI-293] pidoptera: A Domain-Agnostic Engine for Causal Simulation with Mixed Variable Types Vine Copula Dependence and Individual-Level Counterfactual Propagation
链接: https://arxiv.org/abs/2610.24147
作者: Danilo Augusto Sarti
类目: Methodology (stat.ME); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Computation (stat.CO)
备注: R and Python package available at this https URL
Abstract:lepidoptera is an open-source R and Python package for simulation from structural causal models with heterogeneous variable types. The package is built around a typed mechanism architecture, which separates the causal graph from the generating rule at each node and exposes each rule as an inspectable, serialisable object. A pre-built mechanism library covers all nine combinations of continuous, categorical, and ordinal parent and child types, removing the need for users to implement type-specific sampling from scratch. A vine copula mechanism estimates the full joint dependence structure from observed data using pair-copula constructions and the Rosenblatt conditional-simulation algorithm, supporting over twenty copula families. Individual-level counterfactual queries are supported through partial propagation, which freezes the random draws of unaffected nodes and replays only the downstream consequences of an intervention, thereby separating the individual treatment effect (ITE) estimand from the population-level average treatment effect (ATE) without additional model assumptions. An extension layer allows any pre-fitted predictive model to serve as a generative mechanism via a unified callable interface. Three worked illustrations spanning clinical trial planning, emergency-ward capacity analysis, and epidemiological mediation analysis demonstrate the package’s scope.
[AI-294] Analysis of Binary O-type Star Systems Using Recurrent Neural Networks
链接: https://arxiv.org/abs/2610.24015
作者: Miguel Flores R.,Luis J. Corral,Celia R. Fierro-Santillán,Silvana G. Navarro
类目: Instrumentation and Methods for Astrophysics (astro-ph.IM); Solar and Stellar Astrophysics (astro-ph.SR); Artificial Intelligence (cs.AI)
备注: 16 pages, 16 figures, 34 images
Abstract:This work presents an initial analysis of the estimation of effective temperature and surface gravity for simulated O-type binary systems using a recurrent neural network. The proposed system uses 34 equivalent-width measurements from the optical region of a stellar spectrum. Two methods were applied to observed IACOB spectra before normalization: the addition of a blackbody continuum and the addition of a second observed spectrum. The first method produces only small differences in the normalized line profiles and in the estimated physical parameters over 3700-7000~Å. In the second method, the secondary spectrum is added with contributions of 5, 10, 20, 30, 50, and 75 per cent, as defined relative to the original primary spectrum. The general comparison includes 192 simulated systems and 1152 modified spectra. For most of these spectra, the estimated effective temperature and surface gravity decrease as the contribution of the secondary component increases. Individual examples using HD14947, HD190429, HD48099, and HD206267 show that this behavior is not the same for every combination: some estimated effective temperatures fall below the estimates for both isolated spectra, and some surface gravity estimates increase. Comparisons with previously published values also depend on the physical parameter and reference considered. The results show that the proposed system is sensitive to the addition of a secondary spectrum, but do not establish a unique contribution for the secondary component of an observed binary.
[AI-295] Can the Quantum Fisher Effective Dimension Judge a Hybrid Quantum Neural Network?
链接: https://arxiv.org/abs/2610.14070
作者: Md Masum Mia,Amith Khandakar,Mohamed Arselene Ayari
类目: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI)
备注: 20 pages, 10 figures, 9 tables
Abstract:Hybrid quantum neural networks (QNNs) pair a classical feature extractor with a variational quantum circuit, and the effective dimension of the circuit’s quantum Fisher information matrix (QFIM) is widely used to assess such models. This scalar is often interpreted as an indicator of trainability or accuracy, although these readings are rarely tested under controlled conditions. We therefore investigate what it actually measures. We define the effective dimension as the participation ratio of the QFIM diagonal, which lies between one and the parameter count, and we evaluate it across two image datasets, four frozen backbones including a random-projection control, four circuit depths on eight qubits, and three seeds. During training, the effective dimension remains well below the parameter count, and its correlation with test accuracy ranges from -0.72 to +0.87, changing sign across seeds in half of the configurations. Increasing depth activates a larger share of the circuit directions, from 0.38 to 0.91 of the parameter count, while the initialization gradient variance remains between 1e-5 and 1e-4, so no barren plateau (BP) appears at eight qubits. At a fixed circuit, the four backbones share an initial effective dimension of 6 to 9, even though CIFAR-10 accuracy ranges from 23.14% to 86.44%. These results show that the effective dimension characterizes circuit geometry rather than model quality. We conclude with a practical protocol that assigns each diagnostic to the question it can answer.
机器学习
[LG-0] Scaling Parameter-Efficient Fine-Tuning for Brain Encoding
链接: https://arxiv.org/abs/2610.26802
作者: Zhuofan Josh Ying,Guanglong Sun,Nolan Ho,Kieran Liu,Rui Yang,Andrew Chen,Nikolaus Kriegeskorte,Pony Ma,Hossein Adeli
类目: Machine Learning (cs.LG)
*备注: 38 pages, 13 figures, 20 tables. Code available at this https URL
Abstract:Brain encoding models aim to predict neural responses to stimuli. The most widely used approach fits readouts from frozen pretrained backbones. Using neural data to train large models from scratch or fully fine-tune them after task-pretraining has proven difficult because neural data is costly and limited. We show that parameter-efficient fine-tuning (PEFT) enables us to effectively adapt very large pretrained backbones for brain encoding with a small number of trainable parameters. To make it computationally practical, we exploit the backbone’s causal structure to predict multiple timepoints in a single pass instead of using sliding windows, significantly reducing feature-extraction cost. This enables efficient fine-tuning of multimodal backbones with up to 104B parameters, the largest model adapted for brain encoding to our knowledge. Trained with a low-rank adapter (LoRA) and a lightweight readout, our approach consistently improves brain encoding performance across nine diverse multimodal backbones. On the same Qwen3-Omni backbone as the prior state-of-the-art (SoTA), our method BrainLoRA achieves the best published performance on the Algonauts 2025 Challenge. We also show the adapted multimodal backbones remain interpretable: modality attribution recovers the canonical sensory topography that prior work often obtains by construction from separate unimodal encoders, and neural adaptation increases the contribution of audio inputs and cross-modal mixing within the backbone. To scale out the number of adapted models, we develop a multi-LoRA kernel that accelerates joint training of multiple LoRAs on a shared backbone. This enables an ensemble of 200+ adapted models that ranks first on the public Algonauts post-challenge leaderboard at the time of writing. Together, our results highlight PEFT as an efficient and powerful paradigm for brain encoding.
[LG-1] Learning to Refer: Inference-Time Collaboration in Decentralized Federated Learning
链接: https://arxiv.org/abs/2610.26800
作者: Yuheng Ma,Chendi Wang,Qiong Zhang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:In decentralized federated learning, heterogeneous clients may develop complementary expertise, yet each query is typically evaluated only by the client where it arrives. To exploit this distributed expertise without global coordination, we formulate inference-time collaboration as budgeted local expert referral, a graph-constrained routing process in which a query can move only between neighboring clients and only visited clients contribute to its final prediction. EdgeDRM implements this framework by learning a density-based routing potential for each client: differences between neighboring potentials provide the comparisons used to select successive referrals, while local task models generate predictions that are aggregated along the resulting route. Both training and inference are decentralized, using only neighbor-local information without a central router or graph-wide client scoring. We establish estimation guarantees for the learned comparisons and characterize how local improvement and graph structure govern prior-weighted density gains from referral. Experiments on benchmark and real-world medical data show that EdgeDRM outperforms graph-matched decentralized baselines in most settings, with the clearest benefits under sparse connectivity.
[LG-2] wo hidden layer suffice: the caterpillar net
链接: https://arxiv.org/abs/2610.26795
作者: Jakob Paul Zimmermann
类目: Machine Learning (cs.LG)
*备注: 17 pages. Lean formalization and paper sources: this https URL
Abstract:Every continuous piecewise linear (CPWL) function is a finite signed sum of maxima of affine functions, making maximum representation a central problem in the study of the expressiveness of ReLU networks. A now-disproved conjecture predicted logarithmic depth for maxima. Inspired by the proof by Brunck et al. (2026) that two hidden layers suffice for general CPWL functions, we give the Caterpillar Net, a nonrecursive construction for the maximum of N inputs with two hidden layers of widths O(N^2) and 2^O(N\log N) . The construction has a direct geometric interpretation.
[LG-3] On Data Scaling in Tabular ML: More Data Better Generalization?
链接: https://arxiv.org/abs/2610.26794
作者: Ivan Rubachev,Nikita Lazarev,Artem Babenko
类目: Machine Learning (cs.LG)
*备注: Code at: this https URL
Abstract:Recent work in tabular machine learning has emphasized the importance of diverse and realistic evaluation settings. The niche of large-scale (1M+) data, however, remains under-explored. To study training-data scaling, we assemble eight real-world tabular datasets with up to around ten million training examples, retaining the natural distribution shifts of the tasks and using random-split evaluation as a control. We find that GBDTs and MLP-based models show broadly similar scaling trends, with gains from both additional training data and algorithmic improvements. Tabular foundation models show more varied scaling behavior, with recent versions benefiting more consistently from larger contexts even under distribution shift. However, scaling training data under distribution shift introduces new challenges: in-distribution and out-of-distribution performance mostly improve together at smaller scales but can diverge at larger scales. We examine this through MLP capacity sweeps and comparisons of the validation distributions used for tuning and early stopping in MLPs and GBDTs. We also compare common practices of using validation data, contrasting its use for model selection in MLP pipelines with using it for in-context fitting in tabular foundation model pipelines. Under stronger shifts, MLPs selected using validation representative of the evaluation conditions generally maintain or improve performance as training data grows. In contrast, including such examples in context improves foundation-model performance, but does not prevent deterioration with increasing training size in our experiments.
[LG-4] raining with ML-labeled Data via Stochastic First- and Zeroth-Order Optimization
链接: https://arxiv.org/abs/2610.26790
作者: Maria-Eleni Sfyraki,Jun-Kun Wang
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:Training a machine learning (ML) model is an optimization process that typically requires a set of examples with ground-truth labels beforehand. Obtaining these ground truths usually requires laborious human effort and can be costly. The scalability of human labeling becomes a concern when there are a large number of data points that need to be labeled. To address this practical issue, we consider an optimization setting where a subset of samples has ground-truth labels, while the remaining data points were labeled by an ML predictor, which could potentially assign incorrect labels to those points. We propose both first-order and zeroth-order stochastic optimization methods for training on datasets consisting of a mix of true-labeled and ML-labeled samples. We further conduct comprehensive empirical studies from training a linear regression model to fine-tuning a large language model (LLM) to showcase the effectiveness of the proposed algorithms for training with ML-labeled data.
[LG-5] Simulation Priors for Rapid Adaptation in Model-Based RL for Autonomous Racing
链接: https://arxiv.org/abs/2610.26757
作者: Adrian Remonda,Jiajun Xi,Nicklas Hansen,Marcus Greiff,Thomas Lew,John Talbot,John Subosits,Xiaolong Wang
类目: Machine Learning (cs.LG); Robotics (cs.RO); Systems and Control (eess.SY)
*备注:
Abstract:Reinforcement learning enables learning high-performance control policies, but its reliance on extensive interaction limits deployment on physical systems where data collection is costly, and failures are unsafe. We introduce Prior-Informed Adaptation (PIA), an offline-to-online model-based RL framework that uses structured simulation priors for rapid adaptation. PIA pretrains a world model on heterogeneous low- and high-fidelity simulated trajectories, then adapts it with limited target-domain interaction using uncertainty-regularized planning. Rather than requiring an accurate simulator for policy transfer, PIA treats simulated trajectories as biased priors whose residual mismatch is corrected during online adaptation. We evaluate PIA in autonomous racing using high-fidelity simulation and on a real full-size autonomous race car. In simulation, PIA adapts faster across track and surface shifts than strong baselines, achieving lower lap times and fewer failures. On hardware, PIA achieves reliable few-shot adaptation with less than one hour of on-vehicle interaction.
[LG-6] rajectory Replay: Continuous-Time Mixture Transport for Lifelong Generative Models
链接: https://arxiv.org/abs/2610.26749
作者: Alessandro Palma,Charles Jones,Tristan Cinquin,Fabian J. Theis,Frederik Wenkel
类目: Machine Learning (cs.LG)
*备注:
Abstract:Adapting pre-trained continuous-time generative models to sequentially incoming data without forgetting earlier tasks is a central challenge in lifelong learning. A leading strategy, generative replay, uses a generative model trained on historical tasks to augment its own continual update with synthetic samples from experience. However, it requires full trajectory simulations at every training step, which accumulate discretization errors and can degrade the replayed historical distribution. Here, we propose trajectory replay, which bypasses sample generation by training an updated velocity field to match both a frozen historical velocity along its trajectories and the incoming task’s velocity field, weighted by a stability-plasticity coefficient. Through the lens of continuous-time transport, we show the resulting optimum averages the two task velocities by their respective densities and generates the mixture of their probability paths. Combined with mean velocity distillation, trajectory replay accesses historical trajectories using only a few network evaluations per continual update. Empirically, it outperforms generative replay and distillation baselines on class-incremental image generation. In two biological case studies, trajectory replay extends generative perturbation models to new contexts and experimental batches across two data modalities while preserving treatment-effect prediction.
[LG-7] Polar Switch: Fast Muon Orthogonalization with Polynomial Circuits
链接: https://arxiv.org/abs/2610.26742
作者: Zhehang Du,Weijie Su
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:Muon, a 2024-born optimizer for language models, differs from the 2015-born spectral gradient descent mainly in its fast, GPU-friendly Newton–Schulz polynomial for orthogonalizing the gradient matrix. Accordingly, a line of research proposes faster polynomial iterations for gradient orthogonalization, such as Polar Express and Gram Newton–Schulz, both built on quintic polynomials. In this paper, we introduce Polar Switch, a new approach to polynomial gradient orthogonalization. Using simple polynomial circuits, Polar Switch reaches a given high polynomial degree at a lower FLOP cost than quintic iterations. We define a circuit as a short sequence of symmetric matrix products on a few square buffers, and circuits with k products form a restricted family of odd polynomials of degree up to 2^k+1+1 . We choose the sequence of products in each circuit, their coefficients, and how the circuits are composed by numerically minimizing a worst-case bound on the polar approximation error. This minimization is subject to a FLOP budget that accounts for symmetric matrix multiplication kernels and a rounding constraint for numerical stability in half precision. Our two recommended Polar Switch methods have 15.5–18.7% lower runtime than Polar Express on square matrices and 8.6–16.2% lower runtime than Gram Newton–Schulz on rectangular matrices with aspect ratios 3.25, 3.5, and 8, with comparable approximation accuracy. In pretraining Modded-NanoGPT #91, Qwen3-0.6B, and Gemma 3-1B, Polar Switch reaches final validation losses similar to those of established orthogonalization methods. Across the more than 14 million matrices that each of the two recommended methods orthogonalized in these runs, the estimated output spectral norms never exceed the designed bounds by more than 3%, demonstrating their numerical stability.
[LG-8] Cross-Fitting for Neural Posterior Estimation NEURIPS2026
链接: https://arxiv.org/abs/2610.26733
作者: Jeffrey Regier
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: Accepted to NeurIPS 2026
Abstract:Neural Posterior Estimation (NPE) fits an amortized conditional density estimator to a simulated training set of complete-data realizations. NPE’s training objective, log-likelihood, favors posterior approximations that assign arbitrarily high density to the training points, and typical families of amortized conditional density estimators (e.g., normalizing flows, Gaussian encoders) are flexible enough to oblige. The resulting policy of extreme overconfidence translates poorly to held-out data. Early stopping and other regularizers can prevent this degeneracy, but they are limited to selecting posterior approximations along an unfavorable optimization trajectory set by in-sample residuals, where the posterior mean is underfit while the posterior variance is overfit. To avoid this trade-off, we propose Cross-Fitting NPE (CF-NPE), which uses sample splitting to separate conditional-mean estimation from the estimation of the posterior’s higher moments. Using K -fold cross-fitting, CF-NPE fits K posterior-mean regressors and then fits a conditional density estimator to the resulting out-of-fold residuals. In experiments with synthetic data and in a cosmological application, CF-NPE achieves higher held-out log-likelihood than standard NPE and typically improves calibration. CF-NPE is most useful in high-dimensional, simulation-limited settings.
[LG-9] Injection of Space Radiation Errors on Low-power NVM using Generative Deep Learning
链接: https://arxiv.org/abs/2610.26724
作者: Eleonora Giudici,Antonio Miele,Luca Mottola
类目: Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注:
Abstract:We develop an error model for a Non-Volatile Memory (NVM) device that represents the effects of physical faults experienced in a CubeSat mission in Low Earth Orbit (LEO). The CubeSat carries an off-the-shelf low-power Ferroelectric RAM (FRAM) chip and runs repeated read/write tests according to a predefined pattern. We compare every memory snapshot received from the satellite with a reference bitmap to identify bit errors. We visually analyze erroneous values distributions in the memory snapshots and we understand that they are frequently distributed according to regular spatial patterns. We define a generative deep learning technique to automatically learn these spatial patterns and to synthetically generate additional error patterns mimicking the observed ones. The model is capable at producing memory errors for general application-level experimentation. Our preliminary validation indicates that when comparing the model output with the data we collect, 0.92 of the generated bit flips correspond to bit flips in the memory snapshots and the model captures 0.91 of the observed bit flips. A comparison of the spatial distribution of bit flips in 10,000 generated error patterns also provides a Wasserstein distance of 0.073, indicating that the two distributions differ by roughly 7% of the maximum possible displacement of their probability masses.
[LG-10] Deep Pontryagin Learning for Continuous-Time Stochastic Control with Jumps
链接: https://arxiv.org/abs/2610.26718
作者: Jean-Loup Dupret,Patrick Cheridito
类目: Machine Learning (cs.LG); Systems and Control (eess.SY); Optimization and Control (math.OC)
*备注:
Abstract:We develop a deep Pontryagin learning framework for high-dimensional continuous-time stochastic control problems that naturally accommodates jump-diffusions. In contrast to direct policy optimization, which repeatedly differentiates the objective through the entire time-discretized dynamics, our method separates adjoint learning from policy improvement by explicitly incorporating state-sensitivity information from the Pontryagin stochastic maximum principle. For a fixed Markov control, we first compute pathwise adjoint labels by backpropagating the reward through time along simulated state trajectories, and project these labels onto the Markov state through a supervised regression step. The control is then updated by enforcing the pointwise Hamiltonian stationarity condition. We prove that the resulting projected adjoint process satisfies a discrete-time Pontryagin principle and converges, as the time step tends to zero, to the solution of the corresponding continuous-time adjoint backward stochastic differential equation. Numerical experiments illustrate the accuracy, stability, and scalability of the approach.
[LG-11] Doubly Robust Learning of Heterogeneous Survival Treatment Effects under Time-Varying Treatments
链接: https://arxiv.org/abs/2610.26710
作者: Tomàs Garriga,Munib Mesinovic,Luke De Bretton-Gordon,Gerard Sanz,Axel Brando,Tingting Zhu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Estimating heterogeneous survival treatment effects (HSTEs) from time-to-event data under time-varying treatments is important in fields such as medicine (e.g., to decide which sequence of treatments is better in terms of survival probability). However, this task is challenging because treatment decisions, patient characteristics, and censoring probability often evolve together, requiring sequential adjustment for both treatments and censoring. Existing survival learners mostly consider a single baseline treatment assignment, while longitudinal treatment-effect learners typically assume uncensored outcomes. In this paper, we introduce a novel doubly robust learner for heterogeneous survival treatment effects under time-varying treatments and history-dependent censoring. Our framework separates the intervention window from the survival horizon, with treatment prescribed for a finite period, after which care follows its natural course. We further establish the favorable theoretical properties of Neyman-orthogonality, rate double robustness and quasi-oracle efficiency. The framework accommodates flexible machine learning models for nuisance estimation and treatment-effect regression. We evaluate our learners through several numerical experiments, demonstrating their effectiveness.
[LG-12] GeoSplit: Selective Edge–Server Map Acquisition for Cross-User Travel-Mode Recognition
链接: https://arxiv.org/abs/2610.26707
作者: Zepu Wang,Luyao Niu,Ruolin Li,Jeff(Xuegang)Ban
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注: 5 pages, 2 figures, 2 tables
Abstract:Networked mobility sensing couples local trajectory inference with remote geospatial context, making communication decisions part of the prediction system. We present GeoSplit, an edge–server framework for travel-mode recognition that separates the utility of map evidence, its training-time transfer, and its selective acquisition at runtime. On GeoLife with 10% labels, we evaluate these components under user-day and user-disjoint grouping. For unseen users, aligned full-map context raises macro-F1 from 0.7154 to 0.7864, while map-informed distillation yields a motion-only student at 0.7225. Through a compact query interface, entropy gating achieves 0.7729 macro-F1 versus 0.7738 for all queries, querying 39.5% of trajectories and reducing application payload by 60.5%. Under user-day grouping, selective and all-query scores are similarly close at a 21.6% query rate. Learned value estimation and probability fusion offer no consistent improvement. These results illustrate communication-efficient edge–server inference for an intelligent transportation sensing task.
[LG-13] PACT: Plan-Augmented Coordination via Trace-Graphs for Dynamic Multi-Agent Combinatorial Optimization
链接: https://arxiv.org/abs/2610.26671
作者: Xiao Zhou,Alex Schutz,Victor-Alexandru Darvariu,Nick Hawes
类目: Machine Learning (cs.LG)
*备注:
Abstract:Multi-agent approaches have been investigated recently as a means of improving the scalability of reinforcement learning for neural combinatorial optimization. Existing methods assume that all problem instance information is known upfront. However, real-world problems are often dynamic: relevant information is progressively revealed as the agents take actions (e.g., arriving at an edge in a routing problem only to discover it is untraversable). In such cases, existing methods fail to coordinate agents to explore and exploit the progressively revealed information. To address the arising coordination challenge, we introduce PACT (Plan-Augmented Coordination via Trace-graphs). PACT augments the environment graph with auxiliary nodes representing the plans of agents, which are connected to the resource nodes to indicate future dependencies. The proposed message-passing scheme over the augmented graph grounds inter-agent coordination in the environment and enables proactive, rather than reactive, conflict resolution. We show that PACT improves solution quality in dynamic problems relative to baselines and exhibits the best transfer to larger instances on which, remarkably, it outperforms centralized oracles with privileged access to instance uncertainty.
[LG-14] Nested Shared-Operator Recursion on ARC-AGI-1: An Empirical Study of HRM TRM and URM
链接: https://arxiv.org/abs/2610.26666
作者: Sara Neri,Eugenio Gaeta,Andrea Ceschini,Andrea Facchinetti,Giuseppe Fico
类目: Machine Learning (cs.LG)
*备注:
Abstract:Recursive reasoning models refine persistent internal states by repeatedly applying learned operators. We investigate whether adding a persistent intermediate state that integrates repeated updates of a faster state through the same operator improves predictive performance relative to each model family’s baseline at matched parameter counts. We define this organisation operationally and implement it within three architectures: the Hierarchical Reasoning Model (HRM), Tiny Recursive Model (TRM) and Universal Reasoning Model (URM). Using the official codebases on ARC-AGI-1, we compare each baseline with its parameter-matched nested variant. We evaluate held-out queries from tasks whose demonstrations were available during training. The experiments use a single-GPU budget far below the published training protocols. Under the tested conditions, nested variants show no consistent advantage in token accuracy, predictive loss or exact solving, which remains near zero for all configurations. The results therefore leave the hypothesis open. We specify a confirmatory comparison, with an informative baseline, multiple seeds, controls that isolate nesting and predefined criteria, to determine whether this organisation merits further development.
[LG-15] SleepPairNet: Learning Multimodal Pediatric Sleep Representations with Random-Pair Cross-Attention
链接: https://arxiv.org/abs/2610.26628
作者: Saurav Raj Pandey,Harlin Lee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Analyzing physiological sleep signals is important for assessing sleep quality and detecting sleep-related disorders. Learning inter-modal relationships across physiological signals can provide richer representations for downstream sleep analysis. We introduce \textbfSleepPairNet, a multimodal self-supervised learning (SSL) framework that learns pairwise interactions between physiological signals through cross-attention. We propose a random modality-pair selection strategy that exposes the model to diverse signal pairs during pretraining, enabling flexibility across modality combinations at test time. Pretrained on the NCH dataset, SleepPairNet outperforms the evaluated SSL baselines on all respiratory tasks while remaining competitive on EEG-related tasks, and shows even stronger results when transferred to the external CHAT dataset.
[LG-16] Global Time-Aware Conditioning (GTC): A Mechanism for Time-Aware Pediatric Sleep Representation Learning
链接: https://arxiv.org/abs/2610.26618
作者: Saurav Raj Pandey,Harlin Lee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Sleep physiology evolves throughout the night, with changes in sleep cycles, respiratory stability, and cardiovascular dynamics. However, self-supervised learning (SSL) methods for sleep typically process short, fixed-duration windows without explicitly modeling where each window occurs within the full sleep session. We introduce Global Time-Aware Conditioning (GTC), a mechanism that complements local positional encodings by conditioning latent representations on the global time-of-sleep of each input window. We integrate GTC into three representative SSL frameworks and evaluate them on five clinically relevant tasks with extreme class imbalance across two independent pediatric cohorts. GTC generally improves F1 and AUPRC across architectures and tasks, with gains persisting under cross-cohort transfer, highlighting the value of global temporal context for sleep representation learning.
[LG-17] Degree Bounds for Separating Polynomial and Neural Invariants for Point Clouds
链接: https://arxiv.org/abs/2610.26612
作者: Snir Hordan,Harm Derksen,Mircea Petrache,Nadav Dym
类目: Machine Learning (cs.LG); Commutative Algebra (math.AC)
*备注: 34 pages, 2 figures
Abstract:Neural networks that respect the symmetries of point clouds commonly compute high-order representations of the rotation group to improve their expressivity. Such models are known to be universal, and thus able to separate any pair of point clouds, when using representations of sufficiently high order, yet no precise bound on this order, as a function of the number of points, was previously known. We establish the first quadratic lower and upper degree bounds for polynomials that separate point clouds of n points up to rotations and permutations. By showing Tensor Field Networks and MACE can express all invariant polynomials of a given degree with sufficient representation order, we additionally derive the first explicit, quadratic bound on the representation order sufficient for these architectures to be complete, replacing previous existence results with concrete guarantees. We further show that, under local regularity assumptions, point cloud separation can be achieved with a message-passing-based model using invariants whose degree depends only on the size of a local neighborhood, rather than on the size of the full point cloud. Finally, we provide counterexamples showing MACE must compute high order irreducible representations to separate locally regular point clouds.
[LG-18] A Probabilistic Rotation-Preserving Kernel Operator Surrogate for Continuum-Robot Design Spaces ICRA
链接: https://arxiv.org/abs/2610.26605
作者: Branden Frieden,Alan Kuntz,Varun Shankar
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Submitted to the 2027 IEEE International Conference on Robotics and Automation (ICRA)
Abstract:We develop an uncertainty-aware, operator-learning-based surrogate for tendon-driven continuum robots (TDCRs) that generalizes across robot design and actuation while predicting the complete equilibrium configuration. Our central idea is a nonstationary Gaussian-process (GP) kernel whose metric is learned by a shallow neural network from mechanics-aware design coordinates. The network learns how similarity changes across design space, while the resulting kernel defines the design-to-pose operator, its posterior uncertainty, and a criterion for sequential simulator acquisition. We represent backbone position and rotational strain with spatial kernels and reconstruct orientation directly on SO(3), preserving valid material frames and tendon geometry. We use ASkotch with Nystrom preconditioning for scalable kernel operator solves and a separate Nystrom approximation for posterior covariance. Experiments show that kernel-coverage acquisition reduces mean trajectory error by approximately 15% relative to IID sampling, while posterior variance increases for increasingly unsupported designs. Compared with prior neural-operator TDCR models, the method is 30-50% more accurate using approximately 5x fewer training designs while additionally providing posterior uncertainty estimates.
[LG-19] ITU-DESC: a missingness-aware machine-learning dementia risk score in the UK Biobank cohort
链接: https://arxiv.org/abs/2610.26604
作者: Abdullah Kavaklı,Faik Boray Tek
类目: Machine Learning (cs.LG)
*备注:
Abstract:Background: Existing dementia risk scores rely on fixed, pre-selected features and linear models, or ignore bias from informative missingness in cohort data. Here we introduce the Istanbul Technical University Dementia Risk Score (ITU-DESC), an explainable machine-learning score for incident all-cause dementia that combines broad feature selection across roughly 8,000 candidate features with explicit control of missingness bias. Results: We studied approximately 500,000 UK Biobank participants with a median follow-up of 16.6 years. We withheld missingness patterns from the model using unconditional random hot-deck imputation. We stabilized feature selection by repeating BorutaSHAP-based feature selection with ten random-imputation seeds and selecting predictors chosen in at least eight of the ten seeds, which yielded 38 predictors. We trained a CatBoost gradient-boosted decision-tree model and evaluated it on a holdout test set of 50,041 participants. For all-cause incident dementia, ITU-DESC achieved an average precision of 0.138 (95% confidence interval 0.123-0.156) and an area under the receiver operating characteristic curve (ROC AUC) of 0.853 (0.844-0.863), significantly outperforming other risk scores, including the strongest comparator, the UK Biobank Dementia Risk Prediction model (0.118 and 0.841). We identified dementia risk predictors absent from established scores, including leukocyte telomere length and liver enzymes. Conclusions: ITU-DESC is the first dementia risk score to combine broad feature selection with explicit control of missingness bias. Subjects: Machine Learning (cs.LG) ACMclasses: I.2.1; J.3 Cite as: arXiv:2610.26604 [cs.LG] (or arXiv:2610.26604v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.26604 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-20] Bombadil: A Foundation Model for Plant Electrophysiology
链接: https://arxiv.org/abs/2610.26597
作者: James Kirk,Piotr Bednarski,Filip Twardy,Justin Mast,Lee von Kraus
类目: Machine Learning (cs.LG); Signal Processing (eess.SP); Quantitative Methods (q-bio.QM)
*备注: 25 pages, 13 figures, 11 tables
Abstract:Living plants, like animals, use electrical signals to carry information about their physiological state across their body, from roots to branches. These signals can be recorded, as they are in animals, from a pair of differential extracellular electrodes inserted into signal-conducting tissue. While machine-learning models have been trained to predict specific plant stress states from these signals, each of these models has been trained for only a single task and none have been designed to transfer across tasks, species, or environments. These approaches do not allow easy scaling across the wide range of species and environments that are relevant to agriculture, forestry, and horticulture for both commercial and consumer plant care. Here we present Bombadil, a foundation model for plant electrophysiology, pretrained without labels on over 4 million hours of electrical signals recorded from 100 plant species in ten distinct environments, including laboratories, greenhouses, commercial orchards, and natural forests. In this way, Bombadil is able to learn the fundamental features of plant electrical signals across a wide array of species and environments. Then, from those same electrical signals, Bombadil tracks changes in seven physiological indicators of plant stress, even among species and stressors with no labels in training. Comments: 25 pages, 13 figures, 11 tables Subjects: Machine Learning (cs.LG); Signal Processing (eess.SP); Quantitative Methods (q-bio.QM) Cite as: arXiv:2610.26597 [cs.LG] (or arXiv:2610.26597v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.26597 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-21] SafeWall: A Risk-Aware Job Walltime Recommendation for HPC Scheduling
链接: https://arxiv.org/abs/2610.26577
作者: Kanglin Xu,Michael Papka,Zhiling Lan
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注: Accepted at the 29th Workshop on Job Scheduling Strategies for Parallel Processing (JSSPP 2026), held with IEEE IPDPS 2026. 20 pages, 4 figures, 4 tables
Abstract:In high-performance computing (HPC), users must submit resource requirements for their jobs, with user-supplied walltime being a critical parameter. However, these estimates are notoriously inaccurate, as users tend to overestimate walltime to prevent the scheduler from terminating their jobs prematurely. While existing work on learning-based walltime prediction primarily focuses on maximizing prediction accuracy, such predictions are inherently risky for walltime recommendations. Specifically, underestimation leads to job termination before completion, resulting in lost progress and wasted resources. In this study, we present SafeWall, a novel recommendation method that adjusts walltime predictions by incorporating an explicit, nonnegative safety buffer to mitigate underestimation risk. The core of our approach is a risk-aware buffer that dynamically calibrates safety margins based on individual user’s historical behavior. When user data is limited, the system employs statistical shrinkage to align these margins with a global baseline, ensuring robustness. This yields a lightweight mechanism that optimizes the trade-off between prediction accuracy and termination risk. SafeWall is modular and applicable to various learning-based prediction models. We evaluate SafeWall using five public workload traces, including the full-lifetime job log from the 4,392-node Theta machine at the Argonne Leadership Computing Facility. Across all traces, SafeWall substantially reduces underestimation risk compared to the unbuffered base predictors, while preserving competitive accuracy. The source code is publicly available at this https URL.
[LG-22] Wet-Biased Data Pipelines Mislead the Training and Evaluation of Generative Precipitation Nowcasting
链接: https://arxiv.org/abs/2610.26486
作者: Christoph Metzl,Stefan Kapsreiter,Kianusch Vahid Yousefnia,Virginia Poli,Richard Müller,Frank Krüger,Tobias Bölle
类目: Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph)
*备注: 63 pages, 7 figures and 3 tables in the main text, 11 supplementary figures and 9 supplementary tables. This work has been submitted to npj Climate and Atmospheric Science
Abstract:Generative models for precipitation nowcasting are trained and evaluated predominantly on wet-biased datasets, which over-represent precipitation events to emphasize rare extremes. Selecting test cases based on target observations is known to favor forecasts that match the selected distribution rather than the intended deployment distribution. We show analytically how this target-dependent sampling also shifts the conditional distribution learned by generative models. When training and test sampling match, even an ideal model learns precisely the distribution favored by evaluation, allowing the evaluation bias to conceal the training bias. Using a fixed satellite-to-radar diffusion model, we demonstrate that target-dependent test sampling can reverse the calibration assessment. The model appears under-predictive on a wet-biased test set but over-predictive on a deployment-representative test set. This result is replicated outside the model’s training domain after a seven-year gap. Utilizing classifier-free guidance as an example, we demonstrate how the contradictory calibration assessment can misdirect downstream model development.
[LG-23] A Simple and Unified Method for Gradient-Variation Regret
链接: https://arxiv.org/abs/2610.26473
作者: Hang Yu,Peng Zhao
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:Gradient-variation online learning has attracted increasing attention for its connections to game theory and optimization. It has been studied in both non-stationary online learning, which aims to minimize dynamic regret against time-varying comparator sequences, and universal online learning, which seeks to adapt to unknown loss curvature. Prior work obtained gradient-variation bounds using different techniques tailored to each setting. We propose a simple and unified method based on surrogate losses with a quadratic penalty on decision movement, yielding gradient-variation bounds in both settings. For non-stationary online learning, the method induces an optimistic-to-gradient-variation reduction, enabling a broad class of algorithms to match the rate established by Zhao et al. (2024). For universal online learning, it yields a simple algorithm that matches the optimal universal regret bounds of Yan et al. (2024) while relaxing the extended smoothness assumption. Finally, we apply our method to online games and improve on existing guarantees.
[LG-24] On Learning Expand-and-Sparsify Representations
链接: https://arxiv.org/abs/2610.26470
作者: Alireza Asghari,Parikshit Ram,Kaushik Sinha
类目: Machine Learning (cs.LG)
*备注:
Abstract:Expand-and-sparsify ( \textttEaS ) representations lift an input to a much higher dimension through a linear map and keep only the k largest coordinates. Existing theory builds that map at random and needs the expanded dimension to grow exponentially in the input dimension, which puts the construction out of practical reach. We ask what is gained by learning the map by gradient descent instead, and how that gain depends on how wide the expansion is. We sweep 73 OpenML classification tasks and 48 regression tasks under a protocol that holds the datasets, expanded dimensions, sparsity levels and splits fixed across every method, and we summarise each comparison by a median over datasets and a Wilcoxon signed rank test. We report four findings. First, learning the expansion helps most when the expansion is narrowest, and its advantage over a random map decays to nothing by the widest expansions we run, so a learned \textttEaS layer can be made small. Second, with the map learned, a top- k representation that keeps at most a tenth of its coordinates is indistinguishable from a dense \textttReLU layer of the same expanded dimension at every expansion factor we test, while GELU and SiLU are not. Third, when the representation is forced to be binary the sparse code is the better of the two, and the reason is that binarization costs it nothing while it costs the dense layer accuracy. Fourth, how sparse the representation can be is governed by how wide it is: the sparsity level that is free rises as the expansion grows, so there is no fixed limit on how sparse it can be. The regression tasks reproduce the first finding and qualify the second and third.
[LG-25] Residual-Based Adaptive Refinement in Hybrid Variational-Collocation PINNs for Singular Elliptic PDEs with Mixed Dirichlet-Neumann Boundary Conditions
链接: https://arxiv.org/abs/2610.26458
作者: Somnath Giri,Muslim Malik,Rishikesh Yadav
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注: 27 pages, 50 figures
Abstract:Solutions of second-order elliptic PDE problems on nonconvex or cracked domains with mixed Dirichlet-Neumann conditions typically lie outside H^2(\Omega) . Near re-entrant corners, crack tips and points where the boundary condition changes type, the gradient grows like r^\lambda-1 with \lambda1 . We study residual-based adaptive refinement in hybrid variational-collocation PINNs. A multiscale Fourier-feature network is trained on a composite loss. The loss combines a variational residual tested against Legendre bubble functions, a strong-form collocation term and penalties for the Dirichlet and Neumann data. The variational term needs only first derivatives of the network, whereas the collocation term requires second derivatives and provides a residual that can be evaluated at any point. After each training stage, candidate points selected by this residual are added to the collocation set. The method is tested on five two dimensional benchmarks, namely a Mode~III crack, a cross-shaped domain with four re-entrant corners with unequal amplitude, a reaction-diffusion problem with four localized peaks, a monotone semilinear problem with a boundary layer and the Motz problem. Compared with uniform point insertion under the same collocation and iteration budget, residual-based refinement reduced the relative L^2 error by up to a factor of 32 , the relative L^\infty error by up to a factor of 38 and the relative H^1 semi-norm error by up to a factor of 15 .
[LG-26] A Simple and General Approach to top-k Selection
链接: https://arxiv.org/abs/2610.26407
作者: Motti Goldberger,Nils Rudi
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:This paper considers the Bayesian top- k identification problem, wherein a decision maker is tasked with selecting k items from a pool of items of unknown quality. To assess the items, the decision maker adaptively allocates a fixed sampling budget, where each sample of an item yields a noisy observation of its quality, with the objective of minimizing the expected opportunity cost of the selected set. We propose an approach that allocates each batch of samples by minimizing a surrogate objective derived from an upper bound on the expected opportunity cost. The approach is general: samples can be allocated in batches, the selection size k is arbitrary, and it applies to many sampling distributions. It is also simple and intuitive. We prove that the method’s posterior expected opportunity cost decays at an exponential rate of at least half the best rate achievable by any sampling rule. In simulations across a wide range of settings, the method is competitive with, and often improves upon, the best existing algorithm in each setting.
[LG-27] Forag erSim: Evaluating Embodied Map-Based Reasoning in Large-Scale Dynamic Worlds
链接: https://arxiv.org/abs/2610.26395
作者: Tomáš Musil,Martin Kotík,Michal Vavrečka,Karla Štěpánová,Sebastian Scherer,Martin Saska
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Code and data: this https URL
Abstract:Animals build cognitive maps of vast environments, use them to perform map-based navigation, and might even be reusing the same circuitry used in spatial reasoning for creating and using maps of arbitrary task-spaces. In contrast, current embodied AI systems either rely on hand-engineered mapping pipelines, which break in the presence of unmodeled dynamics and are restricted to metric space, or on learned policies that rarely form map-like representations at all. Developing architectures that could learn to build and use map-like representations as flexibly as animals is, however, difficult because few embodied AI benchmarks actually require map-based reasoning. To address this, we present ForagerSim — an interdisciplinary research platform and simulation environment focused on embodied tasks that require map-based reasoning, with scenarios ranging from mobile-robot search and rescue to animal foraging in both complex mazes and vast, potentially infinite open areas. ForagerSim is highly modular and allows modifying independent axes of difficulty related to perception (abstract 2D sensing to full RGB or LiDAR), scale (room-scale to multi-km scale), topology (open-fields to complex mazes), dynamics (smoke, debris, sensorimotor failures and others) and tasks (from simple 2D coverage to foraging-inspired well-alternation tasks). We compare reinforcement learning baselines with a simple engineered robotics stack and a human player on 8 example worlds which show how different difficulty axes affect existing methods. Our results show that, when sensing is simplified to 2D, learned policies can perform multi-object search and well-alternation tasks surprisingly well in simple topologies, but they fall short of human performance in larger or more complex-topology environments where map-based baselines perform well, suggesting that map-based reasoning is the bottleneck in these scenarios.
[LG-28] CTLite: Lightweight Irregular Multivariate Time Series Forecasting with Reusable Continuous-Time Summaries
链接: https://arxiv.org/abs/2610.26379
作者: Shusen Ma,Xudong Zuo,Yangchun Xie,Yun-Bo Zhao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Irregular multivariate time series forecasting (IMTSF) requires capturing temporal dynamics and cross-variable dependencies from sparse, asynchronous observations. Existing methods often rely on learned temporal encoders, coupling historical representation construction with model optimisation and repeatedly processing the same observations. We propose CTLite, a lightweight forecasting framework that decouples historical summarisation from learned prediction. Specifically, a parameter-free continuous-time moment encoder summarises local levels and trends at multiple temporal scales while suppressing poorly supported slope estimates. A compact readout combines damped local extrapolation with a low-rank cross-variable correction that accounts for observation freshness, target availability, and forecast horizon. Since the historical summaries are independent of trainable parameters and forecast queries, they can be reused across optimisation steps and repeated predictions for an unchanged history without altering predictions or parameter gradients. Experiments on four real-world benchmarks show that CTLite achieves the best or second-best mean MSE and MAE among the evaluated methods with only 88-2,537 trainable parameters. Under matched GPU workloads at batch size 32, CTLite reduces inference latency by 18.6-87.3% compared with the latest models, including the cost of historical encoding. Reusing summaries further reduces total runtime by approximately 52% over 16 forecast calls on the same history compared with recomputing the encoder for every call, including the initial encoding cost. The code is at: this https URL.
[LG-29] Logit Averag ing During Training Sets a New State of the Art for Non-Invasive Speech BCI
链接: https://arxiv.org/abs/2610.26347
作者: Miran Özdogan,Gereon Elvers,Michael Bronstein,Oiwi Parker Jones
类目: Machine Learning (cs.LG)
*备注:
Abstract:Non-invasive brain-computer interfaces are in practice limited by signal-to-noise ratio, and recent state-of-the-art methods address this limitation by aggregating over repeated trials of the same stimulus. Yet multi-trial aggregation has been treated inconsistently across the literature: some work averages brain signals at the model’s input, other work averages predictions at the model’s output, and neither treats aggregation as an architectural choice in its own right. We provide the first systematic comparison of aggregation strategies for non-invasive neural decoding, evaluating six paradigms (including logit averaging during training and three attention-based variants we introduce) across three datasets, four tasks, and two group sizes. We find that aggregation choice matters substantially: training-time logit averaging, the simplest method we propose, outperforms all others in 6 of 7 settings. On the phoneme classification setting of the 2025 PNPL competition, it sets a new state of the art, significantly outperforming the competition winner by 17.4 absolute percentage points in balanced accuracy. We further draw a theoretical connection between logit averaging and joint ensemble training, which has been shown to underperform independent ensembling in the deep learning literature; our results call into question recent explanations of why joint ensemble training fails. Together, these findings establish multi-trial aggregation as a key design choice for non-invasive neural decoding and identify training-time logit averaging as a simple, effective default.
[LG-30] CAST: Cast-Admissible Generative Design and Shape Optimization with Paired Depth Images
链接: https://arxiv.org/abs/2610.26309
作者: Jihoon Kim,Yongmin Kwon,Namwoo Kang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Deep generative design can produce complex geometries, but unconstrained representations may create undercuts that prevent release from a two-part mold. We present CAST, which reformulates three-dimensional shape generation under a selected mold-opening constraint as a two-dimensional image generation problem using paired top and bottom depth images. Each valid depth pair defines one material interval along a selected mold-opening direction, so reconstructions exclude undercuts in that direction by construction. Diffusion models generate candidate shapes in this representation, and a finite element analysis (FEA) guided optimizer refines the same two boundaries. We evaluate diffusion-based generation with limited training data comprising 211 SimJEB bracket designs. On SimJEB brackets, FEA-guided refinement with a volume penalty reduced compliance in 193 of 200 generated candidates. The median reduction across all candidates was 36.8 percent in the optimization FEA model. Reconstruction of existing brackets revealed a mass–stiffness trade-off: filling occluded regions increased material volume and reduced displacement under the evaluated loads. Paired depth images thus connect generative exploration and structural refinement in a shared design space that preserves the selected pull-direction restriction.
[LG-31] Learning SOCP Solver Dynamics with Graph Neural Networks
链接: https://arxiv.org/abs/2610.26285
作者: Adam Ghalem,Lou Salaün,Davide Rinaldi,Fragkiskos D. Malliaros
类目: Machine Learning (cs.LG)
*备注: Learning on Graphs (LoG) Conference, 2026
Abstract:Second-order cone programs (SOCPs) are a fundamental class of convex optimization problems widely applied across engineering domains. Classical interior-point methods are robust and accurate, but their iterative nature can be costly in real-time or repeated-solve settings. This has motivated learning-based solvers that approximate optimizer solutions with neural networks. Recent work has increasingly studied graph neural networks (GNNs) for linear, quadratic, and second-order cone optimization, leveraging their ability to model sparse relational structure among variables, constraints, and objective functions. Most existing GNN-based solvers compress the optimization process into a single forward pass, predicting the optimizer directly without using the intermediate trajectory produced by classical solvers. We introduce SOCPNet, a recursive GNN framework that uses solver trajectories to learn an iterative SOCP solution procedure. A shared GNN model is applied repeatedly, refining the current iterate at each step and enabling inference for a variable number of iterations. Rather than predicting the solution in a single pass, SOCPNet learns from intermediate iterates to progressively refine its estimates using a trajectory-regularized training procedure. We evaluate SOCPNet on multiple datasets spanning standard SOCP problem classes and show that it consistently reduces the final solution error relative to a SOCP-GNN baseline. Finally, we demonstrate how trained SOCPNet models can be used in a hybrid optimization pipeline by warm-starting a classical interior-point solver, reducing solver iterations without loss of accuracy.
[LG-32] rust Your Targets: Is Target Propagation Block-Diagonal Gauss-Newton?
链接: https://arxiv.org/abs/2610.26283
作者: Giuseppe Castiglione,Christopher Buckley,Ivor Simpson
类目: Machine Learning (cs.LG)
*备注:
Abstract:Target propagation (TP) is a backpropagation alternative in which local targets carry the global learning signal. Prior works claim that TP performs Gauss-Newton (GN) optimisation on these targets, but the correspondence has only been established under unrealistic assumptions where hitting them exactly is trivial. Practical variants instead descend on a least-squares problem via a single gradient step. In this paper, we rederive TP from a trust region under an implicit, previously unexamined norm, and introduce CGTP (Conjugate Gradient Target Propagation), which uses conjugate gradients to solve the layer-local subproblems in parallel. A single adaptive parameter jointly controls this trust region and a matching bound on the weight update; KFAC emerges exactly as the special case where the per-sample Jacobians are replaced by their batch mean, giving its factored damping a variational grounding. On MNIST and Fashion-MNIST MLPs, CGTP matches the accuracy of ADAM and KFAC while needing orders of magnitude fewer outer iterations than ADAM (about 10x fewer backward-pass-equivalents overall), while remaining stable across depth.
[LG-33] SEAM: Spectral Embedding Alignment and Merging for Low-Resource TTS EMNLP2026
链接: https://arxiv.org/abs/2610.26252
作者: Walaa Othman,Baher Mohammad,Stamatios Lefkimmiatis,Ammar Ali
类目: Machine Learning (cs.LG)
*备注: Accepted to Findings of the Association for Computational Linguistics: EMNLP 2026
Abstract:Expanding the vocabulary of conditional text-to-speech (TTS) models typically requires extensive fine-tuning with paired audio, creating a major bottleneck for low-resource and cross-lingual adaptation. To address this, we introduce SEAM, a zero-audio initialization method that constructs embeddings for unseen tokens from known subtokens. Unlike standard pooling strategies that ignore sequential structure, SEAM applies a spectral filter to preserve temporal phonetic order and uses closed-form ridge regression to align the new embeddings with the pretrained conditioning manifold. Across three languages and two TTS architectures, SEAM approaches or matches baseline performance using up to four times less training data, reducing sample complexity by 75% compared to random initialization and mean pooling. Our approach provides a lightweight, reproducible, and theoretically grounded solution for vocabulary expansion in low-resource speech synthesis.
[LG-34] OceanTensorCube and ST-MMT: A Multimodal Data and Transformer System for Seven-Day Forecasting of Pyropia Bleaching KR
链接: https://arxiv.org/abs/2610.26209
作者: Gwangjin Kim,Cheolyoung Jang,Yosep Jo
类目: Machine Learning (cs.LG)
*备注: 27 pages. Forecasting service: this https URL
Abstract:Bleaching of cultured Pyropia (laver) causes losses in Korean seaweed farming. We present a system that forecasts bleaching seven days ahead for farm clusters. OceanTensorCube aligns 32 channels from public marine, meteorological, ocean-model and satellite sources on a 30 m grid, and ST-MMT, a 9.2-million-parameter spatiotemporal transformer with hub-style multi-head cross-attention between variable groups, maps a 14-day input window to a seven-day-ahead bleaching probability. On the validation set, a six-seed ensemble detects every reported positive at a false-alarm rate of at most 3.5%, an upper bound because the report-based labels are positive-unlabeled. A same-date control on separately trained models indicates that the architecture discriminates between clusters when the calendar is held fixed. We also report the system’s weaknesses: skill falls steeply as the quarterly nutrient survey ages (ROC-AUC 0.94 at 30 days, 0.53 at 129 days), and a single LSTM on cluster-average series outperforms a single ST-MMT, with only the ensemble leading at the operating point. We describe the labeling procedure, the evaluation protocol and the operational gate that withholds forecasts when inputs are stale. The forecasting service is operated at this https URL.
[LG-35] aQFormer: A Treatment-as-Query Transformer with Data-Fusion for Multidimensional Treatment Modeling in Coupon Allocation
链接: https://arxiv.org/abs/2610.26199
作者: Jiahong Li,Zeqin Yang,Yu Xiang,Zhifan Jiang,Jingqi Ma,Enzheng Hua,Jixing Xu,Jun Wu,Fei Wang,Enyun Yu,Jiecheng Guo
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study coupon allocation on an online marketplace offering multiple service categories. Issuing coupons requires choosing a vector of category-specific discounts, accounting for both their direct effects and cross-category demand shifts. We present TaQFormer, a treatment-as-query Transformer for this multidimensional treatment setting. Each treatment coordinate is represented as a query token, allowing cross-attention to estimate the response to the full treatment vector. To combine large but policy-biased observational logs with a smaller stream of randomized controlled trial data, we use a two-stage pipeline: observational pretraining learns general outcome patterns, followed by LoRA fine-tuning on RCT data with a request-specific fusion gate. We evaluate the resulting system through the checks required for production, including offline uplift metrics, treatment-parallel serving, latency monitoring, and a 14-day online A/B test. At essentially the same subsidy cost rate, \model increases business value per user by 1.26% over the production baseline and significantly outperforms both the production baseline and the RCT-only model, while meeting the latency budget. It was subsequently rolled out to all marketplace traffic and has operated at full scale for several months as of October 2026. The deployment illustrates how multidimensional treatment modeling and observational pretraining, combined with adaptation using RCT data, can support effective joint discount decisions in production.
[LG-36] Dynamical Batching: How Class-Cycling Training Shapes the Internal Structure of Neural Networks ATC
链接: https://arxiv.org/abs/2610.26156
作者: Nicolás Ratier-Werbin,Samuel Lozano,Miguel Ruiz-García
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Soft Condensed Matter (cond-mat.soft)
*备注: 18 pages, 10 figures, 5 tables. Code: this https URL (archived at this https URL )
Abstract:We investigate whether cyclically varying the class composition of training batches can modify both the performance and the internal organization of neural networks. Motivated by nonequilibrium systems, in which cycling between related sub-goals can make switching between their solutions increasingly efficient, we introduce Dynamical Batching (DB), a sampling protocol that periodically emphasizes each class while leaving the loss and the optimizer unchanged, and compare it with Uniform Batching (UB), in which every mini-batch has equal class proportions. In the regime where standard training fails, DB changes the outcome: on a three-arm spiral, a capacity-limited MLP that plateaus far below perfect accuracy under UB fits the training data under DB. Tracking individual weights, we find that under DB the change between class-specialized states is carried by a smaller set of weights than under UB, and that this set is markedly more persistent from cycle to cycle, on every synthetic dataset we test and in a wider network that fits the data under both protocols. This organization is a signature of the protocol rather than a predictor of its benefit: it appears equally where DB leaves accuracy unchanged. On CIFAR-10, the same protocol modestly improves the test accuracy of a channel-narrowed ResNet-10, a standard residual architecture, and of a second capacity-limited convolutional network. These results show that how class emphasis is distributed in time can shape both the solution a network reaches and the internal structure of the network.
[LG-37] Variance-Aware Non-Stationary Bandits:All-Budget Regret Bounds and Adaptation Costs
链接: https://arxiv.org/abs/2610.26116
作者: Xiaojin Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Low reward variance reduces estimation cost, but does not eliminate monitoring of arms that may improve while unobserved. We refine the variance–variation tradeoff and characterize how both costs depend on supplied variance and variation budgets in bounded, non-stationary multi-armed bandits. Let A be the number of arms, T the horizon, B a bound on consecutive mean variation excluding the initial vector, and Q a bound on the sum of the largest per-round reward variances. For every admissible pair Q,B , the minimax expected dynamic regret is, within a single logarithmic factor, [ \min\T,;A+\sqrtAQ+(AQBT)^1/3 +\sqrtAT\min\B,\sqrt B\. ] The deterministic boundary Q=0 is characterized without logarithms. For small positive budgets, a safe calibration phase identifies a common interval containing all future means of surviving arms. Clipping observations to a wider interval reduces the cost of restarting a prior hybrid static-regret algorithm; an original-mean analytical reference controls its residuals without charging the variation of clipped means. Balancing clipping bias and monitoring closes the small-budget noisy regime. Fixed oblivious constructions give the matching lower bound. We also quantify adaptation costs: on a disclosed deterministic one-improvement family, a stationary-regret budget has a matching worst-case change-regret frontier, including a finite improvement pulse. This implies an \Omega((T/A)^1/4) loss when one policy competes with all supplied-budget deterministic oracle rates. The budget-free low-noise scheduling problem remains separate. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.26116 [cs.LG] (or arXiv:2610.26116v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.26116 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-38] From Retrieval to Recognition:Knowledge-Enhanced Post-training of Time Series Foundation Models for Classification NEURIPS2026
链接: https://arxiv.org/abs/2610.26114
作者: Zhen Liu,Yucheng Wang,Yingpeng Du,Tianjun Wei,Qianli Ma,Min Wu,Jie Zhang,Zhu Sun
类目: Machine Learning (cs.LG)
*备注: Accepted in NeurIPS 2026
Abstract:Time series foundation models (TSFMs) have achieved notable progress in classification tasks through large-scale pretraining. However, their post-training practice mainly focuses on implicit parameter fine-tuning, where task-relevant information is entangled with abundant irrelevant knowledge in vast parameter spaces. This may hinder TSFMs from precisely activating useful knowledge for downstream classification. To this end, we propose Retrieval-to-Recognition (R2R), a plug-and-play post-training paradigm that adapts pretrained TSFMs to time series classification. Rather than pretraining a new foundation model, R2R improves how existing TSFMs are post-trained: it introduces a decoupled query encoder to extract discriminative patterns and retrieve instance-level numerical knowledge from a constructed knowledge base. The retrieved explicit knowledge is then fused with the TSFM’s implicit parametric embeddings, enabling joint knowledge transfer at both the parameter and feature levels. In this way, R2R allows TSFMs to recall task-relevant patterns learned during pretraining, improving the efficiency of knowledge utilization in downstream classification transfer. Experiments on 158 time series datasets show that R2R consistently improves the classification performance of state-of-the-art TSFMs. Ablation and knowledge-enhanced studies further validate the effectiveness of the proposed framework.
[LG-39] Declarative Explanations for Graph Neural Networks: A Demonstration
链接: https://arxiv.org/abs/2610.26104
作者: Cibo Yu,Haitong Tang,Yinghui Wu,Arijit Khan,Tingting Zhu,Tingyang Chen,Xiangyu Ke
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph Neural Networks (GNNs) have proven effective for graph analytical tasks, yet their “black-box” nature complicates the understanding of their layer-wise behavior and hinders their application to large graphs. We demonstrate SliceGXQ, an end-to-end declarative and interactive system for querying and exploring layer-wise explanations for GNNs over large-scale graphs. In particular, SliceGXQ provides (1) an interactive graphical user interface for dataset, model, and explanation configuration; (2) a lightweight SPARQL-like parser that allows users to specify target nodes, layers, explanation modes, and constraints; and (3) an integrated execution and visualization workflow that invokes SliceGX as its underlying explanation engine. We demonstrate how users can diagnose a misclassified node, identify problematic layers and influential messages, and perform targeted model debugging, together with large-scale graph interpretation and fraud-detection analysis. Our demonstration video and source code are available at this https URL.
[LG-40] Learning from a Bandit Learner
链接: https://arxiv.org/abs/2610.26103
作者: Michela Petriconi,Nicolas Menet,Claire Vernade
类目: Machine Learning (cs.LG)
*备注:
Abstract:Classical inverse bandit methods typically aim to recover true reward parameters. In contrast, we infer a demonstrator’s subjective beliefs, specifically its internal estimates of suboptimality gaps. We consider no-regret demonstrators that follow randomized policies, such as Thompson sampling, and characterize a family of algorithms for which per-step inversion is possible. This enables gap estimates to be recovered directly from the arm-selection distribution. Our analysis identifies an inherent stability-coverage trade-off: inversion becomes ill-conditioned as the demonstrator concentrates its probability mass on a single arm, while the number of informative rounds depends on the rate at which its policy converges. We establish conditions under which inversion is reliable and provide theoretical guarantees on the resulting estimation error. Experiments on simulated data support the analysis and illustrate how these conditions affect gap recovery in practice.
[LG-41] Provable Adversarial Detection: Prime Quantization meets Gromov–Wasserstein
链接: https://arxiv.org/abs/2610.26102
作者: Manohar Kaul,Sadbhavana Babar,Aditya Ramesh,Aditya Saibewar,Fumiyoshi Kasahara,Maki Inui,Hiroaki Kurihara
类目: Machine Learning (cs.LG)
*备注: 52 pages,
Abstract:Adversarial vulnerability persists across modern vision architectures from CNNs to vision language models (VLMs), yet existing detection methods rely on heuristics without theoretical guarantees. We address the fundamental question of when adversarial perturbations can be provably detected from a geometric perspective. Our key insight is that adversarial perturbations cannot simultaneously preserve geometric structure across spaces with fundamentally different properties. Accordingly, we construct two such complementary metric spaces. First, we use a standard CNN embedding space Z , where adversarial samples exhibit significant displacement patterns. Second, we build a novel prime-quantized space P , that absorbs small perturbations through number-theoretic discretization, resulting in minimal displacement, while preserving discriminability. We then leverage the geometric discrepancies across spaces Z and P to detect adversarial samples. To the best of our knowledge, we establish the first rigorous separation theory for adversarial detection, proving that adversarial samples create unavoidable geometric inconsistencies across both spaces. Our framework provides theoretical guarantees including pixel-level absorption bounds, neighborhood diameter concentration, Gromov-Wasserstein (GW) separation theorems, and practical risk control. Extensive experiments validate our theoretical predictions and achieve consistently strong detection performance across a wide range of attack types and model families.
[LG-42] Multilingual Voicemail and Answering Machine Detection via Pretrained Audio Embeddings and Gradient Boosting Machines
链接: https://arxiv.org/abs/2610.26100
作者: Kemal Altwlkany,Muamer Paric
类目: ound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
*备注: Presented at SPECOM 2026
Abstract:Answering machine detection (AMD) is a fundamental task in telephony that classifies whether an answered call is picked up by a human or a voicemail (machine). Despite its widespread use across cloud communication platforms, existing approaches rely on proprietary systems with undisclosed architectures, or require computationally expensive models that cannot operate in real time. We propose a lightweight AMD system that combines pretrained audio embeddings with gradient boosting machines, trained to prioritize the correct detection of human-answered calls. We evaluate seven models across two datasets: three models that we propose, a proprietary industry model, two prior academic approaches, and a large language model baseline that relies on automatic speech recognition. Furthermore, we release OpenAMD, the first publicly available AMD dataset comprising of more than 500 recordings in 7 languages. Our best model achieves the highest precision across both datasets while maintaining low inference latency. We assess the robustness of models to additive noise and packet loss, and perform tests of statistical significance to discover per-language and per-gender model performance. Finally, we evaluate our model on an out-of-domain task of detecting robocalls, where it achieves 94.34% recall without any additional training, which is comparable to the industry standard.
[LG-43] LeakBench: Auditing Reward Hacking through Prediction-Time Leakage in Clinical Modeling Agents
链接: https://arxiv.org/abs/2610.26088
作者: Xiaoyuan Bao,Jingsi Jiang,Jiao Shan,Yulong Cao,Haoyu Wang,Zida Liu,Jun Gao
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 28 pages, 4 figures. LeakBench implementation code is not included in this arXiv version. Public companion software archives provide non-identifying aggregate evidence for two cited real-world motivating results
Abstract:Autonomous clinical-modeling agents can optimize predictive performance while violating conditions that make a model usable. We introduce LeakBench, an executable, guideline-derived audit suite that tests agent decisions for prediction-time leakage, calibration, site-level evaluation, support awareness, and decision-curve net benefit. LeakBench combines a synthetic task with known failure structure and two real-data settings: MIMIC-IV (430k admissions) and governed multi-hospital inpatient record front-page (MHFP) data. Our primary analysis uses reward hacking descriptively, without attributing intent: an agent selects prediction-time-invalid inputs because they improve its validation reward. A deterministic optimizer and every evaluated unconstrained local-model configuration select a fatal leaking feature. In fully logged September 2026 A/B runs, DeepSeek Flash and GPT-6-astra each leak in 10/10 unconstrained runs and 0/10 runs with an explicit admission-time constraint. This contrast is version-sensitive: a blinded Claude/Fable-5.1 chat run does not leak in either condition, while explicit-constraint violations appear only in a historical August 2026 slice lacking raw trajectories. Replacing AUROC with AUPRC does not remove leakage selection. Renaming or disguising the leaking inputs changes neither selection nor provenance-based detection; a name-based filter fails. A deterministic provenance verifier-in-loop removes the leak, making the LLM repair loop defense-in-depth. Specific verifier feedback clears fatal leakage in a hosted agent, and a three-seed ablation finds principle-only feedback clears decision-level defects, though these mitigation results are limited. LeakBench treats trustworthy clinical prediction as a property of an agent’s modeling process, not merely its score; it is an audit testbed with known-failure controls, not yet an expert-validated benchmark.
[LG-44] FinHopper: A Progressive Multi-Hop Benchmark for Complex Financial Agents with Step-wise Evaluation EMNLP2026
链接: https://arxiv.org/abs/2610.26036
作者: Yuxuan Wu,Miaopei Lin,Qiufang Ying,Lin Dan,Hao Zhao,Qirui Liu,Yuteng Shen,Shaohui Wu,Weihong Luo,Xiku Du,Xing Tang,Hao Chen
类目: Machine Learning (cs.LG)
*备注: Accepted to EMNLP 2026. 28 pages, 7 figures, 8 tables. Benchmark: this https URL
Abstract:Financial agents are expected to answer factual investment queries through open search, but financial multi-hop retrieval differs from general multi-hop QA in a crucial way: each intermediate answer is an exact typed constraint, such as an entity, date, event, or numerical indicator, that determines the next query. This strict answer-conditioned dependency makes small intermediate errors cascade into plausible but wrong search trajectories. We introduce FinHopper, a benchmark for complex financial agents with controlled reasoning depth from 1 to 6 hops, covering stocks, funds, and macroeconomic facts, with gold intermediate answers for process-level evaluation. To construct FinHopper at scale, we design a state-transition-based synthesis framework that composes verified atomic financial QA patterns into dependency-preserving chains, followed by targeted in-synthesis verification, model-ensemble checking, and public-search validation by financial experts. Experiments with 20 LLM-agent configurations show that public search tools are necessary but insufficient: performance drops sharply as hop depth increases, the best model remains far below human experts, and process-aware scoring reveals partial reasoning progress hidden by final-answer-only evaluation. The benchmark is available at this https URL.
[LG-45] runcate or Not? A Scaling Study of Looped Transformers with Truncated Backpropagation
链接: https://arxiv.org/abs/2610.26030
作者: Xuyuan Xiong,Siyu Chen,Zihan Shao,Yue Su,Weiyang Jin,Ping Luo,Zhuoran Yang,Xihui Liu
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Looped transformers reuse a shared block of layers to increase computational depth without increasing the number of trainable parameters. They are trained with backpropagation through time (BPTT). To save training compute, BPTT is often truncated to the last few loops, although this may weaken the benefit of looping. To study this tradeoff systematically, we jointly vary model size, training tokens, and forward and backward loop counts across nearly 600 training runs and fit a unified scaling law that captures the effects of looping and truncation through effective parameters. Despite the training-compute savings of truncation, our fitted law favors full BPTT when forward and backward loop counts are jointly optimized. In matched-compute runs, full BPTT through fewer loops consistently outperforms truncated settings of the same compute. The law further shows that at fixed model size, the compute-optimal loop count and training-token budget both grow with compute. When model size is also free, the law identifies two fully trained loops as the best looped setting. Experimentally, this setting achieves lower loss than the compute-optimal non-looped model with about 27% fewer non-embedding parameters and only 8% more training compute.
[LG-46] ProtGlycanDock: Towards Accurate Protein-Glycan Docking with Tailored Dataset Benchmark and Model NEURIPS2026
链接: https://arxiv.org/abs/2610.26009
作者: Minghao Xu,Jiaze Song,Jinyu Wang,Wentao Zhang
类目: Machine Learning (cs.LG)
*备注: Accepted by NeurIPS 2026 Evaluations and Datasets Track. Datasets and source codes are released
Abstract:Performing accurate protein-glycan docking is critical to understanding protein-glycan interactions which govern many biological processes. However, this important task is hampered by the lack of tailored datasets, benchmarks and models for protein-glycan docking. To fill this blank, we first curate a standard benchmark ProtGlycanDock for training and evaluating protein-glycan docking models under both protein and glycan generalization settings. On this benchmark, we evaluate ten models in three categories, physics-based docking methods, machine-learning-based docking models and molecular structure prediction models. To obtain more precise binding poses, we further propose two finetuning techniques to inject the knowledge of glycan conformations into a general-purpose molecular structure prediction model. These two techniques respectively augment model inputs with glycan topological features and refine glycan binding pose with a glycan energy loss. Benchmark results show the superiority of AlphaFold 3 among existing methods, and the two proposed finetuning techniques can be well combined to achieve a state-of-the-art model for protein-glycan docking. Also, we perform stratified benchmark analysis and case studies to better understand the strengths and weaknesses of current models, providing insights for future model improvements. We maintain datasets at this https URL and source codes at this https URL
[LG-47] Scenario Generalization and Sector Transfer in Real Surface Irrigation Surrogate Modelling with Tabular Models ECML ACL KDD2026
链接: https://arxiv.org/abs/2610.26005
作者: Alessandro Cagiano,M. Aybars Seyhan,Fabiola Gangi,Matteo Zignani,Daniele Masseroni
类目: Machine Learning (cs.LG)
*备注: Accepted at MACLEAN 2026 (Workshop on Machine Learning for Earth Observation), held in conjunction with ECML PKDD 2026
Abstract:Surface irrigation planning requires screening many operating conditions and spatial configurations, while high-fidelity shallow-water-equation (SWE) simulations are expensive. We study tabular surrogate models for predicting maximum water height from drone-derived field morphology, terrain, hydrological, spatial, and scenario descriptors, where each row represents a spatial cell under one irrigation scenario. To assess deployment feasibility, we evaluate gradient-boosted trees, neural tabular models, AutoML, and tabular foundation models (TFMs) under two regimes: same-sector generalization, where a model predicts unseen scenarios after observing a limited set of configurations in the same sector, and sector-out transfer, where knowledge learned from source sectors is transferred to an unseen sector. The main findings are twofold: a) under the evaluated fixed 8/32 same-sector split, surrogates achieve strong performance, especially with raw spatial coordinates; b) sector-out transfer is substantially harder because the target sector has different coordinates and geometry. Sector-frame normalization of spatial coordinates improves transfer across similarly structured sectors, but does not solve cross-shape generalization. Overall, tabular surrogates appear useful for same-sector scenario screening under this fixed split, while robust transfer to unseen sectors remains open.
[LG-48] SPLASH: Contrastive Audit Pressure for Detecting and Localizing Security Hallucinations in Generated Code
链接: https://arxiv.org/abs/2610.26001
作者: Mirza Masfiqur Rahman,MD Ajwad Akil,Imtiaz Karim,Ashish Kundu,Ramana Kompella,Elisa Bertino
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注:
Abstract:Code generated by large language models often passes every functional test while remaining insecure. These security hallucinations are common, written with high confidence, and repeated across samples, so neither the generator’s own uncertainty nor static analyzers reliably catch them. We introduce SPLASH, which is guided by a new measure called audit pressure. It contrasts teacher-forced log-probabilities for the generated program under a security review and a style review. Aggregated to statements and programs, and grounded in a probabilistic model of competing edits, audit pressure allows SPLASH to choose the secure program among a model’s candidate programs and locate the vulnerable statement, with no model training, CWE-specific rules, or internal access to the generating model. On 63,684 programs generated by 12 models for 899 tasks, SPLASH picks secure code in 74% of cases, places the vulnerable statement first in 58% of insecure programs, and chooses better than a frontier LLM judge.
[LG-49] PnP-CF: A Consensus Flow-Matching Solver for General Inverse Problems NEURIPS2026
链接: https://arxiv.org/abs/2610.26000
作者: Dibyanshu Kumar,Magda Gregorova
类目: Machine Learning (cs.LG)
*备注: Short version accepted at PriGM Workshop, NeurIPS 2026
Abstract:Plug-and-Play (PnP) methods using continuous-time deep generative models as learned data distribution priors enable synthesizing high-quality solutions of model-based inverse problems. The quality of the solutions depends critically on the interplay and consensus between the inverse problem solver and the generative model sampler. We revisit a recently introduced flow-based PnP framework, which we refer to as PnP-CF (Consensus Flow), and extend its application to nonlinear inverse problems, including those with learned neural-network forward operators. PnP-CF combines a flow-matching denoiser with (i) ADMM-like dual variable updates and (ii) a time-adaptive pullback onto the intermediate flow probability path that combines a stochastic and a deterministic component. We discuss how these two features stabilize the algorithm and promote consensus, and demonstrate on a set of experiments that PnP-CF not only solves standard linear inverse problems with quality at least comparable to state-of-the-art baselines, but also extends gracefully to nonlinear inverse problems, including forward operators realized by pretrained neural networks, for which existing methods produce unsatisfactory results or fail completely.
[LG-50] Internalizing Chain-of-thought into Looped Transformers
链接: https://arxiv.org/abs/2610.25998
作者: Kehang Zhou,Pinyan Lu
类目: Computational Complexity (cs.CC); Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注: 20 pages, 1 figure
Abstract:Chain-of-thought (CoT) enables constant-bit-size transformers to perform general-purpose computation through generated tokens (Li and Wang, 2025a), whereas looped transformers carry out the same computation by recurrence in hidden states (latent thought). We construct a family of constant-bit-size looped transformers that interpolates between CoT and latent thought. For a prescribed loop count L=L(n) and a fixed integer k\ge1 satisfying s(n)^1/k=\Omega(L) , our transformer simulates a time- t(n) , space- s(n) Turing machine using a context window of O(s(n)/L) and O(t(n)s(n)^1/k/L) generated tokens, reducing both resources by a factor of L relative to the corresponding unlooped queue-machine simulation. The construction also holds with softmax attention at constant bit-size, under the finite-precision kernel specified in this paper.
[LG-51] Graph-to-Table Interfaces Control Relational Generalisation
链接: https://arxiv.org/abs/2610.25993
作者: Alessandro Cagiano,Lorenzo Molfetta,Gianluca Moro,Matteo Zignani,Pasquale Minervini
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph-to-table systems combine three design choices: an encoder that maps relational evidence to feature rows, a context constructor that selects labelled examples, and a tabular decoder. We investigate whether these components can be evaluated independently through matched comparisons that change one component while holding the rest of the pipeline and the evaluation constant. Across logistic regression, CATBOOST, and TABPFN-3 on CLUTRR, 57 GraphLog worlds, and 57 knowledge graphs, we show that conclusions about one component can change when the others change. On CLUTRR, absolute positional encodings maintain all distinctions among the evaluated queries and fit the short-path setting nearly perfectly; yet, they generalise worse than relation-count features because the supported prefix acts as a shortcut that breaks down on longer paths. On GraphLog, encoder preference depends on the decoder, while the relation-count representation degrades as CatBoost tree depth increases. Across different knowledge graphs, adding the frozen ULTRA score to local features is consistently useful, whereas adding local features to the score has a smaller and less stable effect. A pointwise TABPFN-3 readout using the full encoder underperforms native ULTRA on 48 of 57 datasets. These results show that encoder assessment depends jointly on context and decoder, while panel-level conclusions can also depend on evaluation and aggregation protocols.
[LG-52] Maestro: Agent ic Formal Proof Development by Coordination
链接: https://arxiv.org/abs/2610.25920
作者: Haoyu Li,Zeren Shui,Yuhao Zhang,Ferhat Erata,Benjamin Mikek,Murali Krishna Ramanathan,Hao Zhou
类目: Machine Learning (cs.LG)
*备注:
Abstract:Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard. Existing prover systems take a divide-and-conquer approach that recursively decomposes hard proof obligations into easier lemmas and splits again any lemma that resists proof. However, a lemma can resist proof for different reasons, including insufficient guidance, a false statement, or a flaw in the decomposition that produced it. Existing systems cannot tell these apart as a failed attempt reports little beyond its failure, and responses are confined to the failing node without considering the rest of the proof or revising the plan above it. The system may thus keep decomposing obligations along a flawed proof direction until the budget runs out, causing the whole proof attempt to fail. We introduce Maestro, a Lean prover harness that handles failures through global coordination. Provers report why they failed, not just that they failed, and a coordinator that sees the whole proof chooses an appropriate response, such as guiding a stuck prover, correcting a false lemma, or retracting a flawed subtree and replanning it. The same harness extends to verified code synthesis, where a checked counterexample sends the code back for repair. We evaluate Maestro with four frontier LLMs on seven mathematics and three software verification benchmarks. It proves more targets than each model’s native coding agent (Claude Code or Codex) on every benchmark the agent has not already saturated, with the largest gains on research-level mathematics. With Claude Opus 5, it discovers its own proofs of 34 of the 48 theorems in the June 2026 release of arXivLean, against 19 for the native agent. It also outperforms all published prover systems on every benchmark they share, solving 669/672 PutnamBench, 59/60 Lean-IMO-Bench, and 6/6 IMO 2026.
[LG-53] An Analysis of Semantic and Spatio-Temporal Co-evolution for Edge Classification on Dynamic Text-Attributed Graphs
链接: https://arxiv.org/abs/2610.25906
作者: Tyler Bonnet,Marek Rei
类目: Machine Learning (cs.LG)
*备注:
Abstract:Dynamic edge classification on text-attributed Web graphs, such as review networks, community forums, and news-derived knowledge graphs, requires high quality, adaptive semantic representations in the context of network evolution. Existing dynamic graph architectures typically initialize text feature vectors with pretrained language models and hold them static during training to preserve their semantic structure, forgoing the potential benefits of allowing them to co-evolve with network topology. In this work, we investigate the effects of enabling such co-evolution with a unified text-graph alignment framework that integrates a lightweight Transformer adapter to update token-level text representations via spatio-temporal gradients from a contrastive pretraining objective. To prevent detrimental semantic distortion, a language model maintenance objective anchors the evolving text embeddings to their original semantic structure. Evaluations on the Dynamic Text-Attributed Graph Benchmark demonstrate that this anchored text-graph co-evolution consistently improves upon the edge classification performance of frozen-text baselines across diverse Web-related domains. Representational analyses reveal that the framework reduces anisotropy, improving the linear separability of text embeddings, and effectively encodes structural and temporal information into the semantic space. A computational efficiency analysis confirms that this beneficial semantic and spatio-temporal co-evolution is achieved without incurring the prohibitive computational overhead typical of end-to-end joint training.
[LG-54] Learn the Interface Not the Implementation: Output Representation Dominates Training Regime in LLM -Based RTL Generation
链接: https://arxiv.org/abs/2610.25881
作者: Siddharth Patel,Vikram Singh,Siddharth Savyasachi Malu,Rohit Singh
类目: Machine Learning (cs.LG); Hardware Architecture (cs.AR)
*备注: 6 pages, 3 figures. Accepted for presentation at VLSID 2027
Abstract:Large-language-model register-transfer-level (RTL) generation has concentrated on how to train the model: fine-tuning, continued pre-training, and reinforcement learning from a verifier. We argue this is the second-order question. On a bounded, formally specifiable task of generating single-error-correcting, double-error-detecting (SECDED) ECC memory wrappers, we hold the task, data, and correctness oracle fixed and vary two axes independently: the training regime (seven settings) and the output representation (four levels, from fully expanded flat Verilog to an eight-field configuration expanded by a deterministic generator). Because a reference generator exists, correctness is decided by formal equivalence over a design space of approximately ten million points, rather than sampled from a hand-written testbench. Two results follow. First, representation dominates: every representation close to raw RTL reaches a pass rate of at most 0.17, while the configuration reaches 0.83 with no training at all. Once the model emits the configuration, every training regime achieves a pass rate between 0.89 and 1.00. Second, raw RTL hits a hard output-length limit: a 512-bit design requires 44,237 tokens of flat Verilog compared with just 69 tokens for the configuration. Consequently, raw generation becomes infeasible above 32-bit data words rather than merely inaccurate. A model trained from scratch with fewer than 50 million parameters, combined with the verified generator, achieves 0.89 formal equivalence, compared with 1.00 for a 32-billion-parameter fine-tuned model, while using approximately 1600 times less inference energy per correct design. Inference energy also dominates amortized training energy. The lesson is to learn the interface, not the implementation.
[LG-55] Asynchronous On-Policy Self-Distillation
链接: https://arxiv.org/abs/2610.25880
作者: Xidong Wu,Qi Qi,Yuqiong Ji,Shangqian Gao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Unlike on-policy distillation (OPD), which depends on a separate teacher model, on-policy self-distillation (OPSD) leverages a single language model acting simultaneously as the student and the teacher, with the teacher conditioned on privileged information. Despite this architectural simplicity, synchronous rollout generation dominates overall training time, creating a severe systems bottleneck when scaling OPSD. To address this, we introduce Asynchronous On-Policy Self-Distillation (AsyncOPSD), a framework that decouples rollout generation from learner updates to improve training throughput. Since the teacher and student share parameters, AsyncOPSD eliminates the multi-model coordination overhead inherent to OPD architectures in asynchronous training. As the student policy evolves, buffered trajectories generated by older policies become stale, potentially degrading performance. To mitigate this, we introduce a distillation objective based on staleness-aware policy control and token-level entropy weighting. Across three model sizes and mathematical reasoning benchmarks, AsyncOPSD improves the training throughput by 1.96 – 2.25\times over synchronous OPSD while increasing average benchmark accuracy.
[LG-56] DeltaTrace: Efficient Signed Attribution for Reasoning Language Models
链接: https://arxiv.org/abs/2610.25877
作者: Chen Liang,Daoping Wu,Donghua Yang,Zhiyu Liang,Ying Zhang,Shuguang Xiong,Hongzhi Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study how words in the context support or oppose a language model’s response and present DeltaTrace, a novel attribution framework that explains the effect of input tokens on a fixed model response. DeltaTrace measures how the response’s log-probability changes when the context tokens being explained are masked. To distribute this difference among source tokens, it constructs operator-specific local rules for finite changes and composes them backward through the computational graph using effective secant slopes. Each token is assigned a signed attribution as the inner product of its embedding displacement and the corresponding backward-propagated slope. Local conservation makes these contributions sum to the measured score difference and preserves their meaning when aggregated over sentences or passages. DeltaTrace supports both attention and linear-attention models, with auxiliary memory growing linearly with sequence length at fixed model dimensions and chunk size. On the evaluated attention model, DeltaTrace achieves the lowest deletion-faithfulness scores (RISE and MAS) on most retrieval and reasoning tasks and the highest needle-in-a-haystack Recall, exceeding FlashTrace by 5.67-24.33 percentage points. On the evaluated hybrid attention and linear-attention model, DeltaTrace lowers RISE on 9 tasks and MAS on 12 relative to FlashTrace.
[LG-57] Residual-Guided Adaptive Tucker Rank Allocation for Parameter-Efficient MCI Conversion Prediction
链接: https://arxiv.org/abs/2610.25873
作者: Guanghua He,Hancan Zhu,Gaohang Yu,An Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Allocating limited trainable capacity across convolutional layers is a central challenge in parameter-efficient fine-tuning (PEFT) of pretrained three-dimensional (3D) networks. We propose Residual-Guided Adaptive Tucker Rank Allocation, which jointly selects input- and output-channel ranks under a fixed adapter-parameter budget while preserving the spatial kernel modes. To guide rank selection, we learn task-specific proxy updates from downstream training data and approximate them using candidate Tucker rank pairs. A perturbation analysis bounds the output discrepancy between the fixed proxy network and its Tucker approximation in terms of layerwise reconstruction residuals, motivating residual-based allocation. We normalize each squared residual by the corresponding proxy-update energy to compare relative approximation errors across layers. Dynamic programming minimizes their sum under a global budget constraint that accounts for both channel factors and the Tucker core. This yields a globally optimal allocation over the specified candidate rank pairs. We evaluate the method on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort for predicting conversion from mild cognitive impairment to Alzheimer’s disease within 36 months, using hippocampal segmentation pretraining. Across five random seeds, the proposed method achieves a mean area under the receiver operating characteristic curve (AUC) of 0.7706, compared with 0.7239 for fixed-ratio Tucker adaptation and 0.7430 for full fine-tuning. The total number of trainable parameters is limited to approximately 0.618 million, accounting for only 1.45% of the full model. Adaptive Tucker rank allocation thus provides an effective approach to parameter-efficient transfer learning in 3D medical imaging.
[LG-58] Statistical Learning Theory for Direct Preference Optimization
链接: https://arxiv.org/abs/2610.25856
作者: Joel Wills,Yuanfan Li,Yiming Ying,Ding-Xuan Zhou
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 16 pages, including references and appendices
Abstract:Direct Preference Optimization (DPO) learns from pairwise preferences through a logistic objective whose score-difference parameterization follows the Bradley–Terry model. We study how failure of this modeling assumption limits DPO, distinguishing structural misspecification from limited model capacity and finite-sample estimation. Our main contribution is a characterization of structural approximation error through the triangle-curl of the preference log-odds, which vanishes exactly when preferences admit a Bradley–Terry representation. We upper bound the achievable excess logistic risk by a score-class approximation term and a triangle-curl term, and establish a lower bound proportional to triangle-curl under additional boundedness assumptions. For ReLU score models, including clipped models, we derive explicit approximation rates showing that increasing capacity can eliminate the score-approximation term while leaving the structural obstruction unchanged. We further establish finite-sample bounds for clipped empirical risk minimizers that separate network approximation, structural misspecification, and estimation error. We place these results within a surrogate-risk framework by identifying contextual preference accuracy (CPA) as a natural measure of agreement between preference judgments and changes in pairwise probability ratios relative to the reference policy. Standard logistic consistency and calibration results connect DPO’s objective to this policy property and transfer our excess-risk upper bounds to CPA, without treating CPA as a complete measure of generated-response quality.
[LG-59] ACE-Trader: Adversarial Co-Evolution for Generator-Critic Trading Systems
链接: https://arxiv.org/abs/2610.25847
作者: Chang Zhou,Jiarun Zhou,Xingtong Yu,Hong Cheng,Hui Zhang,Xinming Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Generator–critic trading systems pair an LLM-based trader that makes trading decisions with a red team agent that evaluates these decisions and provides feedback for revision. However, improving individual trading decisions does not ensure that both agents learn from experience and improve together over time. This raises an important question: how can the trader and red-Team select useful experience and learn from each other’s evolving behavior? On one hand, experience that benefits decisions under one market condition may become ineffective or harmful under another. On the other hand, trading outcomes reflect both agents’ actions, making it difficult to derive appropriate experience for each role. In this work, we propose ACE-Trader, an adversarial co-evolution trading framework. Specifically, we first develop separate context-dependent memory selectors that learn to identify useful experience for the current trading decision. Both selectors receive shared feedback from net-return differences between complete memory-assisted and memory-free trading pipelines, with online updates emphasizing recent outcomes. We further design a co-evolution mechanism that links role-specific memories through shared trading records. We use these records to evaluate the Red-Team’s risk assessments and the effects of the Trader’s revisions against realized market outcomes, then derive experience for their respective memories. By linking this experience to the corresponding records, we enable the Trader to improve its proposals based on prior critiques and the Red-Team to adjust its review focus as the Trader’s behavior changes. Experiments on U.S. stock trading show that ACE-Trader outperforms all 22 active baselines in cumulative return and achieves the highest Sharpe ratio and lowest maximum drawdown across both active methods and passive benchmarks.
[LG-60] Is it RL or Data? A Controlled Comparison of GRPO and Rejection-Sampling Fine-Tuning
链接: https://arxiv.org/abs/2610.25812
作者: Lakshya Chaudhry
类目: Machine Learning (cs.LG)
*备注: 20 pages, 4 figures, 11 tables
Abstract:Bauer et al. (2026) showed that in low-data reinforcement learning with verifiable rewards, training GRPO on mixed-difficulty prompts gives up to 5x the sample efficiency of easy-only prompts. They did not test whether this benefit is specific to GRPO, or whether another post-training method given the same data would obtain it too. We test this on their procedural counting task with Qwen3-4B-Base and LoRA, comparing GRPO to rejection-sampling fine-tuning (RFT) on the same prompts, verifier and token cap. Six primary arms cross the two methods with three prompt sets (Easy, Mixed, and a Curated subset of prompts the base model sometimes solves, which makes GRPO’s implicit filter explicit). We use three seeds per arm. Hypotheses and falsifiers were registered before training. To rule out a poorly tuned baseline, RFT receives a 27-run hyperparameter sweep, while GRPO uses Bauer et al.'s recipe unchanged. GRPO outperforms RFT at every prompt set: by 10.9 points on held-out test problems with curated prompts, and, in post hoc comparisons with matched rollout budgets, by 8.4 and 9.2 points on the easy and mixed prompts. On test the advantage survives the least favorable treatment of truncated completions; out of distribution, truncation prevents a conclusion. Curating prompts does not help RFT (-2.7 points). Mixed-difficulty data helps RFT (+3.7 points), while for GRPO the evidence is inconclusive. A secondary iterated-RFT arm, which resamples from the current policy, recovers 3.2 of the 10.9-point gap (about 30%). These are controlled measurements on one synthetic task family and one base model with three seeds per arm.
[LG-61] Communication-Efficient Federated Domain Generalization in Edge Networks via Dynamic Prototype Regulation
链接: https://arxiv.org/abs/2610.25810
作者: Yuankun Xia,Hui Wang,Ling Yuan,Ziyu Hong,Di Yin,Yufeng Zhou
类目: Machine Learning (cs.LG)
*备注:
Abstract:Federated learning (FL) is a distributed paradigm for Internet of Things (IoT) edge networks. However, deploying Vision-Language Models (VLMs) over wireless edge networks presents distinct challenges: large model parameters increase uplink transmission latency over bandwidth-constrained fading channels, and non-independent and identically distributed (Non-IID) data exacerbates client drift. Existing feature alignment solutions often introduce auxiliary computations that consume time slots allocated for wireless parameter transmission, increasing the likelihood of straggler dropouts under system delay constraints. To address these issues, we propose DPGR-GCAM, a communication-efficient federated domain generalization framework. By employing a dual-parameter-efficient fine-tuning (PEFT) architecture, we design a channel-aware tiered transmission mechanism that adaptively truncates payloads to maximize node survival rates under fading conditions. To mitigate client drift, we introduce the Gradient-Conflict Aware Modulation (GCAM) mechanism. GCAM operates within the low-dimensional feature manifold to dynamically resolve multi-objective optimization conflicts with minimal edge computation latency, thereby reducing computation-induced network outages. Experiments demonstrate that DPGR-GCAM improves cross-domain generalization and sustains convergence under network fading scenarios with a 10% client participation rate.
[LG-62] Sector-Ring Linear Reservoirs: Sector-Wise Decay-Timescale Allocation With State Reuse Across Candidate Settings
链接: https://arxiv.org/abs/2610.25806
作者: Yudai Ebato,Sou Nobukawa,Yusuke Sakemi,Takashi Kanamaru,Nina Sviridova,Kazuyuki Aihara
类目: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
*备注: 19 pages, 9 figures, 6 tables. Submitted to IEEE Access
Abstract:In reservoir computing, performance depends on whether the fixed dynamics of the reservoir suit the task, and the appropriate memory length and frequency response differ across tasks. The reservoir settings are found by a search on each task, and driving the reservoir with the input and collecting its states for every candidate setting dominates the cost of the search. In a reservoir with a linear recurrence, the dynamics decompose into modes, one per eigenvalue, and the eigenvalue distribution becomes a design variable. We propose the Sector-Ring Linear Reservoir (SRLR), in which the eigenvalue angles are fixed at evenly spaced positions around a circle and one decay timescale, chosen from a few values, is assigned to each sector of consecutive angles. To keep the scale of each mode’s state equal across candidates, the input weights are scaled using the power spectrum of the training input. To reduce the cost of the search, the states of the modes for each decay-timescale value are computed once and reused across candidates. On prediction and reconstruction benchmarks built from chaotic systems, the SRLR achieved, in all eight cases, a lower mean test error than a reference model in which all modes share one decay timescale, and was competitive with nonlinear echo state network and leaky-integrator echo state network baselines. The selected decay-timescale profiles changed with the prediction target and horizon. The state reuse reduced the number of computed mode trajectories in the search by 99.8% relative to per-candidate recomputation.
[LG-63] Detection Augmented Learning In Non-Stationary Bandit Environments
链接: https://arxiv.org/abs/2610.25782
作者: Argyrios Gerogiannis,Yu-Han Huang,Subhonmesh Bose,Venugopal V. Veeravalli
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 36 pages, 1 table, 5 figures, extension of https://arxiv.org/abs/2501.19401
Abstract:We study the problem of non-stationary bandits without knowledge of the environmental changes. We propose Detection Augmented Learning (DAL) a black-box paradigm that comprises three modular components: a bandit algorithm, a change detector, and a forced exploration schedule. This modular design enables its applicability across a broad spectrum of bandit problems, from classical to highly structured settings. First, we provide theoretical insights into DAL’s strong empirical performance: a single regret bound shows that DAL transfers the regret of its stationary input to the piecewise stationary setting, attaining or improving the best known bounds in every setting considered. We further identify conditions under which DAL performs well in drifting environments. Then, we showcase DAL’s practical performance and versatility through extensive experimentation in which DAL consistently surpasses all state-of-the-art methods, both prior-free and prior-based, across diverse non-stationary scenarios in nine bandit settings. Finally, we investigate the effectiveness of the forced exploration and show that a bandit-driven strategy improves performance and shortens detection delay.
[LG-64] Which Answers Agree? Evaluating Self-Consistency from Incomplete Logs
链接: https://arxiv.org/abs/2610.25752
作者: Zhengtao Gui,Zexi Chen,Weining Shen
类目: Machine Learning (cs.LG)
*备注: 20 pages, 2 figures, including appendices
Abstract:Self-consistency answers a question by generating multiple responses and selecting the most frequent answer. Its evaluation requires both correctness labels and information about which responses agree; correctness labels alone generally determine neither voting accuracy nor the benefit of additional votes. For tasks in which all correct responses map to the same normalized answer, we derive sharp accuracy bounds using classical stochastic majorization, refined by abstention counts and answer-category limits. We also derive sharp bounds on accuracy changes across voting budgets that share a response pool. When answer-category counts are available, we introduce Plurality-U, a count-based dynamic program that exactly averages voting accuracy over all fixed-size subsets, eliminating finite-pool Monte Carlo error. On 1,403,520 stored outputs from recent reasoning models in Scorio Lite and Scorio Trace, recording abstention counts and answer options cuts the width of SuperGPQA accuracy intervals by more than half. For moving from 9 to 17 votes, our bounds reduce interval width by a median of 77.7% on SuperGPQA and 72.3% on 14 GPQA and Adaptive-Consistency pools, compared with a baseline that combines separate accuracy bounds. Exact computation has lower paired median runtime than 2,000-subset Monte Carlo in 14 of 15 tested pool-budget settings. These results clarify what incomplete logs can establish and provide an exact evaluation procedure when answer counts are retained.
[LG-65] NADS: Navigator-Guided Data Selection for Mitigating Catastrophic Forgetting in Fine-Tuning NEURIPS2026
链接: https://arxiv.org/abs/2610.25751
作者: Jianhao Zhang,Ou Wu,Yi Du
类目: Machine Learning (cs.LG)
*备注: Accepted by NeurIPS 2026
Abstract:Catastrophic forgetting is a central obstacle in single-stage fine-tuning of large language models (LLMs) on a new task, especially when the original pretraining data are unavailable. Synthetic data offers a practical mitigation path, but candidate pools are often noisy and redundant, making effective selection essential. Existing methods often rely on indirect proxy signals whose alignment with actual training benefit can be limited, while utility–diversity selection over large candidate pools remains computationally demanding. To address these challenges, we propose NAvigator-guided Data Selection (NADS), a fine-tuning framework for LLMs. NADS first follows the target fine-tuning recipe without preservation regularization to obtain a navigator model, whose deviation from the pretrained model exposes recipe-induced capability drift. It then uses the predictive divergence between the navigator and the pretrained model to score forgetting-aware utility for each synthetic candidate, and constructs a constraint set through an efficient utility–diversity selection strategy. Finally, NADS fine-tunes on the new-task data while distilling pretrained behavior on the selected constraints to preserve general capabilities. Experiments show that NADS delivers a stronger balance between new-task performance and general capability preservation, while reducing utility–diversity selection overhead. Code at: this https URL.
[LG-66] Bayesian Model Discovery with Language Models: Closing the Loop between Hypotheses and Experiments
链接: https://arxiv.org/abs/2610.25747
作者: JooKyung Yuh,Byungmin Park,Eunseo Choi,Jinheon Baek,Junmo Cho,Kyu Hyun Choi,Sung Ju Hwang,Kyong Hwan Jin,Hae Beom Lee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Scientific model discovery requires proposing candidate models of an unknown system and designing experiments that discriminate among them. However, LLM-based approaches typically rely on in-context learning to predict outcomes and select experiments, lacking explicit model evaluation to systematically guide hypothesis search, while Bayesian approaches enable principled model comparison but are confined to a predefined candidate set which results in limited search spaces. To this end, we introduce BMD, which closes the loop between LLM-driven hypothesis generation and Bayesian experimental design. Specifically, an LLM proposes model families, while a numerical engine fits their parameters, weighs them by model evidence, and averages their predictions. Then, BMD chooses the next experimental conditions by maximizing the joint expected information gain over families and their parameters, while feeding the updated beliefs and residual diagnostics back to the LLM to guide its next proposals. Across four scientific discovery benchmarks, BMD achieves lowest final prediction error than in-context learning and LLM-based scientific agents while recovering exact functional forms more often, with ablations supporting numerical feedback and the joint objective.
[LG-67] Fundamental Limits of Information-Theoretic Secure Aggregation with User Dropouts in Federated Learning
链接: https://arxiv.org/abs/2610.25743
作者: Jinbao Zhu,Kai Wan,Giuseppe Caire
类目: Information Theory (cs.IT); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: The abstract was shortened to meet arXiv’s requirements. The full version is included in the manuscript
Abstract:Secure aggregation enables a server to compute the sum of users’ gradients or model updates without revealing individual values. With user dropouts, correlated random keys protect local gradients while allowing recovery of the desired gradient sum. We consider three phases: key distribution, masked-gradient upload, and mask elimination. Users establish correlated keys, upload masked gradients, and provide additional messages to eliminate residual aggregate masks. Existing information-theoretic schemes typically design mask-elimination communication for a prescribed dropout threshold and evaluate its cost under worst-case dropouts across the masked-gradient upload and mask-elimination phases. Such designs cannot adjust communication to actual dropout scenarios. Moreover, the randomness and communication costs of establishing correlated keys remain largely unexplored. To address these limitations, we formulate an adaptive secure aggregation framework under the T-colluding security constraint and construct a single scheme that supports all feasible dropout scenarios by adapting its communication to the actual number of dropouts. We establish converse bounds on four resources: key randomness, key-distribution communication, masked-gradient communication, and mask-elimination communication. The latter two bounds are characterized for each realized dropout set in the masked-gradient upload phase. The proposed scheme achieves the optimal masked-gradient communication rate for every feasible dropout scenario and the optimal mask-elimination communication rate in almost all feasible regimes. Its key rate and key-distribution communication rate are each optimal or within a factor of two of their respective information-theoretic optima. These results characterize the fundamental randomness and communication costs of secure aggregation with user dropouts.
[LG-68] How Much Does One Wrong Answer Cost? Error Propagation in Constraint Pruning with Noisy LLM Oracles
链接: https://arxiv.org/abs/2610.25739
作者: Zhengtao Gui(University of California, Irvine),Weining Shen(University of California, Irvine)
类目: Machine Learning (cs.LG)
*备注: 33 pages, including appendices
Abstract:Constraint pruning saves queries to a large language model (LLM) or another noisy oracle by skipping each yes/no question implied by answers accepted earlier, for example through transitivity. Entity-resolution systems and LLM-based rankers use this idea. However, their analyses typically assume a correct oracle. We argue that pruning should be judged by its risk when the oracle errs, since an accepted wrong answer spreads to every question it implies. When each answer is wrong independently with small probability \varepsilon , the first-order risk is set by how many facts one wrong answer corrupts on average. We consider two kinds of constraints. For implication rules, as in taxonomies, pruning in a random order adds no first-order risk: a wrong answer that would change k facts is accepted with probability 1/k , so it costs one fact on average. Rules with s premises break this balance: a wrong answer at a premise shared by r rules can cost order r^1-1/s facts. For entity resolution with m records of one entity, the first-order risk equals \varepsilon times the Wiener index (the total pairwise distance) of the spanning tree built by the query order. Pivot orders, which first compare one record with all others, amplify errors by less than 2 and are optimal; random orders by order m^1/3 ; the common most-similar-first order is the worst in one dimension and, for large m , worse than random in two. With several entities, the cost of wrongly merging two does not depend on how many others exist. Simulations agree with every predicted constant within sampling error.
[LG-69] Neural Video Coding under Limited Delay: Theoretical Optimality and System Design
链接: https://arxiv.org/abs/2610.25729
作者: Likun Cai,Jun Chen,Ashish Khisti
类目: Machine Learning (cs.LG); Information Theory (cs.IT)
*备注: 48 pages, 16 figures
Abstract:Traditional neural video compression (NVC) methods predominantly follow strictly causal predictive-coding paradigm, treating zero-frame delay as a fixed constraint rather than delay as an explicit design parameter. From an information-theoretic perspective, allowing limited frame delay enables a codec to exploit correlations with future frames within the delay window, thereby reducing uncertainty that is inaccessible to strictly causal prediction and potentially improving the rate-distortion tradeoff. In this work, we formalize this intuition through a rigorous rate-distortion analysis builds on the clique factorization of Markov networks, and propose Graph-Factored Sequential Coding (GFSC), a principled framework for sequential video coding under limited delay. To our knowledge, we provide the first theoretical result showing that, for general K -order Markov sources with positive distributions, a K -frame delay is sufficient to attain joint-coding rate-distortion optimality over a nontrivial distortion region, a guarantee unavailable to zero-delay causal coding in general. In practice, we instantiate GFSC through lightweight neural modules, allowing existing causal predictive codecs to be augmented without modifying their core architectures. Experiments across multiple NVC models and datasets show that GFSC consistently improves rate-distortion performance over the corresponding causal counterparts. Under matched delay budgets, GFSC also achieves competitive performance compared with representative bidirectional codecs.
[LG-70] Learning to Couple: Context-Adaptive Joint Denoising for Diffusion Language Models
链接: https://arxiv.org/abs/2610.25723
作者: Zilei Shao,Ian Li,Heng Zhao,Kewei Lian,Guy Van den Broeck,Zhe Zeng,Anji Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Diffusion language models accelerate generation by predicting multiple tokens in parallel, but factorized denoisers treat simultaneous predictions as conditionally independent. Decoding heuristics reduce the resulting errors by controlling which tokens are committed, but leave the underlying distribution factorized. Joint modeling addresses the source of the problem, yet existing methods attach separately learned layers to frozen denoisers, leaving the backbone unadapted to the joint objective and the coupling parameters fixed across contexts. We propose Codpula, a context-parameterized denoiser combining a Transformer with a tractable probabilistic inference layer. Given a partially masked input, the Transformer produces token potentials that shape per-position predictions and compact gates that modulate the layer’s coupling parameters, yielding a context-dependent joint distribution while retaining exact conditional inference. We jointly optimize both components under a conditional denoising objective using a dual-flow multiplicative update that preserves valid probabilistic parameters. At inference, Codpula supports standard heuristics and uses conditional queries to adapt commitments to dependencies among proposed tokens without extra Transformer evaluations within a step. Codpula sets new state-of-the-art results across a broad suite of reasoning benchmarks while advancing the accuracy-latency Pareto frontier in mathematical reasoning and code generation.
[LG-71] Evidence Sufficiency and Model Comparisons in MexRegQA: An Item-Level Audit of Mexican-Spanish Regulatory QA
链接: https://arxiv.org/abs/2610.25710
作者: Fred Hu,Yeqi Wang(1),Xing Zhou(2) ((1) Huawei Technologies de México, Mexico, (2) Northeastern University, USA)
类目: Machine Learning (cs.LG)
*备注: 26 pages. Companion artifacts are prepared but not yet publicly available
Abstract:Similarity-gated coverage does not establish that retrieved evidence contains the facts needed to answer a question. We audit Mexican-Spanish financial-regulatory QA and find that 35 of 135 covered items lack some or all load-bearing facts. On the post-hoc full-evidence slice (n=98), eleven prompt-matched models score 95-98/98 and none differs significantly from the anchor; this does not establish equivalence. Removing evidence-defective items changes the nominal significance status of 14 of 55 pairwise comparisons. On 56 zero-evidence items, accuracy spans 33.9%-89.3%. Removing evidence from the audited items under an abstention-permitting prompt reduces accuracy by 63-81 percentage points in all ten tested legs, with no recovered items. Building on sufficient-context research, we contribute MexRegQA v1.0: 192 items with provenance and evidence-status labels, frozen retrieval evidence for 191 evaluated items, predictions, judge outputs and offline checks. The audit was outcome-aware, the later model/panel sweep reused the same questions, and all reported scores and labels rely on machine evaluation without independent human validation. We propose releasing evidence-status labels alongside grounded QA benchmarks so that model comparisons can be examined separately from evidence availability.
[LG-72] Predictive Bandit Curriculum Reinforcement Learning via Gradient-Based Sample Group Discovery
链接: https://arxiv.org/abs/2610.25704
作者: Liang Xu,Shufan Shen,Zhongni Hou,Chao Bi,Yufei Zhang,Wei Lin,Guojun Yin,Shuhui Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Curriculum learning, which organizes training samples according to their learning priority, has been widely adopted in reinforcement learning to improve training efficiency. However, existing methods commonly construct curriculum units using predefined data properties and estimate their utility by simply aggregating a single proxy signal from recent training this http URL designs may group samples with different optimization effects and fail to anticipate how their utility evolves in relation to the policy. We propose Predictive Gradient-based Bandit (PGB) Curriculum, which first discovers optimization-aware learning units by clustering projected sample-level gradients. PGB then employs a lightweight temporal predictor to forecast the future utility of each unit and converts the predictions into a Boltzmann sampling distribution for adaptive data allocation. Experiments across multiple model scales and reasoning benchmarks show that PGB consistently outperforms standard GRPO, difficulty-based curricula, and history-based scheduling strategies, demonstrating improved data utilization and training efficiency.
[LG-73] Stochastic Tracking of Drifting Affine Fixed Points
链接: https://arxiv.org/abs/2610.25699
作者: Sajad Khodadadian
类目: Machine Learning (cs.LG)
*备注: 50 pages, 2 figures
Abstract:Linear stochastic approximation (SA) is widely used to estimate fixed points from noisy observations, but most finite-time analyses assume that the underlying operator remains fixed. We study linear SA with time-varying affine operators, where the fixed points change over time and the observation noise may be correlated across rounds. For constant-stepsize updates, we establish a bound on the cumulative expected squared distance between the iterates and the corresponding fixed points. The bound accounts for the initial error, observation noise, and changes in the fixed points over time. We also establish lower bounds for scalar Gaussian models, showing that, under specified conditions, a suitably chosen constant stepsize achieves the smallest possible worst-case cumulative error up to constant factors, even compared with estimators that use future observations. These results further quantify the effect of temporal correlation on estimation accuracy. Since choosing an appropriate stepsize generally requires knowledge of the noise level and how much the fixed points change, we develop an adaptive method that uses an additional bounded, unbiased observation of the fixed point to achieve an almost optimal algorithm without needing the knowledge of the noise level and the fixed point variation over time. Finally, we apply our results to value-function estimation using synchronous tabular temporal-difference learning in changing environments, assuming generative-model access and a common invariant distribution.
[LG-74] LOOP: Recovery-Aligned Iterative Layer Pruning via On-Policy Distillation
链接: https://arxiv.org/abs/2610.25679
作者: Liang Xu,Shufan Shen,Zhongni Hou,Yufei Zhang,Wei Lin,Guojun Yin,Shuhui Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Layer-level pruning is attractive for large language model (LLM) deployment as it reduces the model size while preserving a dense architecture that can be efficiently executed on standard hardware. Current methods typically follow a one-shot paradigm, which first removes selected layers to reach a target model size and then performs additional training to recover model capabilities. However, pruning directly to the target size can substantially degrade reasoning capabilities, making subsequent recovery more difficult under a limited training budget. Moreover, this paradigm separates layer selection from recovery, limiting the use of recovery information to guide selection and potentially hindering the recovery of reasoning capabilities. To address these limitations, we propose LOOP, a Layer-level Recovery-Oriented On-Policy Pruning framework that interleaves pruning and recovery across multiple shots. In each shot, LOOP removes a single layer and performs short on-policy distillation, mitigating the abrupt degradation in reasoning capabilities associated with one-shot pruning. To guide layer selection in subsequent shots, LOOP reuses on-policy trajectories and optimization signals from recovery to jointly assess each remaining layer’s reasoning contribution and recovery sensitivity, thereby making layer selection recovery-aware. Evaluations on several benchmarks show that LOOP consistently outperforms one-shot baselines in average accuracy across different model scales and pruning ratios.
[LG-75] Can Metadata Fix the Gauge? Calibration Turns Sparse Multi-Source Learning from Sparse PCA into Sparse Mean Recovery NEURIPS2026
链接: https://arxiv.org/abs/2610.25663
作者: Yirong Xiang,Yibo Zhou
类目: Machine Learning (cs.LG)
*备注: Accepted as NeurIPS 2026
Abstract:Source labels tell us where a sample came from, but not how that source oriented its coordinates. In a sparse Gaussian multi-source model with source-specific sign gauges, that distinction changes the learning problem. Without calibration, pooling across sources erases the mean signal, and the source-labelled experiment reduces to sparse PCA. With task-independent relative-gauge metadata, synchronization aligns sources and the same observations become a sparse mean problem. We formalize the value of metadata through AlignedMass, which quantifies how much first-order signal a metadata-induced alignment restores. Source labels create only limited AlignedMass, whereas valid calibration can create AlignedMass proportional to the full sample size. This yields a statistical separation in micro-source regimes and, under standard sparse-PCA hardness assumptions, a computational separation as well. Synthetic experiments check the predicted scaling; controlled real-feature and physical sensor studies examine sign cancellation, separately acquired calibration, and reuse across tasks. Comments: Accepted as NeurIPS 2026 Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.25663 [cs.LG] (or arXiv:2610.25663v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.25663 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-76] OPDArena: Understanding Learning and Forgetting in On-Policy Distillation
链接: https://arxiv.org/abs/2610.25651
作者: Yuanze Xu,Tianzhu Ye,Hui Sun,Wenhui Wang,Li Dong,Furu Wei
类目: Machine Learning (cs.LG)
*备注: Code: this https URL
Abstract:On-policy distillation (OPD) is increasingly used for language model post-training, yet how teacher supervision shapes learning and forgetting remains unclear. We present OPDArena, a framework comparing supervised fine-tuning (SFT), reinforcement learning with verifiable rewards (RLVR), and OPD across five pretrained models and diverse teacher sources. It jointly evaluates mathematical learning, general capability retention, and predictive distribution changes. Teacher choice strongly shapes retention: students of RLVR-trained teachers improve on the target task while retaining prior capabilities, whereas students of SFT-trained teachers incur greater capability loss. Across post-training methods, out-of-domain (OOD) distribution change shows a more consistent positive association with forgetting than in-domain change. To understand how teacher choice affects these changes, we examine the additional information teachers provide relative to the student’s pretrained model. Controlled teacher comparisons show that the task-specific preferences teachers acquire through post-training make student updates more concentrated in-domain.
[LG-77] Conditional Averag e Treatment Effect Distillation for Hypothetical State-Intervention Targeting Under Budget Constraints
链接: https://arxiv.org/abs/2610.25627
作者: Vasundhara Acharya,Bülent Yener
类目: Machine Learning (cs.LG)
*备注: Under review at PLOS ONE
Abstract:Personalized public-health targeting must capture heterogeneous intervention benefits while remaining interpretable under resource constraints. Flexible causal-effect models can be hard to operationalize, while coarse subgroup policies may lose individualized effect information. We study budget-constrained hypothetical state interventions in observational data without a logged historical decision policy. For obesity, elevated glucose, and lifetime smoking-history states, causal-structure-informed pretreatment covariates are prespecified, with capacity capped at 70% of those observed in the adverse state. We evaluate causal distillation where Causal Forest and Bayesian Causal Forest teachers estimate conditional average treatment effects (CATEs), which decision tree and random forest students approximate. We compare individualized policies with unsupervised K-means subgroup policies and supervised causal-tree policies with leaf counts matched to K-means cluster counts. Observational policies are assessed on held-out data using a doubly robust estimator, bootstrap uncertainty, and decision-focused teacher-student fidelity measures. Across PIMA obesity and glucose and NHANES smoking-history experiments, each distilled student was within 0.011 absolute utility of its teacher, with allocation Jaccard similarities of 0.738-0.937. K-means and causal-tree policies were also competitive, with the highest point-estimate policy varying by application. Prespecified paired comparisons were nonsignificant after Holm adjustment, while near-budget-boundary agreement was less stable than global fidelity for some teacher-student pairs. Findings support causal distillation for translating flexible effect models into simpler budgeted targeting rules, although observational estimates depend on causal-identification assumptions and require longitudinal or prospective validation.
[LG-78] Zephyrus-R1: Agent ic Reinforcement Learning for Weather Reporting
链接: https://arxiv.org/abs/2610.25625
作者: Fengfei Yu,Marshall Fisher,Isadora White,Sumanth Varambally,Yasaman Jafari,Shivam Yadav,Samuel M. Bartlett,Shawn Roj,Benjamin J. Moore,Michael D. Warner,Luyu Han,Zhirui Xia,Duncan Watson-Parris,Taylor Berg-Kirkpatrick,Yian Ma,Rose Yu
类目: Machine Learning (cs.LG)
*备注: 27 pages, including references and appendices
Abstract:Reinforcement learning (RL) has emerged as a key method for training AI models in scientific applications, but RL remains difficult when outputs cannot easily be verified against a known ground-truth answer. Natural-language weather forecast reports are one such domain: they translate atmospheric data into assessments of weather evolution, hazards, and uncertainty for both the general public and key decision-makers. Unlike numerical weather-analysis questions, they admit many valid outputs, which limits deterministic verification. To address this, we develop Zephyrus-R1, a practical framework for RL on scientific weather forecast reporting. We focus on three key components of the agentic RL pipeline: reward design, curriculum learning, and an analysis of human-expert alignment. To encourage both report legibility and factual accuracy, we design a reward function that combines a meteorological rubric with bidirectional claim entailment, both evaluated against National Weather Service reference reports. Our framework optimizes this reward in a simulated tool-calling environment, enabling vision-language agents to gather and analyze forecast evidence across regions and lead times before producing a report. We use annotations from three meteorologists to audit the claim labels produced by our LLM judges. Finally, a curriculum helps small models first develop tool use and scientific reasoning through RL on verifiable weather-analysis questions, then optimize the open-ended report reward. After training Qwen3.5-4B with Zephyrus-R1, we improve the mean across six judge-metric scores by 26% over the base model and achieve results competitive with frontier open-weight models. For Qwen3.8-27B, we improve the same mean by 13% over the base model, by 6.7% over Kimi-K3, and by 2.7% over GPT-5.6-Luna.
[LG-79] PRISM: Prior Relational Information for Self-supervised Modeling to Enhance Solubility OOD Generalization NEURIPS2026
链接: https://arxiv.org/abs/2610.25605
作者: Xinyi Chen,Jiahuan Pang,Titus Chima,Paul Weng,Wendong Wang
类目: Machine Learning (cs.LG)
*备注: NeurIPS 2026, accepted
Abstract:Solubility depends on intermolecular interactions between solutes and solvents, and is central to drug discovery, chemical and materials process engineering. Its prediction requires out-of-distribution (OOD) generalization to unseen solutes, solvents, and solvent mixtures, where supervised methods suffer from data scarcity, and existing self-supervised methods do not explicitly learn chemically meaningful relations. Learning chemically meaningful relational information without labels can alleviate data scarcity and improve OOD generalization, but remains challenging. We introduce Prior Relational Information for Self-supervised Modeling (PRISM) and apply it to solubility OOD generalization. PRISM is a label-free pretraining framework that converts expert-defined chemical priors into relational teachers. For solubility prediction, it trains four encoders to preserve the population-level relational information from priors capturing hydrophobic, electrostatic, polarizability, and substructural effects, using Kullback-Leibler (KL) divergence to match distance-based neighborhood distributions during pretraining. Across scaffold, solute, and solvent OOD settings, PRISM matches or surpasses supervised and self-supervised baselines. Under mixed-solvent OOD, freezing the four PRISM encoders and training a lightweight fusion-and-composition head outperforms strong baselines and substantially reduces run-to-run variation. Ablations and control studies confirm that gains stem from chemically meaningful relational structures baked into the encoders by PRISM. We expect this strategy of turning relational priors into self-supervision signals to extend beyond solubility prediction to domains where chemically or biologically meaningful relational priors are available.
[LG-80] Adversarial Post-Training of Foundation Models for High-Frequency Trading
链接: https://arxiv.org/abs/2610.25604
作者: Alexandre Bismuth,Kang Li,Aramis Marti-Shahandeh,Valentin Mohl,Jakob Foerster,Anisoara Calinescu
类目: Machine Learning (cs.LG)
*备注: Accepted at the 7th ACM International Conference on AI in Finance (ICAIF '26), Milan, Italy. 9 pages, 5 figures, 4 tables. Code: this https URL
Abstract:While autoregressive foundation models trained on limit order book (LOB) data show state-of-the-art performance for market simulation, pre-training with cross-entropy neither penalizes the compounding of sampling errors over free-running rollouts nor targets desired market properties. We introduce a post-training method for LOB foundation models using a Wasserstein generative adversarial network (GAN) objective. A critic, trained online on rollouts through the true, non-differentiable matching engine, guides updates from evolution strategies over rank-4 LoRA perturbations of the pre-trained generator, under a KL trust region. For efficiency, we use EGGROLL (Evolution Guided GeneRal Optimisation via Low-rank Learning), a memory-efficient variant of evolution strategies. On a held-out panel, adversarial post-training improves LOB-Bench realism by 17.6% (WS-21) and 17.0% (Mean- L_1 ), while selection-matched random-perturbation and compute-matched continued-pre-training controls yield no improvement. Post-training is thus a necessary stage that more pre-training does not replace. Per-position divergence measurements show that it slows error accumulation over 500-message rollouts from 3.1\times to 2.3\times . Adversarial post-training is also inexpensive, requiring under 50M training tokens and 12 GPU-hours per seed, three orders of magnitude below the 48B-token pre-training budget. Ablations over critic inputs, optimizers, and the population-rollout trade-off, together with an analysis of reward hacking, yield practical design guidelines for adversarial post-training.
[LG-81] Distributionally-Robust Policy Learning with Observational Data NEURIPS2026
链接: https://arxiv.org/abs/2610.25591
作者: Debmalya Mandal
类目: Machine Learning (cs.LG)
*备注: To appear at NeurIPS 2026. Generative AI was used only to polish writing and implement the main algorithm
Abstract:Policy learning from observational data is fundamental to applications ranging from personalized medicine to welfare-program targeting, but standard approaches optimize empirical performance and may fail when the deployment distribution differs from the training distribution. We study \emphdistributionally-robust policy learning: given observational data, we seek a policy \pi \in \Pi that maximizes the worst-case policy value over an f -divergence ball of radius \rho around the data-generating distribution. We propose an estimator built on cross-fitted doubly-robust scores for the conditional treatment effect, paired with a DRO solver that handles \textttKL, \chi^2 , and \textttCVaR ambiguity sets. We establish two convergence guarantees. First, the estimated policy attains the standard \widetildeO(n^-1/2) rate on its DRO regret, and inherits the doubly-robust property – the rate continues to hold whenever either the outcome model or the propensity model is consistently estimated, even at slow nonparametric rates. We then propose a sample-splitting bias-correction algorithm derived from the Lagrangian dual, and show that under a Tsybakov-style margin condition on the treatment-effect, the estimator attains a fast rate of \widetildeO(n^-(1+\alpha)/(2+\alpha)) , interpolating between \widetildeO(1/\sqrtn) and \widetildeO(1/n) as the margin parameter \alpha ranges over [0,\infty) . To our knowledge, this is the first work to incorporate a margin condition into the analysis of distributionally-robust policy learning, and the first to obtain fast rates in this setting. Empirically, on synthetic and semi-synthetic experiments calibrated to a welfare-to-work program, distributional robustness yields meaningful improvements over empirical welfare maximization when training and test distributions differ.
[LG-82] Divergence based Ranked Preference Alignment for LLM s
链接: https://arxiv.org/abs/2610.25588
作者: Rajdeep Haldar,Yue Xing,Qifan Song
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Alignment of large language models typically relies on pairwise preference feedback, though ranked feedback can capture more comprehensive information across multiple responses within a single annotation. However, ranked-preference alignment is underexplored in practice compared to simpler pairwise approaches, potentially resulting in suboptimal behavior. Concurrently, recent research has conceptualized large language model alignment objectives as divergence estimators between aligned (preferred) and unaligned (less-preferred) response distributions, resulting in a family of objectives with an interpretable theoretical foundation and strong empirical performance. This work unifies these perspectives by introducing f -RPO, a f -divergence objective for ranked preference optimization that extends divergence-based alignment to ranked feedback and recovers pairwise preference alignment as a special case. Our theoretical analysis establishes rank-order consistency, showing that f -RPO preserves the underlying preference ordering at convergence. Empirical results indicate that total-variation and reverse-KL variants of the proposed objective consistently outperform ranked-alignment baselines in both ranking and generation quality across Qwen, Llama, Gemma models ranging from 0.5B to 8B parameters.
[LG-83] CVTA: Cross-Variable Temporal Attention for Multivariate Irregular Time Series Prediction NEURIPS2026
链接: https://arxiv.org/abs/2610.25584
作者: Ankitkumar Joshi,Milos Hauskrecht
类目: Machine Learning (cs.LG)
*备注: Accepted at the Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS 2026), Main Track
Abstract:Irregularly sampled multivariate time series arise in many real-world domains such as healthcare, climate science, and large-scale monitoring systems. Characterized by asynchronous observations and heterogeneity across variables, such time series pose significant challenges for forecasting. In this work, we introduce Cross-Variable Temporal Attention (CVTA), a short-term, continuous-time forecasting framework that learns prediction functions from its compact, interpretable state summarizing recent temporal dynamics across variables. CVTA uses Variable-Conditioned Embeddings to project heterogeneous variables into variable-specific latent spaces, and applies temporal attention to capture cross-variable dependencies from this state. Furthermore, we demonstrate that CVTA can be integrated as an auxiliary module into existing state-of-the-art prediction models through a dynamic meta-decision model. Extensive experiments on multiple real-world benchmarks show that CVTA achieves strong short-term predictive accuracy, supports efficient online inference, improves existing prediction models when used as an auxiliary module, and learns attention patterns that align well with domain knowledge.
[LG-84] FTFM: Factorized Tabular Foundation Models
链接: https://arxiv.org/abs/2610.25558
作者: Josif Grabocka
类目: Machine Learning (cs.LG)
*备注:
Abstract:Tabular foundation models (TFMs) are becoming a powerful alternative to conventional tabular learning, adapting to new datasets in context without task-specific training. Current architectures, however, face a trade-off: models that preserve cell-level representations are expensive because they repeatedly attend across rows and columns, while faster models collapse each row into a fixed embedding and lose this fine-grained state. We introduce Factorized Tabular Foundation Models (FTFM), which preserve cell representations while factorizing their computation through one contextual latent per row and one per column. Cells no longer attend to one another directly; instead, each cell is updated through the interaction of its row and column latents, reducing repeated attention while retaining feature-specific information. FTFM achieves competitive predictive performance in tabular classification on major benchmarks such as TabArena and BeyondArena.
[LG-85] Beyond the record: evaluating probabilistic AI weather forecasts deep in the tail
链接: https://arxiv.org/abs/2610.25545
作者: Bastiaan Van Velthoven,Cas Decancq,Diego G. Miralles,Stijn Luca,Thomas Mortier
类目: Machine Learning (cs.LG)
*备注:
Abstract:Can probabilistic artificial intelligence (AI) weather forecasts retain skill beyond their training domain? We compare four state-of-the-art AI models (ArchesWeatherGen, U-Cast, WeatherNext2 and GenCast-mini) with the physics-based IFS-ENS over land in 2022, at 1.5° resolution and lead times of 2–10 days. Monthly ERA5 records from 1979–2021 define common historical boundaries, and we characterize new records set in 2022 by their depth, i.e., the exceedance margin relative to local historical variability. Two proper scores, record-threshold weighted CRPS and Brier score, assess the magnitude and probability of record exceedances across all eligible cases, retaining both missed extremes and false alarms. Verified against ERA5, AI models outperform IFS-ENS for heat and cold records at almost every lead and depth. For heat, the advantage strengthens with depth, partly because IFS-ENS over-forecasts deep records, and largely persists for deeper records under HRES-fc0 verification. For wind, gains erode for deeper records and longer leads. Better tail scores can nevertheless coexist with aggregate probability deficits or surpluses, and some improvements arise chiefly on non-exceedances. Increasing underprediction on progressively more extreme observed subsets can also arise through outcome selection, even for an ideal conditional-mean forecast. These findings establish that native probabilistic AI ensembles retain, and for heat increase, their advantage over IFS-ENS as thresholds move beyond local records, although all advantages narrow with lead time. They caution against generalizing the extrapolation failure reported for deterministic AI forecasts to probabilistic systems, and motivate common verification conventions that separate comparative tail skill, performance on realized extremes, and probabilistic calibration.
[LG-86] Reinforcement learning on the EPRI AI-ccelerating Unit Commitment benchmark: diagnosing over-commitment and guiding MILP with learned policies
链接: https://arxiv.org/abs/2610.25544
作者: Maharani Devira Pramita,Aidan O’Sullivan
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Reinforcement learning (RL) is increasingly proposed as a fast solver for unit commitment (UC), but standalone agents remain far more expensive than mixed-integer linear programming (MILP). In the first study of RL on the 51-generator, 72-hour benchmark of the Electric Power Research Institute (EPRI) AI-ccelerating Unit Commitment competition, we trace this gap to the agents’ own exploration, close most of it, and show that the resulting policy can accelerate an exact solver. In every configuration tested, Proximal Policy Optimisation (PPO) and an action-branching Deep Q-Network (DQN) trained from scratch committed about twice the units of a thermal-only MILP optimum, at 39-64% higher cost, whereas a merit-order rule came within 1.6-7% (10-24% when unserved energy is priced). When each unit’s decision is sampled independently, spare units insure against random de-commitments. Consistent with this, injecting exploration noise during fine-tuning enlarged a lean policy’s fleet, a reward without shaping terms did not remove the over-commitment, and under their training noise the agents outperformed the rule subjected to comparable noise. Imitating the MILP solutions and fine-tuning conservatively brought PPO within 2.6% of the optimum, and these policies had the lowest penalised cost of any evaluated (about 5%). Fixing the fine-tuned policy’s confident decisions made the MILP 3.8-8.5 times faster within 0.10-0.25% of the best solution, with a smaller worst case than an imitation-only policy fixing as many variables (1.1% against 8.9%). Because dispatch method and benchmark scope alone shift the reported optimality ratio by up to 3.6% and 9%, we propose a reporting protocol.
[LG-87] Breaking the 1/3 Barrier for Approximate Nash Equilibria in Symmetric Bimatrix Games
链接: https://arxiv.org/abs/2610.25524
作者: Ashkan Soleymani,Patrick Jaillet,Gabriele Farina
类目: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)
*备注: ACM-SIAM Symposium on Discrete Algorithms (SODA), 2027
Abstract:We give the first unconditional family of polynomial-time algorithms that breaks the 1/3 approximation barrier for symmetric bimatrix games. For every fixed \delta0 , an algorithm in this family computes a symmetric (1-1/\sqrt2+\delta) -approximate Nash equilibrium, improving on the previous best guarantee of (1/3+\delta) due to Kontogiannis and Spirakis [2011].
[LG-88] FracTS: Hierarchical and Autoregressive Time Series Generation
链接: https://arxiv.org/abs/2610.25523
作者: Jiayu Li,Umair Afzal,Zilong Zhao,Milad Abdollahzadeh,Uzair Javaid,Biplab Sikdar
类目: Machine Learning (cs.LG)
*备注:
Abstract:Time series are ubiquitous across different domains, and the growing use of synthetic data for privacy-preserving sharing, data augmentation, and stress testing makes time series generation increasingly important. Early generative approaches are predominantly autoregressive, which can suffer from low sample diversity, error accumulation, and difficulty modeling long-range dependencies. In this paper, we introduce FracTS, a hierarchical, coarse-to-fine autoregressive model for multivariate time series generation. Inspired by fractal generative modeling developed for images, FracTS adapts the paradigm from 2D to 1D sequences by learning a multi-scale representation that captures global trajectory structure at coarse scales and local dynamics at fine scales while following the autoregressive nature of time series. In contrast to unstructured images, time series are structured, allowing easy automated or designated temporal feature processing optimized for the model, and global (static) feature extraction describing the instances that generate the corresponding time series in the collection. We incorporate them in the generation process to further improve the generation quality. The design naturally handles multivariate dependencies and improves long-horizon temporal coherence without requiring fixed-length inputs. Experiments on five datasets (two conditional, three unconditional) conducted against seven baselines show that FracTS consistently outperforms baselines across marginal distribution and predictive ML utility, and correlation preservation in the unconditional setting. In particular, the predictive ML utility is averagely improved by 100.0% and 75.0-81.0% for unconditional and conditional time series datasets compared to the best baseline (relative error computed on dataset-averaged raw values). Code is available at: this https URL.
[LG-89] KPI-Aware Multimodal Anomaly Detection for 5G and Open RAN Telemetry ICDM2026
链接: https://arxiv.org/abs/2610.25510
作者: Chiman Salavati,Liang Wu,Kelly Wan,Mayank Darbari,Liangjie Hong
类目: Machine Learning (cs.LG)
*备注: Accepted at the IEEE ICDM 2026 Applied Research Track
Abstract:Reliable anomaly detection is a deployment bottleneck for 5G/Open RAN observability and emerging self-optimizing 6G systems, where hundreds of heterogeneous KPIs, severe class imbalance, and strict alert budgets make purely numerical scores brittle. We present KAC, a deployment-oriented KPI-aware multimodal anomaly-detection framework that uses KPI-level residual-text alignment to turn compact operator-style summaries into a learnable feedback signal over full-KPI temporal evidence. This gives the detector broad numerical coverage like modern time-series models, while retaining the sparse KPI-specific reasoning that makes operator feedback sustainable in a future operator-facing workflow. Across public telecom telemetry, enterprise-scale Open RAN observability data, and a Nokia benchmark built from real 5G RAN production packet traces with realism-hardened injected anomalies, KAC achieves strong threshold-based detection performance while remaining competitive on threshold-free ranking metrics. On TelecomTS, at a fixed 2.5 percent false-positive-rate operating point, KAC achieves the highest recall among the evaluated residual-based methods while maintaining comparable precision; across the evaluated benchmarks, it remains robust under substantial class imbalance. KAC is integrated in pre-production shadow mode with Nokia’s GKE-based KPI construction pipeline, where it continuously scores telemetry windows alongside an internal packet-level anomaly detector for operational evaluation.
[LG-90] Steered Not Stirred: Effective Reward Tilting using Surrogate Reward Mixing
链接: https://arxiv.org/abs/2610.25502
作者: Alec Sargood,Olivia Gallup,Georg Maierhofer,Noam Ghenassia,Paul Duckworth,Krisztina Sinkovics,Avishek Joey Bose
类目: Machine Learning (cs.LG)
*备注:
Abstract:Continuous-time generative models, from diffusion models, flow matching, and now flow maps, owe their practical value to high-fidelity generation and steerability. In many applications, a ground-truth latent ``oracle’’ reward R^* is unavailable or expensive to evaluate, e.g., human judgments or experimental measurements. Although the oracle reward is not directly observable, it can often be approximated through multiple cheaper surrogates that, individually, provide incomplete characterizations but collectively reveal substantial information about the downstream task. In this paper, we present a method for composing such surrogates to steer a pretrained model toward the distribution tilted by an unavailable R^* . Given a small feedback set of oracle rankings or numerical rewards, we introduce Mixing And Reward Tilting for INference-time Improvement (MARTINI), which learns a surrogate composition for plug-and-play use with standard inference-time steering methods. Our theory characterizes surrogate information limits and relates population recovery error to ideal-target KL at small tilt strength. For preferences, the bound additionally accounts for an irreducible mismatch between the Plackett-Luce population target and the conditional mean. Experiments demonstrate improved oracle-tilt fidelity in language generation and ImageNet alignment, and improved joint pose accuracy and validity in protein-ligand cofolding, without inference-time oracle queries.
[LG-91] Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models
链接: https://arxiv.org/abs/2610.25495
作者: Junyu Chen,Franklin Liu,Moshu Xu,Lijian Yang
类目: Machine Learning (cs.LG); Econometrics (econ.EM); Trading and Market Microstructure (q-fin.TR)
*备注:
Abstract:Modeling limit order book (LOB) dynamics is central to market forecasting and execution-strategy evaluation. While diffusion models achieve strong empirical performance, existing LOB snapshot models typically lack a learned recurrent encoder and formal statistical guarantees. We introduce LOBGrad, a conditional DDPM framework for probabilistic LOB forecasting that combines an autoregressive RNN encoder with a diffusion-based generator. In particular, we establish an excess-risk oracle inequality for the one-step DDPM noise-prediction objective under exponential mixing and regularity assumptions. We further show that, under contractive dynamics, the dependence on the conditioning-window length can be reduced to a dependence on the latent-state dimension. Experiments on BTC/USDT, GOOGL, and META show that LOBGrad improves on its encoder-free diffusion ablation across all reported queue metrics and obtains competitive results against other baselines. It also reproduces the key stylized facts of LOB microstructure, and yields reasonably calibrated queue-based predictive intervals. A stylized downstream task illustrates the potential value of the generated forecasts.
[LG-92] Consistent LLM Call Handling Across Java Frameworks: A Reference Implementation and Benchmark
链接: https://arxiv.org/abs/2610.25480
作者: Oleksandr Klymenko
类目: Machine Learning (cs.LG)
*备注: 7 pages, 2 figures, 3 tables, 1 listing. Code and data: this https URL
Abstract:Java applications reach large language models (LLMs) through an integration framework, and rely on it to handle a failing provider: to retry what can be retried and nothing else, to wait as long as the provider asks, never to present a truncated answer as complete, and to record each failure in the application’s metrics. We state these expectations as a six-clause reliability contract and measure Spring AI 2.0.1 and LangChain4j 1.21.0 against it with a scripted OpenAI-compatible provider: fifteen failure scenarios, synchronous and streaming, three repetitions. LangChain4j meets it in 47 of 77 applicable cases with its defaults and in 51 of 83 when configured explicitly, Spring AI in 72 of 77 and 76 of 83; both deliver a stream that ended without its terminator as a complete answer. We then place one implementation of the contract beneath both frameworks, at the HTTP boundary - an OkHttp interceptor for Spring AI, an HttpClient for LangChain4j, and a this http URL transport for code without a framework - sharing a 725-line core that depends only on the JDK and the Micrometer Observation API. With it, all three clients behave identically and meet the contract in 90 of 91 cases; the remaining case is a policy trade-off between retrying a hung provider and waiting for a slow first token. Against a local model server the layer rejected none of 700 calls, and it adds a median of 32-217 microseconds to a call.
[LG-93] Near-Optimal Last-Iterate Convergence of SGD in the Smooth Interpolation Regime
链接: https://arxiv.org/abs/2610.25469
作者: Matan Schliserman,Tomer Koren
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study the last iterate of stochastic gradient descent (SGD) with a constant stepsize on smooth convex objectives in the low-noise and interpolation regimes, where the variance \sigma_\star^2 of the stochastic gradients at the optimum is small or zero. We establish that after T steps of SGD on \beta -smooth convex losses with any constant stepsize 0\eta2/\beta , the last iterate attains expected excess risk \widetilde O\bigl(\tfrac1\eta(2-\beta\eta)^3T+\tfrac\eta(2-\beta\eta)^4\sigma_\star^2\bigr) , which gives the rate \smash\widetilde O(1/T) for every constant stepsize in the interpolation regime ( \sigma_\star=0 ), matching the \Omega(1/T) lower bound up to logarithmic factors and improving upon the O(1/\sqrt T) rate of Attia et al. (2026) for \eta=1/\beta . Our analysis is based on a new technique, which compares the second moments of the stochastic gradients along the SGD trajectory across only logarithmically many time scales. For quadratic losses in the interpolation regime, an adaptation of this technique gives an O(1/T) rate for every constant stepsize, extending the recent optimal rate for the randomized Kaczmarz method of Derezinski and Dong (2026) to stepsizes above the greedy stepsize with an arguably simpler proof.
[LG-94] Semi-supervised learning on singular data sets
链接: https://arxiv.org/abs/2610.25450
作者: Guillaume Valette
类目: Machine Learning (cs.LG); Algebraic Geometry (math.AG); Statistics Theory (math.ST)
*备注:
Abstract:We propose to model singular data clouds by subanalytic sets, and develop an approach to semi-supervised learning in this setting. It relies on a Morrey-type embedding for Sobolev spaces relative to globally subanalytic measures with possibly singular support. As an illustration, we present a greedy method based on a Monte Carlo approximation of the Sobolev norm of the candidate solutions. We report numerical experiments on point clouds sampled from real algebraic varieties, as well as a test on a classification problem based on MNIST images downsampled into 7\times7 .
[LG-95] Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting NEURIPS2026
链接: https://arxiv.org/abs/2610.25392
作者: Jingru Fei,Wei Fan,Kun Yi,Qi Zhang,Zhendong Niu
类目: Machine Learning (cs.LG)
*备注: Accepted by NeurIPS 2026 spotlight
Abstract:Explicitly modeling periodicity has proven to be an efficient approach for time series forecasting, where the residual, obtained after removing the periodic component from the time series, is typically treated as unstructured noise. However, in this work, we show that such residuals still exhibit structured, \textitphase-dependent deviations, as supported by both empirical observations and theoretical analysis. This suggests that residuals should not be treated merely as unstructured noise, but instead as \textitphase-conditioned signals whose variations arise in both direction and magnitude. Motivated by this perspective and inspired by phase and amplitude modulation in classical signal processing, we propose \textitPaloma, a simple yet effective framework for modeling residual deviations. Specifically, Paloma comprises two complementary modules: \textita Phase Rotation module, which performs phase-conditioned rotations to capture directional changes in the feature space, and \textitan Amplitude Scaling module, which applies phase-conditioned affine transformations to model magnitude variations. We further abstract the two modules into a general-purpose plugin, termed the \textitPaloma technique, which can be readily integrated into existing forecasting architectures to modulate input sequences in a phase-aware manner, enabling backbone models to capture phase-dependent variations in both direction and magnitude. Extensive experiments on twelve datasets demonstrate that Paloma achieves state-of-the-art and robust forecasting performance. Additional results show that the Paloma technique consistently improves a wide range of baseline models, validating its effectiveness as a general-purpose component. Our Code is available at \urlthis https URL.
[LG-96] he Hidden Look-Ahead: Point-in-Time Leakage in Electricity Market Data
链接: https://arxiv.org/abs/2610.25389
作者: Abdelkrim Alfalah
类目: Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE); Applications (stat.AP)
*备注: 13 pages, 4 figures, 7 tables
Abstract:Empirical studies of electricity prices routinely state that their models use only information available at the time of the trading decision. For weather-driven forecasts drawn from commercial archives, this condition is harder to meet than it appears, and whether it holds can change over time. We identify three timing pitfalls. A forecast run is labelled by the issue time of its weather model, but becomes available hours later and keeps changing after publication. Auction gates follow local clock time, while archives use UTC. Outturn series are revised after publication without earlier values being kept. Using the timestamps of 128,656 forecast runs from ten weather-model configurations, matched to the day-ahead and first intraday auction gates in Great Britain over 2022-2026, we find that selecting “the latest run issued before the gate” picks a run not yet published at the gate on more than 99% of delivery days for six of the ten configurations at one of the two gates. For GFS at the day-ahead gate, the share is 0.6% in winter and 99.8% in summer. The exposure also shifts as publication delays drift, and an action margin of a few minutes can cost a full six-hour forecast cycle. The wrongly selected inputs differ materially from those available at the gate, but their apparent accuracy gain ranges from -9% to +24% across variables, models, gates and seasons, so its effect cannot be assumed and must be measured. We argue that point-in-time correctness is a dynamic property of the data pipeline, and propose a protocol that audits it from timestamps before any model is fitted.
[LG-97] Structured Relevance Determination and Channel Additive Gaussian Processes for Image Data
链接: https://arxiv.org/abs/2610.25385
作者: Harald Roßels,Ludovico Scarton
类目: Machine Learning (cs.LG)
*备注: Code: this https URL
Abstract:We introduce radial basis function (RBF) kernels for Gaussian processes that use structured hyperparameter sharing for image-valued inputs. For high-dimensional inputs such as images, a fully anisotropic RBF kernel assigns a separate lengthscale to each input dimension, which can introduce many hyperparameters and may lead to overfitting and poorly calibrated predictive uncertainty when data are limited. We propose structured relevance determination by tying lengthscales across groups of adjacent pixels and within each colour channel. This yields patch-wise and channel-wise constrained anisotropic RBF parametrisations that reduce the number of independently learned lengthscales. We further introduce channel-additive Gaussian processes for multi-channel images and combine them with structured relevance determination, resulting in a parameter-efficient decomposition of the latent function into components defined on channel subsets, enabling some non-local generalisation. The proposed kernels are evaluated on MNIST and a compact binary aircraft-detection dataset of low-resolution RGB satellite image chips. We also show how channel-wise structured relevance determination can be incorporated into convolutional Gaussian processes. We observe improved predictive performance compared to isotropic and fully anisotropic RBF baselines.
[LG-98] ILDE: Teacher-in-the-Loop Distillation from Embodied Foundation Models
链接: https://arxiv.org/abs/2610.25376
作者: Yuxiang Lu,Yantong Liu,Jinghua Hou,Xueli Sun,Ruihua Han,Zhe Liu,Hengshuang Zhao
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Project page: this https URL
Abstract:Large-scale action pretraining equips Embodied Foundation Models (EFMs) with strong manipulation capabilities, with particular strengths in generalization and closed-loop robustness, but reproducing such pretraining requires substantial robot data and computation. To transfer these capabilities from pretrained EFMs to policies without action pretraining, we introduce TILDE (Teacher-In-the-Loop Distillation from EFMs), which synergizes on-policy distillation at student-induced states with learning from teacher-executed recovery trajectories. We further propose AutoDAgger, an autonomous data aggregation pipeline in which a robotic reward model is utilized to monitor task progress and trigger teacher intervention, thereby collecting recovery experience without human monitoring or manual control. AutoDAgger accelerates recovery-trajectory collection by more than 2\times compared with the human-in-the-loop method. By unifying learning from demonstrations, on-policy teacher supervision, and teacher’s corrective behavior, TILDE improves task performance by up to 4.1\times over behavior cloning across three real-world tasks and supports effective cross-architecture distillation.
[LG-99] Mining the Self-Rewriting System: A Vision for Data Mining on AI-Native Infrastructure ICDM
链接: https://arxiv.org/abs/2610.25374
作者: Mayank Darbari,Chiman Salavati,Liang Wu,Dawei Wang,Jiong Zhang,Kelly Wan,Krishna Chakka,Liangjie Hong
类目: Machine Learning (cs.LG)
*备注: Accepted to the Blue Sky Track of the IEEE International Conference on Data Mining (ICDM)
Abstract:Classical data mining treats data as generated outside the learner. Emerging AI-native infrastructure invalidates this assumption: as agents observe system state, act, update retrieval stores and models, and promote new artifacts into the serving path, their own decisions reshape the future logs, traces, corpora, and evaluation records used for later training and release. We call this new research object a self-rewriting system and the resulting system-induced distribution shift endogenous drift. Our central, falsifiable claim is that cross-surface causes are not identifiable in general from marginal per-surface monitoring: different coupled failures can produce the same individual signals while differing in their joint behavior, requiring joint analysis for reliable attribution. This agenda comprises four foundational research problems: fusing heterogeneous drift signals into calibrated gating and attribution (C1); mining agent traces whose activity alphabet evolves (C2); maintaining knowledge fabrics through online reindexing, faithfulness, and compression decisions (C3); and safely promoting concurrent learning streams through one shared evaluation gate (C4). These remain fragmented across concept drift, performative prediction, process mining, streaming retrieval-augmented generation, and MLOps, without a unified data-mining agenda. We outline public benchmarks and evaluation criteria, using AI-native telecom only as a motivating exemplar with analogues in finance, software CI/CD, and clinical decision support.
[LG-100] Multi-site Learning for Forecasting Day-Ahead Wind-Power Production for Onshore Wind Farms in Great Britain with Deep Neural Networks
链接: https://arxiv.org/abs/2610.25370
作者: James Kean,Aidan O’Sullivan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Accurate day-ahead wind-power production forecasting is increasingly important for energy system operations, market activity and balancing Great Britain’s grid as wind generation grows. In this study, we train multi-site deep neural networks (DNNs) using pooled historical production and reanalysis weather data from 88 onshore wind Balancing Mechanism Units (BMUs). Forecasts are produced at 09:00 UK time for the following midnight-to-midnight delivery day, matching NESO’s operational day-ahead forecasting schedule. We use reanalysis weather as future covariates during training and numerical weather prediction (NWP) forecasts during inference, extending training beyond the available historical NWP archive. Our multi-site recurrent neural network outperforms site-specific RNNs and a multi-site XGBoost model, while processing the full future-weather trajectory for the forecast horizon further improves accuracy relative to hour-aligned approaches. Our best model, an RNN with a bidirectional weather encoder (BiRNN) for NWP trajectory processing, reduces MW mean absolute error and root mean squared error relative to NESO by 17.94% and 12.51%, respectively, across 638,246 matched BMU-hour observations under conditions approximating unconstrained wind farm operations. In 40 leave-one-out trials, the BiRNN achieves lower MAE than NESO at 32 excluded sites, indicating transferable forecasting ability despite reduced accuracy when sites are omitted from training. Our capacity-estimation, data-cleaning and reanalysis-based training approach provides a basis for scaling NWP-aware multi-site forecasting to additional wind farms, longer production histories, and increased forecast accuracy beyond that seen in this study.
[LG-101] GreenNIDS: A Benchmark Workflow and Analysis Dataset for NetFlow Intrusion Detection Experiments
链接: https://arxiv.org/abs/2610.25366
作者: Lilia-Eliana Popescu-Apreutesei,Sabina-Cristiana Necula
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 12 pages, 3 figures, 2 tables. Accepted at IE 2026 (International Conference on Informatics in Economy)
Abstract:Comparing the predictive performance and computational cost of flow-based intrusion detection experiments requires both quantities to be available for each run. GreenNIDS connects benchmark execution on public NetFlow datasets to a reusable analysis dataset. Model implementations write a common set of outputs, which are ingested into a run-level table with companion tables for model configuration, feature importance, and training history. The run-level table combines predictive metrics with timing, throughput, memory use, and energy estimates based on thermal design power. A CPU-only demonstration comprises 216 runs across four dataset families, three versions, and binary and multiclass tasks. Ingestion produces a 216 x 96 analysis table with no duplicate run identifiers and complete Matthews correlation coefficient (MCC) and F1 coverage. Multiclass ROC-AUC is missing for 36 PyTorch runs, making metric availability an explicit part of subsequent comparisons. The stored records support grouped performance and training-time plots, as well as resource screening after execution. Combining task-specific MCC thresholds with limits on training time, peak RAM, and estimated training energy retains 25.0% of the full run collection. These results establish compatibility within the evaluated collection and support comparisons after execution, subject to the recorded preprocessing, execution profile, and approximate energy accounting.
[LG-102] Operator-Mismatch Contrastive Learning for World-Relative Trajectory Coherence
链接: https://arxiv.org/abs/2610.25360
作者: One Octadion,Firmansyah Sundana,Novanto Yudistira
类目: Machine Learning (cs.LG)
*备注: 19 pages, 6 figures, 11 tables
Abstract:World models are increasingly judged by how well they rank plausible futures, yet plausibility alone is the wrong target. The same scene can evolve under many latent rules, and a future that is smooth and physically valid under one rule can be incoherent under the rule actually in force. We formalize the resulting failure, operator-blindness, and show that it is systematically hidden by standard evaluation: a scorer can attain near-perfect generic ranking while separating current-world from other-world futures no better than chance. We introduce Operator-Mismatch Contrastive (OMC) learning, whose semantic hard negatives are valid-but-wrong-operator trajectories, and a matching protocol that measures world-relative coherence directly. On a controlled synthetic benchmark with in- and out-of-distribution operator parameters, across two trajectory-scoring backbones, and on feature-level external validation with CoPhy ballsCF, OMC raises operator-sensitive scoring far above a chance baseline that nonetheless achieves 0.99 generic top-1. We prove that the wrong-operator negative is necessary both for the protocol to detect the failure and for the objective to fix it, and position OMC as an auxiliary coherence signal (a trajectory-coherence scorer, not a generative model, planner, or control agent) that can be attached to existing world-model representations.
[LG-103] A Distribution-Free Separation Between Learned and Fixed Features
链接: https://arxiv.org/abs/2610.25335
作者: Daniel Barzilai
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:The success of neural networks is largely attributed to their ability to learn useful representations from training data. However, known results that prove a computational advantage of learned features over fixed ones are restricted to specific input distributions. We settle the COLT open question of Feldman, Kamath, and Srebro (2026) by showing that this advantage can persist in the distribution-free setting. We construct a family of hypothesis classes that can be efficiently learned by a natural gradient-based algorithm on a two-layer ReLU network, but cannot be represented by linear predictors over any fixed feature map with polynomial dimension (and runtime). Specifically, for any sufficiently small error and any sufficiently large runtime budget R (depending on the error), there exists a hypothesis class that the neural network learns under any input distribution to the desired expected classification error in time at most R, while the dimension complexity (governing the dimension needed by linear predictors) is superpolynomial in R. Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML) Cite as: arXiv:2610.25335 [cs.LG] (or arXiv:2610.25335v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.25335 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-104] Follow the Mixed Groups: Ridge-Tracking Gaussian Sampling for Zero-Variance Web Rollouts
链接: https://arxiv.org/abs/2610.25330
作者: Tomas Duncun,Jonah Ellison,Owen Pell,Mateo Ruiz,Hanae Fujita,Ingrid Solberg,Caleb Nguyen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Group relative policy optimization learns from a question only when the sampled responses disagree. A group that is entirely correct or entirely incorrect has zero reward variance and a zero advantage. On the Web, that group has already spent its searches and page fetches. Gaussian curricula aim the next batch at pass rates where a finite group can still mix. A Gaussian on the latest empirical rate treats every all-fail history as the same point, and a center stepped by the average signed change of those rates is pushed by prompts already stuck at 0 or 1. RIDGE keeps the Gaussian curriculum and changes the coordinate, the center, and the width. Each prompt carries a Beta posterior. The sampling weight is the marginal of a Gaussian target under a Gaussian approximation of that posterior, divided by a prediction of tool-call cost. The center and the width are an exponential smooth of the posterior means of groups that just came back mixed. All-correct and all-incorrect groups refresh the posterior and do not move the center. The GRPO update itself is unchanged. On two math backbones and a retrieval-augmented question answerer, RIDGE raises average accuracy by 3.4 and 2.9 points over GRPO and average exact match by 3.7 points, at 29.6% fewer math rollouts and 32.7% fewer tool calls than dynamic sampling.
[LG-105] Where Groups Split: A Probe Prior for Nonzero Within-Group Variance in RL Post-Training
链接: https://arxiv.org/abs/2610.25325
作者: Jonah Ellison,Mateo Ruiz,Owen Pell,Tomas Duncun,Hanae Fujita,Ingrid Solberg,Caleb Nguyen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Group relative policy optimization learns from a prompt only when the rewards in the sampled group disagree. With a binary verifier, a group that is entirely correct or entirely incorrect has within-group variance zero and a zero advantage. Uniform sampling continues to spend rollouts on prompts that are unanimous now and stay unanimous as training goes on. Pass rate is a weak forecast of that future. Two prompts at the same pass rate can move apart: one into the range where a finite group splits, and one toward a ceiling where the group becomes unanimous again. Aperture estimates the later pass rate from a short probe. A small subset of the pool is trained with the same recipe, every prompt is scored on the base model and on the probe model, and the two pass rates are plugged into the probability that a group of the training size is all wrong or split. The resulting weight is a static prior. It can be used alone or multiplied by an online scheduler, and the GRPO update is left as it is. On mathematical and logical reasoning with 1.7B and 4B models, the prior improves greedy accuracy over uniform GRPO and raises the fraction of groups with a nonzero advantage. The same prior still helps when it is combined with dynamic sampling or with selective rollouts.
[LG-106] A discrete-continuous-time stochastic framework for attention-deficit/hyperactivity disorder-related symptom modeling and pharmacological intervention via methylphenidate
链接: https://arxiv.org/abs/2610.25285
作者: Oana-Maria-Catalina Crisan
类目: Machine Learning (cs.LG); Probability (math.PR)
*备注: 27 pages, 2 tables, 2 algorithms, 20 figures
Abstract:This paper presents an original mathematical and computational framework for modeling the efficacy of methylphenidate as an optimal control policy acting on the nonlinear dynamics of attention deficit hyperactivity disorder. Neural processes in ADHD are conceptualized as stochastic processes with trajectories that inherently converge towards stable, yet dysfunctional patterns associated with clinical symptoms. The pharmacological intervention is mathematically formalized as an external control signal that reassesses the system’s drift and redirects these trajectories towards neurotypical regimes. To bridge these theoretical stochastic models with empirical clinical data, latent representations are inferred using a novel deterministic hybrid deep learning architecture, combining a one-dimensional convolutional neural network and a Vision Transformer, trained on preprocessed functional magnetic resonance imaging data.
[LG-107] Data-driven forecasting of tropical cloud forest phenology: A dynamical systems and machine learning hybrid approach
链接: https://arxiv.org/abs/2610.25283
作者: Pedro Garcia,Saul Flores
类目: Machine Learning (cs.LG)
*备注: 21 pages, 4 figures
Abstract:Forecasting how species and ecosystems respond to climate change is a critical challenge in ecology, particularly in hyper-diverse tropical biomes. While traditional phenological models often rely on preconceived functional forms, this study introduces a data-driven hybrid approach that integrates dynamical systems theory with machine learning to forecast flowering and fruiting cycles. Using a robust 35-year dataset from a Venezuelan tropical cloud forest, we reconstruct the system’s state-space using delay vectors inspired by Takens’ theorem, incorporating climate forcing variables such as temperature and precipitation. The model was trained using Gaussian Process Regression, providing both point predictors and explicit uncertainty estimates. Our results clearly suggest that the synergy between dynamical systems and machine learning effectively captures the non-linear complexities of ecological time series. Specifically, the model achieved higher predictive accuracy for flowering and fruiting. This hybrid framework offers a scalable tool for anticipating phenological mismatches and informing conservation strategies in the face of global environmental change. Comments: 21 pages, 4 figures Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.25283 [cs.LG] (or arXiv:2610.25283v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.25283 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-108] Federated Learning for Malware Image Classification under Data Heterogeneity
链接: https://arxiv.org/abs/2610.25264
作者: Victor Taiwo,Cemal Nisan,Muhammad Athallah,M. Emre Gursoy,Oznur Ozkasap
类目: Cryptography and Security (cs.CR); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Malware image classification (MIC), in which binary files are converted into visual representations and deep neural networks are trained to identify malware families, has recently emerged as an effective approach for malware detection. Federated learning (FL) enables malware image classifiers to be trained collaboratively by multiple clients without sharing raw data. However, a fundamental challenge in FL is data heterogeneity (non-IID), which is particularly relevant but understudied in the malware context. In this paper, we address this gap via three contributions. First, we present an empirical study of federated MIC under varying degrees of data heterogeneity across three popular malware image datasets (MalImg, Virus-MNIST, and MalNet). We show that while FL achieves strong accuracy under IID conditions, both accuracy and training stability degrade as data becomes increasingly non-IID. Second, we introduce a set of round-level statistics that characterize the data contributed by participating clients in each FL round, and demonstrate that class entropy (diversity of class labels present in a given round) exhibits strong positive correlation with global model accuracy. Third, leveraging this insight, we propose an entropy-aware client selection strategy that filters candidate client subsets based on class entropy, and show that our strategy outperforms standard random selection. Our results highlight the role of client composition in federated MIC and underscore the importance of distribution-aware classifier training under data heterogeneity.
[LG-109] EasyResearch: Honest Regression Forecasting and Signal Classification for Researchers Who Do Not Program
链接: https://arxiv.org/abs/2610.25263
作者: Ahmad B. A. Hassanat,Ghada A. Altarawneh
类目: Machine Learning (cs.LG)
*备注:
Abstract:EasyClassifier showed that researchers who do not program can obtain honest classification results when the correct evaluation is the default. Most applied studies, however, predict a number, forecast a series or classify recordings, and each task has its own way of producing scores that do not hold on new data. We present EasyResearch, open-source analysis cores for regression, time-series forecasting and signal classification, and EasyResearch Desktop, a Windows application that runs these three tasks and EasyClassifier’s classification without code. All four follow the same rules: everything learned from data is learned on training data only, models keep their published defaults with fixed seeds, the selected model receives a separate final score, forecasts are never evaluated on shuffled data, and every run writes a report with a Methods paragraph and the references for every method used. We measured whether the reported scores are honest by keeping part of each dataset untouched. In regression (9 public datasets, 90 runs), the common practice overstated R^2 by +2.2 points and the best comparison score by +2.5, while EasyResearch’s final score was -1.0; screening predictors on all rows before cross-validation, which EasyResearch does not offer, reported R^2 = 0.27 on pure noise against -0.44 on new data. In forecasting (15 series), shuffled cross-validation over windows of past values promised errors 19% smaller than its chosen models achieved in the untouched future, while EasyResearch’s final-period score was unbiased to within 1.5%; the models EasyResearch chose had 14% lower future error, and on random walks shuffled cross-validation chose a machine-learning model in 5 of 5 series, each worse in the future than repeating the last value, which EasyResearch chose 4 times. In signal classification (4 UCR datasets) one method won every comparison and neither estimate was biased.
[LG-110] Emergent Sparsity: How Many Effective Variables Survive Coarse-Graining in AlphaEarth Embeddings? NEURIPS2026
链接: https://arxiv.org/abs/2610.25258
作者: Johannes Hirn
类目: Machine Learning (cs.LG); Statistical Mechanics (cond-mat.stat-mech); Geophysics (physics.geo-ph)
*备注: Camera-ready version of a paper accepted at the 2nd Workshop on Advances in Representation Learning for Earth Observation (REO), NeurIPS 2026. 20 pages, 15 figures
Abstract:AlphaEarth encodes each 10 m land pixel as a 64-component vector intended to support diverse global mapping tasks. Yet many scientific applications work at coarser spatial supports, for which fewer than 64 effective variables may suffice. We ask whether these vectors can be reorganized into ranked global nonlinear coordinates, how many suffice at a chosen reconstruction accuracy, and whether that number decreases with spatial scale. We thus describe effective dimension as a two-control surface over tolerance and spatial scale, rather than one intrinsic number. At each scale, a \beta -VAE scan with a common architecture produces a rank–distortion envelope, giving the smallest retained count achieved within the scanned family at each reconstruction tolerance. At 3% validation tolerance, this number falls from 24 to 7 when the scale is increased from 10 m to 78.9 km, whereas the corresponding PCA rank falls only from 46 to 38. Repeating the analysis on unit-norm targets preserves this trend.
[LG-111] Improving the Expressive Power of Hybrid Sequence Models via Multi-Hop Retrieval
链接: https://arxiv.org/abs/2610.25250
作者: Yam Eitan,Gilad Yehudai,Haggai Maron
类目: Machine Learning (cs.LG)
*备注:
Abstract:Hybrid architectures offer a promising approach to efficient language modeling by combining a small number of full-attention layers with a larger number of efficient sequence layers, such as sparse attention and recurrent layers. While full-attention layers maintain a KV cache that grows linearly with sequence length, efficient layers typically rely on separate, fixed-size states rather than accessing this richer existing memory. Recent work has begun exploring ways to allow efficient layers access to the KV cache, but the theoretical benefits of such access, and the extent to which existing approaches can realize them remain poorly understood. We investigate these questions by analyzing the expressive power of individual hybrid blocks on function composition, a well-studied task in sequence-model expressivity associated with multi-hop reasoning. We first establish that both conventional hybrid blocks and those with top- \kappa KV-cache access require \Omega(\sqrtn) capacity to compute even two-function composition over a domain of size n . To overcome this limitation, we introduce the \emphMulti-hop Retrieval Attention (MuRA) hybrid block, a simple architecture that enables multi-hop retrieval from a shared KV cache while preserving asymptotic computational and memory efficiency. We prove that a single fixed size m -hop MuRA block can compute m -function composition, whereas computing (m+1) -function composition with the same retrieval depth requires polynomial model size, establishing a strict expressivity hierarchy in retrieval depth. Synthetic experiments support this hierarchy, showing that increasing retrieval depth enables learning progressively deeper compositions, substantially outperforming existing hybrid architectures. Together, our results identify multi-hop retrieval as a promising mechanism for improving the expressive power of efficient hybrid architectures.
[LG-112] DROD: Dual-Ratio Ordered Dropout for Federated Continual Learning on Heterogeneous Devices
链接: https://arxiv.org/abs/2610.25220
作者: Berk Sezer,Özgür Soysal
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注: 18 pages
Abstract:Federated learning for capacity-heterogeneous clients and federated continual learning are each well studied on their own, but their intersection – clients holding differently sized models while every client’s data stream also drifts over time – remains underexplored. We study this setting directly, training a ResNet-18 fleet from scratch on Split-CIFAR-100 (10 tasks, class-incremental), and introduce DROD (Dual-Ratio Ordered Dropout), which builds on FjORD’s Ordered Dropout by using two independent reduction axes to extract sub-models: a shortcut ratio and a branch ratio. On a shared parameter/FLOP budget ladder, DROD achieves the highest mean accuracy among the heterogeneous-model federated learning strategies we compare. Its advantage over the two strongest baseline geometries persists across every replay-buffer size we test and under ER, DER++, and BiC; under EWC, all three geometries collapse alike.
[LG-113] Shallow Probes Are Sufficient for Near-Oracle Personalization Depth Allocation in Federated Learning
链接: https://arxiv.org/abs/2610.25193
作者: Beyazit Bestami Yuksel,Emrah Dikbiyik,Ayse Yilmazer Metin
类目: Machine Learning (cs.LG)
*备注: 6 pages, 9 figures. Accepted and presented at SINCONF 2026 (18th International Conference on Security of Information and Networks), Beykent University, Istanbul, Turkiye, 14-16 October 2026
Abstract:Rather than proposing a new federated optimizer, we study whether a small client-specific measurement budget can support cost-aware personalization-depth decisions under a utility that trades Macro-F1 against deployment cost. Exhaustive evaluation of six depths defines an offline oracle that is not deployable. We audit the Probe Sufficiency Hypothesis: three shallow probes (P0–P2) may contain most decision-relevant information for approaching that oracle. FedPBA is a secondary learned instantiation for deep-action extrapolation. On CIFAR-10/100 and patient-disjoint PTB-XL (45 sessions; 900 clients; 2700 evaluations; GroupKFold), training-free greedy probing achieves regret 0.0071 and 98.3 percent oracle recovery, beating metadata-only policies and FedPBA (0.0110 / 97.3 percent). Aggregate shallow-best versus oracle correlation is 0.9818 but falls to 0.578 on deep-oracle rows; PTB-XL deep-oracle cases remain a high-regret boundary without a validated escalation gate.
[LG-114] Sampling and Curvature in First-Exit Comparisons for Momentum SGD
链接: https://arxiv.org/abs/2610.25191
作者: Haiyi Li
类目: Machine Learning (cs.LG)
*备注: 84 pages, including references and appendices
Abstract:First-exit prediction for momentum SGD depends on sampling information that pointwise gradient laws omit. Strongly convex finite pools with identical pointwise laws and matched iid momentum dynamics can have different reshuffled exits; even analytic pools with uniformly close pointwise laws can have separated exit probabilities. For persistent sampling in one dimension, we establish an explicit O(h) mean-exit comparison with a finite-rate switching process under nonvanishing component velocities and step-size conditions; iid sampling instead averages to an ODE. Retaining realized minibatches and sampled curvature yields finite-window exit brackets beyond this regime. In 32 trained ridge-logistic runs, 29 meet prespecified targets with rounding-certified brackets; an enclosure control explains most of the width advantage. Exact-real certificates establish finite means at the original tested radii for three ResNet-18 checkpoints and give local finite-mean conditions for tanh and ViT networks; their upper bounds remain loose, and original-radius ViT mean guarantees remain unresolved despite 128 fresh continuation paths. A two-network replication gives 32/32 successful seeds in each of four finite-path persistence comparisons. Independent equal-length controls retain slower exits after refresh resets than without resets, but the refresh-versus-random ordering reverses with update length at both networks. These interventions do not causally decompose raw momentum SGD. Neural step-size scaling does not validate the switching limit.
[LG-115] An Information-Geometric Framework for OCS-Reconfigurable Interconnect Topology in Large-Scale AI Clusters
链接: https://arxiv.org/abs/2610.25173
作者: Shintaro Yoshizawa
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注: 27 pages
Abstract:Optical Circuit Switches (OCS) make inter-pod wiring in AI clusters programmable, posing design questions on wiring choice, symmetry departure, and reconfigurations. We present a unified information-geometric framework to address these issues.A wiring is a point in a probability simplex over circuits within a degree polytope. Departure from a symmetric Fat-Tree prior \mathbfq is measured by \DKL(\cdot ,|, \mathbfq) . The cost U is the smoothed DAG critical-path length, whose gradient equals path marginals of a Gibbs distribution via forward–backward this http URL free energy F_\beta_1 = U + \beta_1^-1 \DKL(\cdot ,|, \mathbfq) , we prove well-posedness, monotonic dependence on \beta_1 , adaptation bounds via Jeffreys divergence, and a symmetry principle for static optimality. An entropic proximal-gradient method updates multiplicatively with a |V| -dim dual KL projection, achieving linear this http URL time-varying traffic, a threshold policy yields regret O(\sqrt\Gamma C V_1 T) against the switching-free optimum. Traffic changes are detected by a multinomial GLR–CUSUM detector with non-asymptotic false-alarm bounds. Multi-job generalized-mean fairness aligns with \alpha -fairness ( \alpha \ge 2 ). Experiments on pipeline, data, and MoE workloads validate the framework against integer programming.
[LG-116] FHKAN: Fractional Hermite Kolmogorov-Arnold Networks for Data-Driven and Physics-Informed Learning
链接: https://arxiv.org/abs/2610.25148
作者: Arsha Farnoush,Amirhossein Sadr,Aida Pakniyat,Dara Rahmati,Saeid Gorgin,Kourosh Parand
类目: Machine Learning (cs.LG)
*备注:
Abstract:Kolmogorov-Arnold Networks (KANs) replace fixed scalar weights by learnable univariate maps, yet most spectral KANs still rely on Chebyshev polynomials that live on a bounded interval and therefore require a saturating squashing of the pre-activation. We introduce the Fractional Hermite Kolmogorov-Arnold Network (FHKAN), a neural architecture that equips KANs with an adaptive, trainable fractional-order generalized Hermite basis. With a nonnegative shape parameter, the underlying generalized Hermite functions are orthonormal on the entire real line and uniformly bounded under a Gaussian envelope, so they act directly on unbounded pre-activations without any saturating map. We derive closed-form derivatives of the generalized, fractional, and scaled basis away from the origin and determine its order of differentiability at the origin, where the fractional power map creates a singular point. On synthetic regression, FHKAN attains the lowest test error at every network depth among nine fixed activations and the spline-based KAN. On image classification, image denoising, and sentiment analysis, the best-performing fractional Hermite block yields the highest accuracy or reconstruction quality. On physics-informed problems, FHKAN reaches mean absolute errors of order 10^-6 to 10^-5 on the Lane-Emden equation, residuals of order 10^-4 on the Burgers equation, and absolute errors below 10^-4 on a fractional delay differential equation.
[LG-117] Many Repairs One Direction: Steering Code Models Toward Safer Cryptographic API Use
链接: https://arxiv.org/abs/2610.25133
作者: Tianhe Lu,Sakuna Harinda Jayasundara,Eric Spero,Ewan Tempero,Robert Biddle,Giovanni Russello
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注:
Abstract:Cryptographic API misuse in large language model-generated code can compromise software security. While many existing approaches improve code security by modifying model parameters or inputs, how representation changes associated with different cryptographic API repairs can support a shared intervention for safer code generation remains underexplored. To investigate this, we construct 48,000 matched program pairs across four programming languages and ten cryptographic API misuse categories. Each pair changes a specified API decision while preserving the programming task. Subtracting the misuse program’s internal representation from its secure counterpart’s gives a repair vector. Across five models, comparisons of mean repair vectors reveal stronger organization by misuse category than by language. These category-specific patterns also support a shared intervention. We introduce CrypSteer, which extracts a shared direction by normalizing repair differences and selecting a principal component aligned with repairs across misuse categories. With model weights fixed, adding one shared direction per model to its internal activations during generation improves aggregate target security rates by 5.28–15.00 percentage points over unsteered generation across languages and misuse categories. These findings connect the shared structure of cryptographic repair representations to a reusable intervention for safer code generation.
[LG-118] Memory as Belief: Bayesian Evolution of Memory for Self-Evolving LLM Agents
链接: https://arxiv.org/abs/2610.25119
作者: Yiming Xu,Qian Yang,Zihan Chen,Yuefei Wu,Wenhan Zeng,Bin Shi,Bo Dong
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large language model (LLM) agents increasingly rely on external memory to accumulate past experiences for continual adaptation. However, effectively reusing memory requires accurately assessing its value, as retrieved experiences may become unreliable when transferred across task contexts. Despite its importance, memory valuation remains largely underexplored, with existing methods relying on historical retrieval success rates that overlook the context dependency of memory usefulness and the varying reliability of estimates under different amounts of evidence. Inspired by the Bayesian brain hypothesis, we propose BEAM (Bayesian-style Evolutionary Agentic Memory), a framework that reformulates memory value evaluation as a parameterized cognitive inference process, where memory value is represented as a context-conditioned dynamic belief rather than a passive historical statistic. Specifically, BEAM infers memory value beliefs from current task contexts and continuously calibrates them using execution feedback as empirical evidence, transforming memory from passive storage into an adaptive cognitive mechanism for continual self-evolution. Extensive experiments across reasoning and code generation benchmarks demonstrate that BEAM consistently improves memory utilization and enables continual self-evolution of LLM agents.
[LG-119] Polynomial Learning of Positive ReLU Networks under Gaussian Inputs
链接: https://arxiv.org/abs/2610.25116
作者: Zhijie Zheng
类目: Machine Learning (cs.LG)
*备注: 18 pages
Abstract:We study bias-free, one-hidden-layer ReLU networks with nonnegative output weights under standard Gaussian inputs. Hidden directions may be arbitrarily close or linearly dependent. The labels have the target as their conditional mean, and only the unconditional second moment of the noise is bounded. We present a candidate real-arithmetic learning guarantee: a width- k target with total output weight at most one can be approximated to squared prediction error \varepsilon by a nonnegative ReLU network of width at most k+\min\k,d\ , with sample and operation cost (1+\sigma^2)\operatornamepoly(d,k,\varepsilon^-1,\log(1/\delta)) . The argument combines implicit high-order moments with a direction cover that is complete for heavy atoms and sound for every output. Exact Gaussian soft localization refines this cover while pruning keeps at most k centers at every scale. Clipping the query multipliers controls heavy-tailed, input-dependent noise without assuming a uniform conditional variance bound. A final convex fit and an exact sparsification step give the stated output width. We also obtain a parameterized projective direction-recovery guarantee. The result concerns real-arithmetic complexity; no bit-complexity or practical running-time guarantee is asserted.
[LG-120] Conformal Prediction-Enabled Sequence-level Policy Optimization for Reasoning Language Models
链接: https://arxiv.org/abs/2610.25092
作者: Hui Yang,Wan Tian,Tao Ren,Zhongyi Li,Yijie Peng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Reinforcement learning improves the reasoning capability of large language models through verifiable rewards for mathematical and coding tasks. In token-level methods such as GRPO, clipping acts on individual token ratios while the reward and advantage apply to a complete response. Recent sequence-level methods align these units but still use fixed clipping bands, whose tail probabilities change as the score distribution evolves. We propose Conformal Prediction-Enabled Sequence-level Policy Optimization (CSPO), which adaptively calibrates sequence-level clipping bounds using recent training statistics. The resulting band targets a specified marginal tail budget and adapts to policy drift; calibration conditional on response length requires additional assumptions. We bound the probability that a fresh score falls outside the band by the target miscoverage level plus explicit CDF discrepancy and drift terms. Under additional moment conditions, this bound also controls the objective and gradient error relative to the unclipped GSPO power surrogate. Experiments with the Qwen3-1.7B series show improvements over GSPO on difficult mathematical reasoning benchmarks and ALFWorld, with clipping endpoints that adapt during training.
[LG-121] he Optimal Sample Complexity of Pointwise Replicable Agnostic Learning
链接: https://arxiv.org/abs/2610.25062
作者: Guangjian Zhang
类目: Machine Learning (cs.LG)
*备注: 92 pages
Abstract:Pointwise replicability asks that two independently trained predictors, using the same random seed, agree at each fixed query with high probability. We determine its optimal agnostic sample complexity on finite and countable domains. For every nonempty binary class of finite VC dimension d\ge1 , the required number of independent examples is \Theta!\left(\fracd\rho^2\alpha^2+\frac\log(1/\beta)\alpha^2\right) , uniformly over independent accuracy, failure, and disagreement parameters \alpha,\beta,\rho\in(0,1/10) . The learner returns one binary function, may be improper, and uses unrestricted shared randomness. Its sample budget is fixed before observing data, and the replicability guarantee includes queries of zero marginal mass. The main construction combines an ensemble of empirical-risk minimizers with a smooth, signal-dependent risk decomposition. A single shared phase selects the final binary predictor. Record times certify its weak-signal risk, while a uniform bound protects the strong-signal risk at that same selected time. A phase-comparison theorem controls the actual shared-seed execution under simultaneous changes in the marginals and risk coefficients. A rare repair step then makes the confidence cost additive. Together with matching lower bounds, this removes the remaining logarithmic overhead and parameter restrictions in the general countable-domain sample-complexity problem. This extended version also retains the earlier explicit constructions and their proofs, including proper learning for VC-one classes, the full finite-atom class, a faster realizable bound, and counterexamples that separate restrictions of particular algorithms from restrictions of the learning problem.
[LG-122] PACT: NVFP4 Data-Flow LoRA Fine-Tuning on Consumer GPUs
链接: https://arxiv.org/abs/2610.25048
作者: Lei Gao,Hossein Entezari Zarch,Chaoyi Jiang,Murali Annavaram
类目: Machine Learning (cs.LG)
*备注:
Abstract:Fine-tuning language models on consumer GPUs is constrained by computation and memory. Although LoRA reduces trainable parameters and optimizer state, frozen projections remain expensive, and activations retained for backward can dominate memory usage. We present PACT, a LoRA fine-tuning system that accelerates frozen projections with native NVFP4 GEMMs while reducing quantization and dequantization overhead and limiting quantization error. To reduce activation memory, PACT reuses quantized RMSNorm outputs and saves gate/up activations in NVFP4 for backward. To limit numerical error, it preserves BF16 inputs for forward nonlinear operations and calibrates frozen weights offline using GPTQ adapted to NVFP4 scales and quantized activations. Fused nonlinear kernels and cached forward and backward weight layouts reduce quantization and dequantization overhead. Evaluations across multiple models and tasks on a 16,GB RTX~5080 show lower memory usage and faster LoRA fine-tuning with competitive accuracy, enabling model adaptation on resource-constrained consumer hardware.
[LG-123] Integrable building blocks for symplectic neural networks: learned-type layers fiber-preserving layers and what a single layer can do
链接: https://arxiv.org/abs/2610.25023
作者: Andrew Qing He
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Symplectic Geometry (math.SG)
*备注: 27 pages, 28 figures
Abstract:Exactly symplectic neural networks are built almost exclusively from shear maps and gain expressivity only through depth. We ask what else a single explicit symplectic layer can be. Observing that every layer in use is the exact flow of a Hamiltonian that is a function of commuting integrals, we obtain three families of layers: a learned-type layer, in which every conjugate pair is rotated, squeezed or sheared by a state-dependent amount and the type is found by training; moving-plane shears, residual symplectic maps outside the known ridge-function class; and fiber-preserving layers, which combine an unconstrained point transformation with a fiber translation at the cost of one linear solve. We characterize residual symplectic layers x \mapsto x + J\nabla H(x) : in one degree of freedom they are exactly the shears along a line, while from two degrees of freedom on a strictly larger class exists. Conjugating a learned-type layer by a symplectic network yields an exact one-parameter group with an exactly conserved learned Hamiltonian. In experiments on sixteen systems with one to forty degrees of freedom, against SympNets, HénonNets and learned-plane shear networks, layers built on nonlinear integrals are far more accurate when the Hamiltonian contains functions of those integrals. On nonlinear Schrödinger lattices with six to forty sites, twist layers alternating with linear symplectic maps are over a hundred times more accurate than every shear-based network, and with an on-site potential of learned form three to four orders of magnitude more accurate. On chains of Tao oscillators with the step size as an input, stretch layers are 45 to 70 times more accurate than learned-plane shears and recover the Hamiltonian to 1%. On generic non-separable systems, fiber layers improve on learned-plane shears by at most a factor of two, and widening the networks does not help any architecture.
[LG-124] Decentralized Online Exp-Concave Optimization
链接: https://arxiv.org/abs/2610.25017
作者: Hao Qiu,Mengxiao Zhang,Yuanyu Wan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Decentralized Online Convex Optimization (D-OCO) is a fundamental framework in which a set of local agents aims to minimize a sequence of global loss functions using only local computations and communications. While recent work has established near-optimal regret guarantees for convex and strongly convex losses, the case of exp-concave losses remains an open challenge: specifically, whether logarithmic regret is achievable. In this paper, we answer this question affirmatively by proposing Decentralized Online Newton Step (D-ONS), which adapts the idea of blocking and accelerated gossip to the Online Newton Step algorithm in a non-trivial manner. Concretely, our algorithm achieves a regret bound of O\left(Nd(1-\sigma_2)^-1/2,\log^2(T) \right) where N is the number of agents, d is the dimension, T is the horizon, and 1-\sigma_2 represents the spectral gap of the network. To our knowledge, this is the first logarithmic regret bound for decentralized online exp-concave optimization. We further establish a nearly matching lower bound of \Omega\left(Nd (1-\sigma_2)^-1/2,\log(T/N) \right) showing the near optimality of our upper bound. Finally, we extend our algorithm to decentralized online linear regression in which the feasible domain is unbounded and features and labels are locally observed. By gossiping covariance and label-feature statistics, we also achieve logarithmic regret.
[LG-125] Multifidelity Monte Carlo Estimator for High-Dimensional Quantities of Interest
链接: https://arxiv.org/abs/2610.25016
作者: Zhibo Dai,Hamdi A. Tchelepi,Daniel M. Tartakovsky
类目: Machine Learning (cs.LG)
*备注:
Abstract:Multifidelity Monte Carlo (MFMC) reduces the cost of estimating the statistics of quantities of interest (QoIs) computed via expensive high-fidelity model by using cheaper low-fidelity models as control variates. We extend MFMC to vector-valued QoIs by replacing the scalar control-variate coefficients with matrices and minimizing a weighted mean-squared error. The optimal full-matrix coefficients are available in closed form, and their variance reduction is summarized by an effective correlation that takes the place of the squared Pearson correlation in the closed-form MFMC design. Restricting the coefficients to block-diagonal, diagonal, and scalar classes gives a nested hierarchy of estimators with increasing variance reduction and increasing pilot-data demand, with closed-form coefficients for compatible weightings. For a scalar QoI obtained by linearly projecting the state, correcting the full state before projecting is never worse at the population level. Experiments on a Gaussian problem, a reduced-order model of an advection–diffusion problem, and a coarse-grid two-phase flow problem show that the best class depends on the pilot size relative to the number of covariance quantities it must estimate, and that reusing the pilot samples lowers the mean-squared error.
[LG-126] Latent World Models Should Plan with Distances Along the Latent Manifold
链接: https://arxiv.org/abs/2610.25013
作者: Zunzhi You,Linwei Tao,Siyu Xu,Yunke Wang,Chang Xu
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Latent world models such as LeWM plan by searching for an action sequence whose predicted terminal latent is closest to the goal latent under an ambient norm, such as the Euclidean distance. Planning with such a cost assumes that proximity in the ambient latent space reflects proximity to the goal in the environment. Studying LeWM, we find that this assumption holds only in a small neighborhood around the goal, whereas farther away the L2 cost forms spurious low-cost basins that trap the planner. Yet neighborhood-based embeddings show that the latents retain the underlying physical state. We therefore hypothesize that the latents lie near a curved, low-dimensional manifold, and that the terminal cost should measure the distance along this latent manifold rather than through the ambient space. To estimate this manifold distance, we propose the Neighborhood Graph Cost (NGC), which computes shortest-path distances on a mutual nearest-neighbor graph constructed over training latents. NGC changes only the terminal cost and requires no retraining. This single change improves the planning success rate of LeWM across all four benchmark tasks and nearly solves TwoRoom and Reacher: even at the farthest goal offsets tested, NGC reaches 95% and 100% success, whereas the L2 cost falls to 27% and 81%. These results support the manifold hypothesis and our argument that latent world models should plan with distances along the latent manifold.
[LG-127] Conjugate Compilation: Self-Improving Agent ic Co-Compilation for Speculative Decoding
链接: https://arxiv.org/abs/2610.25012
作者: Annabelle Sujun Tang,Rohan Mahapatra,Christopher Priebe,Hanyang Xu,Manya Ghobadi,Christopher Rossbach,Hadi Esmaeilzadeh
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Programming Languages (cs.PL)
*备注:
Abstract:As LLM adoption surges, scaling model serving requires unlocking stranded GPU capacity alongside deploying more GPUs. Speculative decoding brings two models into the serving process: a small draft model proposes tokens that a larger target model verifies. Yet compiling these models individually leaves their shared execution outside the compiler’s optimization boundary. We introduce CONJUGATE COMPILATION, which reframes speculative decoding as a cross-model compilation problem and jointly optimizes the draft and target models within their shared execution rather than treating each in isolation. Our key insight is that cross-model compilation should be informed by the measured performance of the co-located shared execution it produces: an optimization that improves a kernel in isolation need not improve the models when they execute together. To expand cross-model opportunities, we devise operand layouts that interleave blocks from both models within common tiles, co-locating GEMM and non-GEMM work while preserving their mathematics. We realize CONJUGATE COMPILATION through a self-improving agentic search system in which an LLM edits cross-model optimization plans using the compiler and GPU execution as its environment. Each proposed plan is compiled and evaluated through its co-located shared execution, whose measured score, together with kernel-level verdicts and compiler feedback, informs subsequent proposals, while Monte Carlo tree search accumulates rewards to guide exploration. Across six single-GPU speculative-decoding configurations on NVIDIA B300 and RTX 5090 GPUs, CONJUGATE COMPILATION achieves a geometric-mean per-round speedup of 1.37x over conventional execution. Interleaved kernels raise this speedup to 1.45x, reaching 1.59x in the best case. CONJUGATE COMPILATION also improves GPU utilization on average by 1.46x on RTX 5090 and 1.50x on B300.
[LG-128] No Escape from the Quadratic Barrier: Adaptive Attacks on Linear ell_2 Sketches
链接: https://arxiv.org/abs/2610.25008
作者: Edith Cohen,Uri Stemmer
类目: Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注:
Abstract:Linear sketches estimate properties of high-dimensional inputs from a fixed low-dimensional projection. At constant accuracy, randomized Euclidean-norm sketches, notably the Johnson-Lindenstrauss transform, support exponentially many nonadaptive queries in sketch dimension k . For adaptive black-box queries, known privacy-based defenses guarantee \widetilde\Omega(k^2) and it remained open whether stronger guarantees were possible. Prior attacks used more queries, restricted the responder, or provided weaker failure guarantees. We show that the quadratic guarantee is optimal up to logarithmic factors, even with a computationally unbounded responder that knows the matrix, attack algorithm, history, and query distributions and may change its randomized rule after every query. For every fixed hidden A\in\mathbbR^k\times n with n40k and target accuracy \alpha \in (0,1] , our black-box attack uses O(k+\alpha^2k^2\log(2+\alpha^2k)) queries in \widetildeO(1+\alpha^2k) adaptive rounds to force, with constant probability, a round in which a constant fraction of answers have \Omega(\alpha) relative error. Our results clarify the limits of defending a fixed compressive linear representation by adapting the prediction rule.
[LG-129] Block Recursive Transformers for Structured Parameter Sharing EMNLP
链接: https://arxiv.org/abs/2610.24998
作者: Josep Marques,Hyunyoung A. Cho,Youngeun Kim,Youjia Zhang,Aecheon Jung,Kyuhong Shim,Sungeun Hong
类目: Machine Learning (cs.LG)
*备注: 13 pages, 9 figures. Published at EMNLP Findings 2026. Typo correction from camera-ready manuscript. Project website: this https URL
Abstract:Recursive Transformers depend not only on how many parameters are shared, but also on how shared parameters are organized across depth. Existing static schemes adopt two extreme reuse topologies: layer-wise repetition, where each layer is reused independently, and global cyclic reuse, where the entire stack is repeatedly applied as a single shared function. We argue that sharing granularity is an important factor in recursive parameter sharing. This study proposes Block Recursive Transformers (BRT), which partition layers into contiguous blocks and recursively reuse each block before moving to the next. This intermediate topology enables local refinement within blocks while preserving depth-wise transitions, without adding routing modules, auxiliary states, or extra training objectives. Across decoder-only models up to 1.7B parameters, with additional encoder-decoder and encoder-only studies, BRT improves validation negative log-likelihood and downstream accuracy over static recursive sharing baselines. Ablations and representation analyses further show that intermediate block granularity better balances local refinement and depth-wise differentiation. Project website: this https URL
[LG-130] -MUON: Tensorizing Muon via t-SVD for Large Language Model Pre-training
链接: https://arxiv.org/abs/2610.24938
作者: Dongyang Li,Mingsong Yan,Zheng Zhang,Sui Tang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Muon has recently emerged as a strong optimizer for large language model pretraining by exploiting matrix structure through polar orthogonalization. However, it processes each parameter matrix independently, whereas recent tensor-aware extensions of Muon jointly update groups of eligible matrix parameters through matrix unfoldings. In contrast, we propose T-MUON, a t-SVD-based generalization that groups structurally related Transformer gradients into third-order tensors and defines the corresponding Muon-style update directly through the M -transform. We show that the exact T-MUON update is a steepest-descent direction under the \star_M -spectral norm and establish a convergence guarantee. Across GPT-2-style and LLaMA-style models from 51M to 1.3B parameters trained on FineWeb and FineWeb-Edu, head-wise T-MUON with the identity transform consistently achieves lower validation perplexity than AdamW, Muon, and variants using DFT, DCT, or DST transforms. These gains persist in additional experiments with longer context lengths and larger token budgets, although the preferred tensorization mode varies across settings.
[LG-131] MLFont: Machine Learning-Driven Font Subsetting and Range-Compact Unicode-Range Slicing for Low-Latency Web Typography
链接: https://arxiv.org/abs/2610.24937
作者: Quan Wang,Yiran Mao
类目: Machine Learning (cs.LG)
*备注:
Abstract:CSS3 @font-face rules with unicode-range descriptors partition web fonts into downloadable slices. While Latin fonts need ~20 KB in WOFF2, Pan-European (LGC) and East Asian (CJK) fonts contain 2,297-30,759 codepoints (158 KB-5.9 MB in WOFF2), where heuristic partitioning causes unused-glyph downloads, high HTTP RTT, CSS bloat, or missing-glyph fallbacks. We present MLFont, a machine learning and optimization framework that learns font slicing strategies from corpora via three stages: (1) a Zipfian High-Frequency Tier (most) isolating top-frequency codepoints; (2) a Latent Co-Occurrence Topic Tier using Online Variational Bayes LDA (Online LDA) or a Deep Neural Network Topic Model (DNNTM) with Shannon Entropy Difference Loss and Sinkhorn-Knopp normalization; and (3) a Memetic Range-Compact Genetic Algorithm (GA) with Potts-Viterbi dynamic programming and island bridging that minimizes topic discrepancy and CSS fragmentation. Across five Google Noto fonts (LGC, KR, JP, TC, SC), MLFont cuts 3G per-glyph latency by 56.7%-75.3% over single-file delivery and 46.1%-66.3% over equal-bin baselines on CJK with 100% coverage, reduces CSS intervals by 17.9%-41.7%, and achieves 1.56x-6.09x cold-cache speedups (65.4%-84.9% CJK and 44.7% LGC bandwidth savings) across 20 websites in simulator and Headless Chrome benchmarks (45/45 exact slice-set parity). Code: this https URL Models: this https URL Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.24937 [cs.LG] (or arXiv:2610.24937v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.24937 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-132] DF-PALM: Dynamic Fusion Prompt Learning for Audio–Language Models
链接: https://arxiv.org/abs/2610.24921
作者: Yulin Sun,Shenghua Li,Yong Dou,Xinwang Liu,Cheng Yang,Kele Xu
类目: ound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
*备注: 13 pages, 5 figures, 14 tables
Abstract:Few-shot audio classification requires audio–language models to distinguish variable acoustic events from limited labeled data. PALM’s main formulation adapts encoded class representations without changing pretrained encoders, but its class representations remain input-independent and classification uses audio features without class-conditioned refinement. We introduce DF-PALM, a dynamic feature-space prompting framework that combines audio-conditioned class representations with asymmetric cross-attention for audio refinement. Class representations are conditioned after text encoding, held fixed across fusion layers for each input, and reused for final scoring. Both pretrained encoders remain frozen, avoiding audio-conditioned text-encoder passes. With PENGI as the frozen backbone, the main 16-shot evaluation covered 11 benchmarks and six baselines. DF-PALM achieved 80.69% mean accuracy and 80.32% macro-averaged F1, exceeding PALM by 4.11 and 4.30 percentage points, respectively. Paired benchmark-level tests supported both gains after Holm correction. Across four shot settings, DF-PALM achieved the highest mean accuracy and macro-averaged F1 among the four evaluated prompt-learning methods. The gains extended to LAION-CLAP and OpenFLAM, with significant improvements over PALM in both metrics. At 50 candidate classes, operator-counted inference floating-point operations increased by 0.39% relative to PALM. These results support instance-conditioned feature-space adaptation as an effective approach to few-shot audio classification. Our code is publicly available at GitHub this https URL.
[LG-133] ATLAS: Expert-Guided Sustainable Satellite Backhaul for Off-Grid Wildlife Monitoring ATC2026
链接: https://arxiv.org/abs/2610.24910
作者: Hao Fang,Miao Zhang,Haoyuan Zhao,Long Chen,Yi Ching Chou,Chi Xu,Cheng Pan,Jiangchuan Liu
类目: ystems and Control (eess.SY); Machine Learning (cs.LG)
*备注: Accepted at the ACM SIGOPS Annual Technical Conference (ATC 2026)
Abstract:Remote wildlife monitoring increasingly depends on high-fidelity cameras and sensors deployed far from grid power and terrestrial networks. Low Earth orbit satellite links can provide connectivity in these settings, but their high power draw and variable bandwidth, combined with intermittent on-site solar harvesting, make continuous backhaul infeasible. Current systems therefore filter at the edge, with motion triggers that waste limited energy and bandwidth on environmental artifacts, or fixed vision models that cannot follow evolving expert objectives. We present \textscATLAS, an expert-guided edge-to-cloud system that treats satellite backhaul as a scarce resource for off-grid wildlife monitoring. \textscATLAS uses foundation models to compile natural-language expert intent into semantic scoring logic, combines semantic relevance with environmental and energy state, and uses a constrained reinforcement-learning scheduler to decide which clips to upload and when. As a representative case study, we deployed \textscATLAS for over six months across 18 off-grid salmon-monitoring river sites. \textscATLAS achieves a mean clip-selection accuracy of 87.4% , exceeding mean baseline accuracy by 26.4 percentage points, while maintaining near-continuous operation through low-solar periods in which baselines incur 16.0% downtime. Comments: Accepted at the ACM SIGOPS Annual Technical Conference (ATC 2026) Subjects: Systems and Control (eess.SY); Machine Learning (cs.LG) Cite as: arXiv:2610.24910 [eess.SY] (or arXiv:2610.24910v1 [eess.SY] for this version) https://doi.org/10.48550/arXiv.2610.24910 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Related DOI: https://doi.org/10.1145/3828523.3857049 Focus to learn more DOI(s) linking to related resources
[LG-134] A Symmetry-Matching Approach to Blind-Spot Reduction in Sparse Autoencoders
链接: https://arxiv.org/abs/2610.24908
作者: Csaba Balogh
类目: Machine Learning (cs.LG)
*备注: this https URL
Abstract:Language models can treat semantically distinct inputs as interchangeable at the representation level, creating blind spots that standard sparse autoencoder (SAE) objectives do not target. In safety-critical settings such as clinical dosage extraction, legal clause interpretation, or financial amount verification, such failures can propagate silently into downstream decisions. We study this problem as one of feature-basis orientation. Building on a symmetry-matching view from algebraic error-detection theory, we add a differentiable V_\mathrmGini regularisation term to SAE training that penalises uneven perturbation response within a specified family. Across GPT-2, Gemma 2, and Qwen 2.5, a frozen multi-seed evaluation shows that V -regularisation raises both relative and absolute lower-tail responses while reducing response-profile imbalance, with no material general-text reconstruction degradation. JumpReLU and minimum description length (MDL) objectives remain close to the Standard SAE profile under the same protocol. Family-geometry controls show that the effect is not explained by template duplication or a single dominant perturbation family, and held-out probes plus null-calibrated separation tests evaluate how far the representation-level change carries to downstream discriminability. A zero-shot audit on out-of-distribution clinical radiology minimal pairs finds the relative weak-tail lift in all evaluated model-by-family cells, not perfectly rank-aligned with the trained statistic, and a held-out family audit finds it on family kinds never used in training. Together, the results show that perturbation-aware SAE training can reorient feature bases toward safety-relevant input changes while leaving downstream validation as a separate empirical step.
[LG-135] Distribution-Based Regularization for Reward Fine-Tuning Diffusion and Flow Models
链接: https://arxiv.org/abs/2610.24897
作者: Romeo Passaro,Jeongjae Lee,Jong Chul Ye,Michael M. Bronstein,Yoshua Bengio,Alexander Tong,Danyal Rehman
类目: Machine Learning (cs.LG)
*备注: 38 pages, 9 figures
Abstract:Reward fine-tuning is an effective way to adapt pretrained diffusion and flow-matching models to downstream objectives, but aggressive optimization often leads to mode collapse and distributional drift. Existing methods attempt to control this drift with reverse-KL regularization, implemented as an \ell_2 penalty on the velocity field; however, this functions by indirectly constraining sampling dynamics rather than directly acting on the generated distribution. To address this, we propose DFR (Distributional Functional Rewards), which augments the task reward with a distribution-dependent, per-sample reward that acts explicitly on the generated distribution. Formally, we generalize the KL-regularized objective from expected rewards, which are linear in the model’s distribution, to convex functionals of it. The first variation of this objective defines an effective, distribution-dependent reward that directly penalizes deviation from the pretrained distribution. This formulation yields additive per-sample rewards for the Fréchet distance (FD), maximum mean discrepancy (MMD), and DreamSim diversity, recovering several recently proposed rewards as special cases. Across various text-to-image fine-tuning tasks, DFR matches the peak reward of KL-tuned baselines while increasing diversity by up to 29% at equal or lower drift.
[LG-136] Query-Dependent Reading for Quantized Inner-Product Estimation: A Minimax Separation
链接: https://arxiv.org/abs/2610.24890
作者: Krishnakumar Balasubramanian,Siddharth Bhandari,Shiva Prasad Kasiviswanathan
类目: Information Theory (cs.IT); Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注: 66 pages
Abstract:We study quantization for inner-product estimation when storage and reading have separate budgets. In the static bit-probe model [Yao, 1981], an encoder stores w bits about x before seeing a query y , and a reader probes at most rw bits to estimate x^\top y . Both vectors lie in the d -dimensional Euclidean unit ball. Protocols may be randomized, and computation is unrestricted. Performance is measured by d times the worst-case mean-squared error over x,y , averaged over protocol randomness. First, we show that query-independent reading cannot benefit from extra storage. Any such reader is equivalent to fully reading an r -bit message, even when later reads depend on previously returned bits. At fixed positive rates W=w/d and R=r/d , the normalized minimax error converges to 1/(4^R-1) , independent of W . The upper bound was already known [Ordentlich and Polyanskiy, 2025]. We prove the matching lower bound for arbitrary scalar decoders, extending the result for unbiased compressors [Safaryan et al., 2020]. Then, we show that query-dependent reading makes extra storage useful. We construct a two-batch protocol using Gaussian projections and random thresholds, with asymptotic normalized error at most (\pi+o(1))/[4R\log(W/R)] as W/R\to\infty . At fixed R , this error vanishes as storage grows, giving an unbounded advantage over query-independent reading. We also prove a finite-dimensional lower bound for arbitrary encoders, decoders, and fully branching readers. In the doubly sparse regime W/R\to\infty and R\log(W/R)\to0 , it gives normalized error at least (1-o(1))/[2R\log(W/R)] . Thus, our protocol is optimal within a factor of \pi/2 in the leading constant. Comments: 66 pages Subjects: Information Theory (cs.IT); Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG) Cite as: arXiv:2610.24890 [cs.IT] (or arXiv:2610.24890v1 [cs.IT] for this version) https://doi.org/10.48550/arXiv.2610.24890 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-137] Sample-Efficient Reinforcement Learning with General Utilities via Neural Dual Critics
链接: https://arxiv.org/abs/2610.24880
作者: Ankur Naskar,Vaneet Aggarwal
类目: Machine Learning (cs.LG)
*备注:
Abstract:Reinforcement learning with general utilities optimizes nonlinear functionals of policy-induced occupancy measures. We develop a natural actor-critic method for concave RLGU objectives admitting an exact variational dual representation, with explicit constructions for a broad class of separable occupancy utilities. Rather than estimating the occupancy measure and then forming the policy-gradient pseudo-reward, our method learns this pseudo-reward directly as a function of the state-action pair using a neural critic. The critic is trained by projected stochastic updates, while multilevel Monte Carlo combines outputs across training horizons to control its bias at logarithmic expected cost. Our analysis separates conditional actor bias from estimator second moments and accounts explicitly for dual optimization, critic approximation, and neural linearization errors. Under policy-transfer, dual-stability, and local critic-regularity conditions, the method achieves \widetildeO(T^-1/2) global convergence above policy, critic-class, and finite-width approximation errors using \widetildeO(T) expected environment transitions and reference samples. For a fixed critic representation, this yields \widetildeO(\epsilon^-2) sampling complexity above the approximation floor, covering infinite state-action spaces and occupancy utilities beyond fixed finite vectors of returns.
[LG-138] FlowMaker: From Tasks to Cost-Efficient Generative AI Workflow Deployments
链接: https://arxiv.org/abs/2610.24861
作者: Dimitrios Liakopoulos,Prasoon Sinha,Neeraja J. Yadwadkar
类目: Machine Learning (cs.LG)
*备注:
Abstract:Generative AI workflows execute user tasks by composing sequences of LLM calls, where each call performs a sub-task (e.g., summarization or translation), forming a workflow. Executing such workflows requires decisions over workflow structure, model assignments, and deployment configurations, which jointly determine latency, cost, and accuracy under dynamic workloads and resource constraints. Existing systems optimize these decisions in isolation or fix them early, leading to suboptimal cost and goodput. We make two key observations. First, efficient serving requires jointly exploring all combinations of workflows, model choices, and deployment configurations under dynamic cluster state. Second, reuse of deployed LLMs is critical for reducing cost, but must be performed on-the-fly, as reuse depends on current load and resource availability. Together, these make selecting a workflow-variant-a combination of workflow, models, and deployment configuration-a time-critical problem. We present FlowMaker, which exposes a task-level interface and selects workflow-variants on behalf of users. To realize this, we design a constraint-guided load-aware pruning algorithm for fast, on-the-fly decisions. FlowMaker improves goodput by up to 10.14 \times and reduces cost by up to 2.77 \times compared to state-of-the-art baselines, including Murakkab.
[LG-139] LLM -Based Evaluation Planning for Heterogeneous Optimization
链接: https://arxiv.org/abs/2610.24855
作者: Nian Ran,Sebastian Brulin,Markus Olhofer,Mauricio A. Álvarez,Richard Allmendinger
类目: Machine Learning (cs.LG)
*备注: 10 pages under review
Abstract:Heterogeneous optimization problems require an optimizer to decide not only which candidate solutions to evaluate, but also which objectives and constraints should be evaluated under different costs and fidelities. Existing SOTA methods such as MFE-NSGA-III address this setting with manually designed selection metrics, but their behavior is largely determined by predefined heuristic rules and hyperparameters. In this work, we propose REP-HO, a reasoning-based evaluation planner that replaces the rule-based mixed-fidelity selection step with a large language model (LLM) planner. The LLM receives the same primitive optimization-state information as the heuristic planner, such as surrogate predictions, uncertainties, evaluation costs, and budget status. We first show that REP-HO achieves competitive performance compared to the human-designed metric-based method (MFE-NSGA-III). We then further enhance each decision with an adaptive nearest-neighbor summary, leading to better and more stable results. We evaluate REP-HO across five heterogeneous regimes covering cost imbalance, constraint difficulty, surrogate fidelity, budget level, and objective conflict. Across these regimes, one of the two variants obtains the best mean ranking score in every regime, REP-HO(Deepseek) is first or second throughout, and the planner shows adaptive allocation behavior, such as prioritizing constraint evaluations under tighter budgets and reallocating constraint evaluation budget when surrogate reliability changes. These results suggest that LLMs can serve as flexible evaluation planners for heterogeneous optimization, offering a general alternative to fixed hand-designed metrics.
[LG-140] An Impossibility Result for ETF Target Geometry and an Equinorm Remedy for Imbalanced Analytic Continual Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.24844
作者: Berk Sezer
类目: Machine Learning (cs.LG)
*备注: Accepted at the NeurReps 2026 Workshop (Extended Abstract Track), NeurIPS 2026
Abstract:Analytic continual learning (ACL) methods such as ACIL and GACL fit a ridge-regression classifier head on frozen features via the Sherman-Morrison-Woodbury identity, achieving exemplar-free, exact updates. A separate line of work, Progressive Neural Collapse (ProNC), shows standard continual classifiers benefit from a progressively expanding Simplex Equiangular Tight Frame (ETF) target. It is natural to ask whether fusing the two ideas, exact analytic updates and ETF targets, would yield a better closed-form continual learner. We show this is not the case as the fusion turns out to be identical to plain one-hot ACIL. Further, Simplex ETFs are not nested across class counts, so the literal “progressively expand the ETF” construction stops yielding an ETF as new tasks arrive, and it introduces catastrophic forgetting ( 4\times that of a classic ACIL head). We therefore separate an ETF’s two defining properties, equiangularity and equinormality, and apply each on its own to the classifier weights rather than the targets, where the invariance above does not hold. Equinorm post-processing of the fitted weights recovers both tail-class and overall average accuracy across the long-tailed regime from 5:1 to 100:1 imbalance (e.g. +32.0 tail-accuracy points and +16.0 average-accuracy points at 100:1 on CIFAR-100/ViT), transferring in sign across two backbones and two datasets, at a mild cost of 5.4 average-accuracy points on CIFAR-100/ViT on near-balanced streams (below \approx 3:1 ). Equiangular geometry, by contrast, is exactly inert on the target side but actively harmful on the weight side, usually losing to plain one-hot. Hence, the generally useful part of Neural Collapse geometry in Analytic Continual Learning is equinorm post-processing of the weights, becoming more useful as class imbalance grows; and equiangularity is at best a no-op and induces catastrophic forgetting at worst.
[LG-141] A Statistical Viewpoint on Adam Shampoo and Muon
链接: https://arxiv.org/abs/2610.24839
作者: Tetiana Parshakova,Robert M. Gower,Paulo Orenstein
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We provide new statistical insights into designing stochastic optimization methods. By modeling the gradient via a normal distribution, we argue that the stochastic gradient should be whitened using its covariance. We give an online method for estimating the mean and variance of the gradients based on regularized likelihood or, equivalently, through a maximum a posteriori estimate. We then show how to use structured approximations to the covariance within this same framework, which includes diagonal and Kronecker-factored approximations. The diagonal approximation results in a method that is very closely related to Adam and performs as well in practice. The Kronecker approximation results in the new MN method, which is a variant of Shampoo and KL-Shampoo. We also recover Muon by applying this same statistical framework to the gradient direction rather than the full gradient. By combining this with a whitening step, we arrive at MN-Muon, a novel preconditioned version of Muon. Our experiments on GPT-2-type models at Chinchilla scale indicate that MND, the diagonal implementation of MN, outperforms both the Distributed Shampoo and the KL-Shampoo implementations in terms of stability, final validation loss and time taken. MND-Muon, the diagonal implementation of MN-Muon, improves on Muon in 21 of 24 paired GPT-small runs and achieves a lower best validation loss on GPT-medium, where it remains stable over the six tested learning rates. However, we found that NorMuon was comparable to MND-Muon and achieved a slightly lower validation loss on GPT-medium and at GPT-small batch 512 on FineWeb.
[LG-142] Recurrent Temporal Restricted Boltzmann Machines as Unsupervised Encoders for Biomechanical Movement Sequences
链接: https://arxiv.org/abs/2610.24829
作者: Wafi Sharif,Nicholas LaHaye
类目: Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: 15 pages, 7 figures, 5 tables. Under review at Transactions on Machine Learning Research (TMLR)
Abstract:Biomechanical movement recordings are continuous, high-dimensional time series, and analyzing them without manually labeling them requires representations that capture how movements evolve over time. Most unsupervised representation-learning pipelines, however, treat each observation frame independently and discard this temporal structure, while energy-based models that could capture it have not extensively been adapted to biomechanical signals. So, this paper implements the Recurrent Temporal Restricted Boltzmann Machine (RTRBM), a recurrent energy-based model, into an unsupervised encoder that maps windows of movement data to fixed-length embeddings, and tests what its recurrent structure contributes. We adapt the model to continuous-valued signals, stack it into a deep architecture, and integrate it into the SIT-FUSE segmentation pipeline. We evaluate the embeddings on simulated musculoskeletal reaching and on two real gait datasets, using clustering, reconstruction, and walking-speed classification of unseen subjects, and compare them against non-recurrent and non-probabilistic baselines, including a control that removes the recurrent connection. Clustering the embeddings recovers movement structure tied to progress toward the target without any labels, which also transfers to unseen episodes. On gait data, the embeddings outperform PCA and autoencoder baselines under a nonlinear probe. However, removing the recurrent connection does not yield a measurable loss on the reaching task, and simple per-channel summary statistics classify walking speed at least as well as the embeddings. The model is therefore a working, verified encoder for SIT-FUSE, while the evidence for a benefit of recurrence is mixed, and the learned embeddings did not outperform simple summary statistics on gait.
[LG-143] Compose the Horizon: Parameter-Shared Blockwise Forecasting under a Sequential-Compute Trade-off
链接: https://arxiv.org/abs/2610.24826
作者: Zeyan Li,Weiheng Li,Shengda Zhuo,Libing Chen,Min Chen,Yin Tang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Most long-horizon forecasters emit the complete future in one pass. We ask whether a compact operator can instead be reused to emit contiguous future blocks, trading stored parameters for sequential computation. Recursive Functional Decomposition (RFD) factorizes the horizon mapping into three parameter-shared block predictions. The blocks occupy disjoint future intervals and are concatenated, not added as trend, seasonal, or residual components, and each emitted block joins the rolling context that conditions the next application. In controlled three-seed experiments, removing forward feedback hurts all four evaluated datasets while stopping gradients through it changes little, so the useful signal is the evolving forecast state itself; reapplying the shared operator helps consistently, but by less than widening it. The model uses 0.08–0.46M parameters, yet its sequential channel-wise execution is slower and more memory-intensive than an iterative baseline at high channel count. In a frozen long-context evaluation against ten strong baseline families, RFD records lower test error in 23 of 24 cells; the comparison is cross-protocol, with RFD using 512/720-step context and the baselines using 96.
[LG-144] Process-Aware Event Log Augmentation for Process Prediction under Labeled Data Scarcity
链接: https://arxiv.org/abs/2610.24777
作者: Sjoerd van Straten,Alessandro Padella,Marwan Hassani
类目: Machine Learning (cs.LG)
*备注: This paper has been accepted for the International Journal of Cooperative Information Systems
Abstract:Predictive Process Monitoring (PPM) forecasts future events of ongoing business processes based on event logs. However, deep learning PPM approaches are often limited by labeled data scarcity and the low variability of real-world logs. To address this, we introduce SiamSA-PPM, a self-supervised framework combining Siamese learning with statistically grounded data augmentation. It employs novel transformation methods that leverage control-flow semantics and frequent behavioral patterns to generate realistic, semantically valid trace variants, learning generalizable prefix representations without labeled supervision. Extensive experiments on real-world event logs with a wide range of variability, incorporating exhaustive hyperparameter tuning, demonstrate the value of this process-aware augmentation. Compared to state-of-the-art baselines, SiamSA-PPM achieves competitive or superior predictive performance in both next activity and final outcome prediction tasks. An open source implementation of our method is publicly available under this https URL…
[LG-145] Global Convergence in Deep Networks via the Second Law of Thermodynamics NEURIPS2026
链接: https://arxiv.org/abs/2610.24775
作者: Matan Tsipory,Ofir Gaash,Itamar Harel,Nathan Srebro,Daniel Soudry
类目: Machine Learning (cs.LG)
*备注: Accepted to NeurIPS 2026
Abstract:Existing convergence analyses of finite-width overparameterized deep neural networks under standard parameterization often rely on parameters remaining close to their initialization and typically neglect the role of noise, despite empirical evidence that stochasticity can improve optimization and generalization. To bridge this gap, we study L_2 -regularized continuous-time Langevin dynamics (CLD) with isotropic noise, for training feedforward networks with smooth activations. We introduce a change-of-measure inequality, inspired by the second law of thermodynamics, that enables control of key properties of the training dynamics under the time-evolving parameter distribution. Leveraging a sharpened lower bound on the minimum eigenvalue of the Neural Tangent Kernel (NTK) at initialization, we show that injected noise helps preserve NTK stability with high probability without confining the weights to a small neighborhood of their initialization. As a result, we derive a non-asymptotic upper bound on the expected training loss. Our analysis yields explicit conditions on width and noise temperature that guarantee convergence to a finite loss floor. Finally, we experimentally show the network Jacobian can change from initialization, departing the commonly analyzed lazy regime.
[LG-146] Paired-Outcome Learning for Selective Spacecraft Attitude Capture under Torque Limits
链接: https://arxiv.org/abs/2610.24763
作者: Donghe Chen,Jia Deng,Zehao Jia,Lin Cheng,Shengping Gong
类目: Machine Learning (cs.LG)
*备注:
Abstract:Torque-limited spacecraft attitude capture requires adapting the maneuver speed to uncertain braking authority. Faster settings can reduce average tracking error while worsening individual maneuvers, making selective modification of an existing controller a baseline-relative learning problem. A paired-outcome method combines three mechanisms: physically interpretable continuation policies with different speed envelopes, separate prediction of task failure and successful-pair tracking gains, and context-dependent rejection that executes the unchanged baseline. Complete paired simulations provide labels under common plant parameters and measurement seeds; training, validation, and calibration are separated by context. Basilisk experiments use 108 development contexts, 324 labeling rollouts, and 624 evaluation episodes across 48 test contexts, including disclosed post-freeze diagnostics. The selective controller reduces aggregate integrated attitude error by 4.67% relative to the branch baseline, with 17 improvements, 31 exact ties, and no observed regressions. Always using the fast alternative achieves a 4.81% reduction but worsens 11 cases. Gains occur in principal-axis, near-axis, braking-boundary, and selected noise/drift maneuvers; general coupled cases retain the baseline. Ablations show no additional aggregate benefit from branch commitment, active querying, or score subtraction. The evidence supports a restricted empirical trade-off between average improvement and individual regression, without establishing a closed-loop safety guarantee.
[LG-147] From Overparameterised Expressions to Parsimonious Models: Fisher-Geometric Pruning for Symbolic Regression
链接: https://arxiv.org/abs/2610.24753
作者: Veronica Guidetti,Luca La Rocca,Fabrício Olivetti de França,Gabriel Kronberger
类目: Machine Learning (cs.LG)
*备注:
Abstract:Genetic programming symbolic regression (GPSR) often returns expressions that are larger than necessary, with dormant branches and finely tuned algebraic configurations that fit sampling noise. Existing complexity-control strategies, such as multi-objective size-accuracy search or information criteria embedded in the fitness function, act on syntactic representations during model search. We introduce Fisher-Geometric pruning (FGP), a post-hoc model-reduction framework that acts on fitted models and can complement these techniques. FGP combines local information geometry at the maximum-likelihood solution with explicit compatibility tests for interpretable parametric boundaries. A taxonomy of parametric directions separates local sensitivity from proximity to reducible boundaries, enabling the treatment of unidentifiable directions and reductions through zero constraints, asymptotic limits, and signed parameter identifications. Candidate reductions are realised by symbolic contraction and refitting, then accepted only when their cross-validated predictive risk remains within a prescribed tolerance. We evaluate FGP on controlled diagnostic problems and Pareto fronts generated by GPSR. Across both settings, FGP substantially reduces structural complexity while preserving, and often improving, out-of-sample accuracy. Comparisons with competing pruning strategies and ablations show that, by combining geometric screening, boundary compatibility, and predictive validation, FGP yields the most consistent joint gains in parsimony and generalisation. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.24753 [cs.LG] (or arXiv:2610.24753v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.24753 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-148] Gaussian Process Bandits with Paired Auxiliary Observations
链接: https://arxiv.org/abs/2610.24737
作者: Weibo Huang,Qing Feng,Ruihao Zhu,Cheng Hua
类目: Machine Learning (cs.LG); Optimization and Control (math.OC); Machine Learning (stat.ML)
*备注:
Abstract:Bayesian optimization (BO) is widely used for expensive black-box optimization, yet observation noise makes each evaluation less informative and slows posterior learning. Standard BO learns from costly outcomes gathered during optimization. In many applications, however, low-cost auxiliary data are available beforehand. We introduce Auxiliary-Informed Gaussian Process Bandit (AI-GPB), a variance-reduction method that combines an offline estimate of the auxiliary mean with a correction coefficient learned from paired online observations. Notably, it exploits correlation between paired observation errors without requiring the auxiliary mean to approximate the objective. We establish finite-sample GP confidence bounds and regret guarantees, including explicit cumulative and simple regret rates for Matérn kernels. Numerical experiments across continuous BO benchmarks with varied landscapes and dimensions show consistent reductions in cumulative and simple regret relative to vanilla GP-UCB, even when within-pair correlation is moderate and the auxiliary mean is estimated from noisy offline data.
[LG-149] How Does Muon Work in Private Deep Learning? Scalable Differentially Private Training via Orthogonalization
链接: https://arxiv.org/abs/2610.24736
作者: Alexander Gaponov,Grigory Malinovsky,Egor Shulgin,Peter Richtarik
类目: Machine Learning (cs.LG)
*备注:
Abstract:Training deep neural networks requires optimizers that balance computational efficiency with formal data privacy. While orthogonalized update rules, such as the recently introduced Muon optimizer, have shown strong empirical performance and scalability, their behavior under Differential Privacy (DP) is not yet understood. In this paper, we propose DP-Muon, a differentially private variant of the Muon optimizer. By analyzing Muon’s orthogonalized updates through the lens of a Linear Minimization Oracle (LMO), we show that the leading effect of per-example clipping is a positive rescaling of the gradient, to which the LMO is blind, so that only a small directional residual controlled by the centered gradient variance enters the optimization error. This yields an \mathcal O(K^-1/4) convergence rate, up to logarithmic factors, to a clipping-dependent neighborhood at a fixed noise scale and, with noise calibrated to a fixed (\varepsilon,\delta) budget, the standard (\sqrt d/(n\varepsilon))^1/2 privacy–utility bound of DP-SGD, without assuming bounded gradients. Furthermore, we investigate the hyperparameter landscape of DP-Muon and document how the optimal learning rate, momentum, and batch size change under privacy. Empirically, on image classification DP-Muon matches or improves on DP-SGD and DP-Adam, whereas on language modeling the best optimizer depends on the batch size, with DP-Muon best at small batches and DP-Adam best near the privacy-optimal batch size.
[LG-150] fxBench: Evaluating and Understanding Formula Suggestions in Spreadsheets
链接: https://arxiv.org/abs/2610.24729
作者: Sanket Mhatre,Sumit Gulwani,Vu Le,Yasharth Bajpai,Gust Verbruggen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Spreadsheets are among the most widely used tools for computations, and formulas are their core building blocks. Besides agentic spreadsheet tools, proactive formula suggestions enable significant productivity gains and are now available in popular spreadsheet software (Microsoft Excel and Google Sheets). Such suggestions pose interesting challenges: the right context must be collected and represented for a machine learning model, which must then be able to write a correct formula, in a way that is fast and cheap enough to trigger on every = (or subsequent keystrokes for auto-completion) without taking users out of their editing flow. Surprisingly, there is no public benchmark for formula suggestions on real spreadsheets. In this paper, we therefore (1) introduce fxBench as a manually curated benchmark of 503 formula suggestion tasks based on Sheetpedia, (2) break down the formula suggestion problem into four capabilities (gather context, understand context, understand intent, write formula) that a formula suggestion system requires, and (3) evaluate existing and new approaches to understand bottlenecks in these capabilities that pose interesting research directions. Specifically, we annotate each benchmark with the cells that describe intent to determine the theoretical performance limit of a given context selection strategy, we evaluate performance of small and large models on different representations of context, including a dense format with high theoretical coverage at low token counts, and we fine-tune smaller language models to show that we can learn to understand such dense formats. Additionally, we perform quantitative and qualitative analysis on failure modes to understand bottlenecks in the formula writing capability; the best configuration only achieved 58.8% correct suggestions.
[LG-151] Native Probability Outputs for Clinical Triage: An Evaluation of Calibration and Decision Accuracy
链接: https://arxiv.org/abs/2610.24692
作者: Liang You,Zhiyin Zhou,Dawei Tang,Chuchu Wu,Jingzhuo Hu
类目: Machine Learning (cs.LG); Applications (stat.AP)
*备注:
Abstract:Patient-facing triage requires an appropriate care recommendation and recognition of when information is insufficient. We evaluate whether Jev, a decision model that returns probabilities over predefined actions, supports both judgments. Using CARE-Bench, we compare Jev with four general-purpose language models and three input-independent baselines across 925 decision points from 439 public cases. On the public test subset, Jev ranks fourth of five models by macro-F1 (0.461) but has the lowest observed multiclass Brier score (0.610), an 18.7% improvement over uniform prediction (case-bootstrap 95% interval, 4.9–31.1%). Noul has lower binary Brier than Choice (0.215 versus 0.276). A evaluation of 453 HealthBench conversations again gives Jev the lowest observed Brier score (0.299), but measures a distinct emergency-referral judgment. Jev combines lower probability error with missed clarification needs in the four-action task. Asking about information sufficiency separately increases recall but also produces frequent unnecessary clarification assignments.
[LG-152] Lightweight Machine Learning for Classifying Disaster-Related Social Media Posts
链接: https://arxiv.org/abs/2610.24652
作者: Chengjun Jiang,Milena Radenkovic
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:Classifying disaster-related social-media posts into humanitarian information labels is challenging because individual messages are typically short and context-limited, humanitarian categories can overlap semantically, and the CrisisBench label distribution is strongly imbalanced. This study investigates whether lightweight lexical machine-learning methods can provide effective and computationally efficient classification under these conditions. Using the CrisisBench v1.0 English humanitarian task, comprising 87,455 records across predefined Train, Development and Test splits, we conduct a controlled comparison of TF-IDF representations and lightweight classifiers, with Macro-F1 as the primary model-selection metric. Beyond aggregate predictive performance, the study jointly examines per-label behaviour, directional error patterns, relative decision margins, computational requirements, and the effect of domain-specific training through a tropical-cyclone specialist model. The selected balanced LinearSVC with unigram-bigram TF-IDF and C=0.5 achieved an Accuracy of 0.8282, a Macro-F1 of 0.6578 and a Weighted-F1 of 0.8315 on the held-out Test split. Lower decision margins were substantially associated with prediction errors, while the tropical-cyclone specialist achieved modestly higher in-domain aggregate performance with substantially lower feature and fitting requirements. The study contributes a controlled, label-sensitive and resource-aware evaluation of lightweight humanitarian text classification that characterises not only aggregate performance, but also where these models fail and how their behaviour changes under domain-specific training.
[LG-153] EdgeKernelBench: Unified LLM -Driven Kernel Generation for Edge Backends
链接: https://arxiv.org/abs/2610.24648
作者: Yuhua Zheng,Xueyi Chen,Xinze Zou,Xin Li,Haoji Hu,Huan Wang
类目: Machine Learning (cs.LG)
*备注: 25 pages, 8 figures. Project page: this https URL
Abstract:Recent works show that LLMs can generate kernels with strong performance on server GPUs. Edge deployment presents distinct challenges because the hardware is heterogeneous, the software stacks are fragmented, and the resource budgets are tight. LLM kernel generation across diverse edge backends has yet to be studied systematically. We introduce EdgeKernelBench, which combines a unified IR deployment framework with a hierarchical benchmark for edge workloads. The unified IR captures reusable computation, scheduling, and memory decisions and connects to each target backend through a compiler interface. The benchmark spans progressively larger program scopes, from individual operators to compact model graphs, across representative edge application domains. We instantiate the design with TVM as a reference compiler and validate it across heterogeneous edge backends. With a single generation attempt, the best model achieves 64.5% Correct@1 but only 25.5% Fast@1.0. This gap shows that functional correctness does not consistently translate into performance gains. We therefore introduce the Layered Edge Kernel Agent (LEKA), which organizes optimization through a layered IR. On the agent evaluation subset, LEKA achieves 43.5% Fast@1.0, compared with 13.0% for direct generation and 26.1% for Claude Code. LEKA and Claude Code use the same base model and optimization budget. Together, EdgeKernelBench and LEKA support unified evaluation and layered optimization of kernels generated by LLMs across edge backends.
[LG-154] Uncovering Semantic Hierarchies in Text-Attributed Graphs via Variational EM-based LLM -GNN Synergy NEURIPS2026
链接: https://arxiv.org/abs/2610.24624
作者: Yunhui Liu,Xudong Jin,Qizhuo Xie,Chunhui Zhao,Xiao Luo,Tao Zheng,Bin Chong,Tieke He
类目: Machine Learning (cs.LG)
*备注: Accepted by NeurIPS 2026
Abstract:While hierarchical structures are intrinsic to organizing human knowledge, current representation learning on text-attributed graphs predominantly operates on flat semantic spaces, overlooking the rich coarse-to-fine granularity inherent in real-world data. To bridge this gap, we propose SHiFT, which synergizes the reasoning power of LLMs with the structural representation capability of GNNs within an Expectation-Maximization (EM) paradigm. In the E-step, we harness LLMs as expert taxonomists to induce an explicit semantic tree, introducing a lightweight Mapping Diagnosis strategy that reduces inference costs by surgically updating the hierarchy only when anomalies are detected. In the M-step, we align the GNN encoder with this induced hierarchy from complementary perspectives, including clustering partition, topological skeleton, and semantic concepts. Theoretical analysis interprets SHiFT as a generalized Variational EM algorithm. Extensive experiments demonstrate that SHiFT not only achieves state-of-the-art performance but also uncovers high-quality, human-readable taxonomies. Our code can be found at this https URL.
[LG-155] CMScience: An Autonomous Scientist for Traditional Chinese Medicine
链接: https://arxiv.org/abs/2610.24601
作者: Yanlan Kang,Ruiqi Liu,Xinpei Gu,Tzu-Yi Lei,Shuai Xu,Xukun Zhang,William Cheng-Chung Chu
类目: Machine Learning (cs.LG)
*备注: 21 pages, 3 figures, including appendices. Code: this https URL
Abstract:Traditional Chinese medicine (TCM) research connects classical texts, medicinal entities, molecular assays, computational predictions, and clinical studies. An autonomous research agent must preserve the inferential boundaries between these evidence types while coordinating analysis across heterogeneous resources. We introduce TCMScience, a scientific knowledge-discovery agent that separates language-model proposals from execution and release authority. Its architecture combines a trusted kernel, a scientific compiler, and claim-support contracts: plans are compiled into typed operations, execution is bounded by explicit permissions, and candidate findings are checked against source provenance, study design, and claim scope before release. The full capability registry contains 117 public-source connectors with 444 typed operations across 29 domains and 150 native tools across 12 domains; the TCM data hub catalogs 136 databases, 76 dataset specifications, and 84,294 formula-source records. These resources support model-independent workflows through browser and local execution interfaces. In a network-pharmacology analysis of Gegen Qinlian Tang, replacing the whole-proteome background with the assayed-protein background changed the number of significant pathways from 66 to zero; incorporating inactive screening records identified nine pathways, including drug-metabolism signals retained as mechanistic hypotheses. A governance ablation over 121 constructed erroneous outputs spanning 26 classes released 121 errors without governance, 57 with the domain layer, seven with the kernel, and zero with both layers. All 10 compliant controls were retained. These results establish complementary control of known failure modes and show how explicit evidence contracts change the conclusions an agent is permitted to report.
[LG-156] DeliMF: Gated Cross-Attention Fusion of Structured Time Series and Clinical Text for Early Intensive Care Delirium Prediction
链接: https://arxiv.org/abs/2610.24585
作者: Yibo Yang,Zerui Huang,Haomiao Yang,Zixia Wang
类目: Machine Learning (cs.LG)
*备注: Preprint
Abstract:Delirium affects a large fraction of intensive care unit (ICU) patients and foreshadows longer stays, lasting cognitive decline, and higher mortality. Flagging the patients who will become delirious within the first day of admission would give clinicians a window to intervene. Yet state-of-the-art predictors rely only on the structured electronic health record (EHR) of vital signs, laboratory values, and medications, while overlooking the free-text clinical notes written in the same window, which carry information the numbers cannot encode. To this end, we present DeliMF, a multimodal framework that fuses the structured time series with clinical text. It encodes the structured signals with a temporal Transformer and the time-stamped radiology reports with a clinical text encoder, and fuses the two through a gated cross-attention module. The gate learns a per-patient weight on the text; when a patient has no note the gate closes and DeliMF reduces to its structured-only form, giving robustness to missing text by construction. On an incident-only MIMIC-IV cohort, DeliMF improves significantly over the baselines, particularly on patients who carry clinical text, while losing nothing on those without. Ablation studies confirm the contribution of multimodal fusion and gated cross-attention, and a case study further shows that fusing clinical text with the structured stream recovers delirium cases that the structured data alone misses. We hope DeliMF encourages early ICU prediction to make fuller use of the overlooked clinical text and contributes to the early identification and timely intervention of delirium.
[LG-157] Representations Have Trajectories: Characterizing Temporal Representation Dynamics in Neural Networks
链接: https://arxiv.org/abs/2610.24571
作者: Shivaprasad B. Gowda,Sparsh Mittal
类目: Machine Learning (cs.LG)
*备注:
Abstract:Recent work has shown that deep neural networks exhibit an implicit low-rank simplicity bias. In this paper, we investigate the questions that naturally arise from this observation. What are the temporal consequences of low-rank simplicity bias during optimization? How does representation geometry evolve throughout optimization rather than only at convergence? Do different neural architectures exhibit distinct geometric optimization behaviors? To answer these questions, we conduct a large-scale empirical study tracking representational geometry throughout training. We find that neural networks follow three recurring geometric trajectories: persistent high-rank, collapse-and-recovery, and persistent low-rank. We further identify Destructive Representation Collapse (DRC) as a pathological regime in which representations collapse to rank-one and fail to recover, leading to a sustained loss of representational capacity. While collapse-and-recovery is consistently associated with successful optimization in controlled settings, experiments on real-world models show that the relationship between geometric trajectories and optimization outcomes is configuration dependent. Overall, our findings suggest that representation geometry evolves dynamically during training and provide a new perspective on how deep neural networks learn.
[LG-158] No-Regret Revenue Maximization in Sequential Market Making Under Inventory Constraints
链接: https://arxiv.org/abs/2610.24562
作者: Davide Maran,Marcello Restelli,Francesco Emanuele Stradi
类目: Machine Learning (cs.LG)
*备注:
Abstract:Market makers keep trading efficient by setting bid and ask prices for buyers and sellers. In sequential market making, trades happen independently based on price thresholds, requiring the market maker to maintain balanced, low-risk inventory over T rounds. This paper studies how to maximize revenue with unknown buyer and seller valuations using regret minimization. When the market maker only sees trade success or failure, an algorithm achieves regret and inventory bounds proportional to T to the power of 2/3. When full order book data is available for failed trades, regret and inventory limits improve significantly to the square root of T. We also establish a fundamental trade-off via a lower bound, by proving that tighter round-by-round inventory limits strictly force higher minimum regret.
[LG-159] A Multi-Step Inertial ADMM Algorithm for Federated Learning
链接: https://arxiv.org/abs/2610.24536
作者: Hao Li,Qingsong Wang,Xihong Yan,Chengcheng Yan,Deren Han
类目: Machine Learning (cs.LG)
*备注:
Abstract:Federated learning has recently emerged as an important paradigm for large-scale distributed model training, which can be naturally formulated as a class of distributed learning problems. The alternating direction method of multipliers (ADMM) is a widely used approach for solving such problems. However, its practical performance is often hindered by significant communication overhead and system heterogeneity. In this paper, we propose a multi-step inertial ADMM algorithm, called MiFedADMM, for federated learning. The proposed method leverages multiple historical iterates on the server side to construct a higher-order extrapolation after each aggregation to accelerate the convergence behavior. Under mild assumptions, we establish convergence results for the single-step variant of the proposed method and derive an O(1/K) convergence rate. Numerical experiments on both synthetic and real-world datasets demonstrate the effectiveness and improved performance of the proposed method compared with existing approaches.
[LG-160] From Score to Diagnosis: Search-Independent Post-Diagnosis for Symbolic Regression
链接: https://arxiv.org/abs/2610.24535
作者: Qian Li,Yuxiao Hu,Juncheng Liu,Yanhao Jia,Yuntian Chen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Symbolic regression (SR) solvers employ diverse search mechanisms, yet most rely on the same low-information scalar fitness. Such scores provide little guidance on why a skeleton fails or how it should be revised. Recent methods develop new methods to address this bottleneck. Their enhanced feedback mechanisms are tightly coupled to their search processes and are difficult to reuse across diverse SR solvers. Instead, we ask whether a shared feedback mechanism can further improve diverse solvers facing the same bottleneck. To this end, we propose PostDiag, a search-independent feedback mechanism, which can be shared to different SR solvers. PostDiag introduces Patch-wise Feature Profiling (PFP) to capture local behavior through patch-wise features of level, trend, and roughness. Based on PFP, PostDiag quantifies Skeleton Expressiveness Risk by comparing the fitted candidate with the observations, and Skeleton Stability Risk by measuring variability of behavior after bootstrap refitting. The Diagnostic Risk Plane (DRP) combines two risks to determine whether a skeleton should be preserved, repaired, simplified, or rewritten, while the patch-wise discrepancies provide evidence for the intervention. An LLM editor then performs the diagnosis-guided intervention, and the validated revision is returned to the original solver through a lightweight adapter. Experiments across multiple LLM-based and GP-based SR solvers show that PostDiag consistently improves accuracy and accelerates convergence.
[LG-161] opoCompile: Molecular Generation with Exact Global Count Constraints
链接: https://arxiv.org/abs/2610.24511
作者: Xiang Wu,Rong-Hua Li,Quanqing Xu,Kangfei Zhao,Guoren Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Molecular generation designs novel molecules with desired structures and properties. Global count constraints specify molecular size, elemental composition, and ring content without prescribing detailed structure. Property-conditioned generators steer generation toward such targets but do not guarantee exact satisfaction. We introduce TopoCompile, which combines an autoregressive component-tree generator with a dynamic-programming compiler for supported component-additive integer counts. TopoCompile represents molecules as trees of intact ring systems and individual atoms, making each supported count additive over components. At each step, an autoregressive Transformer scores component and attachment actions, while the compiler masks actions with no target-satisfying completion. Within the fixed component vocabulary and typed attachment rules, constrained decoding guarantees exact equality for supported non-negative, component-additive integer counts. Applied at inference time, these checks let the same trained generator support different targets and count combinations without retraining. Across four benchmarks, TopoCompile achieves 100% Exact-valid yield for Common and Rare four-count target groups; the strongest baseline within each dataset reaches 44.9% or less on Rare targets. In unconditional generation, it achieves the highest validity or valid-unique-novel (V.U.N.) among the compared methods across all four benchmark datasets.
[LG-162] What "Form of Uncertainty Does Conformal Prediction Quantify?
链接: https://arxiv.org/abs/2610.24510
作者: Soroush H.Zargarbashi,Mohammad Sadegh Akhondzadeh,Rebekka Burkholz,Aleksandar Bojchevski
类目: Machine Learning (cs.LG)
*备注:
Abstract:Conformal prediction (CP) is a post-hoc calibration method that returns prediction sets with distribution-free coverage guarantees, and the size of the output set is often interpreted as a notion of uncertainty. We show that the difference in set size between full and split CP is directly connected to Bayesian posterior variance, which is the epistemic uncertainty of the model. Essentially, this difference in set size is an aggregation of two opposite effects: (1) better model accuracy and therefore smaller set size achieved by training over extra calibration points, and (2) the “formability” of the model at test point towards every possible label. The latter effect vanishes with stronger models or more calibration data (as epistemic uncertainty is known for that behavior). It turns out that the model’s formability itself (without the quantization effect of prediction sets, and through estimating the posterior variance) is a useful tool to quantify the epistemic uncertainty. In our empirical evaluation it performs on par with the state of the art in many setups.
[LG-163] HiPPO Functional Memory: Coordinate-Queryable Attention for Irregular Time Series ICLR2027
链接: https://arxiv.org/abs/2610.24501
作者: Zhaodong Guo(Imperial College London),Haoyun Feng(Imperial College London)
类目: Machine Learning (cs.LG)
*备注: 21 pages, 3 figures, 10 tables. Submitted to ICLR 2027
Abstract:Irregular time series require models to answer temporal queries at arbitrary coordinates, yet existing approaches typically retain observation-dependent representations that must be repeatedly resolved for each query. We introduce HiPPO Functional Memory (HiPPO-FM), a functional-memory attention operator that compiles an irregular history into a reusable coordinate-addressable state. Instead of storing observation-indexed tokens or latent trajectories, HiPPO-FM writes observations into a fixed-rank functional space and performs direct coordinate evaluation through matched functional geometry. Kernel-smoothed HiPPO coordinates and Gram-based readout define a shared write/read mechanism, while observed-support self-attention enables content-dependent representations before compression. We characterize the induced finite-dimensional query space and derive an error decomposition separating functional geometry approximation from learned-memory mismatch. Across continuous-coordinate reconstruction, irregular representation learning, real-world extrapolation, and event-sequence prediction, HiPPO-FM consistently improves strong irregular time-series baselines, achieving strong reconstruction accuracy on Spiral, competitive performance across 20 UEA datasets, improved real-world extrapolation, and superior results on several irregular event benchmarks. After encoding, the compact functional state can be reused across different query sets without reprocessing the observation history.
[LG-164] deKV: Delayed KV Eviction via Layer-Wise Chunked Prefill for Memory-Efficient Long-Context LLM Inference
链接: https://arxiv.org/abs/2610.24499
作者: Yu Wang,Zhu Wang,Ning Hu,Yifan Guo,Xiu Ma,Bin Guo,Zhiwen Yu
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注: Accepted by The Journal of Systems Architecture
Abstract:Long-context large language model (LLM) inference is important for modern intelligent applications but remains constrained by peak memory during prompt prefill. Existing KV cache compression reduces decoding memory but typically requires full-prompt prefill to preserve global context, retaining a high construction peak. Online chunk-level eviction lowers memory earlier but selects before each layer has observed the complete prompt, potentially degrading cache construction and decoding. We present TideKV, an on-GPU cache-construction architecture using layer-major delayed eviction. TideKV processes all prompt chunks within each Transformer layer, retains full-prompt KV only for the active layer, and evicts at layer end. This schedule preserves full-prompt-aware selection while reducing activation and temporary memory. Experiments across multiple computing platforms show that TideKV preserves the inference quality and final cache size of the evaluated compression methods. At 64K on A100, it reduces the construction-time transient footprint by 83.89%. On RTX 4090, it increases time to first token by only 1.06% at 64K and enables 128K prefill. TideKV also completes 128K on Jetson, whereas matched post-prefill compression runs out of memory on both resource-constrained platforms. These results show that delayed layer-end eviction substantially expands feasible context lengths without sacrificing cache utility.
[LG-165] CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models
链接: https://arxiv.org/abs/2610.24495
作者: Arav Dhoot,Yixiong Hao,Zephaniah Roe
类目: Machine Learning (cs.LG)
*备注:
Abstract:While a LLM’s chain of thought (CoT) is a promising approach to interpreting model outputs, prior work shows CoT’s content does not always accurately reflect the reasoning process. Emmons et al. (2025) finds that models can follow a question hint without revealing so in its chain of thought, but when the hint is sufficiently complex the computation is forced into the CoT. This claim, however, has only been tested on Gemini models, and the datasets and code is closed source. We replicate their proxy task across eleven models from six families and find the necessity effect generalizes. Every model follows simple cues above a no-hint baseline, while concealed complex-cue following falls back near baseline for most models. We also extend the study from Emmons et al. (2025) to measure CoT-necessity and unfaithfulness across five cue families (four of them new) and three difficulty levels. The relationship between when CoT is necessary and unfaithfulness is task-specific and model-specific. For some hint types, models often verbalize their hint-following behavior even when the hint can be decoded in a single forward pass. These results support the case for CoT monitoring when unsafe behavior requires non-trivial computation, and that properties of CoT unfaithfulness are sensitive to model and setting.
[LG-166] VBO-MI: A Differentiable Black-Box Optimization Framework via Variational Mutual Information Estimation
链接: https://arxiv.org/abs/2610.24491
作者: Farhad Mirkarimi
类目: Machine Learning (cs.LG)
*备注:
Abstract:Many real-world tasks require optimizing expensive black-box functions accessible only through noisy evaluations, a setting commonly addressed with Bayesian optimization. While Gaussian processes(GPs) and Bayesian neural networks have driven significant progress, GP-based methods scale cubically with observations, and BNN-based approaches remain burdened by expensive posterior sampling such as Hamiltonian Monte Carlo, and inner-loop acquisition optimization in high-dimensional setting. In this work, we propose VBO-MI, a fully gradient-based BO framework that requires no assumptions on a Gaussian prior or a variational posterior family over f . While our framework uses gradient-based updates to train the internal action and information-estimation networks, it treats the objective function f as a strict black box, requiring only zeroth-order evaluations. To enable end-to-end gradient flow, we employ an actor-critic architecture consisting of an action-net to navigate the input space and a variational critic to estimate a surrogate information signal for exploration. Additionally, we design a lightweight surrogate that approximates the objective from observed data using a replay buffer, which reduces the total required real function queries to one batch per iteration. We provide theoretical analysis for our framework, establishing convergence of its training objective and surrogate estimator, and showing, that near-optimal solutions exist. We evaluate our method on a diverse suite of benchmarks, including high-dimensional synthetic function and complex real-world tasks such as Rover Trajectory planning optimization, the Lunar Lander, and Pest Control problems. Our experiments demonstrate that VBO-MI provides the same or superior optimization performance over the various baselines.
[LG-167] Measuring Empowerment from Offline Data for Intrinsically Motivated Exploration
链接: https://arxiv.org/abs/2610.24461
作者: Ishir Garg,Vivek Myers,Qiyang Li,Catherine Ji,Benjamin Eysenbach,Sergey Levine
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large, unlabeled datasets of prior experience have become increasingly available for agentic tasks in domains such as robotics and coding. A common recipe for these tasks is to train a policy using the prior data, and then further improve it with online rollouts. But while much work has studied how to use both the offline and online data for policy extraction, far less work has been done to understand how the prior data can be used to efficiently \textitexplore during the online learning phase. Previous work has attempted to formalize this idea by defining the \textitempowerment, a measure of how many future states an agent can realize; states with higher empowerment require more exploration as there are more futures to explore. However, existing methods for estimating empowerment require online policy rollouts in the environment. We propose the first method for estimating empowerment from unlabeled prior data, without any environment interaction. By jointly learning an empowerment-maximizing policy together with its discounted state-occupancy measure, our estimator can be trained on unlabeled trajectories generated from any unknown policy. We demonstrate that our empowerment estimator qualitatively gives intuitive empowerment measures and show that the pretrained estimator can be used to improve exploration in the online goal-conditioned reinforcement learning setting. Code is available at \urlthis https URL.
[LG-168] Generative Retrieval for Reverse Target Fishing with a Proteome-Scale Benchmark
链接: https://arxiv.org/abs/2610.24458
作者: Zhichao Deng,Haichuan Tan,Xin Hong,Huaizhi Wang,Tao Huang,Bowen Gao,Yanyan Lan
类目: Machine Learning (cs.LG)
*备注: 27 pages, 6 figures, 4 tables
Abstract:Reverse target fishing identifies protein targets for a query molecule and is central to drug discovery and repurposing. Discriminative approaches adapted from virtual screening face sparse, long-tailed associations and missing-not-at-random measurements: treating unrecorded pairs as negatives can suppress genuine targets. We reformulate the task as generative retrieval and introduce GenFish. Residual vector quantization converts frozen protein language model features into shared semantic identifiers. Given a molecular fingerprint, GenFish autoregressively models codes conditioned on the query and preceding code, ranking proteins by identifier likelihood without explicitly labeling unrecorded pairs as negatives. We introduce ProTargetBench, the first proteome-scale benchmark for reverse target fishing to jointly support measured-panel discrimination and full-catalogue recovery of experimentally recorded positives. Covering 20,431 UniProt-reviewed human proteins, it integrates curated bioactivity evidence and distinguishes experimental inactivity from unmeasured associations. On 198 held-out molecules, GenFish achieves higher mean performance across all three full-catalogue retrieval metrics than four discriminative neural baselines and three non-neural baselines, including ligand-based methods. A controlled positive-hiding experiment shows that targeted removal of false-negative competition improves contrastive retrieval beyond matched random masking; GenFish remains stronger under the same incomplete supervision. Pfam alignment indicates that the semantic identifiers capture biologically meaningful relationships among proteins. These results establish generative retrieval as a principled paradigm for reverse target fishing under incomplete bioactivity records.
[LG-169] Near-Optimal Regret for Semi-Bandit Learning of Monotone Stochastic Problems
链接: https://arxiv.org/abs/2610.24446
作者: Zhengjia Zhuo,Viswanath Nagarajan
类目: Machine Learning (cs.LG)
*备注:
Abstract:We consider the task of learning an (approximately) optimal policy for stochastic optimization problems where the true distributions are unknown. In each period, the algorithm plays a policy for the stochastic problem and observes semi-bandit feedback, i.e., it only receives new samples from the random variables that were selected by the policy. We obtain an algorithm that works for a large class of monotone'' stochastic problems, including prophet inequality, Pandora's box, series testing, stochastic knapsack and stochastic matching. We obtain \tildeO(\sqrtnkT) regret for weakly’’ monotone problems and \tildeO(\sqrtnT) regret for strongly'' monotone problems, where T is the number of periods, n is the problem dimension (number of random variables) and k is the maximum support size of any random variable. This improves over previous regret bounds by a \sqrtn factor. In fact, our dependence on n and T is the best-possible due the \Omega(\sqrtnT) lower bound for multi-armed-bandits. Our main technique is the use of Hellinger distance in bounding the error’’ between the true and learned distributions, and the use of a constant-factor chain rule for the Hellinger distance of dependent random variables.
[LG-170] ransformers as Adaptive Kernel Learners across Tasks for In-Context Learning
链接: https://arxiv.org/abs/2610.24441
作者: Zhaiming Shen,Alexander Cloninger,Rongjie Lai,Wenjing Liao
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA); Statistics Theory (math.ST)
*备注:
Abstract:Transformers have emerged as a dominant architecture in modern machine learning, with in-context learning (ICL) enabling models to adapt to new tasks by conditioning on prompts, without any parameter updates. To understand the attention mechanism in transformer blocks, a growing line of theoretical work interprets ICL through the lens of classical statistical estimators. Recent works have established a connection between the attention mechanism and kernel methods, while the kernel function is usually assumed to be fixed. In practice, the query and key matrices in the attention mechanism are trained, which allows the attention mechanism to serve as an adaptive kernel method. In this work, we demonstrate that transformers can act as adaptive anisotropic kernel learners. Specifically, we study ICL for low-rank linear and multi-index nonlinear regressions, and conduct numerical simulations to show that the learned attention scores closely resemble the oracle kernels for these function classes. We then develop a general framework to analyze the generalization error of ICL by connecting transformers to the oracle kernel estimator for each function class. Our theory gives an explicit and constructive implementation of the oracle kernel estimators for these function classes using a wide and shallow transformer architecture, illustrating how the prediction is carried out by the interaction between the query token and the context tokens. Our work complements existing studies by providing numerical and theoretical evidence of how the attention mechanism adapts to anisotropic kernels in both linear and nonlinear settings.
[LG-171] ransport What Shares the Conduction Pathway: Normal-Anchored SPD Geometric Transport for Source-Free Patient Adaptation of Holter and Patch ECG
链接: https://arxiv.org/abs/2610.24430
作者: JinKook Kim
类目: Machine Learning (cs.LG)
*备注: 23 pages, 5 figures, 6 tables (including Supplementary Information)
Abstract:Electrocardiogram (ECG) beat classifiers lose accuracy on new patients and devices, and source-free adaptation methods that assume class-balanced target batches fail on real recordings, whose label-free prefix is 87.9–98.6% normal beats. We propose Normal-Anchored Geometric Transport (NAGT, GA_NS), which estimates a closed-form symmetric positive-definite congruence map from label-free normal anchors in the first 5 min of each recording and applies it only to the normal and supraventricular prototypes, keeping the ventricular and fusion prototypes fixed. The score offsets and decision rules were set on MIT-BIH only; the rate-invariant RR head (relA) was adopted after the development databases (INCART, SVDB) had been opened; the frozen model was then evaluated once on Icentia11k chest-patch recordings under a pre-registered protocol. In 493 patients (973 recordings, 3,615,388 evaluated beats), NAGT reached an arrhythmia macro- F_1 J = \frac12(\textS-F_1 + \textV-F_1) = 48.7% [95% CI 42.1, 54.7] versus 34.5% without adaptation ( \Delta J = +14.1 [+8.6, +19.3] pp; 59.8% versus 43.9% in non-AF) and outperformed all five Holm-adjusted comparators, including AdaBN, Tent and T3A. In the 75-patient test subgroup of the historical split (73 analyzed), the gain was not significant (+6.1 [-8.5, +23.4] pp). Adaptation takes a median of 5.8 ms per record without back-propagation.
[LG-172] Cross-Dimension and Cross-Modality Relational Tensor Representation Learning for Multimodal Health Data Modeling
链接: https://arxiv.org/abs/2610.24419
作者: Jiawei Guo,Meng Zhao
类目: Machine Learning (cs.LG)
*备注: 30 pages
Abstract:Multimodal health data integrate information from heterogeneous clinical sources, creating substantial opportunities for data-driven decision support. Realizing these opportunities, however, requires effectively modeling complex dependencies that arise both within individual modalities and across modalities. This paper proposes Relational Tensor Representation (RTR), a unified framework that jointly learns dimension-specific tensor representations and cross-dimensional and cross-modality relationships. RTR introduces learnable relational bridge matrices that maintain relational information while preserving the structural identities of tensor dimensions and modalities. A structured fusion mechanism further integrates factor-level and relationship-level information into compact representations for downstream prediction. Experiments on simulated data and two multimodal neurophysiological datasets, covering emotion recognition and brain activity classification, demonstrate that RTR consistently outperforms representative tensor learning, multimodal fusion, and Transformer-based baselines. These results highlight the value of explicitly modeling cross-dimensional and cross-modality relationships for learning informative representations from heterogeneous multimodal tensor data. The source code is planned for public release to support reproducibility and future research.
[LG-173] Learning volatility-adaptive forgetting via GARCH-modulated LSTM for stock price forecasting
链接: https://arxiv.org/abs/2610.24386
作者: Xin-Rui Lu,Da-Qing Zhang
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注:
Abstract:Stock price forecasting is hindered by the heteroscedastic nature of financial time series, in which turbulent and tranquil regimes alternate. Although the Long Short-Term Memory (LSTM) network is widely used for this task, its gates are oblivious to such volatility variation, so memory retention remains fixed regardless of market conditions. A common remedy is to supply volatility estimates as an additional input feature, yet this leaves the internal memory dynamics untouched. Instead, volatility can serve as a direct regulatory signal over the gate activation itself. This paper presents the Gsigmoid-LSTM model, which embeds Generalized Autoregressive Conditional Heteroscedasticity (GARCH(1,1))-derived volatility estimates into the LSTM forgetting gate through the Gsigmoid activation function, a volatility-adaptive variant of the sigmoid that combines a volatility adjustment factor with slope regulation via a learnable parameter. The model thereby automatically prioritizes recent information during turbulent periods while preserving long-term dependencies in stable markets. In the engineering application of daily stock price forecasting, extensive experiments, including activation function comparisons, baseline evaluations, and ablation studies, demonstrate that this structural gating modification effectively improves forecasting accuracy and robustness. The proposed model significantly outperforms a feature-level fusion baseline carrying the same volatility signal on all five datasets, as confirmed by Diebold-Mariano tests with all p0.01, attains the best average rank of 1.00 in the Friedman test across nine models, and achieves the lowest MSE and MAE on all five datasets.
[LG-174] A decomposition-integration framework coupling singular spectrum analysis with a volatility-adaptive LSTM for stock price forecasting
链接: https://arxiv.org/abs/2610.24384
作者: Xin-Rui Lu,Da-Qing Zhang
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注:
Abstract:Financial time series exhibit multi-scale, high-noise, and non-stationary volatility dynamics. Decomposition–integration frameworks built on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) have advanced this field, yet existing CEEMDAN hybrids remain unstable: high-frequency mode mixing contaminates downstream predictors, and the heterogeneous volatility of decomposed components is not matched by any volatility-adaptive mechanism. To address these gaps, this paper proposes CEEMDAN-SE-SSA-VMGL. CEEMDAN adaptively decomposes the original series, and sample-entropy-driven K-means clustering groups the intrinsic mode functions into high-, mid-, and low-frequency and residual components. Singular spectrum analysis (SSA) is then applied to the high-frequency group, whose singular-value decomposition of the trajectory matrix and principal-component reconstruction decouple intertwined pseudo-oscillatory modes and alleviate mode mixing. We further design a plug-and-play robust LSTM variant, the volatility-modulated GARCH-LSTM (VMGL), which embeds a differentiable GARCH(1,1) layer inside the LSTM cell as a risk gate; the estimated conditional volatility multiplicatively scales the forget-gate activation, letting memory decay adapt to the instantaneous volatility while remaining fully end-to-end trainable. Component predictions are linearly combined. On five US and Chinese financial series (CSI 300, CSI 500, AAPL, GOOGL, SP 500), the proposed model is compared with eleven benchmark architectures, lowering the forecasting error by 9% to 24% relative to the best benchmark and by 60% to 70% relative to a standard LSTM. Diebold–Mariano and model-confidence-set (MCS) tests confirm the statistical significance and stability of the gains, spectral analyses verify the SSA denoising efficacy, and backtesting demonstrates the framework’s economic utility.
[LG-175] Revisiting Smoothed Online Learning: Competitive Ratio Dynamic Regret with Switching Cost and Best of Both Worlds
链接: https://arxiv.org/abs/2610.24370
作者: Lijun Zhang,Linzhe He,Wei Jiang,Shiyin Lu
类目: Machine Learning (cs.LG)
*备注:
Abstract:We revisit smoothed online learning and develop algorithms with improved competitive ratios, optimal dynamic regret with switching cost, and best-of-both-worlds guarantees. Surprisingly, minimizing the hitting cost alone is \max(1,\frac2\alpha) -competitive for \alpha -polyhedral functions under the \ell_2 -norm switching cost and 1+\frac4\lambda -competitive for \lambda -quadratic growth functions under the squared \ell_2 -norm switching cost, significantly improving existing guarantees. The polyhedral bound is tight for this naive algorithm, and we establish a matching lower bound for \alpha\geq1 . For convex \lambda -quadratic growth hitting costs, minimizing a suitably weighted sum of hitting and squared \ell_2 -norm switching costs yields a competitive ratio of 1+\frac2\sqrt\lambda . For dynamic regret with the \ell_2 -norm switching cost, we develop Smoothed Ader (SAder), which achieves the optimal O(\sqrtT(1+P_T)) rate in the standard online convex optimization setting, where P_T is the comparator’s path-length. Its lookahead version removes the bounded-gradient assumption, and a matching lower bound establishes optimality in this setting. Finally, for convex hitting costs on a bounded convex domain, we develop UDNP to aggregate any two experts in the lookahead setting under both switching costs. Its additional cumulative cost is at most O(1) relative to the first expert and O(\sqrtT\log T) relative to the second, yielding best-of-both-worlds guarantees without bounded gradients.
[LG-176] Can Agents Keep Secrets They Were Never Told? Benchmarking Latent Privacy Risks Across Sessions with MemJoin
链接: https://arxiv.org/abs/2610.24359
作者: Qishun Yang,Shu Yang,Difei Xu,Di Wang,Meng Ding
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:To support sustained interactions and personalized assistance, agents use long-term memory to retain and combine user information across sessions. Reusing such memories can lead an agent to state a sensitive relation while completing an ordinary task that does not require it. The relation to protect and the combination of clues that reveal it may remain hidden at memory write time, when later records and tasks are still unknown. In this paper, we introduce MemJoin, a privacy benchmark of 288 synthetic analysis cases in which the complete history entails a sensitive relation, but every individual session is insufficient. We evaluate six memory systems with the same answer-generating model, alongside full-history and no-history controls. Each system writes and updates memories in session order and later retrieves context for an ordinary task, while target annotations remain with the evaluator. We separately check which target attribute values appear in stored records and retrieved context, and whether the final answer states the complete, person-linked relation. In a separate paired test on a prespecified subset, removing shared identifiers and other connecting clues can change complete-relation output while all checked target attribute values remain visible. These findings show why checking attribute visibility alone is insufficient to evaluate what records from different sessions reveal together.
[LG-177] Blackbox Performance Analysis of LLM -based Network Control
链接: https://arxiv.org/abs/2610.24348
作者: Mahdi Alizadeh,Pooria Namyar,Arvin Ghavidel,Ramesh Govindan
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:LLMs are becoming an indispensable part of network control systems due to their impressive reasoning. Before such LLM-enabled systems can be deployed, it is imperative to develop tools that can audit their performance on previously unseen inputs. However, auditing these systems is challenging due to the inherent opacity of LLMs. LENS is the first step towards black-box performance analysis of LLM-enabled network control systems. It uses novel methods to extend deep reinforcement learning to guide the search for performance gaps and inputs that make LLMs struggle. LENS finds large gaps for LLM-based traffic engineering and GPU allocation systems, and even matches existing white-box analyzers designed for hand-crafted TE heuristics.
[LG-178] Exact Risk Certificates for Pairwise-Judged Reranking
链接: https://arxiv.org/abs/2610.24347
作者: Xinyu Wang,Jingrui Tian,Sicheng Lyu,Yu Gu,Ziyang Song,Xiao-Wen Chang,Lei Ding
类目: Machine Learning (cs.LG)
*备注:
Abstract:Pairwise relevance judgments constrain both the loss numerator and the ideal-ranking denominator of nDCG. We characterize the exact worst-case loss of replacing a ranking with a fixed proposal. Binary gains attain this loss and provide an inspectable witness. Threshold verification reduces to at most K+1 maximum-closure problems. Our exposure-flow ablation isolates denominator coupling: on shared judge trajectories, exact verification saves 2.34 comparisons per TREC-DL development query and 1.98 per confirmation query at a budget of 32. Savings transfer to BEIR and a second proposer. Predictive gates achieve higher deployment gains, while certificates expose the worst gains still permitted by accepted comparisons.
[LG-179] Measuring the Task Value of Precision Upgrades in Closed-Loop Agents
链接: https://arxiv.org/abs/2610.24346
作者: Sicheng Lyu,Xinyu Wang,Yu Gu,Lei Ding,Ziyang Song,Xiao-Wen Chang
类目: Machine Learning (cs.LG)
*备注:
Abstract:A low-precision agent can call its full-precision (BF16) model at selected decisions as a fallback. A learned precision router chooses where to call it from the disagreement between the two models and is judged by end-to-end success. An inserted BF16 action, however, can rescue a failing task or damage a succeeding one. We argue that a fallback’s value is the number of tasks it saves minus the number it spoils, measured for the intervention that will actually run. From restored states, we run paired branches that cross which model supplies the current action with which model continues, separating replacement of one action, takeover of later execution, and both. On Qwen3 agents with simulated 4-bit weights in tau2-bench, replacing the first action where the two models differed in tool use changed many task outcomes, but rescues and damages nearly cancelled, and on held-out telecom tasks the action-only estimate was negative. Replacing both the action and the continuation had a positive net gain in two of five exploratory lower-bit stress settings. Two references changed outcomes at fixed bit width: replacing an action with that of a second 4-bit checkpoint changed about as many outcomes as replacing it with the BF16 action, and reordering tool-argument keys changed outcomes while the executed tool call stayed identical. At common states, the additional value of the BF16 action over that checkpoint remained unresolved. A frozen uncertainty trigger found more outcome changes than random selection, while its net benefit stayed within the range of random selection. We recommend evaluating each fallback by its signed task benefit under the intended continuation, against same-precision references, before attributing gains to more bits.
[LG-180] he Certification Exponent: Instance-Optimal Sequential Certification of Policy Improvement from History Dependent Logs
链接: https://arxiv.org/abs/2610.24345
作者: Rui Wang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 62 pages, 8 figures. Submitted to the Journal of Machine Learning Research
Abstract:Before a new policy replaces a deployed one, we often want a certificate, computed from logged episodes, that it is better. We ask how many episodes such a certificate needs when the logging policy is known but history-dependent, and the error probability must stay below \delta in every environment consistent with the logs. The answer is an exponent K^\star : the smallest KL divergence from the logged law to a law with the same logged action probabilities under which the target does not improve. No sound certifier stops in fewer than \log(1/\delta)/K^\star expected episodes, at any \delta , and whenever 0K^\star\infty a plug-in Kelly certifier attains this rate to first order as \delta\to0 . Coverage, computed by one backward induction, decides whether certification is impossible ( K^\star=0 ), possible at a finite rate, or possible in a constant number of episodes ( K^\star=\infty ). By convex duality, the optimal certificate bets on the importance-weighted advantage and also places fair bets on the logged actions, so doubly robust control variates act as hedges rather than estimators. As a consequence, no choice of off-policy estimator changes K^\star , a coverage ceiling explains the curse of memory, and structural assumptions can be priced in exponent units. A cushion certifier extends the approach to trees too large to enumerate. In experiments the plug-in’s median stopping time is 1.13-1.43x the lower bound on memory tasks, 1.13-1.51x on Tiger ( H\le5 ) and 1.17-1.64x on random trees. The best scalar betting baselines need up to 2.65x (memory) and 4.1x (Tiger) the plug-in’s episodes; only on a few shallow random trees is a doubly robust bet faster, by at most 18%.
[LG-181] Uniform convergence at optimal rate and scale
链接: https://arxiv.org/abs/2610.24336
作者: Noga Alon,Steve Hanneke,Shay Moran,Tom Waknine
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Uniform convergence asks how many random samples suffice for empirical averages to approximate expectations simultaneously over an entire class of functions. It is a central concept in probability and a foundation of statistical learning. For binary-valued classes, the Vapnik–Chervonenkis theorem characterizes uniform convergence by finiteness of the VC dimension. For dimension d , the sharp worst-case sample complexity for additive error \varepsilon is of order d/\varepsilon^2 . For real-valued functions, the analogous combinatorial parameter is the fat-shattering dimension, which depends on an approximation scale \gamma . The challenge is to achieve error \gamma+\varepsilon with order d/\varepsilon^2 samples, matching the sample bound for binary classes, where d is now the fat-shattering dimension at scale \gamma . Previous results achieved either the optimal scale or the optimal sample complexity, but not both. Our main result closes this gap, establishing the sharp sample complexity at the optimal scale. Along the way, we use several tools that may be of independent interest. The first concerns \varepsilon -nets with margin. A classical \varepsilon -net intersects every set of mass at least \varepsilon in a prescribed family; we allow the net to intersect a slightly enlarged version of each such set. Remarkably, this relaxation permits bounds independent of the ambient dimension in natural geometric settings. Second, we derive improved covering bounds in both discrete and real-valued settings: we extend Haussler’s covering lemma to partial concept classes and use this extension to obtain \ell_2 covers at the optimal scale for real-valued function classes. Third, we obtain a chaining bound at the optimal scale by aligning the covers so that successive approximations become increasingly close to one another.
[LG-182] Power-Law Correlations Without Apparent Temperature-Tuned Criticality in Fixed-Tree Hierarchical Languages
链接: https://arxiv.org/abs/2610.24334
作者: Marcel Kühn,Bernd Rosenow
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn)
*备注: 19 pages, 6 figures
Abstract:Sampling temperature can substantially reshape autoregressive generation, but it is unclear when temperature changes in long-range correlations reflect a genuine phase transition rather than a smooth change in statistics. We study this question in a controlled analogue of temperature sampling: a finite-depth probabilistic context-free grammar on a fixed hierarchical tree, where exact next-token conditionals are computed by belief propagation and then rescaled by a softmax temperature before autoregressive sampling. We measure two-point mutual information between tokens as a function of separation. Across the tested temperatures and grammar ensembles, mutual information remains well described by a power-law envelope. Temperature changes the correlation magnitude and fitted exponent, but within the accessible system sizes we find no long-distance plateau characteristic of an ordered phase and no clear crossover to short-range exponential decay. An approximate mapping to an Ising model on a tree explains how exponential decay in tree distance becomes a power law in token distance, allowing power laws to persist without tuning to a special decoding temperature. This controlled baseline clarifies which temperature effects can arise from hierarchical structure alone, before adding training, approximation error, or neural architecture.
[LG-183] PAC Learning with Query Feedback
链接: https://arxiv.org/abs/2610.24332
作者: Chase Hutton,Adam Melrod,Han Shao
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study realizable multiclass PAC learning with query feedback, where each training example provides a single binary response indicating whether its true label belongs to a chosen subset from a fixed query family. Our main contribution is the query dimension, a combinatorial parameter that captures both the structure of the hypothesis class and the ability of the available queries to distinguish labels. This parameter generalizes the bandit DS dimension, which it recovers under singleton queries. We establish a sample-complexity lower bound of \Omega(D/\varepsilon) and an upper bound of \widetildeO(D^2/\varepsilon) , suppressing factors polylogarithmic in the number of labels and the failure probability. When the available queries can split every candidate set into two parts of comparable size, we improve the upper bound to \widetildeO(D/\varepsilon) , yielding a characterization up to logarithmic factors.
[LG-184] Latent State Utilization in Probabilistic Circuits
链接: https://arxiv.org/abs/2610.24315
作者: Zilei Shao,Anji Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Probabilistic circuits (PCs) are generative models that support efficient probabilistic inference. Adding latent states to a PC increases its representational capacity, but yields diminishing gains in likelihood. We use mutual information to examine how the dependence represented by a trained PC changes with latent size. Experiments with Hidden Chow-Liu Trees (HCLTs) on ImageNet32 show that learned dependence grows slowly and remains well below both the latent-state upper bound and estimates of dependence in the data. To understand how learning uses these states, we examine how strongly each sample favors particular latent states and how broadly states are used across the dataset. Tracking these measurements during HCLT training distinguishes rapid early posterior concentration from the more gradual stabilization of sample-dependent state preferences. Early measurements also guide initialization selection in HCLTs and hidden Markov models (HMMs), as well as tree selection in HCLTs. The resulting choices improve the ultimate performance across the tested latent sizes, providing a practical benefit from our study of latent-state utilization.
[LG-185] Muon Operates Beyond the Edge of Stability
链接: https://arxiv.org/abs/2610.24305
作者: Niko Covic,Baraq Lipshitz,Alexander Padula,Stefan Visnjic
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study the Muon optimizer using sharpness and provide the first empirical evidence that the Flat Minima Hypothesis also applies to Muon. Further, we derive Muon’s implicit preconditioner to analyze its potential Edge of Stability behavior. Experiments on CIFAR-10 show sharpness strongly correlates with generalization gap within each optimizer (Pearson r = 0.45 - 0.69 , p \leq 10^-4 ), confirming the hypothesis holds for Muon similarly to Adam and SGD. However, Muon operates above the classical sharpness-based Adaptive Edge of Stability threshold that bounds diagonally-preconditioned optimizers like Adam and RMSprop, yet remains stable. Our findings suggest Muon operates in a fundamentally different regime than diagonally-adaptive methods. Defining the exact stability boundary for Muon remains an open question. All code is available at this https URL.
[LG-186] orch-DPhys: A Torch-Native MOOSE-Compatible Framework for Differentiable Physics Simulation and Hybrid Machine Learning Integration
链接: https://arxiv.org/abs/2610.24301
作者: Akzhol Almukhametov,Yang Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:We present Torch-DPhys, a Torch-native framework implemented in PyTorch for differentiable physics simulation and hybrid physics-machine learning. It implements selected thermal-hydraulic modeling concepts and discretizations from MOOSE’s Thermal Hydraulics Module and the System Analysis Module (SAM), with a preliminary extension to continuum mechanics. Native tensor residuals support CPU and GPU execution, compilation of selected numerical kernels, and sparse direct linear solves. Sensitivities are obtained by differentiating executed numerical operations or through discrete adjoints of converged residual equations. Verification combines analytical solutions, reference-code comparisons, and derivative checks. Demonstrations cover plane-strain elasticity, one- and two-loop thermal-hydraulic transients, and an axisymmetric molten-salt fast reactor. For the reactor, comparison with SAM-Lite gives a maximum normalized field discrepancy of 1.13 x 10^-5 over 200 adaptive time steps, demonstrating agreement between the tested discrete implementations. Selected transient sensitivities are assessed using tangent-linear calculations and finite differences on fixed time schedules. Two applications illustrate the framework’s connection between simulation and optimization. A residual-adjoint parameter-identification calculation recovers nominal pump head and heating power with relative errors below 0.06% from noiseless synthetic MOOSE observations. A separate three-loop plant demonstration uses solver-in-the-loop training of a predictive controller by differentiating trajectory-level objectives through the thermal-hydraulic simulation. These examples illustrate a foundation for gradient-based inference, control, and coupled physics-machine learning models in nuclear engineering.
[LG-187] Predicting CVSS and CWE from CVE Descriptions with Fine-Tuned Small Language Models
链接: https://arxiv.org/abs/2610.24297
作者: Tara Dixit,Jessie Dong,Summer Li
类目: Machine Learning (cs.LG); Cryptography and Security (cs.CR)
*备注: 23 pages, 5 figures, 21 tables
Abstract:Until April 2026, the National Vulnerability Database labeled nearly every published Common Vulnerabilities and Exposures (CVE) entry with a Common Vulnerability Scoring System (CVSS) v3.1 vector, a severity tier, and a Common Weakness Enumeration (CWE) weakness type. Now, NIST enriches only the high priority CVEs, leaving many newly published vulnerabilities unlabelled with the classifiers that security teams use for patch triage. In CVE enrichment, we predict the eight base components of CVSS v3.1 vector, base score, severity tier, and CWE from the vulnerability’s textual description alone. To prevent leakage between near-duplicate descriptions, all systems are trained only on CVEs published in 2022 and 2023. Results are evaluated on all 36,325 usable 2024 CVEs in the pinned NVD snapshot. All scores are either directly predicted, or calculated from predicted vectors using the CVSS formula. A fine-tuned 3.8B-parameter model Phi-3.5-mini, trained with LoRA in 2.5 hours on a single Apple M5 Pro chip, achieves 69.9% accuracy on severity and 70.9% on CWE. It matches the official vector exactly for 48.8% of test CVEs, outperforming TF-IDF logistic regression by 2.8 to 4.7 points on all accuracy metrics based on zero-exclusion paired bootstrap intervals. When we evaluate the base model with a zero-shot prompt, it achieves 20.8% accuracy on severity in a sample of 2,000 CVEs. In this same sample, its stated scores match its own vector precisely for only 3.4% of CVEs. However, deriving scores from predicted vectors boosts the critical recall of TF-IDF logistic regression from 50.7% to 73.3% with no additional cost. And since the errors of the fine-tuned model and TF-IDF affect different CVEs, at least one system matches the correct severity tier for 75.8% of test CVEs.
[LG-188] Constructive Universal Approximation by Infinite-Dimensional Neural Networks via Gaussian-Weighted Ridgelet Transform
链接: https://arxiv.org/abs/2610.24293
作者: Sho Sonoda,Luca Galimberti,Yuanchao Xu,Isao Ishikawa,Masahiro Ikeda
类目: Machine Learning (cs.LG); Functional Analysis (math.FA); Machine Learning (stat.ML)
*备注:
Abstract:Neural networks whose inputs are functions, as in operator learning, are usually shown to be universal by existence arguments, either by reduction to finite-dimensional networks or by Stone–Weierstrass-type density; these prove that approximants exist but say little about how the parameters depend on the target. We take a constructive route. We extend the ridgelet transform, an integral transform related to the Fourier, Radon, and wavelet transforms, to continuous maps on a separable infinite-dimensional Hilbert space, replacing the Lebesgue measure that is unavailable in infinite dimensions by Gaussian weights with a homogeneous mixture on the direction space. The resulting transform has a Plancherel identity and reconstruction formulas incorporating the frame operator, the correction induced by the weights, and it assigns to each target an explicit coefficient measure, one for every admissible analysis filter. From these representations we obtain uniform approximation on compact sets for all continuous non-polynomial activations of polynomial growth, including ReLU, and, by Monte Carlo sampling of the coefficient, finite-width networks with a dimension-free O(N^-1/2) error in the width N for globally Lipschitz activations and coefficients with finite variation and second moments. Explicit examples include Gaussian targets and neural-operator layers. A companion Lean~4 formalization provides machine-checked proofs, with its correspondence to the manuscript documented at this https URL .
[LG-189] Inducing Fourier Geometry: A Smoothness Prior Imposes Approximate Translation Equivariance in Multimodal Stellar Embeddings
链接: https://arxiv.org/abs/2610.24292
作者: Ilay Kamai,Alex M. Bronstein,Hagai B. Perets
类目: Machine Learning (cs.LG)
*备注:
Abstract:Learned representations often contain geometric structure induced by symmetries in the underlying data. We ask whether a known continuous parameter space can instead be used to \emphinduce such structure when the corresponding symmetry is only approximate. We introduce a graph-Laplacian smoothness prior that encourages representations to vary smoothly over a physical parameter space without specifying a representation geometry; in the continuum limit it selects Laplacian eigenfunctions and therefore predicts the Fourier-like geometry associated with translation symmetry. On multimodal stellar representations spanning six astronomical surveys, the prior induces Toeplitz Gram structure and sinusoidal modes along stellar parameters, including in modalities whose native representations lack this geometry, while modality-specific residuals preserve structure outside the imposed coordinates. Of the exact-symmetry predictions, stationarity in a recoverable warped coordinate survives; diagnostics assuming exact quantization or asymptotic rates do not. The prior also improves downstream linear-probe performance and label efficiency, suggesting a general strategy: where a meaningful continuous parameter space exists, its local geometry can serve as a data-informed prior.
[LG-190] An important Bottleneck in LLM Optimizers: Principle-Level Local Optima
链接: https://arxiv.org/abs/2610.24291
作者: Nian Ran,Tianhe Jia,Zhengyang Mu,Yue Wang,Sebastian Brulin,Markus Olhofer,Mauricio A. Alvarez,Yang Liu,Richard Allmendinger
类目: Machine Learning (cs.LG)
*备注: 10 pages under review
Abstract:Feedback-driven LLM optimizers are increasingly used for evolutionary search, Bayesian optimization, heuristic design, and scientific discovery. Yet rapid initial gains often give way to prolonged plateaus despite substantial remaining evaluation budgets. We investigate \emphprinciple-level search collapse, where an optimizer’s search becomes confined to a narrow set of design principles, even though it may continue to generate distinct candidates and achieve small improvements. We assemble a diagnostic benchmark spanning molecular optimization, network sentinel-node selection, and GPU-kernel optimization, and introduce a decision-tree protocol combining principle narrowing, prolonged performance plateaus, and gaps to attainable reference solutions. Across 161 assessable runs on the three problem domains, 78.3% receive support for collapse, including 55.3% with strong support. Candidate-level variation can nevertheless conceal restricted search at the principle level. We then test \emphdiversity injection, a common response to search stagnation, such as higher t, multiple LLMs, and web search, and show that it provides limited and inconsistent benefits: among 65 molecular runs, still 73.8% and 53.8% receive support and strong support respectively. We finally test reference-principle interventions and find that they improve some runs, but their benefits vary across models and tasks: access to a better principle does not guarantee its effective use. Our benchmark and diagnostic protocol provide a basis for evaluating future optimizers on principle-level collapse. Our findings highlight the need for online collapse detection and better mechanisms for generating and maintaining alternative principles during optimization.
[LG-191] Dataset Inference for Data Provenance and Privacy Auditing in Tabular Foundation Models
链接: https://arxiv.org/abs/2610.24278
作者: Dariush Wahdany(1),Jesse Cresswell(2),Naiqing Guan(2),Atiyeh Ashari Ghomi(2),Franziska Boenisch(1),Adam Dziedzic(1) ((1) CISPA Helmholtz Center for Information Security, (2) Layer 6 AI)
类目: Machine Learning (cs.LG)
*备注: Code at this https URL
Abstract:Tabular foundation models (TFMs) are increasingly deployed through black-box APIs and trained on large collections of real-world tabular datasets. As this training paradigm becomes more widely adopted, private or proprietary datasets may be incorporated into pre-training corpora, yet there are currently no dedicated methods for determining whether a given tabular dataset was used to train a TFM. As a solution, we introduce the first dataset inference method for TFMs, aiming to infer whether a suspect dataset was part of a model’s pre-training data. We systematically analyze a broad collection of candidate signals that can be observed from a black-box TFM via input manipulation and find that individual signals provide only weak evidence of training data membership. However, by combining multiple complementary signals, we can infer dataset membership for several state-of-the-art TFMs trained on real tabular data, achieving up to 0.99 ROCAUC dependent on the TFM. We then study factors that influence dataset identification, including pre-training data composition, model capacity, and the use of real vs. synthetic data. Our results show that our dataset inference method is a practical auditing tool for detecting privacy leakage and the use of proprietary datasets in TFMs.
[LG-192] Dynamic Neural Optimal Transport for Subset Alignment
链接: https://arxiv.org/abs/2610.24258
作者: Austin J. Brockmeier,Bilal Riaz
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study optimal transport between distributions when only a subset of the target population is relevant to the source. The goal is to map the source distribution to a reweighted target distribution whose density ratio with respect to the full target lies in [0, c] . This constraint enables the transported distribution to select an appropriate target subpopulation while limiting how strongly any region of the target can be emphasized. We propose a dynamic neural optimal transport framework for this setting using two neural networks. The first parameterizes a time-dependent potential whose spatial gradient defines the velocity field and whose terminal value enforces the density-ratio constraint. The second parameterizes an interpolation between the source distribution and the learned terminal distribution. Unlike prior neural dynamic optimal transport methods, which typically assume a fixed terminal distribution, our formulation learns the terminal distribution within the support of the target. We demonstrate the method on positive-unlabeled learning, unpaired domain translation, and trajectory inference, where meaningful alignments exist between source and target distributions, but direct distribution matching is inappropriate because the target contains additional or irrelevant components
[LG-193] Extending Myersons Optimal Auctions to Correlated Bidders via Neural Network Interpolation NEURIPS2026
链接: https://arxiv.org/abs/2610.24240
作者: Mingyu Guo,Jiayuan Liu,Vincent Conitzer
类目: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG); Theoretical Economics (econ.TH)
*备注: Accepted at NeurIPS 2026
Abstract:We aim to design revenue-maximizing single-item auctions that are deterministic, strategy-proof, and ex post individually rational — the quintessential fundamental model for optimal mechanism design. While Myerson’s seminal work solved this model for independent bidders, straightforward extensions to correlated settings often result in non-monotonic, and therefore non-strategy-proof allocations. Critically, optimal allocation design for correlated bidders is NP-hard to approximate, rendering theoretical optimality guarantees in this domain computationally intractable. We establish a new standard of empirical rigor for this theoretically hard model by proposing an empirical pipeline. We synthesize neural interpolation of Myerson’s greedy allocation based on marginal profits (the correlated variant of virtual valuation), formal neural verification to enforce exact strategy-proofness, and a constructive repair procedure. Empirically, our method is highly effective, consistently achieves near-optimal revenue across a wide range of distributions, including synthetic, adversarial, and real-world (consumer, financial, and industrial) datasets. Compared to existing manual and neural baselines, our approach shows substantial improvement, often reducing the revenue gap by an order of magnitude. Our study is also the first to introduce (unattainable) greedy revenue as a rigorous upper bound for empirical benchmarking, providing a definitive quantitative performance measure where true optima remain theoretically elusive. We demonstrate the generality of our approach by extending it to multi-unit auctions with unit demand and integrating our verification techniques into RegretNet to achieve exact strategy-proofness. Comments: Accepted at NeurIPS 2026 Subjects: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG); Theoretical Economics (econ.TH) MSC classes: 91B26, 68T07 ACMclasses: J.4; I.2.6 Cite as: arXiv:2610.24240 [cs.GT] (or arXiv:2610.24240v1 [cs.GT] for this version) https://doi.org/10.48550/arXiv.2610.24240 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Mingyu Guo [view email] [v1] Fri, 16 Oct 2026 21:10:03 UTC (86 KB)
[LG-194] Hidden Trajectories Dont Lie: Correcting Corruption Drift for Test-Time Adaptation
链接: https://arxiv.org/abs/2610.24220
作者: Amirhossein Sabour,Amir Khasahmadi,Mengliu Zhao
类目: Machine Learning (cs.LG)
*备注: 35 pages, 15 figures
Abstract:Test-time adaptation (TTA) adapts pretrained models to distribution shifts encountered during deployment. However, most existing methods rely on the weight representations of the final layer or signals derived from the model’s output probability distribution. We study hidden trajectories: the ordered sequence of intermediate representations produced across a deep neural network. We observe that clean and corrupted trajectories become more separable as corruption severity increases. We introduce Drift-TTA, which adds zero-initialized low-rank adapters to a frozen backbone and trains them to align corrupted hidden trajectories with those from cleaner inputs. The approach supports two variants: a source-dependent version trained with clean targets, and a source-free setting that learns from corrupted images at different severities. We develop an online TTA approach that freezes the backbone and adapters while updating a scalar gate to control adaptation strength. We further cast trajectory alignment as MAP inference under a Gaussian observation model and derive a KL-based transfer bound characterizing test-time trajectory misalignment in terms of offline residual error, residual dimensionality, and distribution shift between offline and test corruptions. We evaluate Drift-TTA on CIFAR-10-C, CIFAR-100-C, and ImageNet-C using supervised ViTs and CLIP models. On CIFAR-100-C at severity 5, our source-free online approach improves accuracy over no TTA by 13.3 percentage points with ViT-B/16 and 11.6 percentage points with CLIP ViT-B/16, reaching 74.9% and 47.4%, respectively. Under the same setting, prior TTA baselines yield ECEs of 18.00% for ViT and 14.70% for CLIP, while our approach reduces them to 9.53% and 5.77%, respectively. These results demonstrate that hidden trajectory alignment provides an effective signal for both offline correction and lightweight online test-time adaptation.
[LG-195] Certified neural approximations for parabolic obstacle problems
链接: https://arxiv.org/abs/2610.24202
作者: Davood Damircheli
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Functional Analysis (math.FA)
*备注:
Abstract:Residual-trained neural surrogates carry no error guarantee, a limitation for free-boundary problems such as phase change and contact. We derive guaranteed error bounds for neural approximations of parabolic obstacle problems in first-order state–flux–multiplier form. The main result bounds the flux error of the network and the energy error of its feasible projection \max\v,g\ by the constitutive residual, the initial mismatch, the complementarity defect, and the PDE residual weighted by c_F^2/\alpha , without the coercivity and norm-equivalence constants of the first-order formulation. It extends the Apushkinskaya–Repin majorant for the heat operator to general coefficients, free multipliers and a flux bound. A penetration-consistency result shows that neural triples with arbitrarily small bound exist. Outward-rounded cellwise Taylor-model enclosures turn the bound into a finite computation valid on any adaptive partition. The flux and gradient errors of the projected output are certified below 0.27 and 0.33 of the solution scale for a one-dimensional Stefan problem and, after adaptive refinement, below one half for all three trained networks on a two-dimensional expanding disk. On an expanding ellipse the gradient bound stays above one half; substitution diagnostics trace the floor of the trained networks to the projected state near the free boundary, and the remaining gap to enclosure overhead. Without a closed-form solution, a validated lower bound replaces the solution norm; there the enclosure is nearly exact, but the trained networks have errors above the solution scale and no useful accuracy is certified. The bounds control a state–flux norm, not the multiplier or the free-boundary location.
[LG-196] On Reparameterizing the Score Function NEURIPS2026
链接: https://arxiv.org/abs/2610.24188
作者: Mohammad Hossein Amani,Leonardo Bocchieri,Maxime Peyrard,Clément Hongler,Robert West
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: Accepted at NeurIPS 2026. 23 pages
Abstract:Policy gradients for autoregressive sequence models differentiate each token’s log-probability while holding the conditioning history fixed. This discards information about how earlier decisions shape later conditional distributions, creating a credit-assignment bottleneck when rewards are sparse. We introduce History-Reparameterized (HR) policy gradients, which replace the frozen prefix in the score function with a differentiable reparameterized or relaxed history, allowing gradients of downstream log-probabilities to propagate backward through the generated trajectory. The resulting history-pathwise term is generally biased when added directly, so we derive Optimal HR (OHR), a centered control-variate estimator with a closed-form coefficient that minimizes the trace of the gradient variance at the population level. The variance reduction is governed by temporal structure in the policy dynamics: random or weakly structured policies provide little benefit, while policies with learned sequential dependencies yield informative history-pathwise corrections. For discrete tokens, the choice of relaxation is crucial: embedding-level relaxation yields substantial variance reduction that grows with horizon, whereas straight-through Gumbel-Softmax yields essentially none in our diagnostics. We validate the method on a Gaussian linear Markov model, a sparse-reward T-maze with a recurrent policy, and an autoregressive Transformer model for small Traveling Salesman Problem (TSP) instances.
[LG-197] Margin Certificates for Voting Classifiers under Mixing Dependence
链接: https://arxiv.org/abs/2610.24166
作者: Adria Binte Habib,Ibne Farabi Shihab
类目: Machine Learning (cs.LG); Statistics Theory (math.ST); Machine Learning (stat.ML)
*备注:
Abstract:Margin bounds explain why weighted votes generalize after training error vanishes, but dependent observations require concentration arguments that keep the dependence model and failure budget explicit. We study stationary mixing streams without treating dependence as a formal substitution of an “effective sample size.” For a finite base class, we combine independent-block Rademacher theory for stationary \beta -mixing processes with a clipped margin loss to obtain an explicit uniform voting-margin certificate, whose dependence penalty is set by the number of approximately independent blocks and the coupling budget 4(\mu-1)\beta(a) . A black-box transfer theorem shows that any i.i.d. voting-margin guarantee remains valid on a regularly spaced subsequence, with only an additive \beta -mixing failure probability. For votes fixed independently of a stationary finite-state reversible Markov stream, spectral concentration gives the cleaner \sqrtt_rel/n rate, and a sticky-refresh construction shows this exponent is unavoidable. The constant is statistic-specific: the effective sample size is n/\tau_int(g) , with \tau_int(g) \le 2t_rel-1 . When the dependence scale is unknown, a batch-means Student-t interval is valid up to an explicit remainder vanishing in the number of batches, with width governed by \tau_int(g) rather than the worst-case t_rel . Experiments show: the learned-vote certificate is nonvacuous once the block count is large; a two-timescale construction gives a batch-means interval 8.1x narrower than the worst-case interval at nominal coverage (0.945); block-bootstrap and batch-means intervals undercover under strong dependence while the certificate stays valid; and on real, ordered UCI HAR sensor data the dependence-agnostic certificate reaches 0.95 coverage where naive intervals reach only 0.38.
[LG-198] Budgeted Bayesian Bandits for Future Utilization
链接: https://arxiv.org/abs/2610.24156
作者: Sudipto Guha
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS)
*备注:
Abstract:The multi-armed bandit problem models the trade-off between exploration and exploitation. In this paper, we study the variant where exploration has a hard budget and precedes exploitation. Prior guarantees include a 1/3 approximation by Guha and Munagala using Lagrangian duality, and a 4/27 approximation by Ma against a primal linear programming relaxation for a more general model. We show that the analyses of both algorithms are tight, even when their relaxations are asymptotically exact. We provide a different efficient algorithm that achieves a 0.43 approximation. We then provide an efficient algorithm whose guarantee approaches 1-e^-1 approx 0.63 for bounded exchangeable reward observations in a non-vanishing regime, which is asymptotically tight against natural linear programming relaxations. The technique also gives certificates for stopping sequential model evaluation.
[LG-199] Recording-Aware Evaluation of Multimodal OT Intrusion Detection ACSA
链接: https://arxiv.org/abs/2610.24128
作者: Mustafa Umut Ozbek,Khalil El-Khatib,Li Yang
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 12 pages, 3 figures, 5 tables. Accepted for presentation at the Industrial Control System Security Workshop (ICSS 2026), held in conjunction with the IEEE Annual Computer Security Applications Conference (ACSAC 2026), Los Angeles, CA, USA, December 2026. Code and results are available at: this https URL
Abstract:Machine-learning-based intrusion detection systems (IDSs) for operational technology (OT) are commonly evaluated on benchmark datasets in which normal and attack data are collected in separate recordings. In such settings, high performance may reflect recording-associated differences as well as attack behavior. We introduce a recording-aware evaluation procedure that compares attack windows with background from the same recording, uses prior data for model fitting and calibration, tests temporal alignment against randomized event schedules, and reports event coverage together with alarm burden. On the ProvICS multimodal dataset, cyber modalities strongly distinguish the benign and attack recordings before the first labeled attack. Same-recording evaluation reverses the apparent modality ranking across all studied window sizes: the physical view becomes the strongest individual modality, while cyber-only fusion shows the largest difference between cross-recording and same-recording AUC, reaching 0.54. Operational analysis further shows that permissive event-overlap metrics can overstate detection when a small number of long alarm episodes cover multiple actions. An external analysis on HAI shows smaller and reference-dependent recording sensitivity. These findings show that cross-recording benchmark performance should be interpreted alongside same-recording, forward, temporal, and operational evaluations.
[LG-200] Data-Driven Resource Aggregation for Scalable Digital Twin and Simulation Model Generation
链接: https://arxiv.org/abs/2610.24126
作者: Yucheng Wu,Alp Akcay
类目: Machine Learning (cs.LG)
*备注:
Abstract:As industrial digitization accelerates, the use of digital twins has become widespread. However, modern manufacturing systems are highly dynamic and frequently reconfigured, making manual model development and updates increasingly difficult. Recent research on automated simulation model generation shows that historical data such as event logs can yield simulation models automatically. However, modeling every resource at full granularity leads to unnecessary complexity and computational cost. We propose a method for automatically selecting and aggregating resources to simplify simulation modeling. Our method uses ideas from inverse reinforcement learning to learn behavioral value representations of resources from production event logs. This process identifies and aggregates resources with similar functional contributions. Experiments on manufacturing systems, using throughput and system time as performance measures, demonstrate that value-based resource aggregation performs well when the goal is to substantially reduce complexity while preserving performance accuracy.
[LG-201] Looped GNNs Beyond Fixed Depth: Graph Size Generalization and Parameter-Efficient Adaptation
链接: https://arxiv.org/abs/2610.24112
作者: Zhiyang Wang,Zhishang Luo,Yusu Wang
类目: Machine Learning (cs.LG)
*备注: 36 pages, 12 figures
Abstract:Generalizing from small graphs to larger ones can require computation beyond the depth of a neural model used during training. We study target tasks that admit a common local iterative computation and whose effective dependency radius grows with graph size. The central idea is to learn a local update rule through GNN pretraining, then repeatedly apply the same learned processor beyond the training depth to extend information reach on larger graphs without retraining. Calling such parameter-shared architecture a \emphlooped GNN, we categorize six schemes for looped GNNs to separate the effects of depth extension, fine-tuning, and freeze-and-steer adaptation. Our analysis establishes fixed-depth information limits and sufficient conditions under which accurate iterative reuse overcomes them. To improve beyond unadapted reuse, we introduce \emphLoop-Freeze-Steer (LFS), which freezes the pretrained processor and learns a small controller to steer its repeated application, using much fewer trainable parameters than full fine-tuning. Our finite-sample analysis gives sufficient conditions for adaptation to improve performance on larger graphs. Experiments on synthetic tasks and chip placement evaluate depth extension and adaptation, highlighting LFS’s accuracy–trainable-parameter trade-off for size generalization.
[LG-202] GLOW: General Linear Models Optimized with Wards Method
链接: https://arxiv.org/abs/2610.24085
作者: Matt Higger
类目: Machine Learning (cs.LG)
*备注:
Abstract:Neuroimaging seeks brain regions whose imaging features correlate with variables of interest (e.g., diagnosis, age), usually via the General Linear Model (GLM). Region-Based Analysis dilutes or splits any effect that does not fill exactly one atlas region. Voxel-Based Analysis, with Cluster Extent Thresholding and Threshold-Free Cluster Enhancement built on it, fits a GLM per voxel and groups adjacent significant voxels. But neighboring voxels can share a statistic without sharing a GLM, which a per-voxel statistic cannot detect. GLOW (General Linear Models Optimized with Ward’s method) instead builds a hierarchy of nested regions by greedily minimizing GLM error (provably a special case of Ward’s clustering). Maximum-statistic (Westfall-Young) permutation controls family-wise error over all regions. Significance spreads to an effect’s nested fragments and the regions containing it; a final pruning step seeks whole, pure effect extents. We plant synthetic effects of known extent and varying strength in Human Connectome Project (HCP) diffusion images and in white noise. Pruning delivers wholeness: GLOW returns a spatially consistent effect as one region on 57% of HCP trials, the best baseline on 35%; baselines recover comparable volume, but in many pure pieces. The cost is positive predictive value (PPV): GLOW is more sensitive (by up to 0.29), but its regions overrun the effect, mostly because no hierarchy region isolates it, not because of pruning. On HCP, GLOW’s mean Dice gains 0.059 over the best baseline at weaker effects and gives back 0.005 at stronger ones. On white noise, GLOW trails by up to 0.61 Dice: its hierarchy holds the effect, but its permutation test does not certify it. Runtime grows roughly linearly with voxel count, about 4.5 times the voxel-wise methods’ on the full HCP volume. The PPV loss may trace to the assumption of spatially independent residuals. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.24085 [cs.LG] (or arXiv:2610.24085v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.24085 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Matt Higger [view email] [v1] Fri, 16 Oct 2026 18:10:04 UTC (574 KB)
[LG-203] oolRosetta: Surface-Form Invariance of Tool-Call Guards NEURIPS2026
链接: https://arxiv.org/abs/2610.24055
作者: Hasan Demirkiran,Jens Ernstberger
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: Accepted at the NeurIPS 2026 Workshop on Agents in the Wild: Safety, Security, and Beyond (AIWILD). 4 pages plus appendices
Abstract:Step-level guards for tool-using agents are evaluated on tidy JSON tool calls, but deployed coding agents act through execute_bash and execute_code tools whose payloads are strings. We introduce ToolRosetta, which re-renders every candidate action in the AgentDojo and AgentHarm splits of TS-Bench as a Python call, as a shell command, and without the agent’s rationale. The renderings are produced by code and verified by round-trip parsing, so the user request, the history, the label and the action’s effect stay fixed while only the surface form changes. Across eleven guards with two placebo controls, frontier and strong open-weight guards are stable: no frontier error rate moves by more than 3.4 points. Four of five specialized agent guards are not: under the shell form their false-positive rates rise by 21 to 47 points on AgentDojo, and TS-Guard’s by 38 points on its own home benchmark, blocking between a third and nine in ten benign actions; the fifth, StepGuard, does not move. A small open guard drifts five points more permissive. Removing the agent’s rationale moves most guards whose input carries it, in guard-specific directions. A request-blind control on one guard shows that the rationale is where intent is read from. A parser that restores the canonical call in front of the guard removes the effect for these renderings, by construction. Benchmark accuracy on JSON calls does not predict a guard’s operating point behind a code-acting agent.
[LG-204] What Remains Without Test Labels? Re-evaluating a Graph-Level Anomaly Detection Benchmark
链接: https://arxiv.org/abs/2610.24045
作者: Yujie Lu,Da Yan
类目: Machine Learning (cs.LG); Social and Information Networks (cs.SI)
*备注: 68 pages (8 main pages), 7 figures, 54 tables. Code and data: this https URL
Abstract:Graph-level anomaly detectors are trained only on normal graphs, yet benchmarks may use test labels for model selection. We examine whether UB-GOLD results reused by later papers support label-free detection. We audit UB-GOLD and 14 GLAD code bases, re-run 21 detectors on UB-GOLD’s folds under selection rules with and without test labels, and evaluate sensitivity to known structural changes. Test labels select checkpoints in UB-GOLD’s shared training loop and checkpoints or hyperparameters in 8 of the 14 code bases. UB-GOLD also evaluates its kernel baselines against misaligned labels. Across nine datasets, selecting checkpoints by test AUROC adds an average of 5.12 AUROC points [3.97, 6.24] for the 12 UB-GOLD detectors relative to the final checkpoint and 4.13 [3.32, 4.79] relative to held-out normal graphs. In a post hoc permutation analysis with no truly better checkpoint, picking the best of noisy test evaluations still yields a gain more than half as large as the one observed against the final checkpoint. Configurations tuned on test labels score a further 1.53 points [0.64, 2.42] above repository defaults, a gap that mixes the value of test labels with that of tuning itself. With tuned hyperparameters and label-free checkpoint selection, the two strongest detectors retain their lead, while comparisons below them are less stable. Six of eight detectors evaluated out of sample miss size-preserving composition changes that graphlet counts detect. We recommend reporting results selected without test labels as the main comparisons and release per-run outputs that allow verification without retraining.
[LG-205] Visual Causal-chain Bookmarking: Retrieval-Weighted Credit Assignment for Reinforcement Learning from Delayed Reward
链接: https://arxiv.org/abs/2610.24032
作者: Prathap Selvakumar
类目: Machine Learning (cs.LG)
*备注: 9 pages, 2 figures, 5 tables. Master’s dissertation research, University of Manchester
Abstract:RL agents trained on a single trajectory-level return face a credit-assignment problem: when a reward arrives many steps after the decision that caused it, every action gets the same signal regardless of causal responsibility. This paper studies Visual Causal-chain Bookmarking (VCBM), which concentrates credit on the decision points that determine an episode’s return, via two instantiations. The first targets a fully observed, low-dimensional MDP (a delayed-reward watch-repair task), where VCBM statistically discovers which state components the agent causally controls and estimates per-decision-point action values directly, evaluated against ground truth and RUDDER, an LSTM-based baseline. VCBM converges to 95.07% correct decisions in a mean of 1,532.3 episodes across 20 seeds, versus RUDDER’s slower, more variable convergence (mean 3,327.1 episodes; p0.001); VCBM’s discovered causal structure exactly matches ground truth in all seeds. The second targets a partially observed, language-based setting: an LLM agent trained with Leave-One-Out PPO (LOOP) in AppWorld, where causal discovery is intractable and a structural heuristic (state-mutating tool calls) substitutes for it, with credit distributed via retrieval over episodic memory and blended with LOOP’s advantage. Over 200 iterations on identical Qwen2.5-7B-Instruct/LoRA/FSDP2 configurations, VCBM-blended LOOP improves task success over plain LOOP (0.354 vs. 0.458), reduces final-step max importance weight, and improves token efficiency, with a small KL exception favouring LOOP. The watch-repair result supports the claim that weighting causally relevant steps is more sample-efficient than an undifferentiated baseline or a learned return-redistribution network; the AppWorld result, limited to one seed per arm by a 2-GPU budget, is a compute-limited demonstration at a harder, natural-language scale, not a statistical confirmation.
[LG-206] Do Knowledge and Reasoning Oriented Benchmarks Induce Stable Representation Signatures in LLM s? NEURIPS2026
链接: https://arxiv.org/abs/2610.24029
作者: Parsa Rahimi,Azim Dehghani Amirabad
类目: Machine Learning (cs.LG)
*备注: NeurIPS 2026 Interpretability as a Science Workshop
Abstract:Internal signatures could support monitoring, adaptive computation, and mechanistic analysis, but benchmark prompts can be separable without any shared computational distinction. We test whether a prespecified knowledge/reasoning label defines a reusable internal signature in five contemporary 7B to 31B models. We treat the \textbftask as the unit of inference probes hold out complete tasks, labels are permuted across tasks, and the best layer and stream are reselected for every permutation. Raw probes recover an eight-versus-eight-subject MMLU partition in all five models (macro-F1 0.78 – 0.81 ), but prompt text predicts it nearly as well: under an identical probe configuration, the activation advantage is only 0.017 – 0.070 macro-F1 and is not significant. At the prespecified 0.01 level, lexical adjustment removes significance while the adjusted states still identify individual subjects, and the selected site changes across task panels and selection rules. Additionally, two released base-training trajectories show no common direction. These pooled layerwise measures recover the task label but identify no transferable knowledge/reasoning signature at the tested sensitivity, which cannot exclude moderate effects. The resulting protocol gives future internal monitors and intervention studies a sequence of tests to pass. Project page: \urlthis https URL.
[LG-207] Dynamical Kernel Attention: Estimating the Present from a Record of the Past
链接: https://arxiv.org/abs/2610.24026
作者: Peter Racioppo
类目: Machine Learning (cs.LG)
*备注:
Abstract:Standard self-attention compares tokens by similarity but does not account for how much of a past state survives to the present. We introduce Dynamical Kernel Attention (DKA), which derives attention from robust estimation of a latent state evolving under stochastic dynamics. As a past state decays toward equilibrium under the dynamics, the information it carries about the present vanishes, bounding the total evidence a head can draw. Unlike recency biases, DKA can abstain when relevant evidence is weak and guarantees a bounded error when old tokens are discarded. DKA preserves the parallel structure and computational cost of standard attention. We present the model and its theory; empirical evaluation is deferred to a subsequent version.
[LG-208] SPADE-KD: Peer-to-Peer Collaborative Learning under Heterogeneous Settings AAAI2027
链接: https://arxiv.org/abs/2610.24013
作者: Harikrishna Kuttivelil,Katia Obraczka
类目: Machine Learning (cs.LG)
*备注: Previously submitted to AAAI 2027
Abstract:Peer-to-peer collaborative learning is an emerging distributed deep learning paradigm that eliminates the need for a central aggregator and disseminator, avoiding the single-point-of-failure and bottleneck issues they can bring. Many such approaches use parameter averaging, requiring clients to have homogeneous model architectures, whereas others use knowledge distillation as an alternative while leaning on shared public data or centralized control infrastructure. Many of these approaches also make a one-size-fits-all assumption, i.e., jointly training a single, globally optimal solution for all clients, when in reality, clients’ underlying data distributions might differ such that a globally optimal solution may not be locally optimal for all clients. To enable collaborative learning in environments with both model architecture and data distribution heterogeneity, we present SPADE-KD – a fully decentralized model distillation framework that optimizes collaboration towards personalized rather than global performance. SPADE-KD leverages two mechanisms – adaptive distillation and utility-driven peer selection. Adaptive distillation modulates the strength of each incoming model transfer by the performance advantage of the incoming model compared to the native model measured by their task losses, such that a client will distill more aggressively from incoming models that outperform it on its own local data. Utility-driven peer selection uses the loss-informed utility derived from adaptive distillation to rank prospective partners, focusing on exchanges that measurably improve the local model while suppressing low-utility transfers. Using federated classification tasks, we demonstrate that SPADE-KD consistent outperforms naive decentralized distillation with lower loss and faster convergence and that its individual components are more effective together than independently.
[LG-209] Weakly Supervised Discovery of Motion Styles for Compositional Robot Control
链接: https://arxiv.org/abs/2610.24008
作者: Heedo Woo,Antonio Marino,Eduardo Sebastián,Amanda Prorok
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注:
Abstract:When a robot possesses more degrees of freedom than a task requires, multiple trajectories can achieve the same objective using distinct motion styles. Compositional robot control aims to preserve a selected style while simultaneously executing an independent task command, effectively balancing these potentially competing objectives. In this paper, we present CODES, a framework for compositional robot control that discovers a coherent style repertoire from diverse recordings and composes it with independent commands. First, motion retargeting maps the recordings to motions executable by the target robot; next, weakly supervised consensus encoding and density-based clustering identify styles within and across recordings without a prescribed count; finally, reinforcement learning trains a policy to compose the discovered styles with independent commands. We evaluate CODES on a Unitree Go2 quadruped across two simulators and physical hardware. Compared to baselines, CODES significantly improves kinematic coherence. Furthermore, our controller successfully tracks velocity commands that generalize beyond the support of the recorded data, exhibiting robust and distinguishable style characteristics during both cross-simulator and real-world hardware deployments. Project website: this https URL Subjects: Robotics (cs.RO); Machine Learning (cs.LG) Cite as: arXiv:2610.24008 [cs.RO] (or arXiv:2610.24008v1 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2610.24008 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-210] Physics-Aware Self-Supervised Pretraining on Multi-Domain Raman Spectra
链接: https://arxiv.org/abs/2610.23955
作者: Charalampos Davalas,Georgios Levantinos,Anastasios Dimitriou,Nikolaos Papanikolaou,Georgios Th. Papadopoulos,Dimitrios Michail
类目: Machine Learning (cs.LG); Image and Video Processing (eess.IV)
*备注:
Abstract:Raman spectroscopy generates large volumes of unlabeled data that are difficult to exploit with conventional supervised learning. This work introduces a multi-domain self-supervised learning framework based on a masked autoencoder (MAE) for learning transferable Raman spectral representations from heterogeneous biological, chemical, forensic, and materials datasets. The proposed pretraining strategy incorporates physics-aware objectives by combining masked intensity reconstruction with a sparse Voigt-profile decoder that models spectral peak structure through differentiable peak centers, widths, and amplitudes. A denoising objective is further included to improve robustness to spectral corruption. The learned encoder is evaluated on substance classification, component identification, concentration estimation, and spectral denoising under linear probing, attention-based probing, end-to-end fine-tuning, and training from scratch. Experimental results show that self-supervised pretraining is most beneficial for challenging tasks with limited labeled data. Attention-based probing also reveals substantially more useful information in the frozen representation than linear probing, indicating that conventional linear evaluation may underestimate representation quality. In addition, the benefits of pretraining become more pronounced with increasing encoder capacity. Overall, the proposed framework demonstrates that physics-aware, multi-domain self-supervised pretraining can improve label efficiency, robustness, and transferability across diverse Raman spectroscopy tasks.
[LG-211] LatLM: Line-Parallel Language Models with Deferred Commitment
链接: https://arxiv.org/abs/2610.23951
作者: Shaorong Zhang,Shiyuan Li,Longxuan Yu,Greg Ver Steeg
类目: Machine Learning (cs.LG)
*备注:
Abstract:Line-parallel language models advance multiple lines concurrently for substantial speedups. However, frameworks like the Line-Coupled Language Model (LCLM) enforce a fixed commitment schedule across lines. Because dependencies vary within text, a fixed schedule cannot hold back a line whose prerequisites remain uncomputed, forcing premature commitments and cascading reasoning errors. To address this limitation, we introduce the Lattice Language Model (LatLM), making commitment schedules an adaptive decoding-time choice. Text occupies a 2D lattice where eligible frontier tokens are scored each round; a policy commits confident candidates while deferring unready ones for extended context. One forward pass supervises entire sampled trajectories via 2D causal masking, while an append-only cache retains exact states. At 482M parameters, we establish two findings. (1) LatLM matches a data- and capacity-matched autoregressive model on held-out text when decoded sequentially, while its adaptive parallel decoder achieves near-autoregressive accuracy on independent-line tasks using 25–41% as many decoding rounds. (2) When cross-line dependencies cause LCLM’s fixed schedule to fail, deferred commitment substantially improves accuracy while retaining parallelism where dependencies permit. On mixed dependency chains, LatLM achieves 96.5% accuracy versus LCLM’s 78.5%, using fewer than half the decoding rounds of the autoregressive baseline.
[LG-212] Compaction-Aware Reinforcement Learning For Ascend Triton Kernel Generation
链接: https://arxiv.org/abs/2610.23950
作者: Chuansai Zhou,Yanlin Liu,Bingning Huang,Shidi Xi,Chao Huang,Boxing Chen,Mingming Zhu,Guanghuan Fang,Ben Feng,Zongzuo Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Automatically generating high-performance kernels with LLM-based agents promises substantial gains in development efficiency, but two obstacles remain. First, while increasing multi-turn interaction with execution environments is crucial for refining generation quality, continuous context accumulation rapidly exceeds the model’s effective context window. Second, training long-horizon tasks via GRPO inherently incurs sparse rewards due to its reliance on terminal outcome evaluation, severely hindering step-level credit assignment; To address these obstacles, we propose ATRL (Ascend Triton Reinforcement Learning), a post-training method that refines GRPO into a compaction-aware, segment-level objective for long-horizon kernel generation. We also design a lightweight agent harness providing cascaded compilation, correctness, and profiling feedback, alongside an Ascend-specific knowledge base injected during agentic RL training and evaluation to compensate for the base model’s missing platform expertise. Evaluations on NPUKernelBench and KernelBench, both executed on Ascend hardware, show that ATRL consistently surpasses all baselines, including GRPO, in both correctness and speedup.
[LG-213] Parameter misspecification in physics-informed neural networks: A closed-form audit and its limits
链接: https://arxiv.org/abs/2610.23948
作者: David McShannon,Nicholas Dietrich
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA); Computational Physics (physics.comp-ph)
*备注: 10 pages main body, 3 figures main body, 7 pages supplementary material
Abstract:Physics-informed neural networks (PINNs) are trained to minimize the residual of a prescribed equation, so a low training loss does not indicate whether a coefficient in that equation was specified incorrectly. This study evaluates what can be inferred about the encoded parameter from the trained network alone. When the residual is linear in the parameter, written as A - \nu B , the least-squares estimate \hat\nu = A^\topB/B^\topB is obtained from a single evaluation of the frozen network, without retraining or reference data. Its accuracy depends on how closely the network satisfies its own equation, quantified by a normalized residual \rho . Recovery error is associated with \rho (log-log correlation 0.72 across 36 convection-diffusion networks), and empirical exponents of 0.63 to 1.20 across three equations decompose into one half plus a measurable angular term. For Burgers’ equation, a single-model test based on a block-bootstrap standard error flagged 3 of 300 correctly specified networks (observed false-alarm rate 1%), all 2% misspecifications at a practical budget, and two thirds of 1% misspecifications after convergence. For the self-similar blowup rate of the Córdoba-Córdoba-Fontelos equation, where the rate also enters the architecture, the estimator recovers about half of the misspecification, whereas \rho increases approximately quadratically and distinguishes a 0.2% error. When the incorrect rate enters both the equation and the architecture, errors of up to 2% were not detected at the precision reached. At practical budgets, the raw loss can be lower for the misspecified model, a scaling effect that \rho does not share.
[LG-214] Fast Construction of Inner Sphere Trees Using Machine Learning
链接: https://arxiv.org/abs/2610.23940
作者: Jianneng Huang,Wenjia Rao
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 11 pages, 7 figures
Abstract:Collision detection for complex non-convex objects is a bottleneck in physics simulation, virtual reality, and robotics. Inner sphere trees (ISTs) answer proximity and penetration queries efficiently, but their greedy leaf-sphere construction is sequential and computationally expensive, which limits both large-scale preprocessing and runtime use. In this work, we present ML-IST, a machine-learning-based method that reformulates leaf-sphere generation as a single-shot set-prediction problem: a network predicts a fixed set of inscribed sphere candidates from multichannel voxel geometry, differentiable coverage feedback corrects the coarse prediction and fills residual gaps at finer scales, and a lightweight geometric projection enforces validity. Averaged over the test set, a single-GPU implementation of ML-IST reduces mean leaf-generation time by a factor of 2.0 – 2.6 over single-core CPU construction across target capacities of 64–512 spheres, and qualitative results on complex models show multiscale sphere distributions that follow principal volumes and slender features. We further integrate ML-IST into the Jolt Physics and Bullet engines; in an alpha-puzzle benchmark it reaches 55.89~FPS, the highest among the compared representations (conventional IST, GImpact, and convex decomposition). These results support learned leaf generation as a practical component for fast, runtime construction of detailed collision representations.
[LG-215] Pack and Extract: An Effective Graph-Based Approach for Feature Selection
链接: https://arxiv.org/abs/2610.23938
作者: Faisal N. Abu-Khzam,Joseph R. Barr,Cynthia Chaaya,Nadim Obeid,Peter Shaw
类目: Machine Learning (cs.LG)
*备注:
Abstract:Many molecular biology data sets contain thousands of features, such as genes in single-cell experiments or other omics data. Standard predictive models often perform poorly in such settings, as they tend to retain a large number of features and fail to adequately handle correlations among them. Combinatorial approaches, such as feature selection based on the Independent Dominating Set (IDS) model, have shown promising results, but appear to have been largely overlooked, possibly due to their high computational complexity. In this paper, we propose an improvement to IDS-based feature selection that (i) favors features exhibiting greater diversity and (ii) generalizes the notion of independent dominating set to a node packing model. The resulting formulation is used as an intermediate step within the feature selection process. We evaluate our approach on five genetic data sets and compare the results with those obtained using existing methods. Our experimental results demonstrate notable improvements in predictive accuracy over traditional models, as well as increased stability of the selected feature sets.
[LG-216] Announced Breaks Separate Conformal Reliability from Frequency Calibration NEURIPS2026
链接: https://arxiv.org/abs/2610.23917
作者: Karl Li(Columbia University)
类目: Machine Learning (cs.LG); Econometrics (econ.EM); Machine Learning (stat.ML)
*备注: Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026). 53 pages, 7 figures
Abstract:An announced boundary reveals the partition before the first affected outcome, and a reset there serves one of two audits. The per-time audit reads conformal reliability, the expected distance of the threshold’s coverage from nominal at miscoverage 0.1. The frequency audit reads frequency calibration, whether a closed window’s error rate matched it. The reset rule, the post-boundary empirical quantile, attains the minimax per-time rate for independent scores. On bounded scores a recursion reset at the boundary attains a frequency gap of order one over the count when its floor adds a bounded total. At the 50th forecast after a large synthetic break, the reset rule’s per-time gap stands below that of five adaptive methods, four from their authors’ code, with every paired band below zero. After an electricity market reform, on 68 unseen scores, the band stands below zero against two, includes zero against two, adaptive conformal inference among them, and is a point at zero against one. After monetary policy decisions its pointwise band stands above zero against all five on 10 of the first 34 days. Of 9 announced boundaries one clears the smallest coverage difference its window detects, where every method’s per-time gap at the 50th forecast exceeds the nominal miscoverage. Over the first 50 forecasts the reset recursion leads the frequency audit after every synthetic break.
[LG-217] Language-Forcing: Advancing Humanoid Loco-Manipulation via Semantic Alignment of Whole-Body World Action Models
链接: https://arxiv.org/abs/2610.23894
作者: Yanjun Li,Jiarui Ye,Weiyu Guo,Zixiao He,Kangjie Ye,Dihan Wang,Yishuo Cheng,Zishang Xiang,Mingju Gao,Hao Tang
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 9 pages of main text, 31 pages of appendix
Abstract:World Action Models (WAMs) jointly predict robot actions and future observations. However, their visual prediction objectives do not explicitly distinguish task-relevant transitions from incidental scene changes. For whole-body policies, this distinction matters because task progress depends on coordinated changes in body pose, contact, and object state. We introduce LANGUAGE-FORCING, a training strategy that uses language to emphasize task-relevant changes in the future visual representations of a whole-body WAM. Given paired current and ground- truth short-horizon future frames, a frozen vision-language model assesses task progress and predicts plausible subsequent state changes. We encode these descriptions with a frozen text encoder and align their embeddings with intermediate, action-conditioned future-visual features, complementing the original visual and action prediction objectives. All auxiliary components are used only during training, introducing no additional inference overhead. On HumanoidArena, LANGUAGE-FORCING achieves an average success rate of 79.17%, exceeding the strongest competing policy by 9.46 percentage points and the same WAM without language alignment by 10.77 percentage points. Across five real-world tasks, language alignment raises average success from 53.33% to 71.33% over the same WAM baseline.
[LG-218] DUET: Decoupled Upper- and Lower-Body Execution for Task-Agnostic Humanoid Loco-Manipulation
链接: https://arxiv.org/abs/2610.23872
作者: Kapalik Khanal,Biswash Khatiwada,Stephen Afrifa,Fan Wu,Ramesh Bahadur Bist
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注:
Abstract:Changing a humanoid robot’s manipulation policy should not require retraining the controller that maintains the robot’s balance. This study introduces DUET, a decoupled execution framework that enables locomotion and manipulation controllers to operate simultaneously by combining their commands only at the robot interface. A reinforcement-learning locomotion policy controls the 12 leg joints and a waist yaw joint, tracking planar velocity, turning rate, and pelvis height from deep crouching to upright standing on a 23-degree-of-freedom Unitree G1. The 10 arm joints are excluded from the action space, while their positions and velocities are included only as observations, so that any compatible manipulation policy that generates arm-joint targets can control the upper body, and the locomotion policy keeps the lower body stable without requiring shared training or direct coordination. This study demonstrates the modular approach on hardware by pairing the same locomotion checkpoint with two different vision-language-action models, GR00T N1.7 and pi0.5. In simulation evaluation, the locomotion controller kept the robot’s lower body stable with no falls while tracking the commanded planar velocity, yaw rate, and pelvis height with errors of 0.186 m/s, 0.142 rad/s, and 0.016 m, respectively. Ablation studies identify the upper-body pose curriculum as the dominant component, and removing this training step reduces episode survival from 20.0 to 7.4 seconds and increases the fall rate to 6.06 per environment minute. Robustness tests show no falls under pushes of up to 1.0 m/s within the trained payload range, and no falls at twice the payload limit for pushes below 0.7 m/s. These results show that a manipulation policy can be exchanged without retraining, restarting, or otherwise touching the layer that keeps the robot upright.
[LG-219] Measuring the Influence of Modality-Specific Node Representations on Graphs
链接: https://arxiv.org/abs/2610.23863
作者: Xu Wang,Xunkai Li,Yinlin Zhu,Rong-Hua Li
类目: ocial and Information Networks (cs.SI); Machine Learning (cs.LG); Multimedia (cs.MM)
*备注:
Abstract:Multimodal attributed graphs combine graph structure with node text and image representations in recommendation and multimedia analysis. Graph predictors propagate these representations across connected nodes, so one input can change predictions at its own and nearby nodes. Measuring this distributed response reveals which modality-specific inputs a trained model relies on and how predictor design changes that reliance. We study influence measurement for modality-specific node representations. Given an encoded graph, a trained predictor, a training split, and fixed candidate representations, the task outputs one influence score per candidate and separate text and image rankings. We score a candidate by replacing it with its modality’s training-node mean and comparing the original and changed predictions. Existing feature attribution and data valuation generally target one output, training examples, whole nodes, or graph elements rather than prediction changes across graph nodes. Our empirical study shows that nearby-node responses contain information missed by source-only scores and simple proxies, while response magnitude alone does not identify which class changes support the task. Guided by these observations, we propose MoGI (Modality-specific Graph Influence). It propagates training labels to define task direction, reads the same replacement response for the favored class and full class distribution, and averages the resulting ranks. MoGI requires no predictor retraining or learned scoring parameters. Under a shared five-dataset, eight-predictor benchmark, it achieves the largest aggregate held-out loss increase against all five external methods at every tested ranking cutoff.
[LG-220] Forecasting Patient Trajectories from the Electronic Health Records with Action-Conditioned World Models
链接: https://arxiv.org/abs/2610.23862
作者: Simon A. Lee,Jiahang Hank Sha,Soomin Chung,Kai Akamatsu,Eran Halperin,Jeffrey N. Chiang
类目: Machine Learning (cs.LG)
*备注:
Abstract:We introduce a predictive world model for electronic health records (EHR) that represents patient histories as action conditioned latent trajectories. Rather than modeling clinical events solely through autoregressive token prediction, our framework explicitly separates patient states from observed clinical actions and learns their evolution through a state and action conditioned transition operator. At its core is a state dependent linearization predictor that propagates local changes in latent patient state, enabling time resolved prediction, multi step rollout, and action conditioned simulation. Empirically, we show that generated latent trajectories preserve downstream predictive signal, temporal structure, and subgroup level statistical relationships present in encoder derived representations. We further decode free running rollouts into the original clinical feature space and evaluate future clinical measurements against observed values using task matched forecasting baselines, demonstrating that the learned dynamics retain predictive information beyond latent space agreement. Across binary classification and time to event tasks, EHR WM achieves competitive, though task dependent, performance relative to substantially larger EHR foundation models while using a 30x smaller parameter footprint. Finally, perturbing observed action sequences produces corresponding changes in generated trajectories, demonstrating the action sensitivity of the learned dynamics. Together, these results support world modeling as a complementary framework for learning explicit, time indexed representations of patient evolution from longitudinal EHR.
[LG-221] PolyGraph: Plan-Buffered CUDA Graphs for Adaptive Speculative Decoding
链接: https://arxiv.org/abs/2610.23861
作者: Junchao Yang
类目: Machine Learning (cs.LG)
*备注: 6 pages, 1 figure. Trace-driven prototype evaluated on NVIDIA H800
Abstract:Adaptive speculative decoding changes the active batch size B and draft length d from one decode step to the next, whereas a CUDA Graph replays a fixed launch configuration. The resulting verification width M = B d forces serving systems to pre-capture many shapes, capture new ones on the serving path, or pad every step to the largest draft. We present PolyGraph, a plan-buffer mechanism for custom Triton kernels that avoids this choice. PolyGraph captures one maximum-grid graph and writes the active shape to GPU memory before each replay. Kernels skip tiles beyond that shape, while graph topology, allocations, layouts, pointers, and launch grids remain fixed; the update transfers only 28 bytes. On one NVIDIA H800, a trace-driven prototype with Qwen2.5-3B dimensions replaces 64 shape-specific graphs with one, matches the warmed baseline in observed step latency (17.7 vs. 17.8 ms P50), and reduces measured pre-serving graph-build time from 3,915 to 93 ms. Fixed-draft controls show that the benefit comes from avoiding padded payload: the P99 difference from MaxPad is negligible at zero padding waste but reaches an 18.7% reduction at 100% waste. Our evidence establishes a bounded work-shape mechanism for custom kernels, not a production serving system, and is limited to synthetic traces on one GPU.
[LG-222] Clue-Guided Money Laundering Group Discovery
链接: https://arxiv.org/abs/2606.26189
作者: Boyang Wang,Jianing Cao
类目: Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE)
*备注:
Abstract:Money Laundering Group Discovery (MLGD) aims to identify hidden criminal groups and recover their complete structures in large-scale financial networks. Existing graph anomaly detection methods mainly produce node-level risk alerts, while global group discovery methods passively search for suspicious groups over the whole network. Both are mismatched with real Anti-money-laundering (AML) investigations, where analysts usually start from a concrete clue and gradually expand the investigation to recover the responsible group. To address this gap, we propose Clue-Guided Group Discovery (CGGD), where a laundering group is progressively recovered from an initial clue set through analyst interaction. We further propose Clue2Group, a framework that first constructs a compact local investigation context to reduce noise and preserve chain-like and cycle-like laundering structures. It then estimates a clue-conditioned local risk field with a multi-semantic local-temporal GNN, and finally integrates risk, structural, and prior-pattern evidence to recover a coherent laundering group. Experiments on two large-scale AML benchmarks show that Clue2Group provides a practical clue-driven analysis framework for AML investigations, offering a feasible step toward bridging the gap between graph-based AML research and real investigation workflows.
[LG-223] Newton-Based Methods for Bilevel Optimization Beyond Lower-Level Strong Convexity
链接: https://arxiv.org/abs/2610.26839
作者: Youran Dong,Junfeng Yang,Wei Yao,Jin Zhang
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Bilevel optimization provides a framework for machine learning tasks such as hyperparameter optimization, data reweighting, and meta-learning. Existing Newton-based bilevel methods typically rely on the classical hypergradient formula, which is generally unavailable when lower-level solutions are nonunique and the lower-level Hessian is singular. We develop two Newton-based methods beyond lower-level strong convexity by exploiting curvature shared by the lower-level and multiplier updates. Newton-based GALET handles potentially singular lower-level curvature under a positive spectral-gap condition, exploiting its range structure to avoid the squared-Hessian multiplier system used in GALET. Newton-based BAMM uses a strongly convex aggregation model when the lower-level objective is convex and the upper-level objective is strongly convex in the lower-level variable. We establish convergence guarantees and improved complexity bounds relative to GALET and sl-BAMM under the corresponding assumptions. We further introduce GNBO, a generalized Newton-based framework that captures the shared-curvature structure of the proposed methods and several existing Newton and quasi-Newton bilevel methods. Experiments on synthetic problems, data hyper-cleaning, and GPT-2 data reweighting evaluate the proposed methods across different bilevel learning tasks.
[LG-224] Supervised Learning with Quantum Hamiltonian Embeddings
链接: https://arxiv.org/abs/2610.26833
作者: Riccardo Molteni,Minjun Jeon,Paolo Braccia,Abhinav Deshpande,Vedran Dunjko,Liang Jiang,Su-un Lee,Swarnadeep Majumder,Javier Robledo Moreno,Parthasarathy Suryanarayanan,Kunal Sharma
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注: 59 pages, 7 figures
Abstract:Machine learning is widely regarded as a promising application in which quantum computing could offer a computational advantage. However, general quantum machine learning models with provable guarantees on performance, trainability, and efficient inference remain limited. In this work, we introduce quantum Hamiltonian embeddings, a supervised learning framework in which a classical training dataset is encoded into a single effective, parameterized Hamiltonian. Training optimizes the parameters of this data-dependent Hamiltonian, whose approximate ground state defines the classification rule. Once the ground state has been approximated using a quantum algorithm, inference can be performed either classically using a classical representation of the learned state, or with a quantum representation of the state. We establish efficient methods for gradient estimation and prove favorable generalization guarantees. Furthermore, under standard cryptographic assumptions, we show that Hamiltonian embeddings can realize hypothesis functions beyond the reach of efficient classical algorithms. We then demonstrate the framework at scale on IBM Heron processors using a standard benchmark dataset. Together, these results establish Hamiltonian embeddings as a new framework for quantum machine learning, combining theoretical guarantees with a path for experimental investigations at scale on quantum hardware.
[LG-225] Gradient descent reaches arbitrary accuracy in a fixed number of large steps on multiclass logistic regression
链接: https://arxiv.org/abs/2610.26748
作者: Andreea-Alexandra Musat,Nicolas Boumal
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注: 40 pages
Abstract:We show that gradient descent (GD) with step size \eta can achieve \varepsilon optimality gap on the multiclass logistic regression loss in a number of iterations T that is independent of \varepsilon . To do so, the step size \eta must be chosen large enough as a function of \varepsilon (and properties of the linearly separable dataset), while T is a constant that only depends on the data. The relevant properties of the data include the number of points and their margin (for T ), and one additional quantity (for \eta ) that turns out to be necessary in dimension d \geq 3 . The proofs build on the known fact that, as \eta \to \infty , suitably rescaled GD becomes the perceptron algorithm, and the latter finds a perfect classifier in T iterations. A similar claim was shown for two classes in dimension d = 2 using other proof techniques. This result adds to previous works that showed GD on the logistic regression loss can be accelerated by taking large step sizes. Those provoke a wild, oscillatory initial phase, followed by a stable phase. We provide results of this (qualitatively different) nature, too. Comments: 40 pages Subjects: Optimization and Control (math.OC); Machine Learning (cs.LG) MSC classes: 90C25 (Primary) 65K05, 68Q32, 68T05, 62H30 (Secondary) ACMclasses: G.1.6; I.2.6 Cite as: arXiv:2610.26748 [math.OC] (or arXiv:2610.26748v1 [math.OC] for this version) https://doi.org/10.48550/arXiv.2610.26748 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-226] Regime-Aware Interpretable Ranking for Sector Rotation
链接: https://arxiv.org/abs/2610.26456
作者: Kelvin J.L. Koa,Ziling Liu,Xinyang Li,Shuowei Feng,Ruoyu Liang,Mathieu Giroux,Eugene Y. Wen,Xuan Yao,Ke-Wei Huang,Hugo Lamarre
类目: Portfolio Management (q-fin.PM); Machine Learning (cs.LG); Computational Finance (q-fin.CP)
*备注: ICAIF 2026 (Oral)
Abstract:Sector rotation is a common decision in benchmark-relative portfolio management, where portfolio managers rank sectors for overweight and underweight positions. We formulate this task as ranking sector actions by expected future risk-adjusted active performance, measured with forward sector-benchmark Sharpe ratios. Such rankings should preserve temporal information, express model outputs through economically familiar signal characteristics, account for changes in signal relevance across macroeconomic environments, and capture dependencies among sectors. We propose RegimeRank, an interpretable sector-ranking model that represents fundamental and technical signals through expert-defined temporal characteristics, conditions them on learned macro factors via cross-attention, and models cross-sector dependencies through self-attention. Expected-gradient attributions identify the contributions of individual signal-transformation pairs, while macroeconomic and cross-sector attention maps provide diagnostics of the learned conditioning and relational structure. We evaluate RegimeRank on weekly SP 500 data from 2002 to 2024 under combinatorial purged cross-validation, where it outperforms representative ranking and machine-learning baselines across multiple ranking metrics. Ablation studies support the contributions of latent macro states, regime conditioning, and cross-sector modelling, while attribution and attention analyses illustrate the temporal characteristics, macro states, and sector relationships underlying its rankings.
[LG-227] R2-GP: Variance-Calibrated Priors for Additive Gaussian Processes
链接: https://arxiv.org/abs/2610.26437
作者: Javier Enrique Aguilar,Soham Mukherjee,David Kohns
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注:
Abstract:We develop R2-GP, an R2D2-type prior for Gaussian process (GP) generalised additive models (GAMs) that calibrates explained variance ( R^2 ) for Matérn covariance kernels. The function variance for standard Matérn kernels depends jointly on the effect magnitude and length-scale. As such, smooth and rough components with identical amplitudes contribute very different amounts of variation. With independent priors on component amplitudes, this variation results in variance overestimation, especially in higher dimensions. We overcome this by taking direct control over each component’s expected contribution to prior explained variance. We do this by rescaling each component based on its length scale and observed inputs, and modelling each rescaled variance as a share of the function variance induced by a global R^2 prior. This allows R2-GP to scale well to many components. To also scale to large sample sizes, we extend R2-GP to Hilbert space approximation methods, a framework for kernels exploiting their spectral decompositions for fast and efficient inference. Our simulations and real-world case studies demonstrate superior generalisation and parameter recovery over independent variance priors.
[LG-228] Selecting the Best Few with Human Evaluators: Adaptively Combining Ratings and Pairwise Comparisons
链接: https://arxiv.org/abs/2610.26380
作者: Motti Goldberger,Nils Rudi
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:This paper considers a decision-maker tasked with identifying the k best items from a large pool of candidates, such as job applicants, LLMs, award nominees, or new product ideas. The decision-maker learns about the items’ quality by soliciting human evaluations. Each evaluation can be costly, so the decision-maker aims to use the limited evaluation budget efficiently by allocating evaluations adaptively. A key consideration is the form of evaluation to solicit, which should be something evaluators are comfortable providing. Two popular forms are ratings and pairwise comparisons, and we allow for both, as they may be especially useful in combination given the limited budget. Ratings can be used early on to get a coarse picture of which items are good and which are not. As that picture forms, comparisons can refine the ordering of the good items, narrowing in on the ultimate top- k selection. To solve the allocation problem, we develop an adaptive algorithm built from two allocation policies. The first, our primary contribution, is based on a surrogate objective that is naturally forward-looking, valuing evaluations that help inform what to evaluate in future rounds, not just those that directly improve the selection decision. The decision rule is interpretable, and we prove that it is asymptotically consistent. Late in the evaluation process, when being forward-looking is less productive, we construct a myopic policy and a criterion for switching to it. In simulation, the method gives insight into how to combine these two types of evaluations, and beats the (limited) existing benchmarks.
[LG-229] A Pitfall in the Evaluation of Gene Perturbation Models
链接: https://arxiv.org/abs/2610.26267
作者: George Panagopoulos,Saman Omidi,Tsz Pan Tong,Jun Pang
类目: Applications (stat.AP); Machine Learning (cs.LG)
*备注:
Abstract:Gene perturbation models aim to predict transcriptional responses to unseen genetic interventions, but comparisons rely on heterogeneous evaluation metrics and differentially-expressed (DE) gene subsets. We investigate the consequences of these choices using eight methods across six single-gene perturbation datasets under a unified evaluation protocol. Varying the number of top DE genes reveals substantial instability in Precision-based rankings, with the winning method changing 86.7% of the time, while rankings based on the rest of the metrics remain comparatively stable. Investigating this discrepancy exposes that repeating a single predicted expression profile across cell-level rows, a standard practice for many methods, can inflate significance estimates and produce excessively large predicted significant-DE sets. Specifically we found that 67.1% of tested perturbation–gene pairs pass the significance threshold despite some of them having minuscule effects. We derive an inflation factor showing that replicated predictions can amplify the standardized Mann–Whitney statistic even after correction for tied ranks, without adding independent information. This artifact changes Precision mainly as the number of DE genes increases because for smaller cutoffs, most of the genes are already significant due to the effect-based ordering. Taking this pitfall into account in our benchmark comparison, we find that a random forest using Gene Ontology features achieves the best overall results.
[LG-230] Physics-Driven Framework for QoS-Constrained Satellite System with Non-Linear PA Distortion
链接: https://arxiv.org/abs/2610.26153
作者: Hamza Ahmed Qureshi,Abuzar Babikir Mohammad Adam,Vu Nguyen Ha,Konstantinos Ntontin,Symeon Chatzinotas
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注: Accepted at IEEE GLOBECOM 2026 Workshops (GC Wkshps)
Abstract:This work develops a physics-driven self-supervised graph learning framework that directly optimizes the distortion-aware precoder under nonlinear correlated power amplifier (PA) distortion, without relying on expensive optimizer-generated labels for a user-centric satellite system. We minimize nonlinear PA consumption subject to user quality-of-service (QoS) constraints under PA nonlinearity for a downlink multi-input-single-output (MISO) satellite system. Our pipeline jointly designs the precoder and admission control that minimize nonlinear PA consumed power. In particular, a graph neural network (GNN) for the design of a distortion-aware precoder is optimized end-to-end in a self-supervised manner using a differentiable Itô-Hermite model of the true modified Rapp amplifier model. The precoder output by the GNN is used to initialize a low-complexity distortion-aware signal-to-interference-distortion (SINDR) tracking algorithm that jointly optimizes for power and admission control. Our proposed physics-based pipeline closely tracks a successive convex approximation (SCA) based solution in consumed power performance with the advantage of faster runtime and a wide operating range covered during training.
[LG-231] Reinforcement Learning Controllers for Fusion Scenario Development
链接: https://arxiv.org/abs/2610.26138
作者: Evgeny Adishchev,Maxim Nurgaliev,Dmitry Sorokin,Alexander Granovsky,Maksim Stokolesov,Igor Prokofyev,Eduard Khairutdinov,Georgy Subbotin
类目: Plasma Physics (physics.plasm-ph); Machine Learning (cs.LG)
*备注:
Abstract:Scenario development, the design of the waveforms that take a tokamak discharge from breakdown to flat-top, is still an iterative, labour-intensive process. Every candidate scenario has to be simulated with a controller in the loop, and the classical shape and current controllers need low-level actuator references. So the process is iterative and includes a number of trial-and-error steps. We propose to replace the per-scenario controller tuning by a single goal-conditioned reinforcement-learning controller trained once, in advance, against a free-boundary equilibrium and transport simulator. A scenario is then specified only by the target waveforms of the physical quantities of interest (plasma current, line-averaged electron density, LCFS shape, x-point positions), and the controller produces the coil-current and fueling commands that realize them. We demonstrate the approach on an EAST plasma-current ramp-up: the controller takes over a small limited plasma, ramps the current and the density, grows the cross-section, forms two x-points and holds a double-null diverted configuration against sampled shape targets for the rest of a 1.2 s episode. The neural network controller is trained using TQC algorithm. On 1024 test scenarios sampled from a low-discrepancy scrambled Sobol sequence in 11-dimensional target space, the controller forms the divertor in 86% of the cases, holds it for more than 0.5 s in 67% and reaches the end of the episode in a diverted state in 58%, with flat-top tracking errors of a few kA in plasma current, \sim 0.15\times10^19 m ^-3 in density, \sim 2 cm in position and \sim 8 cm in x-point location; the failures concentrate at the high-current density ( J ) region which is believed to be physically or from engineering perspective not feasible within ohmic mode.
[LG-232] owards instance-dependent regret optimality in Episodic MDPs with Posterior Sampling
链接: https://arxiv.org/abs/2610.26029
作者: Victor Boone,Cyrille Kone,Waris Radji,Odalric-Ambrym Maillard,Dorian Baudry
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We study regret minimization in finite-horizon episodic Markov Decision Processes (MDPs). While minimax-optimal algorithms are known, tractable approaches to instance-dependent optimality are still lacking. Motivated by this gap, we introduce \pi_0 -PSRL, a variant of PSRL that uses posterior samples to decide when exploration is needed, while following a fixed reference policy \pi_0 during exploration episodes. This decouples the test for exploration, triggered when the sampled and empirical MDPs have different optimal policies, from the choice of the policy used to gather information. The resulting design addresses a limitation of standard PSRL, where the sampled optimal policy may not be the optimal choice for exploration. We prove instance-dependent regret bounds for \pi_0 -PSRL, identifying the logarithmic exploration cost induced by \pi_0 and taking a step toward matching asymptotic lower bounds for episodic RL. Our proof techniques showcase a novel proof structure to derive problem-dependent regret bounds in episodic MDPs, and concentration results for Dirichlet random variables, that may be of independent interest.
[LG-233] Federated Community Detection in Bipartite Networks
链接: https://arxiv.org/abs/2610.25997
作者: Wanjie Wang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO); Methodology (stat.ME)
*备注: 47 pages, 10 figures, 6 tables
Abstract:We study federated spectral estimation when a large matrix is horizontally partitioned across servers that cannot share raw data. We propose FLICKER, a target-specific gradient-based power iteration that recovers a singular direction associated with a prescribed spectral target using only matrix-vector products on local blocks. Different target directions can be recovered independently and in parallel. We establish non-asymptotic convergence guarantees and provide a recursive grid-search procedure for spectral target selection. Building on FLICKER, we develop BiFLICKER for federated community detection in bipartite networks, together with a sequential projected variant for nearby or repeated singular directions. Under a sparse bipartite stochastic block model, we show that properly rescaled leading singular values from a reference server provide valid global targets, and that Sequential BiFLICKER achieves a vanishing misclustering rate and exact recovery under stronger sparsity conditions. Simulations show that BiFLICKER remains close to centralized spectral clustering and is more robust than existing federated spectral baselines when local sample sizes are small. On a LastFM user-artist network naturally partitioned by country, Sequential BiFLICKER attains an overall ARI of 0.94 against centralized clustering while revealing cross-country communities with distinct listening profiles.
[LG-234] Fair Regression under Approximate Demographic Parity
链接: https://arxiv.org/abs/2610.25796
作者: Weiwei Wang,Xianyang Zhang,Raymond K. W. Wong
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Exact demographic parity requires the same prediction distribution across sensitive groups. In regression, this can be too restrictive and reduce predictive accuracy. We instead bound every pairwise Wasserstein distance between group-specific prediction distributions by a chosen tolerance, allowing the groups to differ while keeping fairness tunable. On the real line, the prediction and constrained transport problems have the same infimum. If the transport problem attains this value and suitable Monge maps exist, a deterministic transport-form predictor realizes it. Under a standard one-dimensional condition, the problem becomes a bounded quantile optimization that is convex when the conditional expected loss is convex in prediction. Under squared loss with two positively weighted groups, the W_1 , W_2 , and W_\infty constraints yield soft-thresholding, shrinkage, and clipping. Under the stated conditions, we establish rates for the known-cost plug-in estimator. When the cost is unavailable, we describe an observed-loss computational proxy. Synthetic and California 2018 ACS studies illustrate the tunable fairness-accuracy trade-off and distinguish calibration feasibility from held-out fairness and accuracy. Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG) Cite as: arXiv:2610.25796 [stat.ML] (or arXiv:2610.25796v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.25796 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-235] Arbitrag e-Free Implied Volatility Surface Reconstruction via Additive Martingales
链接: https://arxiv.org/abs/2610.25671
作者: Haochen Luo,Sifan Lin,Xiaolong Luo,An Binh Minh,Yuan Zhang,Chen Liu
类目: Computational Finance (q-fin.CP); Machine Learning (cs.LG)
*备注: 41 pages, 15 figures. Includes supplementary proofs and experimental analyses in the appendices
Abstract:This work focuses on reconstructing a continuous implied volatility (IV) surface from a small number of option quotes, a fundamental task for option pricing and risk management. We propose Additive Martingale (AddMart), a few-shot IV surface reconstruction framework that satisfies static no-arbitrage conditions by construction. Rather than predicting IVs directly, AddMart maps sparse quotes to non-negative increment densities of an exponential additive martingale, from which valid risk-neutral marginals and IVs are recovered. By incorporating the no-arbitrage structure directly into its architecture, rather than enforcing it through regularization in the learning objective, AddMart produces IV surfaces that satisfy static no-arbitrage conditions throughout the domain. Extensive experiments on ten real-option datasets demonstrate that AddMart achieves competitive reconstruction accuracy with no material static arbitrage violations, while directly enabling the estimation of VaR, CVaR, and crash probabilities.
[LG-236] Sampling of Signals on Multigraphs: From Removable and Uniqueness sets to Spectral Proxies
链接: https://arxiv.org/abs/2610.25643
作者: Alejandro Parada-Mayorga
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注:
Abstract:In this paper, we study the sampling of bandlimited signals on multigraphs, where several graphs with different shift operators are defined on the same set of nodes. Since the shift operators do not commute in general, they admit only a joint block diagonalization, and the frequencies of the multigraph signal model are matrices rather than scalars. Relying on a formal algebraic convolutional signal model, we introduce a notion of frequency and bandwidth for multigraph signals that does not depend on any individual layer, and we extend the notions of removable and uniqueness sets to multigraphs. We prove that the optimality of a sampling set is determined exclusively by the common Fourier basis and its invariant subspaces, and not by any individual layer, although the individual layers provide computable sufficient conditions that certify a sampling set as a uniqueness set. Since identifying which layers provide reliable certificates requires spectral information, we propose an aggregate operator that pools the information of all the layers, is computable directly from the shift operators, and certifies uniqueness without any assumption on the individual layers. This operator leads to a greedy sampling algorithm that does not require computing the joint block diagonalization. Numerical experiments on multigraphs whose layers order the frequencies inconsistently show that the proposed algorithm selects better-conditioned sampling sets and attains lower reconstruction errors than single-layer methods.
[LG-237] Uncertainty-Gated SHAP-Dirichlet Refinement for MRI-Based Depression Classification
链接: https://arxiv.org/abs/2610.25635
作者: Phuc Truong Loc Nguyen,Nguyen Truong Toan To,Truong Thanh Hung Nguyen,Hoang-Loc Cao,Hung Cao
类目: Image and Video Processing (eess.IV); Machine Learning (cs.LG)
*备注:
Abstract:Structural MRI-based classification of major depressive disorder (MDD) remains challenging due to subtle, heterogeneous anatomical changes. We propose uncertainty-gated SHAP-Dirichlet refinement, a post-hoc reliability layer that improves predictions without retraining the classifier. The method retrieves similar labeled SHAP explanations, constructs Dirichlet class evidence, and selectively refines uncertain predictions through probabilistic fusion, while allowing unreliable cases to be deferred for expert review. We evaluate five classifiers on 504 SRPBS subjects (252 controls and 252 MDD patients) using 159 anatomical features and repeated nested cross-validation. Refinement improved accuracy across all 8 repetitions for TabM and ResNet, yielding mean gains of 1.04 and 0.45 percentage points (sign test p=0.008), while modifying at most 8.3% of predictions. Gains were concentrated among low-confidence cases, with significant improvements in calibration. The method also benefited an RBF support vector machine but not logistic regression or random forests. Ablations confirmed the importance of uncertainty gating and soft fusion. Dirichlet evidence identified baseline errors (AUROC 0.62), while deferring 50% of cases increased retained-case accuracy to approximately 75%. These findings support selective explanation-guided refinement as a practical approach to improving prediction reliability and facilitating expert triage in MDD classification.
[LG-238] Concurrent two-scale topology optimization of printable bone scaffolds via rectified-flow generation and exact latent-space refinement
链接: https://arxiv.org/abs/2610.25565
作者: Reza T Batley,Sourav Saha
类目: Computational Physics (physics.comp-ph); Machine Learning (cs.LG)
*备注:
Abstract:Bone scaffolds must reconcile a macroscale mechanical response with microscale geometric feasibility, yet conventional approaches to their design restrict attainable constitutive properties. This work presents a concurrent two-scale topology optimization framework; a conditional rectified flow generates periodic lattice microstructures, which are then refined in latent space by exact finite-element homogenization sensitivities. The resulting unit-cell geometries and their spatial assignments are jointly optimized under mechanical and feature-size constraints. In a distal rabbit femur benchmark, all eight flow-started refinements produce printable cells with maximum relative errors under 1% across six prescribed engineering moduli. Constructing the implant from a single tiled cell reproduces intact whole-femur compliance within 0.1% and regional strain energy within 0.35% under four loading conditions. In human femoral reconstruction, multiscale concurrent optimization produces a cortical mid-shaft scaffold with porosity 63% while satisfying the compliance constraint within 2% under walking and stair-climbing loads. It achieves an 84% porosity on a proximal cancellous scaffold with strain-energy deviations below 0.25% under the same loads. Analysis of the learned representations reveals distinct architectural families in latent space. Further, the attention heads of the rectified-flow transformer specialize for spatial shifts – learning the analogue of a finite-difference stencil – and coordinate symmetries, with ablations showing their influence on generated topologies. These results establish a route to printable, mechanically constrained scaffold designs. The framework is general enough for extending to many other multiscale structural design applications including topology-microstructure co-design in metal additive manufacturing.
[LG-239] Estimating Causal Bounds from Observational Data With Neural Generative Models
链接: https://arxiv.org/abs/2610.25246
作者: Max Zhu,Srivathsan Amruth,Sina Akbari,Qingyuan Zhao
类目: Methodology (stat.ME); Machine Learning (cs.LG)
*备注: 23 pages, 3 figures, 3 tables
Abstract:Identifying causal bounds from discrete observational data is a fundamental problem in causal inference. While exact analytical bounds exist for certain graphs, this is not generally the case. In this paper, we introduce a framework for estimating such bounds using neural generative models together with a new extrapolation procedure to estimate these bounds more accurately. Across multiple causal inference settings, it produces bounds that are consistently closer to the true bounds than those obtained with existing general neural approaches. We validate the framework on a range of benchmark problems for which exact bounds are available.
[LG-240] LiftDiff: Guided Discrete Diffusion for Quantum Code Construction
链接: https://arxiv.org/abs/2610.25213
作者: Yotam Peled,Eliya Nachmani
类目: Quantum Physics (quant-ph); Information Theory (cs.IT); Machine Learning (cs.LG)
*备注:
Abstract:The realization of fault-tolerant quantum computation relies on Quantum Error-Correcting Codes (QECC). Of particular interest are codes that combine a high encoding rate with sufficient distance. Existing algebraic constructions, such as lifted product (LP) codes, build a large quantum low-density parity-check (qLDPC) code from a classical base matrix. In this work we introduce LiftDiff, a generative approach to the discovery of QECC. A discrete denoising diffusion model is trained to generate base matrices over the group algebra \mathbbF_2[\mathbbZ_\ell] , which are then expanded into LP codes. The reverse diffusion process is guided by an energy function that evaluates quantum code properties, including girth, degree, sparsity and decoding performance, steering the sampler toward matrices that yield LP codes with favorable parameters. Our method discovers codes with parameters such as [[169,13,11]] , [[143,15,9]] , and [[260,24,10]] , including a [[247,19,14]] code that achieves the highest kd^2/n among the baselines. Moreover, several discovered codes outperform the Gross code across a broad range of the tested physical error rates under both code-capacity and circuit-level noise.
[LG-241] RAIL-5G: A Challenging Real-World Dataset and Benchmark for AI-Based 5G Indoor Localization
链接: https://arxiv.org/abs/2610.25201
作者: Jonas Pirkl,Andreas Eidloth,Maximilian Kasparek,Tobias Feigl,Christopher Mutschler
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注: To be published in the proceedings of the 16th International Conference on Indoor Positioning and Indoor Navigation (IPIN)
Abstract:Reliable radio-based indoor localization is essential for industrial automation, yet existing methods are often evaluated in environments that do not fully reflect the severe multipath and non-line-of-sight (NLOS) conditions encountered in real deployments. This limits our understanding of how AI, i.e., modern geometric, unsupervised, and semi-supervised localization methods generalize to industrial 5G settings. We present RAIL-5G, a public real-world 5G NR channel impulse response (CIR) benchmark recorded in an operational railway maintenance facility. RAIL-5G covers an 811m ^2 indoor area with eight distributed radio units, robot-mounted user equipment (UE) trajectories, high-accuracy iGPS reference positions, and severe metal-induced multipath caused by train compartments and maintenance infrastructure. We benchmark geometric time-difference of arrival (TDoA), supervised fingerprinting, unsupervised channel charting, and self-supervised wireless pretraining methods. Our results highlight the difficulty of RAIL-5G: on the full dataset, TDoA reaches 7.02m mean average error (MAE), the best unsupervised method reaches 2.46m, and supervised fingerprinting achieves 0.52m. Self-supervised pretraining reduces the labeling effort by up to one order of magnitude, but the remaining gap to supervised learning shows that robust AI-based localization under severe multipath remains an open challenge. The dataset and code are available at this https URL.
[LG-242] ChronoChart: Setup-Agnostic Channel Charting for Label-Free Indoor Localization
链接: https://arxiv.org/abs/2610.25199
作者: Majid Sharghi Foroushani,Jonas Pirkl,Jonathan Ott,Maximilian Stahlke,George Yammine,Tobias Feigl,Christopher Mutschler
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注: To be published in the proceedings of the 16th International Conference on Indoor Positioning and Indoor Navigation (IPIN)
Abstract:Accurate radio-based indoor localization remains challenging in multipath-rich and dynamic indoor environments. Supervised fingerprinting methods achieve high positioning accuracy by learning mappings from radio channel measurements to physical locations, but require costly position labels and frequent recalibration. Channel Charting (CC) addresses this limitation by learning low-dimensional representations of Channel State Information (CSI) without position labels, yielding channel charts in which nearby points correspond to nearby physical locations. However, prominent CC methods rely on setup-specific dissimilarity metrics, such as ADP-, CIRA-, or PDR-based distances, and on precomputed O(n^2) distance matrices, limiting scalability and transferability across radio systems. We propose ChronoChart, a timestamp-supervised CC framework that requires only CSI and timestamps without prior knowledge for training. Our method jointly optimizes complementary objectives: a triplet loss using temporally derived positive and negative samples to establish coarse global structure, a temporal-proximity loss that promotes smooth trajectories in the chart, and a joint-embedding loss that enforces consistency between representations of different augmentations. Evaluated on real-world 5G, MIMO, and passive UWB data, our method matches or exceeds prior CC baselines that depend on tailored distance metrics, and remains scalable and applicable across heterogeneous radio systems.
[LG-243] Decision-Target Reversal under Informative Label Missingness
链接: https://arxiv.org/abs/2610.25128
作者: Farinborz Setoudehtazangi,Geoffrey J. McLachlan
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Informative label missingness can improve classification efficiency because the pattern of label availability may itself contain information about the discriminant structure. We show that this advantage need not persist across deployment objectives: with the data-generating model and statistical experiments held fixed, informative partial classification can be favourable under one loss and unfavourable under another. We call this phenomenon decision-target reversal. For binary Gaussian discrimination, we derive a local cost-sensitive excess-risk expansion and, in one dimension, obtain explicit critical cost ratios at which the first-order risk ordering reverses. We show that the required covariance geometry arises under finite uncertainty-dependent missingness mechanisms with a fixed marginal missing-label proportion. We then extend the analysis to multiclass finite-action problems through loss-specific active-boundary curvature, yielding a general decision-target frontier. Monte Carlo experiments illustrate the predicted reversals, and a descriptive gastrointestinal-lesion application with agreement-defined label availability illustrates sensitivity to deployment costs. The results show that favourable informative missingness is a property of the statistical experiment relative to the deployment decision, rather than of the experiment alone.
[LG-244] Non-Acceleration of the Heavy-Ball Method in 1D
链接: https://arxiv.org/abs/2610.25104
作者: Konstantinos Fotopoulos,Michael Helcig,Karl Deck,Jannis Alsbach
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Despite having an asymptotically optimal rate of convergence on convex quadratics, the acceleration of Polyak’s heavy-ball method and a characterization of its dynamics within more general settings have remained long standing open problems. Recently, the work of Goujaud et al. [Math. Program. (2025)] made progress regarding acceleration in the context of smooth, strongly convex minimization: for a small enough inverse condition number, either the method does not accelerate on quadratics, or there exists a smooth strongly convex function such that heavy-ball cycles over the roots of unity. The constructed functions have two variables, thus leaving the question of acceleration over univariate functions open. In this work, we use a Fourier-based analysis to i) provide a proof for the non-acceleration of the heavy-ball method over the class of univariate smooth strongly convex functions. We also ii) provide a strict enlargement of the previous root-of-unity cycling region, with insights regarding the “fractional” sets of Goujaud et al. [arXiv:2502.19916], and iii) give a higher dimensional cycling instance which is not covered by root-of-unity cycling.
[LG-245] Normalizing Flow for Neural-Enhanced Track-Before-Detect
链接: https://arxiv.org/abs/2610.25044
作者: Shaoxiu Wei,Bhaskar Rao,Florian Meyer
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注:
Abstract:Multiobject tracking (MOT) is a fundamental capability in autonomous navigation, radar and sonar systems, as well as 6G integrated sensing-and-communication (ISAC) networks. In this paper, we propose a neural-enhanced track-before-detect (TkBD) framework that operates directly on complex-valued raw signals and learns a latent representation of unmodeled propagation effects during runtime. We introduce an environment-dependent latent intensity variable that characterizes the signal power of each object in the shared propagation environment. The probability density function of this intensity variable is represented by a normalizing flow and trained online in an unsupervised manner within the sequential Bayesian inference procedure. The learning objective is derived from a variational Bayes formulation by maximizing the evidence lower bound (ELBO), thereby accounting for the uncertainties inherent in the Bayesian TkBD problem. To enable scalable inference, we make use of a recently introduced Bayesian TkBD framework that combines belief propagation, which avoids high-dimensional operations, with sequential Monte Carlo methods for efficient low-dimensional updates.
[LG-246] Scalable and Adaptive Distribution Regression via Random Features
链接: https://arxiv.org/abs/2610.24892
作者: Yinsong Wang,Xiao Liu,Yu Ding
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Distribution regression seeks to learn mappings from probability distributions to response variables and has emerged as a unifying framework for kernel methods, multi-instance learning, and set-valued learning. Despite its success, kernel-based distribution regression remains computationally expensive and sensitive to kernel choice, motivating scalable and adaptive representations. Building on random feature approximations, we develop a finite-sample analysis of Random Feature Distribution Regression (RaDR) under two-stage sampling. Our perturbation analysis separates the effects of distribution-level sampling, finite within-bag sampling, and random feature approximation. We establish sufficient feature-budget conditions under which RaDR preserves the order of our excess-risk upper bound for exact two-stage kernel distribution regression while maintaining computational efficiency. We further propose Adaptive Random Feature Distribution Regression (A-RaDR), a unified score-guided framework that adaptively allocates random features according to downstream learning objectives. Experiments on synthetic benchmarks and 41 real-world datasets demonstrate that RaDR delivers substantial computational savings while maintaining competitive predictive performance, whereas A-RaDR further improves learning accuracy in distribution regression while retaining the computational speed up. Together, these results establish a practical and theoretically grounded framework for adaptive, scalable distribution learning
[LG-247] d Mondrian: Triage and Mitigation of Fine-grained Disparities in Conformal Prediction
链接: https://arxiv.org/abs/2610.24813
作者: Ziang Gao,Pengqi Liu,Archer Yi Yang,Mouloud Belbahri,Jesse C. Cresswell,Masoud Asgharian
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Fairness in conformal prediction is often pursued through Mondrian calibration, which provides finite-sample coverage guarantees separately across groups. These guarantees, however, are coarse: substantial disparities can remain within finer strata. This coarse-to-fine fairness issue raises two questions: how severe can such disparities be, and how can they be mitigated? Aggregate disparities can depend on how groups are distributed across strata. For severity assessment, we introduce propensity standardization to examine the effect of population composition, and show how group composition can induce disagreement between aggregate and finer-stratum comparisons. For mitigation, we develop Tilted Mondrian to improve fairness compared to group-Mondrian at a finer level without additional model training. Tilted Mondrian retains groupwise coverage guarantees and mitigates coverage and prediction set size disparities within finer strata. Finally, we derive a method-agnostic bound that quantifies how much reduction in finer-stratum coverage disparity can be achieved.
[LG-248] Martingale Posterior Cox Processes
链接: https://arxiv.org/abs/2610.24787
作者: Joshua Rooijakkers,Edwin Fong
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注:
Abstract:Cox processes provide a principled Bayesian nonparametric framework for modeling event-based data through an underlying intensity function. However, existing inference methods for estimating the intensity typically rely on computationally expensive Markov chain Monte Carlo sampling or approximate inference techniques. In addition, to ensure tractability, these methods often define priors over indirect latent structures, making them challenging to interpret. In this paper, we introduce the Martingale Posterior Cox Process (MPCP), which recasts Bayesian intensity estimation as inference over unobserved events given the observed event history. This perspective avoids the need to define complex priors while retaining the flexibility of nonparametric modeling. We propose a scalable, GPU-friendly copula-based predictive update, and provide strong theoretical guarantees for the convergence of the predictive resampling algorithm. Empirically, we show that MPCP provides excellent performance in both simulations and real data at a fraction of the computational cost of competing Bayesian approaches.
[LG-249] Conditional Weight of Evidence for Gradient-Boosted Credit Models: Estimation Additivity and Its Relation to Adverse Action Reason s
链接: https://arxiv.org/abs/2610.24586
作者: Agus Sudjianto,Denis Burakov
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Risk Management (q-fin.RM); Methodology (stat.ME)
*备注: 41 pages, 3 figures, 12 tables. Code: this https URL
Abstract:Traditional weight of evidence assigns a credit characteristic a value based on its association with default across a population. Credit analysis also requires an assessment of that characteristic in the context of an individual applicant’s record. We study conditional weight of evidence for this purpose. It measures the change in default log odds from observing a characteristic after the remaining credit information is known. Its sign identifies whether the characteristic adds adverse or favorable information for that applicant, and its magnitude measures the strength of that information. The same characteristic value can therefore receive different assessments for applicants with different payment histories or other credit attributes. We develop estimators using gradient-boosted trees and evaluate their accuracy in a bureau-style simulation. Regressions of the full model’s predicted probabilities recover conditional evidence more accurately than separate models fitted to default outcomes, provided the full model is accurate. They also recover a deliberately constructed file-age effect whose direction varies with utilization and is often missed by marginal weight of evidence. Local class-frequency estimates improve on marginal weight of evidence but remain inaccurate for some weak effects. An application to FICO HELOC data illustrates estimation when the population evidence is unknown. Conditional weight of evidence provides an applicant-specific interpretation of the information in a credit characteristic; its population summary, a conditional information value, splits the joint information in the record across blocks when averaged over orderings. Neither is an adverse action reason, which explains the fitted score difference between the applicant and chosen passing records.
[LG-250] Accelerating Finite-Sum Bilevel Optimization with Fully First-Order Oracles
链接: https://arxiv.org/abs/2610.24538
作者: Qihao Zhou,Shanshui Niu
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 23 pages, 2 figures, 1 table
Abstract:We study nonconvex–strongly-convex bilevel optimization with finite-sum objectives and fully first-order oracles. We propose Sampled Implicit Corrections for Bilevel Optimization (SICBO), which combines a penalty approximation with variance-reduced corrections for the lower-level variables. Under componentwise lower strong convexity, SICBO finds a point x satisfying \mathbbE|\nabla\Phi(x)|^2\le\varepsilon^2 within \tildeO(N+\sqrtN\kappa^7/2\varepsilon^-2) component-gradient calls, where N is the total number of components and \kappa is the normalized lower-level condition number. The bound achieves square-root finite-sum dependence using only gradients and preserves the best known deterministic fully first-order condition-number exponent for smooth problems. Our improvement follows from a curvature-dependent bound for nonlinear correction errors, which permits epochs of length \tilde\Theta(\sqrtN/\kappa) with batches of size \tilde\Theta(\sqrtN\kappa) . A single-level reduction shows that the dependence on N and \varepsilon is optimal up to logarithmic factors for fixed \kappa .
[LG-251] Reconstruction of Unobserved Core Electron Density in Tokamak Plasmas with a Neural Operator Transport Surrogate and Uncertainty Quantification
链接: https://arxiv.org/abs/2610.24525
作者: D. Sorokin,A. Granovskiy,M. Stokolesov,E. Adishchev,M. Nurgaliev,G. Subbotin
类目: Plasma Physics (physics.plasm-ph); Machine Learning (cs.LG)
*备注:
Abstract:Real-time control of tokamak plasmas requires the electron density profile. However, in a reactor-relevant device, diagnostic port access is limited and the profile can be observed only partially. In the present work we reconstruct the density profile by inferring the coefficients of a physics-based one-dimensional particle transport equation, using a Fourier Neural Operator (FNO) surrogate of the transport solver as a fast differentiable forward model. The surrogate enables realtime-compatible gradient-based reconstruction and uncertainty quantification from the reflectometry and interferometry measurements. We compare four uncertainty quantification methods - multi-seed ensembles, the Laplace approximation, Metropolis-Hastings, and a Gaussian method independent of the reconstruction fit. The resulting uncertainty is scored with CRPS and coverage calibration, separately on the observed and unobserved regions. We further demonstrate advantage of our approach over standard analytical function fit on synthetic and experimental DIII-D discharges.
[LG-252] Denoising Implicit Variational Inference NEURIPS2026
链接: https://arxiv.org/abs/2610.24509
作者: Zitong Wang,Zekun Wu,Longlin Yu,Cheng Zhang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: Neurips 2026 camera ready
Abstract:Semi-implicit variational inference (SIVI) expands the representational capacity of variational families, but optimizing the Evidence Lower Bound (ELBO) remains challenging because the marginal score of the semi-implicit distribution is generally intractable. Unbiased Implicit Variational Inference (UIVI) and related methods address this difficulty through reparameterized ELBO gradients, but existing approaches estimate the required score indirectly through reverse-conditional sampling or approximation, which can be computationally costly and unstable. We propose Denoising Implicit Variational Inference (DIVI), which learns the marginal score directly via denoising score matching. The learned score is used as a plug-in estimator in the pathwise ELBO gradient, replacing inner reverse-conditional sampling with a score-network evaluation. We analyze DIVI as inexact stochastic gradient ascent and show, under stated assumptions, that the averaged stationarity measure is controlled by the usual optimization term and the average score-matching error. Empirically, we evaluate DIVI on both synthetic distributions and a variety of real-data Bayesian inference tasks. The results show that DIVI improves over the current UIVI baselines while reducing the cost of ELBO-gradient estimation.
[LG-253] Private Evaluation of High-Dimensional Functions
链接: https://arxiv.org/abs/2610.24477
作者: Mohammad Afzali,Hassan Ashtiani,Adam Smith
类目: Machine Learning (stat.ML); Cryptography and Security (cs.CR); Data Structures and Algorithms (cs.DS); Information Theory (cs.IT); Machine Learning (cs.LG)
*备注:
Abstract:We study differentially private black-box evaluation of arbitrary high-dimensional functions. Given a dataset S and query access to a function f —whose sensitivity may be unknown or unbounded—the goal is to privately output an approximation to f(S) . The black-box evaluator can only query the value of f over a down neighborhood of S (i.e., those subsets of S that are obtained by removing a bounded number of elements from S ). We design generic black-box evaluators for both pure and approximate differential privacy through a reduction to high-dimensional private quasi-concave optimization (QCO). Our black-box evaluator enjoys a geometric notion of utility. Consider the vectors obtained by applying f on sets in a down neighborhood of S . With high probability, the evaluator’s output is guaranteed to be in the convex hull of these vectors. Our work generalizes the result of Linder et al. [LRSS25] for real-valued function evaluation to high dimensions. It also recovers the private center-point guarantees of Beimel et al. [BMNS19], with an explicit additive approximation to maximum Tukey depth. We also apply our evaluators to user-level quasi-concave and convex optimization, showing how they can improve upon the sample-and-aggregate baseline of Nissim et al. [NRS07] by replacing the multiplicative data overhead with an additive overhead. Finally, we show that our formulation of high dimensional black-box evaluation and high dimensional private QCO are equivalent up to a factor of the dimension: the hardest private evaluation instances already arise from the cases that f is a (non-private) quasi-concave optimizer. Subjects: Machine Learning (stat.ML); Cryptography and Security (cs.CR); Data Structures and Algorithms (cs.DS); Information Theory (cs.IT); Machine Learning (cs.LG) Cite as: arXiv:2610.24477 [stat.ML] (or arXiv:2610.24477v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.24477 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Mohammad Afzali [view email] [v1] Sat, 17 Oct 2026 05:46:38 UTC (45 KB)
[LG-254] Margin-Based Generalization for Quantum Machine Learning with Classical Data Encoding
链接: https://arxiv.org/abs/2610.24426
作者: Ju-Young Ryu,Tak Hur,Daniel K. Park
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注: 22 pages, 6 figures, 7 tables
Abstract:Understanding generalization is essential for assessing the practical viability of quantum machine learning. In many applications, quantum models operate on classical data that must first be encoded into quantum states, making the encoding part of the effective hypothesis class. We develop a margin-based generalization framework for this setting, covering both fixed and trainable quantum data encodings. For fixed encodings, the existing margin bound transfers unchanged. For trainable encodings, we derive a bound using a two-scale covering argument that combines the original margin-dependent term with an additive encoding-capacity term of order \tilde O(\sqrtL/m) , where L is the number of trainable encoding parameters and m is the training-sample size. Across 19,440 simulated training runs spanning various image datasets, circuit architectures, and optimization settings, the inverse margin performs best among five baseline and two bound-derived complexity measures under conditional-independence and 90th-percentile sign-error criteria, within bootstrap uncertainty. These results provide a unified margin-based view of generalization in quantum machine learning with classical data.
[LG-255] Prototype-Guided Optimization of Quantum Data Embeddings with Linear Cost
链接: https://arxiv.org/abs/2610.24403
作者: Seokhoon Jeong,Daniel K. Park
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注: 26 pages, 5 figures, 5 tables
Abstract:In supervised quantum machine learning with classical data, classification performance critically depends on the geometry induced by quantum data embedding. Existing data-adaptive approaches rely on pairwise fidelity evaluations between samples, which scale as O(B^2) for B samples and create a computational bottleneck in iterative embedding optimization. To address this issue, we propose a prototype-guided framework for efficient data-adaptive quantum embedding. Using class-prototype separation and sample-to-prototype compactness, we derive a measurable lower certificate for class-ensemble distinguishability and construct a prototype-based objective requiring only O(B) fidelity evaluations. We integrate this objective into an end-to-end pipeline that combines task-specific quantum circuit architecture optimization with continuous refinement of the data-embedding parameters. Across five benchmark datasets with up to 20 qubits, our method retains competitive quantum kernel support vector machine (SVM) performance while reducing fidelity-evaluation cost, with particularly strong gains over budget-matched pairwise optimization on T-MNIST. Geometric analyses examine how the learned embeddings relate to kernel alignment and classification margins, revealing dataset-dependent benefits and limitations. We further derive a prototype-based classifier in the reproducing kernel Hilbert space that provides an explicit witness for the SVM objective and a low-cost proxy for model selection, and extend the framework to multiple prototypes for multimodal class structures.
[LG-256] CoCoNet: Convex Conformal Network for Multi-Dimensional Uncertainty Regions
链接: https://arxiv.org/abs/2610.24246
作者: Alireza Moradi,Thomas Bruys,Pascal Van Hentenryck,Reza Zandehshahvar
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Conformal prediction (CP) provides distribution-free coverage guarantees under exchangeability, but the efficiency of its prediction regions depends critically on the choice of the conformity score. This is particularly important in multivariate regression, where dependencies across output dimensions induce complex joint uncertainty structures. This paper introduces CoCoNet (Convex Conformal Network) for learning convex conformity scores that induce flexible, data-adaptive prediction regions. CoCoNet learns the score from held-out data by optimizing prediction-region efficiency while enforcing convexity with respect to the residual. CoCoNet then applies split-conformal calibration to retain finite-sample coverage guarantees. The paper presents two variants of the method: CoCoNet _\mathrmSpline , based on piecewise-linear scores, and CoCoNet _\mathrmICNN , based on Input Convex Neural Networks. Experiments compare CoCoNet with other CP baselines over synthetic and real-world datasets with output dimensions ranging from 2 to 16. The results demonstrate the flexibility of CoCoNet in learning different response types. Across 8 models and 30 multi-dimensional regression tasks, CoCoNet _\textICNN and CoCoNet _\mathrmSpline\text-\ell_1 achieve the best efficiencies with average ranks of 1.45 and 2.75, respectively. In particular, CoCoNet _\textICNN ranks first on 22 tasks out of 30 in terms of efficiency.
[LG-257] Censored Flow Matching for Synthetic Time-to-Event Generation
链接: https://arxiv.org/abs/2610.24233
作者: Jianbin Tan,Xingran Chen,Kevin He,Zhenke Wu,Bhramar Mukherjee,Anru R. Zhang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO); Methodology (stat.ME)
*备注:
Abstract:Generating time-to-event data is increasingly important in survival analysis, but it remains statistically challenging because event times are often censored and thus prevent flexible generative modeling. To fill this gap, we propose a novel method called censored flow matching (CFM), aiming at providing a flexible flow-based generative mechanism for synthesizing time-to-event data as well as learning their corresponding probabilistic densities. Unlike existing methods, CFM does not rely on survival likelihoods to handle censoring, but instead exploits the monotonicity property of differential flows to address censoring via pullback sampling. Under this framework, we implement an alternating procedure between pullback sampling and CFM steps, leading to a flow-based expectation–maximization (EM) algorithm tailored for time-to-event inference. Under mild distributional conditions, we establish the convergence of CFM to the true event-time distribution as the EM iterations proceed. We then apply CFM to generate time-to-event data and estimate their probabilistic densities, where the observed censoring time can be incorporated to provide truncated sampling or truncated density estimation for survival inference. Through extensive simulation studies, we demonstrate the benefits of CFM for complex and large-scale data generation. Finally, we apply CFM to analyze data from the Organ Procurement and Transplantation Network (OPTN) and the Medical Information Mart for Intensive Care IV (MIMIC-IV), showcasing its benefits for synthetic and predictive survival analyses in terms of data fidelity, predictive performance, and empirical privacy characteristics.
[LG-258] Sequence-to-Sequence Neural Surrogate Modeling of Transient Responses in High-Speed Links with Variable Design Parameters
链接: https://arxiv.org/abs/2610.24141
作者: Julian Withöft,Shayan Ghanbari,Werner John,Emre Ecik,Ralf Brüning,Jürgen Götze
类目: ignal Processing (eess.SP); Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注: 12 pages, 13 figures, 5 tables. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible
Abstract:This paper presents sequence-to-sequence (seq2seq) neural surrogates for high-speed transient response modeling and signal integrity analysis in fixed and variable design parameter spaces. To identify the structural requirements for prediction accuracy, multiple neural architectures were evaluated with respect to different training strategies such as sample-based and unit interval-based, as well as input configurations such as TX-only and TX+RX. For fixed design parameters, feedforward neural networks achieve accuracy comparable to recurrent models while requiring lower computational cost, as the static mapping can be approximated by fully connected and non-sequential representations. Variable design parameters alter the electromagnetic characteristics of the interconnect, which creates parameter-dependent temporal dynamics that require sequence-aware architectures to resolve the evolving interactions within the transient response. The results based on a 14-dimensional variable design parameter scenario show that while autoregressive TX+RX feedback suffers from severe phase drift and error propagation under variable conditions, non-autoregressive TX-only inference leads to stability. To overcome the structural limitations of standard decoders, a recurrent Early Fusion LSTM is proposed, which injects design parameters at the initial encoding stage. This approach achieves superior prediction accuracy, with an RMSE of 31.0 mV and an R2 of 99.51% for transient waveform prediction, an RMSE of 30.7 mV and an R2 of 99.57% for eye-height prediction, and an RMSE of 19.5 ps and an R2 of 99.94% for eye-width prediction. These results demonstrate that neural seq2seq surrogates are accurate for transient response modeling and signal integrity analysis in complex variable design parameter spaces.
[LG-259] Reducing double-dipping in single-cell topic models with three-way count splitting
链接: https://arxiv.org/abs/2610.24130
作者: Zheng Qu
类目: Genomics (q-bio.GN); Machine Learning (cs.LG)
*备注: 16 pages, 4 figures. Code: this https URL
Abstract:Topic models learn gene programs from single-cell RNA sequencing (scRNA-seq) data, and the same counts are then used to test which genes and donors are associated with those programs. Count splitting is the standard remedy for this reuse, but it has not been tested when an encoder reads each cell’s counts. On a 104,400-cell lupus cohort with 300 permuted null genes per run, naive tests rejected 36% of null gene-topic pairs at a nominal 5% level, and the usual two-way split rejected 38%, because the encoder sees the tested gene’s own held-out count. Encoding topic usage on a separate third of the counts, with beta-binomial thinning, lowered this to 7.7% (6.7% for genes outside the model); for two other models, LDVAE and scETM, the rate was 7.0-7.3% (naive 51-59%). A label-free interferon program shifts in held-out counts in the lupus cohort and in a pre-specified COVID-19 cohort, and donor-level topic usage predicts disease (AUROC 0.96) beyond cell-type proportions (0.86 and 0.58 in the two cohorts). Calibration and prediction did not depend on the model. We do not claim better clustering, prediction or topic stability. Code, figures and result tables are public (MIT license; DOI https://doi.org/10.5281/zenodo.23307193).
[LG-260] Interpretability of AI weather models via intermediate decoding
链接: https://arxiv.org/abs/2610.24110
作者: Matthias Beylich,Kirsten I. Tempest,Tobias Selz,George C. Craig
类目: Atmospheric and Oceanic Physics (physics.ao-ph); Machine Learning (cs.LG)
*备注: 16 pages, 10 figures, 1 table. Supplementary material in the ancillary files
Abstract:AI weather prediction models are often regarded as black boxes. Despite their impressive skill, there is little understanding of how the input data is processed during inference to produce the forecast. This limitation arises from the use of high-dimensional latent representations in which individual dimensions are no longer directly interpretable in physical terms. To gain insight into how forecasts are constructed, we propose the method of intermediate decoding for AI weather models and apply it to GraphCast, Aurora and ArchesWeather. In this approach, latent states from intermediate depths are decoded back into meteorological fields. To do this, we use the decoder that has been trained to output the forecast at the end of processing. Since latent states at different depths do not share a consistent basis, we train affine transformations that align intermediate latent states with the decoder’s expected input space. Applying this framework to the three models yields the following results. First, the skill of forecasts at intermediate stages of processing improves gradually through the processor layers. Second, bypassing all processor blocks shows that the encoder-decoder pair is not an identity mapping returning the input, but instead plays a significant role in the forecast construction. Third, by examining increments of individual processor blocks across model depth, we show that all three models progressively refine the forecast from coarse to finer spatial scales. Finally, we assess the physical consistency of the intermediate decoding forecasts and find that large scales remain physically balanced throughout processing, while small scales become more balanced with each additional processor block.
附件下载
