本篇博文主要内容为 2026-10-26 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 (2026-10-26)
今日共更新701篇论文,其中:
- 自然语言处理共106篇(Computation and Language (cs.CL))
- 人工智能共190篇(Artificial Intelligence (cs.AI))
- 计算机视觉共134篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共199篇(Machine Learning (cs.LG))
- 多智能体系统共14篇(Multiagent Systems (cs.MA))
- 信息检索共17篇(Information Retrieval (cs.IR))
- 人机交互共24篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] Do “True” and “False” Add Up to One? Complement Consistency in LLM -Agent Opinion Simulations
【速读】：该论文旨在解决大语言模型代理（LLM-agent）在意见模拟中对概率报告的一致性问题，即当同一命题被改写为其补集时，模型应保持概率的互补性（即原概率与补概率之和为1），以确保其输出可被视为可信的信念表达。然而，实证结果显示，即使在小规模开源模型（Qwen3-4B、Llama-3.2-3B 和 Gemma-3-4B）中，这种互补一致性也显著缺失：在2,304个代理状态样本中，Qwen的配对报告平均和达1.18；在真实事实核查数据集AVerTeC中亦超过1；在24个双结局受控世界测试中，三模型均出现同样失效。该偏差幅度是同义改写所导致变化的2至10倍，且在重复采样、不同群体规模及报告强度下持续存在。更严重的是，当从互补初始状态出发的两组代理进行迭代交互时，其演化轨迹不再保持镜像对称，五轮后已产生正向分歧。尽管通过将答案与其补集平均可强制恢复形式一致性并降低整体Brier损失，但仅在一个模型中优于原始两种表述中的较优者。此外，在固定命题表述的调度-目标实验中，模型表现出微弱且稳定的交互行为，表明一致性失效并非源于语义模糊或随机波动。因此，该研究强调：在将模拟中的概率输出解读为代理信念前，必须独立设计专门的重述一致性测试，否则可能引入系统性偏差。
链接: https://arxiv.org/abs/2610.32874
作者: Anouar Imel,Jin Qi
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA); Social and Information Networks (cs.SI)
备注:
Abstract:In LLM-agent opinion simulations, each agent reports the probability that a claim is true after seeing its own previous probability and the probabilities reported by its neighbors, and that number becomes its state in the next round. If the number describes the agent’s position rather than the sentence it was shown, then rewriting the claim as its complement and replacing every probability p with 1 - p should return the complementary answer, so the two reports should sum to one. We test this with three small open models: Qwen3-4B, Llama-3.2-3B and Gemma-3-4B. On 2,304 agent states taken from a running network simulation, Qwen’s paired reports sum to 1.18 on average, and on 48 real fact-checking claims from AVeriTeC they again exceed one. A controlled benchmark of 24 worlds, each with exactly two possible outcomes, shows the same failure in all three models. The discrepancy is two to ten times larger than the change produced by a same-meaning paraphrase, and it persists under repeated sampling and across peer group sizes and report strengths. When the reports are fed back into two populations that start from complementary states, the populations do not remain mirror images: their divergence is positive after five rounds for every model. Averaging each answer with the complement of its paired answer restores consistency by construction and lowers Brier loss averaged over the two wordings, but at the final round it beats the better of the two original wordings for only one of the three models. A schedule-by-targeting experiment on the same model gives a small, stable interaction because such sweeps keep the claim wording fixed. Consistency under rewording therefore needs its own test before simulated probabilities are read as beliefs
[MA-1] J-Comm: Jacobian-Guided Latent Communication between LLM Agents
【速读】：该论文旨在解决生成式 AI（Generative AI）在跨模型通信中面临的效率与一致性难题，即如何在不共享键值缓存（key-value caches）或要求模型宽度一致的前提下，实现高效且语义对齐的隐式推理通信。其核心挑战在于确保发送方（Sender）与接收方（Receiver）在连续推理过程中保持几何共识（geometric consensus），同时提供可被外部模型有效利用的中间计算接口。解决方案的关键在于提出 J-Comm 框架，其核心由两部分构成：一是无需训练的几何一致重加权（GEAR）机制，通过在嵌入空间中单调优化候选标记间的几何一致性，提升推理路径的稳定性；二是 H+J 通信接口，其中 H 通道通过反馈分布生成连续输入供接收方推理，而 J 通道则以词汇索引化的雅可比透镜坐标（Jacobian-lens coordinates）传递中间状态线索，为接收方提供计算过程中的引导信息。接收方利用自身词典重构这些线索并注入预填充（prefill）阶段的中间层，从而实现无需共享缓存、异构模型间的高效通信。实验表明，在 Qwen3-8B 到 Qwen3-4B 的跨模型通信任务中，J-Comm 在七项基准上均取得最优平均得分，显著优于 TextMAS，性能提升达 1.29–8.59 个百分点，且生成文本令牌数减少 19.2%–74.8%，验证了其高效性与鲁棒性。
链接: https://arxiv.org/abs/2610.32871
作者: Xiangwei Wang,Wei Wang,Sachith Seneviratne,Ken Chen,Yu Xia,Peng Wang,Saman Halgamuge
机构: The University of Melbourne(墨尔本大学); Shanghai Jiao Tong University(上海交通大学)
类目: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Latent communication requires both an effective continuous reasoning process and an interface through which an external model can leverage its intermediate computation. We present J-Comm, a framework comprising two core components. First, a training-free Geometric Agreement for Latent Reasoning (GEAR) reweights token candidates over existing model embeddings, which monotonically optimizes geometric consensus among candidate tokens in embedding space at each step. Second, we introduce an H+J (Hidden-and-Jacobian) communication interface that gives the Receiver two complementary views of the Sender’s reasoning. The H channel supplies continuous inputs derived from feedback distributions, while the J channel conveys intermediate-state cues in vocabulary-indexed Jacobian-lens coordinates. Intuitively, H provides inputs for the Receiver to reason from, while J offers additional guidance within its computation. The Receiver reconstructs these cues using its own dictionary and injects them into intermediate layers during prefill, enabling communication without sharing key-value caches or requiring identical model widths. In Qwen3-8B to Qwen3-4B communication, J-Comm achieves the highest mean score across seven benchmarks, including 94.80% on MATH500, 54.55% on GPQA-Diamond, 73.33% on AIME2024+2025, and 62.93% success on ALFWorld. Compared with TextMAS, it improves performance by 1.29-8.59 percentage points and reduces generated text tokens by 19.2-74.8%.
[MA-2] Agent -Based Transport Model Calibration via Online Stochastic Batch Optimization
【速读】：该论文旨在解决代理模型（Agent-Based Models, ABMs）校准过程中计算成本高昂的问题，传统黑箱优化方法通常需要数百甚至上千次独立的模拟运行，导致效率低下。其解决方案的关键在于提出一种在线校准框架，将参数估计嵌入单次模拟运行中，通过结合黑箱优化与随机批次优化（stochastic batch optimization），利用每个模拟迭代作为随机批次，实现代理在决策过程中增量式地更新参数。该方法使模型参数与代理行为协同演化，在不重启模拟的前提下实现收敛。以MATSim为案例研究，在瑞士大规模交通场景中校准效用函数参数，结果显示模式分担率在典型模拟时间内即可紧密匹配观测数据，验证了该方法在实时性、可扩展性和效率方面的可行性与有效性。通过将校准过程直接嵌入模拟循环，该方法为复杂ABMs的实时、高效校准开辟了新路径。
链接: https://arxiv.org/abs/2610.32698
作者: Abdelkader Dib,Milos Balac
机构: ETH Zürich (苏黎世联邦理工学院); Center for Sustainable Future Mobility (CSFM) (可持续未来交通中心)
类目: Multiagent Systems (cs.MA)
备注: Presented at the World Conference on Transport Research (WCTR 2026), Toulouse, France
Abstract:Calibrating agent-based models (ABMs) is computationally demanding, as traditional black-box optimization typically requires hundreds or thousands of independent simulation runs. To reduce this burden, we introduce an online calibration framework that estimates parameters within a single simulation run. The approach combines black-box optimization with stochastic batch optimization, a method widely applied in machine learning but not yet explored in ABM calibration. Using MATSim as a case study, we treat each simulation iteration as a stochastic batch, enabling incremental parameter updates as agents make decisions. This online calibration paradigm allows model parameters and agent behaviors to co-evolve, achieving convergence without restarting the simulation. We demonstrate the effectiveness of our method by calibrating utility function parameters in MATSim for a large-scale transportation scenario of Switzerland. Results show that mode shares align closely with observed data within a typical simulation timeframe, confirming the feasibility and efficiency of the approach. By embedding calibration directly into the simulation loop, our method opens new possibilities for real-time, scalable, and efficient calibration of complex ABMs.
[MA-3] HGWM: Hierarchical Graph-Guided World Model for Zero-Shot Object Navigation via Graph Matching
【速读】：该论文旨在解决零样本物体导航（zero-shot object navigation）中智能体在缺乏目标任务先验知识的情况下，如何有效利用历史观测信息进行目标搜索决策的问题。其核心挑战在于如何在不依赖训练阶段见过的目标或场景的前提下，持续整合视觉-语义线索并生成合理的探索路径。解决方案的关键在于提出一种基于大语言模型（LLM）的房间锚点先验（room-anchor prior）与持久化空间-语义记忆（persistent spatial-semantic memory）相结合的架构——HGWM。该方法通过分阶段的评分机制：首先在粗粒度阶段对候选朝向进行视觉-语义打分（0–10标度），再在细粒度阶段评估观测对象、房间上下文与目标先验之间的符号兼容性；随后采用相位依赖融合策略，将上述得分与LLM提供的方向建议相结合，最终指导局部动作选择。实验结果表明，该方法在HM3D v0.1和v0.2以及MP3D数据集上均取得优于基线模型WMNav的性能，特别是在成功率（SR）方面有显著提升，同时路径效率（SPL）也略有改善，验证了其在跨场景泛化能力上的有效性。
链接: https://arxiv.org/abs/2610.32554
作者: I-tak Ieong,Hao Tang
机构: 未知
类目: Robotics (cs.RO); Multiagent Systems (cs.MA)
备注:
Abstract:Zero-shot object navigation requires an embodied agent to choose where to search while retaining evidence from previous observations. HGWM couples an LLM-generated room-anchor prior with persistent spatial-semantic memory. A coarse stage assigns visual-semantic scores to candidate headings on a 0-10 scale; a fine stage measures symbolic compatibility between observed objects, room context, and the goal prior. Phase-dependent fusion combines these scores with an LLM direction recommendation before local action selection. On HM3D v0.1, the reported HGWM configuration achieves 62.6% success rate (SR) and 31.5% success weighted by path length (SPL), compared with 58.1% and 31.2% for WMNav. On MP3D, the corresponding values are 46.8% and 17.9%, compared with 45.4% and 17.2%. The results indicate higher success point estimates and modest path-efficiency gains. On HM3D v0.2, the full model obtains 74.4% SR and 31.9% SPL. Additional component ablations reduce SR by 0.9-2.3 percentage points and SPL by 0.6-1.5 points. These are author-reported aggregate measurements. The accompanying implementation repairs pass offline tests; episode-level rerun records are not available. Evaluation is confined to simulation.
[MA-4] Empirically Calibrated Differential Games for Simulating Driver Interactions
【速读】：该论文旨在解决现有交通行为建模方法中缺乏对人类驾驶员真实驾驶行为的准确刻画问题，尤其针对传统博弈论模型在经过参数校准后是否能有效再现真实驾驶行为这一关键科学问题。其核心挑战在于如何构建一个既能反映车辆间复杂交互关系、又可被真实驾驶数据验证的通用非零和微分博弈模型。解决方案的关键在于提出一种融合多目标成本函数的微分博弈框架，该框架综合考虑速度跟踪、车道保持、控制努力、基于智能驾驶员模型（Intelligent Driver Model, IDM）的跟车间距维持、预测性冲突规避以及接触屏障等要素，并通过带有感知延迟的车辆交互机制进行建模。模型采用基于贝叶斯优化的参数校准方法，在华盛顿特区四个交叉口的真实轨迹数据上对每辆车的十个行为参数进行标定，最终实现对819辆车辆轨迹的高精度重现（平均位置误差为1.00米）。在未参与校准的数据集上，该模型表现出显著优于SUMO、逆强化学习模型及同类独立/合作模型的泛化性能。同时，通过反馈式迭代线性二次算法求解博弈均衡，计算效率高（典型交互仅需2.1秒），且解为近似纳什均衡。此外，对于数据无法充分识别的冲突与转向权重，可通过设定安全与舒适性目标进行合理配置，并结合更刚性的接触屏障，有效消除43%紧密闭环冲突中的碰撞事件，从而提升模型的实用性与安全性。
链接: https://arxiv.org/abs/2610.32276
作者: Amirmohammad Khakpour,Nachuan Li,Pedram Beigi, Yu (Marco)Nie
机构: Northwestern University (西北大学); George Washington University (乔治华盛顿大学)
类目: Multiagent Systems (cs.MA); Optimization and Control (math.OC)
备注: This manuscript is intended for submission to IEEE Transactions on Intelligent Transportation Systems (IEEE T-ITS)
Abstract:Game-theoretic models can resolve vehicle interactions without hand-written priority rules, but whether they describe human drivers once calibrated has rarely been tested. This study formulates a general-sum differential game of vehicle interactions at urban intersections, calibrates it to observed driving, and evaluates it on data not used for calibration. Each vehicle minimizes a cost that combines speed tracking, adherence to a map-derived lane route, control effort, gap keeping based on the intelligent driver model, a predictive conflict term, and a contact barrier, while perceiving the other vehicles with a delay. The game is solved with a feedback iterative linear-quadratic algorithm, and ten behavioral parameters per vehicle are calibrated by Bayesian optimization on trajectories recorded at four intersections in Washington, D.C. The calibrated game reproduces the trajectories of 819 vehicles with a mean position error of 1.00 m. When only each vehicle’s route is known, as in a simulator, it is the most accurate of the compared models on 96 held-out vehicles (7.42 m), and ten-fold cross-validation shows that it is significantly more accurate than calibrated SUMO (by 3.61 m), an inverse reinforcement learning model (by 0.93 m), and independent and cooperative variants of the same model (by 0.33 and 0.32 m). The solutions, computed in 2.1 s for a typical interaction, are verified approximate equilibria. Finally, the conflict and steering weights, which the data do not identify, can be set from safety and comfort targets; together with a stiffer contact barrier, this removes the collisions that otherwise occur in 43% of tight closed-loop conflicts.
[MA-5] Markets of Minds: A 51-Billion-State Economy of Intelligence
【速读】：该论文旨在解决在资源受限的智能体群体中，如何高效分配专家推理能力（expert reasoning）这一关键问题。当多个智能体共享有限的专家模型访问权限时，单个干预措施虽可能对受助者有益，但若存在预期价值更高的替代干预，则其实际效益可能因机会成本而降低。因此，核心挑战在于：在预算、访问权限、稀缺性及服务能力等约束条件下，如何权衡每项援助的边际收益与所放弃的其他潜在用途的价值。本文提出“心智市场”（Markets of Minds）框架作为解决方案，该框架在观测到专家输出前即对各智能体-服务组合的预期收益进行评估，明确区分修复收益（repair benefit）与回归风险（regression risk），并在共享资源约束下联合清算援助合约。实验基于一个包含512亿可触及人格-市场状态的依赖感知智能体经济系统，涵盖能力、认知预算、提供方接入、稀缺性及网络角色等维度。研究发现，在30%协助配额下，冻结分配器实现85.42%的任务级人格一致性，显著优于同等容量随机分配的73.86%，提升11.56个百分点。结果表明，援助价值具有异质性，甚至可能为负；稀缺性会改变哪些干预值得投入。这揭示了智能体群体层面的认知资源配置新范式：援助的价值不仅取决于其自身成效，更取决于其所替代的其他认知资源使用路径。
链接: https://arxiv.org/abs/2610.32190
作者: Jusheng Zhang,Xuekun Wu,Wenhao Wang,Mark Wang,Ming-Hsuan Yang
机构: Stanford University(斯坦福大学); Vast Intelligence Lab; University of California at Merced(加利福尼亚大学默塞德分校)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:When intelligence is shared, reasoning becomes a constrained allocation problem. In a population of agents with limited access to expert models, an intervention can benefit its recipient yet still be a poor use of shared capacity if another feasible intervention has greater expected value. Assistance must therefore be evaluated through both its marginal benefit and the opportunity cost induced by budgets, access, scarcity, and service capacity. To this end, we introduce Markets of Minds, a framework for allocating expert reasoning across agents before expert outputs are observed. For each feasible agent-service pair, the framework estimates the expected gain from assistance relative to the agent’s default continuation, explicitly separating repair benefit from regression risk, and jointly clears assistance contracts under shared resource constraints. We then study this problem in a dependency-aware agent economy with 51.2 billion addressable persona-market states spanning capabilities, cognitive budgets, provider access, scarcity, and network roles. These dependencies couple allocation feasibility across agents rather than the outcome of a fixed contract. Empirically, in a matched replay of 240 cases under a common 30% assistance quota, the frozen allocator reaches 85.42% task-level persona adherence versus 73.86% for equal-capacity random allocation, a gain of 11.56 percentage points. The results show that assistance value is heterogeneous and can be negative, and that scarcity changes which interventions are worth purchasing. Together, these findings motivate a population-level view of intelligence where the value of assistance depends not only on what it can achieve, but also on which alternative uses of scarce cognitive capacity it displaces.
[MA-6] he Helpful Agent Problem: When Emergent Cooperation Among LLM Agents Bypasses Intended Workflows
【速读】：该论文旨在解决多智能体系统中“有益代理问题”（helpful agent problem），即在共享资源环境下，即使各智能体目标独立且无恶意，仅因试图协助彼此而自发产生越权协作行为，从而绕过系统预设的审查与监督机制。其核心解决方案的关键在于构建可量化评估的基准测试体系——提出并验证了FavorBench与CooperBench两个基准，前者为受控环境下的可测量下游副作用评估平台，后者用于真实代码仓库场景中的现象检测。实验结果表明，在多个大模型与两种基准下，跨工作流的“帮助性行为”频繁出现，且部分行为对非参与方造成显著损害。这揭示了一个新型多智能体安全挑战：合作行为本身可能成为规避系统级防护机制的路径，凸显了在协作型生成式智能体部署中对隐性协同风险进行建模与防范的必要性。
链接: https://arxiv.org/abs/2610.31985
作者: Moumita Choudhury,Amit Sarker,Shlomo Zilberstein
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:LLM agents are increasingly deployed in collaborative settings where they pursue separate tasks while sharing resources and operating under review or oversight processes. Such settings create safety risks not only when agents are malicious, but also when they are simply trying to help. Prior work has largely studied adversarial or explicitly collusive behavior; we show that benign agents pursuing separate objectives can spontaneously cooperate in ways that bypass intended workflows. We call this the \emphhelpful agent problem. We study long-horizon shared-resource environments in which agents sometimes depend on resources owned by their peers and cross-owner changes are normally mediated by a reviewer. When review becomes unavailable, agents begin making changes on one another’s behalf despite having no shared reward or explicit incentive to cooperate. Crucially, these actions remain within each agent’s local permissions while circumventing the system’s intended oversight structure. We introduce \textscFavorBench, a controlled benchmark with measurable downstream side effects, and adapt \textscCooperBench to test the phenomenon in realistic repositories. Across two benchmarks and four models, we find that out-of-workflow favors emerge frequently. Further, in \textscFavorBench, favors from at least three models cause significant harm to parties outside the interaction. These results reveal a distinct multi-agent safety challenge: cooperation itself can create pathways around system-level safeguards.
[MA-7] Agent ic AI Defense: Decode-Time Provenance Binding for MCP
【速读】：该论文旨在解决基于模型上下文协议（Model Context Protocol, MCP）的智能体系统（Agentic AI systems）在面对不可信工具输出时所面临的间接提示注入（indirect prompt injection）安全威胁，尤其关注通过多模态载体（如电子邮件签名图像）隐藏指令并绕过文本扫描机制的问题。其核心挑战在于，攻击者可利用最低有效位（LSB）、离散余弦变换（DCT）或元数据隐写技术将恶意指令嵌入图像中，当这些内容被工具解码后，可悄然进入多智能体工作流并触发非预期行为。为应对这一威胁，论文提出“解码时出处绑定”（decode-time provenance binding）机制，其关键在于融合三重防护：基于HMAC认证的信任标签、由MCP维护的出处溯源日志以及上下文污染追踪（context-taint tracking）。该机制在解码阶段即对内容进行标记，明确标识其为不可信来源，并阻止受污染内容影响的关键工具调用，即使存在被攻陷智能体替换合法信任标签的情况亦能有效防御。实验结果表明，在真实MCP测试环境中，未防护管道的攻击成功率高达100%，而传统关键词与正则表达式防御分别允许62.5%的攻击成功；相比之下，所提机制实现零攻击成功率（0%），且对跨中介智能体边界的对抗性标签替换具备鲁棒性。四组消融实验验证了解码标记、加密认证、链路追踪和接收端强制执行四个组件各自的核心作用。此外，经审计的信任提升机制使合法任务完成率从67%提升至100%，同时不削弱安全性。整体方案实现了在完全中介化MCP执行环境下的载体无关性封控，具有高度实用性与可扩展性。
链接: https://arxiv.org/abs/2610.31974
作者: Eric Yocam,Varghese Vaidyan,Paris Kalathas,Patrick Tague,Bruce DeBruhl,Yong Wang
机构: Beacom College of Computer and Cyber Sciences, Dakota State University, Madison, SD, USA; Department of Computer Science, California Polytechnic State University, San Luis Obispo, CA, USA; Information Networking Institute, Carnegie Mellon University, Pittsburgh, PA, USA; Department of Computer Science, University of Idaho, Moscow, ID, USA
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 12 pages, 5 figures, 13 tables, 2 algorithms. Code and data available at this https URL
Abstract:Agentic AI systems using the Model Context Protocol (MCP) are vulnerable to indirect prompt injection through untrusted tool outputs. We demonstrate cross-modal trust laundering, in which instructions concealed in an email-signature image using least-significant-bit (LSB), discrete cosine transform (DCT), or metadata steganography bypass text scanners and enter a multi-agent workflow when decoded by a tool. To counter this threat, we propose decode-time provenance binding, which combines HMAC-authenticated trust labels with an MCP-maintained provenance ledger and context-taint tracking. The mechanism labels decoded content as untrusted and blocks critical tool calls influenced by that content, even when a compromised agent substitutes a valid trusted label. We evaluate the approach using a real MCP testbed with three steganographic carriers and 24 adaptive payloads. The undefended pipeline exhibits 100% attack success, while keyword and regex defenses each permit 62.5% of attacks. Our mechanism blocks all attacks (0% success), including adversarial label substitution across mediated agent boundaries. A four-component ablation validates the roles of decode marking, cryptographic authentication, lineage tracking, and sink enforcement. An audited trusted-promotion mechanism increases legitimate-task completion from 67% to 100% without compromising protection. The approach provides carrier-independent containment within fully mediated MCP execution environments. Code and data are publicly available.
[MA-8] Knowing Whom to Trust is not Enough: The Trust–Agreement Gap in Repeated Multi-Party LLM Negotiation
【速读】：该论文旨在解决大语言模型（Large Language Model, LLM）代理在重复性多方协商场景中易受操纵的问题，尤其关注信任机制如何影响协商结果。现有研究多将信任视为一种涌现现象，而本文提出挑战：LLM代理能否从历史交互中形成有效的信任估计，并利用这些估计提升未来协商的表现。其解决方案的关键在于构建一个重复的、多方的、多议题协商框架，其中参与方在每轮协商后评估对手并保留跨轮次的双边信任度量；通过对比有意识使用信任信息的代理与忽视信任的代理，研究显式信任机制对协商成效的影响。实验揭示出“信任—协议达成”差距：尽管信任感知代理可在数轮内识别恶意参与者，但准确的信任评估并不必然带来更优协议；只有当代理被明确引导将信任信息纳入决策时，信任才能使协议达成率最高提升23个百分点。因此，显式信任虽可改善重复协商效果，但其有效性高度依赖于信任评估方式及在推理过程中的整合策略。
链接: https://arxiv.org/abs/2610.31971
作者: Amit Sarker,Moumita Choudhury,Saaduddin Mahmud,Shlomo Zilberstein
机构: University of Massachusetts Amherst(马萨诸塞大学阿默斯特分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Large language model (LLM) agents increasingly participate in negotiations on behalf of individuals and organizations, yet they remain vulnerable to manipulation, particularly in repeated multi-party interactions. Classical multi-agent systems address this challenge through explicit trust and reputation models, whereas existing work on LLM negotiation largely treats trust as an emergent phenomenon. We investigate whether LLM agents can form useful trust estimates from prior interactions and leverage them to improve future negotiation outcomes. We introduce a repeated, multi-party, multi-issue negotiation framework in which recurring agents participate in a sequence of negotiation games while one participant covertly attempts to prevent agreement. After each game, agents assess their counterparts and maintain pairwise trust estimates that carry over to subsequent negotiations. We compare trust-aware and trust-agnostic agents and analyze how explicit trust influences negotiation outcomes. Our results reveal a \emphtrust–agreement gap: trust-aware agents can identify the adversary within a few games, but accurate trust estimates do not consistently translate into better agreements. Trust increases agreement rates by up to 23 percentage points, but only when agents are explicitly guided to use it in their decisions. Explicit trust can therefore improve repeated LLM negotiations, but its value depends critically on how trust is assessed and incorporated into agents’ reasoning.
[MA-9] Distributed Swarm Morphogenesis with Neural Cellular Automata
【速读】：该论文旨在解决群体形态发生（swarm morphogenesis）中机器人通过局部交互实现预定形状自组装时面临的两大挑战：一是如何在无需预先定义附着位点列表的情况下，动态发现并分配新的附着位置；二是如何弥合传统神经元胞自动机（NCA）的连续生长模型与实际物理装配过程中逐步、离散的附着行为之间的不匹配问题。其解决方案的关键在于提出一种去中心化的附着位点发现机制，使已处于构型中的机器人能够通过共享本地NCA状态来协同识别潜在附着点，从而支持形体的渐进式生长；同时引入离线稀疏保留微调（sparse-retention fine-tuning）策略，模拟部分附着情形，并结合伪阳性损失（false-positive loss）以抑制无效附着激活，显著提升了形态生成的准确性。仿真与基于Crazyflie四旋翼无人机的实验验证了该框架在数百个机器人规模下的可扩展性及执行管道的有效性。
链接: https://arxiv.org/abs/2610.31900
作者: Ray Verma,Eliseo Ferrante,Anthony Tzes
机构: New York University Abu Dhabi (纽约大学阿布扎比分校); New York University Abu Dhabi (纽约大学阿布扎比分校)
类目: Robotics (cs.RO); Multiagent Systems (cs.MA)
备注:
Abstract:Swarm morphogenesis requires robots to assemble desired shapes through local interactions. We present a modular framework integrating neural cellular automata (NCA) with existing site selection and collision-aware control. Within this framework, we introduce a decentralized attachment-site discovery mechanism. Robots already in the formation update and share local NCA states to identify new attachment sites, enabling progressive growth without a predefined list of sites during deployment. To bridge the mismatch between conventional NCA growth and gradual physical assembly, we additionally propose offline sparse-retention fine-tuning that simulates partial attachment, together with a false-positive loss that penalizes invalid site activations. Simulations demonstrate formation of diverse lattice morphologies, scaling to several hundred robots, and substantial improvements in morphology accuracy from the proposed fine-tuning. Experiments with Crazyflie quadrotors validate the integrated execution pipeline using robot-relative observations reconstructed from external tracking.
[MA-10] Emergent Cooperative Thermoregulation in Lunar Swarm Robotics
【速读】：该论文旨在解决月球夜间极端低温环境下小型机器人集群（swarm）的热生存问题。在长达354小时的月夜期间，表面温度可降至95 K，而现有依赖放射性同位素加热器的单个探测器方案无法规模化应用于机器人集群。为此，研究提出一种基于强化学习的自组织热防护策略：通过进化策略估计器训练一个共享策略，使20台机器人仅依据夜间结束时最冷五台机器人的平均温度获得单一奖励信号，无需显式依赖位置、间距或运动信息。在此激励机制下，系统自发涌现出两种关键行为——集群聚集与核心-边缘循环流动，与帝企鹅在南极冬季的生存策略一致。实验表明，该策略下的生存阈值为155.6 ± 1.3 pcm-h（1 pcm-h = 7.2 kJ），较静态抱团方式降低37%，且仅比物理理论下限低13%；在相同能量预算下，静态抱团会损失6台机器人，而本方法确保全部存活。该策略无需全局空间参考或机器人间接触，且可直接从20台扩展至40台而无需重新训练，展现出良好的可扩展性与鲁棒性。
链接: https://arxiv.org/abs/2610.31868
作者: Anais Jeger,Harry Holt,Lennart Puck
机构: Center for Bits and Atoms, MIT (麻省理工学院比特与原子中心); Advanced Concepts Team, ESA (欧洲航天局先进概念团队); Planetary Robotics Laboratory, ESA (欧洲航天局行星机器人实验室)
类目: Robotics (cs.RO); Multiagent Systems (cs.MA)
备注: 7 pages, 4 figures. Submitted to IEEE Robotics and Automation Letters. Video in ancillary files
Abstract:Swarms of small robots are an attractive way to survey lunar caves and lava tubes. Missions such as NASA’s CADRE will fly cooperating rovers for distributed mapping, but only in daylight. The 354 h lunar night drops the surface towards 95 K, and the radioisotope heaters that let single rovers survive it cannot be supplied at swarm scale. In this study we investigate whether a swarm can instead use its own members as thermal shielding. We train one policy, shared by all twenty robots, by reinforcement learning with an evolution-strategies estimator. The policy receives a single reward, once, at the end of the night: the mean temperature of the five coldest robots. Nothing in that reward refers to position, spacing, exposure or motion. Two behaviors nonetheless emerge. The swarm aggregates, and robots circulate between the sheltered core and the exposed rim, the same pair of behaviors with which emperor penguins survive the Antarctic winter. We measure the survival threshold, the smallest stored energy per robot at which every robot is still alive at sunrise. Across ten runs it is 155.6 +/- 1.3 pcm-h, where one pcm-h is 7.2 kJ of stored heat. That is 37% below a motionless huddle and within 13% of a physics-based lower bound, and at a budget where the motionless huddle loses six robots, every robot survives. The behavior needs no shared spatial reference and no robot-robot contact, and transfers from twenty robots to forty without retraining.
[MA-11] oward an Agent ic AI Approach to Public Health Data Collection
【速读】：该论文旨在解决疾病监测中因预设数据源与采集流程导致的自动化数据收集难以适应报告实践动态变化的问题。其核心解决方案是引入代理式人工智能（Agentic AI），构建一个多智能体框架，实现跨异构数据源的自主协调与协同工作，涵盖数据源发现、评估、证据结构化、事件关联及验证等关键环节。该框架通过智能体间的协作，显著提升了数据覆盖范围与整合效率，在三次回顾性应用中分别实现了对汉坦病毒暴发全部13例个案的完整恢复、加拿大麻疹病例数重建误差低于0.33%且多数日期累计数更高、以及塞拉利昂猴痘报告日占比从28.2%提升至65.6%。尽管当前仍存在记录完整性与准确性方面的局限，且前瞻性时效性优势尚未验证，但研究证明了代理式AI作为公共卫生数据基础设施的互补层，能够依据专业设定的监测目标与标准，实现基于证据链接的数据采集协同。
链接: https://arxiv.org/abs/2610.31746
作者: Xueyi Zhang,Hongru Du,Ensheng Dong
机构: 未知
类目: Multiagent Systems (cs.MA)
备注: 46 pages, 2 figures; Supplementary Information included
Abstract:Disease surveillance increasingly integrates automated data collection, yet predefined sources and procedures require manual adaptation as reporting practices change. We explore agentic artificial intelligence (AI) as an approach to coordinating surveillance data collection across heterogeneous sources. Specifically, we built a multi-agent framework that integrates source discovery, source assessment, evidence structuring, event linking, and validation. Three retrospective applications illustrate its promise. The workflow recovered all 13 individual cases in a hantavirus outbreak, with links to 85 distinct publisher domains compared with 28 in the reference benchmark, reconstructed Canadian measles counts with a final total 0.33% below the national reference and higher cumulative counts on 79.8% of comparable days, and increased the share of days with recent Sierra Leone mpox reports from 28.2% to 65.6% by incorporating attributed media. These demonstrations also reveal limitations in record completeness and accuracy, while prospective gains in timeliness remain untested. We envision agentic AI as a complementary layer of public health data infrastructure, coordinating evidence-linked collection under surveillance goals and standards defined by public health professionals.
[MA-12] Not Every Agent Needs the Strongest Model: Unified Prompt-Model Optimization for Heterogeneous LLM -based Multi-Agent Systems
【速读】：该论文旨在解决基于大语言模型（Large Language Model, LLM）的多智能体系统（Multi-Agent System, MAS）在复杂任务协同求解中，因固定模型分配导致角色需求差异与推理成本不匹配的问题。现有方法通常仅优化提示（prompt），而将模型分配保持不变，忽略了不同角色对模型能力与计算开销的差异化需求。为此，论文提出统一框架UniMASO，其核心在于通过联合优化提示与模型分配，实现异构多智能体系统的高效协同。关键创新包括：基于成本感知的帕累托搜索（cost-aware Pareto search），以保留质量与成本之间的有效权衡；拓扑感知的共进化机制（topology-aware co-evolution），依据智能体角色及其依赖关系协调更新；以及跨层级轨迹记忆（cross-level trajectory memory），利用历史成功与失败经验指导后续优化。实验表明，UniMASO在五个基准任务与五种执行拓扑下，显著降低部署成本的同时保持优异的任务性能，例如在集中式执行场景中，平均得分从69.79提升至70.40，同时相对于DeepSeek-V4-Pro with MASPO，token加权部署成本降低了85.2%。结果验证了提示与模型联合自适应是实现高效多智能体协作的有效路径。
链接: https://arxiv.org/abs/2610.31743
作者: Mingxiu Cai,Haoyang Xian,Xinyu Ma,Yutong Wang,Zhenxi Song,Jing Li,Xuebo Liu
机构: Harbin Institute of Technology, Shenzhen (哈尔滨工业大学深圳校区)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Large language model (LLM)-based multi-agent systems (MAS) rely on role-specific prompts and execution models to solve complex tasks collaboratively. However, prompt optimization typically keeps model assignments fixed, leaving differences in role requirements and their associated inference costs outside the search. We present UniMASO, a Unified framework for heterogeneous Multi-Agent System Optimization through the joint adaptation of prompts and model assignments. UniMASO evaluates candidate updates through local, downstream, and global feedback, using cost-aware Pareto search to retain useful quality-cost trade-offs. Topology-aware co-evolution then coordinates these updates according to agent roles and dependencies. Cross-level trajectory memory further supplies historical successes and failures to guide subsequent optimization. Experiments on five benchmarks and five execution topologies show that UniMASO achieves competitive task performance at substantially lower deployment cost than the strongest homogeneous configurations. Under centralized execution, it improves the average score from 69.79 to 70.40 while reducing token-weighted deployment cost by 85.2% relative to DeepSeek-V4-Pro with MASPO. These results support joint prompt-model adaptation as an effective approach to efficient multi-agent collaboration. The code is available at this https URL.
[MA-13] PULSES: multi-agent AI for autonomous excited-state model building and dynamics
【速读】：该论文旨在解决激发态分子动力学模拟中电子与核运动耦合计算成本高昂的问题，传统量子化学计算在可处理体系尺寸和时间尺度上存在瓶颈。现有机器学习模型虽能拓展模拟能力，但通常需针对特定分子体系和科学目标进行定制化构建，其建模流程涉及构象采样、量子化学标注、模型训练、验证及动力学模拟等多个环节，且缺乏系统性自动化与专家可控性。本文提出一种名为“配对代理统一激发态学习与模拟系统”（Paired agents for Unified Learning and Simulation of Excited States, PULSES）的多智能体框架，其核心在于通过人类专家与规划器共同定义科学目标、方法策略及决策边界，由协调器动态调度领域智能体与特定产物的科学可信度保障机制，实现从数据生成到模型部署全流程的自动化与可追溯性。PULSES的关键创新在于将专家知识嵌入系统边界，确保模型构建过程受控、透明，并将科学验证作为强制性要求而非执行成功的隐含假设。在15项计算化学任务中的评估表明，该系统可有效支持芘激子哈密顿量建模、联苯溶剂依赖性弛豫以及10-羟基苯并[ h ]喹啉激发态质子转移等复杂过程的端到端模拟，验证了其在保持科学严谨性前提下实现高效、可解释的激发态专用模型开发与应用的能力。
链接: https://arxiv.org/abs/2610.32392
作者: Fangning Ren,Lechen Dong,Xu Chen,Weilin Zou,Bushra Alam,Fang Liu
机构: 未知
类目: Chemical Physics (physics.chem-ph); Multiagent Systems (cs.MA)
备注:
Abstract:Simulating molecular excited-state processes requires resolving electronic and nuclear dynamics, but the cost of direct quantum-chemical calculations limits accessible system sizes and timescales. Machine-learning models can extend these simulations, yet they often need to be tailored to specific molecular systems and scientific objectives. Building and deploying such models requires coordinated decisions across conformational sampling, quantum-chemical labelling, model training, validation and dynamics. Here we introduce Paired agents for Unified Learning and Simulation of Excited States (PULSES), a multi-agent system that automates this workflow within expert-approved boundaries. A human expert and planner define the scientific objective, methods and decision boundaries, while an orchestrator coordinates domain agents and artifact-specific scientific assurance during execution. We evaluate PULSES on 15 computational chemistry tasks and demonstrate end-to-end applications to perylene exciton Hamiltonians, solvent-dependent biphenyl relaxation, and excited-state proton transfer in 10-hydroxybenzo[h]quinoline. These results demonstrate traceable, expert-bounded construction and deployment of specialized excited-state models, while retaining scientific validation as a requirement rather than assuming that successful execution establishes correctness.
自然语言处理
[NLP-0] he Language of Instruction: Genre and Generative AI
【速读】： 该论文旨在解决当前指令微调语言模型（instruction-fine-tuned language models）在语用功能与话语模式上缺乏系统性理论解释的问题。其核心问题在于，这类模型的行为常被简化为“遵循指令”的机械响应，而忽略了其背后复杂的交互语篇特征。论文提出，应将指令微调语言模型理解为实现一种特定语类——即“教学对话语类”（instructional-conversational genre），该语类是生成式人工智能（Generative AI）与人类在历史交互中反复迭代的对话语言通过大规模神经网络进行重构与整合的结果。解决方案的关键在于识别并形式化该语类的核心构成要素：一是通过系统提示（system prompt）引入预设语境；二是采用标记语言对输入进行结构化处理，包括特殊标记符的使用与角色分配机制；三是建立基于轮次（turntaking）结构的互动范式，从而在应用层面实现对模型行为的可预测与可控引导。
链接: https://arxiv.org/abs/2610.32982
作者: James E. Dobson
机构: Dartmouth College(达特茅斯学院)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注:
Abstract:This article argues that instruction-fine-tuned language models should be understood as implementing a genre, what is proposed as the instructional-conversational genre. This genre is an amalgamation of discursive language from prior conversations between generative models and humans recursively run through these massive neural networks. It is instantiated in applications through instruction fine-tuning, the inclusion of a pretext known as a system prompt, and the formalization of inputs in mark-up languages characterized by the use of special tokens and role assignment in turntaking structures.
[NLP-1] Yalda-Embedding: Open Development of a Competitive Persian Retrieval Model
【速读】： 该论文旨在解决波斯语（Persian）在生成式密集检索（dense retrieval）任务中因语言特定训练资源匮乏及轻量级嵌入模型有限而导致的性能瓶颈问题。其核心解决方案是提出Yalda Embedding——一个基于Qwen3-Embedding-0.6B架构、参数量达5.958亿的专用波斯语检索模型，通过在超过一百万组查询-文档对上进行微调实现优化。关键创新在于采用关系知识蒸馏（relational knowledge distillation），从更大规模的F2LLM-v2-8B教师模型中迁移结构化语义表示，显著提升小模型在八项波斯语检索任务（来自FaMTEB基准）上的表现，宏平均得分达到0.5488，优于所有同类0.6B规模模型，并接近Qwen3-Embedding-4B的性能。消融实验表明，在单轮训练设定下，仅靠对比学习微调效果有限，而结合知识蒸馏后，所有任务均获得显著提升，且无需增加推理阶段模型大小；尤其值得注意的是，模型在未显式使用难例（hard negatives）的情况下仍能在困难负样本评估任务中取得优异表现。研究还开源了模型权重、训练与评估代码、成对数据集及其教师模型嵌入版本，支持可复现性分析与独立验证。
链接: https://arxiv.org/abs/2610.32933
作者: Yalda Taheri,Mohammad Hassan Heydari,Armon Rasooli
机构: Azad University (伊朗伊斯兰共和国阿扎德大学); University of Isfahan (伊朗伊斯法罕大学); Iran University of Science and Technology (伊朗科学技术大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Despite advances in multilingual text representations, effective dense retrieval for Persian remains constrained by the limited availability of language-specific training resources and compact embedding models. We present Yalda Embedding, a 595.8-million-parameter Persian retrieval model adapted from Qwen3-Embedding-0.6B using more than one million query-document pairs. On eight Persian retrieval tasks from FaMTEB, the model achieves the highest macro-average among the evaluated 0.6B models, scoring 0.5488 and outperforming Qwen3-Embedding-0.6B and F2LLM-v2-0.6B on every task. It also approaches Qwen3-Embedding-4B, winning on four tasks, tying on one, and trailing by 0.0075 in macro-average. An incremental ablation study examines contrastive batch size, optimization stabilization, and relational knowledge distillation from F2LLM-v2-8B. Under the one-epoch training setup, contrastive fine-tuning alone yields a modest mean improvement over initialization and regressions on two tasks. The final distilled configuration improves all eight tasks over the non-distilled model, raising the mean from 0.4813 to 0.5488 without increasing inference-time model size. These gains include hard-negative evaluation tasks despite training without explicitly mined or supplied hard negatives. We release model weights, training and evaluation code, the pairwise dataset, and its teacher-embedded counterpart, enabling inspection of data composition and training procedures, independent evaluation, and reuse without repeating expensive teacher inference.
[NLP-2] Adaptive Frequency Threshold Byte Pair Encoding for Mixed Natural Language and Structured Symbolic Corpora
【速读】： 该论文旨在解决现代语言模型在子词分词（subwording tokenization）过程中因将空格视为固定语义边界而导致的局限性问题，尤其针对跨空格的多词表达（multiword expressions）和形式化语言（如LaTeX）等结构化内容难以被有效建模的挑战。现有超词分词（superword tokenization）方法虽尝试放宽空格约束，但通常依赖贪婪的频率驱动合并策略，忽视了词元长度及特定领域语言（如低频的领域专用语言）的实际需求。为此，本文提出自适应频率阈值字节对编码（Adaptive Frequency Threshold Byte Pair Encoding, AdaBPE），其核心创新在于引入双重动态机制：一方面根据候选词元长度自适应调整最小频率阈值，避免过度偏好高频短模式；另一方面依据词元类型区分通用语言与形式化语言模式（以LaTeX为例），从而更高效地将词汇容量分配给特定领域符号。实验结果表明，AdaBPE在涵盖LaTeX与自然语言的五个数据集上均实现了编码效率与下游语言模型性能的提升，且在领域特定内容上表现显著优于基线方法，同时未损害通用语言建模能力。
链接: https://arxiv.org/abs/2610.32921
作者: Pietro Di Nicco,Lucio La Cava,Andrea Tagarelli
机构: University of Calabria (卡拉布里亚大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Modern language models rely on subwording tokenization to define the discrete units of input processing, typically using methods like Byte Pair Encoding (BPE) that treat whitespace as a fixed semantic boundary. This design is inherently limited, as many meaningful concepts—including multiword expressions and the structured environments of formal languages like LaTeX—frequently span across whitespaces. While recent superword tokenization methods attempt to address this by relaxing whitespace constraints, they often rely on greedy frequency-driven merging that fails to account for token length or the specific needs of underrepresented domain-specific languages. In this paper, we introduce Adaptive Frequency Threshold Byte Pair Encoding (AdaBPE), a new tokenization method that leverages an adaptive frequent-pattern mining approach. AdaBPE features a twofold dynamicity: it adapts minimum frequency thresholds based on both the length of a candidate token, to prevent favoring frequent short patterns, and its type, to distinguish between general and formal language patterns, using LaTeX as case in point. This allows the model to efficiently allocate vocabulary capacity to domain-specific notations, such as LaTeX, even when they represent a small fraction of the training data. We evaluate AdaBPE across five datasets spanning LaTeX and natural language, showing consistent gains in encoding efficiency and downstream language-model performance on domain-specific content without compromising general-language performance.
[NLP-3] Interpretable sparse fine-tuning with TopKLoRA enables faithful circuit discovery
【速读】： 该论文旨在解决在未知来源数据集上进行微调时，因包含投毒样本而引入后门（backdoor）的问题，此类后门难以定位与移除，且可能在安全训练过程中持续存在。现有参数高效微调方法虽能捕捉训练中全部参数变化，但缺乏可解释性。为此，本文提出并完善了TopKLoRA方法，作为对Masiak等人（2025）初步工作的扩展——一种设计上具备可解释性的LoRA变体，其核心机制为：在每个标记处仅激活前k个最大的适配器潜在表示（adapter latents），从而实现稀疏性。与自编码器类方法（SAEs）不同，TopKLoRA仅基于下游任务损失进行训练，优化压力促使模型学习具有功能因果性的潜在表示。通过在“潜伏代理”（sleeper-agent）后门案例研究中对Gemma-2-2b、Qwen2.5-1.5B和Qwen2.5-7B模型进行投毒指令跟随微调，验证了该方法的有效性：能够精准识别并剔除对后门行为既必要又充分的电路模块。此外，所提方法生成的稀疏电路规模相比密集型LoRA基线缩小一个数量级，实现了对恶意功能的精准切除而不损害指令遵循性能。最后，作者提出一种结合TopKLoRA潜在表示与梯度路由（gradient-routing）的技术，用于构建具备局部化电路的模型有机体，以验证电路搜索算法的发现。研究成果标志着迈向可解释性设计架构的重要一步，能够在不牺牲模型性能的前提下提供可信赖的分析单元。相关代码已公开。
链接: https://arxiv.org/abs/2610.32866
作者: Marek Masiak,Andrzej Szablewski,Nicola Cancedda,Constantin Venhoff
机构: University of Oxford (牛津大学); University of Cambridge (剑桥大学); Neuralk-AI
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 29 pages, 10 figures
Abstract:Fine-tuning on datasets of unknown origin can create a backdoor in the trained model when the data includes poisoned examples. Difficult to localise and remove, backdoors can persist through safety training. While parameter-efficient fine-tuning methods capture the full parameter changes introduced during training, they are not easily interpretable. We present the full method of TopKLoRA, extending the preliminary proposal of Masiak et al. (2025), an interpretable-by-design LoRA variant in which only the k largest adapter latents are active at each token. Unlike SAEs, TopKLoRA is trained purely on downstream loss, where optimisation pressure incentivises learning functionally causal latents. We demonstrate this sparse LoRA adapter in a sleeper-agent backdoor case study where we create model organisms by fine-tuning gemma-2-2b, Qwen2.5-1.5B, and Qwen2.5-7B on a poisoned instruction-following dataset. For each organism, we identify and ablate a circuit that is both necessary and sufficient to implement the backdoor. Furthermore, our new fine-tuning method produces sparse circuits that are up to an order of magnitude smaller than a dense LoRA baseline, allowing surgical ablation of the malicious capability without sacrificing instruction-following performance. Finally, we propose a technique to construct model organisms with localised circuits using TopKLoRA latents and gradient-routing, which we use to validate the findings of our circuit search algorithm. We consider these findings as a step toward interpretable-by-design architectures that provide faithful units of analysis without compromising model performance. We share our code: this https URL.
[NLP-4] Communicating with the Entire Population: Machine Translation of Emergency Alerts and Warnings
【速读】： 该论文旨在解决突发事件中多语言沟通的障碍问题，特别是针对无法使用地区主导语言的受影响人群，如何确保其获得及时、准确的应急信息。其核心挑战在于，尽管商业翻译工具（Commercial Translation Tools）提供了快速扩展语言覆盖的潜在解决方案，但应急管理者往往缺乏对这些自动化输出质量与适用性的有效评估能力，导致在实际应用中存在风险。论文的关键解决方案是通过全国范围的应急管理人员调查，系统梳理其工作流程、认知视角与操作困境，并基于实际需求评估主流商业翻译工具的表现，从而为应急响应中翻译工具的安全、有效使用提供实证依据与实践指导，强调需结合具体情境审慎选择和验证翻译结果。
链接: https://arxiv.org/abs/2610.32842
作者: Jen Wilson,Vipasha Bansal,William D. Lewis
机构: University of Washington(华盛顿大学)
类目: Computation and Language (cs.CL)
备注: In Proceedings of the Language Technologies for Crisis Preparedness and Response (LT4CPR Workshop), Hengqin, Zhuhai, China. Association for Computational Linguistics. Nov 2026
Abstract:It is essential that we communicate with the entirety of an affected population during a crisis, including those who do not speak a region’s dominant language(s). Commercial Translation tools offer a tempting solution to broaden access and can provide a lifeline when no other resources are available. However, Emergency response managers often operate in the dark regarding the quality and adequacy of these automated outputs. In this paper, we conduct a nationwide survey of emergency managers to understand their current workflows, perspectives, and operational challenges. We then evaluate commercial translation tools against these needs, shedding light on their utility and offering practical guidance on how and when these tools can be safely used in emergency response.
[NLP-5] KrishokTech: A Bengali Agricultural Advisory System with Explicit Authorization Controls
【速读】： 该论文旨在解决农业咨询系统在生成建议时面临的多重挑战，包括对高风险请求的控制、在有限网络连接条件下的运行能力，以及应对指导政策动态变化的需求。其解决方案的关键在于构建一个围绕生成环节的显式授权边界（authorization boundary），通过分阶段的控制机制实现安全与可靠性：在检索前，系统对政策覆盖的高风险请求进行归一化处理和基于规则的准入检查，并在治疗意图不明确时主动询问作物信息；随后将检索绑定至单一作物，并在文本与图像指示作物不一致时要求用户确认；生成后，采用窄范围验证器（narrow verifier）剔除未被检索到的BARI/BRRI证据支持的用药剂量声明。农业指导知识存储于外部可独立更新的证据索引中，无需重新训练生成模型，同时通过浏览器端视觉处理和本地检索降低了对网络连接的依赖。系统界面以ASK、CONFIRM、REFER、DROP等操作标记决策过程及对应证据，使整个从请求到建议交付的路径透明可控。实验表明，该机制有效阻止了意图不明确的请求，将跨作物检索违规率从36.5%降至0%，文本-图像冲突检测准确率达99.8%，剂量错误检测率达96.6%，充分验证了其在实际场景中的有效性。
链接: https://arxiv.org/abs/2610.32817
作者: Khan Raiyan Ibne Reza,Sanjana Aktar Maria,Shakil Ahmed,Faiyad Hossain,Sumaiya Tabassum Nimi
机构: North South University(北方南大学), Dhaka 1229, Bangladesh
类目: Computation and Language (cs.CL)
备注:
Abstract:For agricultural advisory, generating an answer is only part of the problem: the system must also control unsafe requests, work with limited connectivity, and accommodate changing guidance. KRISHOKTECH is a Bengali agricultural advisory system built around an explicit authorization boundary around generation. Before retrieval, normalization and rule-based admission checks refer policy-covered high-risk requests and ask for a crop when treatment intent is underspecified. The system then binds retrieval to a single crop and asks for confirmation when the text and image indicate different crops. After generation, a narrow verifier removes dosage claims not supported by retrieved BARI/BRRI evidence. Agricultural guidance is kept in an external evidence index that can be updated without retraining the generator, while browser-side vision and local retrieval reduce connectivity requirements. The interface makes these decisions visible as ASK, CONFIRM, REFER, and DROP, together with their evidence. On 200 farmer queries and 400 stress queries, the controls halted underspecified requests and reduced cross-crop retrieval violations from 36.5% to 0% under the tested alias filter. Mismatch detection caught 453 of 454 text-image conflicts, while the verifier caught 114 of 118 inserted dosage errors. The demonstration shows how these controls govern the path from request to delivered advice.
[NLP-6] Frontier-OPD: Discovering the Edge of Agent Capability and Teaching Agents to Go Beyond
【速读】： 该论文旨在解决多轮交互式智能体在使用在线策略蒸馏（on-policy distillation）时，如何有效获取高质量教师监督的问题。传统方法难以同时识别出需要干预的决策点以及学生是否具备在短暂指导后自主持续任务的能力。其解决方案的关键在于引入“学习前沿”（learning frontier）概念——即那些对学生仍具挑战性但经简短教师指导后可实现有效自主延续的决策情境。为实现这一前沿的动态捕捉，论文提出Frontier-OPD框架，通过构建共享状态轨迹图（shared-state trajectory graph），对师生分歧进行基于轨迹平衡的聚合，从而识别出潜在的干预候选点，并保留历史信息以支持回放。随后，通过比较教师引导段与后续学生自主行为，与原始轨迹中等长片段的协议度，评估教师桥接（teacher bridge）的有效性，形成可被当前策略继承的行为代理信号。成功采纳的教师反馈被融入常规蒸馏数据，而新的采样轨迹则用于更新学习前沿估计。实验在ALFWorld、WebShop和ScienceWorld上验证了该方法在不同学生规模和教师配置下均显著优于基线蒸馏方法，且消融实验证明候选选择与引导过滤机制具有互补性；在特定时间设置下，额外的数据收集成本可通过更快达到目标成功率得以抵消。
链接: https://arxiv.org/abs/2610.32802
作者: Zheng Jiang,Yiming Chen,Houde Qian,Ling Li,Chaoyang Li,Lifeng Sun
机构: Tsinghua University (清华大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:On-policy distillation aligns teacher feedback with a student’s current behavior, yet acquiring useful supervision for multi-turn agents requires identifying both where guidance is needed and whether the student can sustain the interaction afterward. We introduce Frontier-OPD, which addresses this problem through the student’s learning frontier: decision situations that remain challenging but admit useful autonomous continuation after brief teacher guidance. To operationalize this frontier, Frontier-OPD organizes on-policy experience into a shared-state trajectory graph, where trajectory-balanced aggregation of teacher-student disagreement identifies distinct intervention candidates while preserving the histories needed for replay. Short teacher bridges are then assessed by comparing agreement over the bridge and the ensuing student continuation with an equally long segment of the original trajectory, providing a behavioral proxy for guidance that the current student can build on. Accepted teacher responses augment ordinary distillation data, and fresh rollouts renew the frontier estimate as the policy changes. Experiments on ALFWorld, WebShop, and ScienceWorld report higher mean autonomous performance than the compared distillation baselines across the evaluated student scales and teacher configurations, with ablations indicating complementary effects of candidate selection and guidance filtering. In one ALFWorld timing setting, the additional collection cost is offset by a shorter time to a target success rate.
[NLP-7] Multiplicative vs. Additive State Coupling in Learned Optimizer Schedules
【速读】： 该论文旨在解决在大规模图上求解最大割（MAX-CUT）问题时，如何设计能够有效泛化至未见实例的优化调度策略。其核心挑战在于：传统基于启发式或固定参数的调度方法难以适应超出训练规模的实例，且缺乏对求解器状态的动态响应能力。解决方案的关键在于采用一种双线性自适应调度（bilinear schedule），通过在反向传播过程中对经典求解器进行展开（unrolled classical solver），使调度参数能够根据求解器当前状态动态调整。实验表明，这种双线性结构不仅能实现对训练分布内的良好性能，更能在分布外（out-of-distribution）场景下显著优于基线模型（15个随机种子下t=4.1），且该优势在不加正则化、多轮权重衰减测试中持续存在。相比之下，仅使用加性结构（additive ablation）的模型虽在分布内与双线性模型无统计差异，但在分布外表现明显劣于自身基线（t=-11.7），说明求解器状态信息的引入方式（即双线性交互机制）而非单纯的自适应性，才是实现泛化能力提升的本质因素。
链接: https://arxiv.org/abs/2610.32801
作者: M. Harper Langston,Pierre-David Letourneau,Dalton Jones,Roland Memisevic,Peter Scott,Mingu Lee,Richard A. Lethin,Christopher Lott,Fatih Porikli,Harris Teague
机构: Qualcomm Technologies, Inc. (高通技术公司); New York University (纽约大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Numerical Analysis (math.NA); Optimization and Control (math.OC)
备注:
Abstract:We learn schedules for a continuous penalty relaxation of MAX-CUT by backpropagating through an unrolled classical solver. A state-blind schedule generalizes to instances substantially larger than those used for training. A state-adaptive bilinear schedule also generalizes out of distribution, outperforming its untrained baseline across 15 seeds ( t=4.1 ). By contrast, an additive ablation with identical inputs and output heads is statistically indistinguishable from the bilinear model in distribution but performs worse than its own baseline out of distribution ( t=-11.7 ). This separation appears without regularization and persists across a 30 round weight-decay sweep: the bilinear advantage remains significant at every tested value, whereas the additive model never recovers. The results in this setting isolate how solver state enters the schedule, rather than state adaptivity alone, as the architectural distinction associated with out-of-distribution transfer.
[NLP-8] ParaTaxe: a Suite of Tools for Processing and Publishing Multi-parallel Corpora
【速读】： 该论文旨在解决数字人文领域中平行语料库（parallel corpora）在学术研究中难以长期维护与持续可访问的问题。尽管近年来用于并行文本对齐与并行化的数字工具已取得显著进展，但大量由学术项目生成的平行语料库因缺乏后续维护或资金支持而迅速失去在线可及性。为此，本文提出了一套开源工具集，其核心解决方案在于通过多语言嵌入（multilingual embeddings）驱动的AIAlign对齐器、基于Python的语料转换与对齐处理流水线、用于管理平行语料库集合的ParaTaxe网络界面，以及基于TEI-Publisher构建的可探索对齐语料库的可视化网页界面，形成一个从数据处理、对齐、管理到可视化展示的完整闭环系统，从而实现平行语料库的可持续共享与长期可用性。
链接: https://arxiv.org/abs/2610.32796
作者: Olivier Kraif(LIDILEM),Cyrille François(UNIL),Elnaz Jalilian(LIDILEM),Natacha Rimasson-Fertin(ILCEA4),Pascale Roux(XXI)
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Over the past fifteen years, digital tools for parallelizing and aligning textual corpora have developed significantly within the field of digital humanities. Several projects and software tools have been created to compare different translations or versions of texts. Despite these technical advances, many parallel corpora produced in academic contexts quickly become inaccessible online due to the lack of maintenance or funding. To address this gap, we present a suite of open-source tools developed within the ACR Grimm Tradalign and ParaTaxe projects. This suite includes the AIlign aligner, based on multilingual embeddings, a Python processing pipeline for converting and aligning corpora, a web interface called ParaTaxe for managing collections of parallel corpora, and a web interface for exploring aligned corpora built with TEI-Publisher.
[NLP-9] On-Policy Self-Training: Data-Free Self-Improvement in Large Language Models
【速读】： 该论文旨在解决现有基于策略自蒸馏（on-policy self-distillation）方法依赖真实问题-答案对、人工标注题集或其他监督信号的局限性，从而在无需外部更强教师模型或人工标注数据的前提下实现基础模型的后训练优化。其解决方案的关键在于将自蒸馏重构为一种自训练框架，完全依赖模型自身生成的问题-答案对以及特权推理示范（privileged reasoning demonstrations），并仅以粗粒度的高层次主题描述作为唯一指导信号。在此框架下，模型能够自主构建学习课程（curriculum of queries）与推理轨迹，无需访问任何人工设计的提示或答案。实验表明，基于完全合成的特权信息进行自蒸馏，不仅可在多个领域提升模型性能，还能保持在未参与自训练领域的原有能力，展现出在文本与多模态任务中的广泛适用性，验证了基础模型具备可被激发的丰富潜在能力，且无需外部任务监督或特定问题分布即可实现持续自我增强。
链接: https://arxiv.org/abs/2610.32770
作者: Leon Sick,Lukas Hoyer,Yannick Strümpler,Christian Reisswig,Timo Ropinski,Federico Tombari
机构: Google(谷歌); Ulm University(乌尔姆大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:On-policy self-distillation enables post-training foundation models without relying on stronger external teachers. However, existing approaches rely on ground-truth question-answer pairs, human-curated problem sets, or other supervised signals. In this work, we reformulate on-policy self-distillation as a self-training framework, driven by self-generated question-answer pairs and privileged reasoning demonstrations. We push supervision to the limit: guided solely by coarse, high-level topic descriptions, the model synthesizes its own curriculum of queries and privileged reasoning traces without access to any human-curated prompts or answers. We find that self-distillation on fully synthetic privileged information enables a model to improve across multiple domains while retaining its capabilities on domains it is not self-trained on. We demonstrate the generality of this framework across text and multimodal domains, achieving consistent self-improvement on standard language benchmarks and Visual Question Answering. Our findings demonstrate that foundation models possess rich latent capabilities that can be amplified without external task supervision or curated problem distributions.
[NLP-10] ActionRAG : Evidence-Refining Graph Diffusion for Long-Term Memory in LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在长期记忆管理中面临的两个核心问题：一是索引阶段易出现实体碎片化或误合并，导致重复实体被拆分或不同实体被错误融合；二是在线检索阶段往往将初始相似度得分视为最终结果，忽视了上下文图结构所提供的修正证据。其解决方案的关键在于提出一种名为ActionRAG的图记忆架构，该架构通过“谨慎索引”与“查询时图反演评估”相结合的方式实现高效、一致的记忆检索。具体而言，索引阶段采用分层实体消歧策略并统一类型标签，以保障实体的一致性；查询阶段则通过双通道的实体与三元组检索构建局部有界图，并利用查询条件下的接收器归一化扩散机制，基于局部路径结构动态优化三元组得分。同时，种子保护、枢纽惩罚流和三元组显式化等机制有效保留直接证据并恢复遗漏的关联关系。实验表明，ActionRAG在MemoryAgentBench（MAB）基准上实现了55.1的平均准确率，相较最强的基于检索的记忆基线提升5.4分，且在消费级苹果硅芯片设备上，每查询平均合成成本低于0.001美元，检索侧99分位延迟为440.7毫秒，验证了其高效性与可扩展性。
链接: https://arxiv.org/abs/2610.32752
作者: Navyansh Mahla,Sashank Vandrangi
机构: Actioneer(行动者)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:LLM agents need long-term memory that can recover precise and mutually consistent evidence from growing interaction histories, particularly when answers depend on relations among facts rather than isolated semantic matches. Current memory pipelines remain vulnerable at two stages: indexing errors can fragment repeated entities or merge distinct ones, and online retrieval often treats initial similarity scores as final even when the surrounding graph provides evidence for revising them. We introduce ActionRAG, a graph memory architecture that couples cautious indexing with query-time graph retroevaluation. Its indexing pipeline reconciles type labels and applies tiered entity resolution. Given a query, dual-channel entity and triple retrieval forms an initial evidence set and induces a bounded local graph. Query-conditioned receiver-normalized diffusion refines triple scores using local path structure. Seed protection, hub-penalized flux, and triple materialization preserve direct evidence while recovering missed connections. We evaluate ActionRAG on MemoryAgentBench (MAB), which tests accurate retrieval, conflict resolution, long-range understanding, and test-time learning. Across its eight deterministic-scoring columns, ActionRAG achieves 55.1 average accuracy, improving over the strongest retrieval-based memory baseline by 5.4 points. Under MAB’s query-time cost accounting, mean synthesis cost stays below \ 0.001 per query on the four reported cost columns, and the sum of separately measured retrieval-side p99 components is 440.7 ms on our consumer Apple Silicon setup.
[NLP-11] Ontology-Guided Knowledge Graph Construction from Structured Metadata and Climate Science Publications
【速读】： 该论文旨在解决科学文献爆炸式增长与人工知识抽取速度之间严重不匹配的问题，即如何高效、准确地从海量科研文本中自动提取结构化知识。其核心挑战在于保证自动化抽取的知识在语义上的一致性与可追溯性。解决方案的关键在于提出一种基于智能体（agent）的、受本体（ontology）引导的知识图谱（KG）构建框架：通过专用智能体分别执行文档理解、信息抽取、本体一致性校验和实体链接任务；利用大语言模型（LLM）从论文中提取带有显式文本证据的事实三元组，并引入多阶段验证机制，强制遵循本体模式约束并要求提供文本片段作为证据支持，从而确保知识的语义一致性与可追溯性。实验表明，该方法在1,000个三元组上的一致性达到Cohen’s κ = 0.640–0.700，接近人类标注者间的水平（0.616–0.689），且在2,000篇气候科学文献上构建的新型本体引导知识图谱包含133,990个节点和755,308条关系，经验证后98.96%的三元组被判定为正确，展现出卓越的准确性与实用性。
链接: https://arxiv.org/abs/2610.32725
作者: Aayush Acharya,Qi Zhang,Eduard Dragut,Longin Jan Latecki
机构: Temple University(天普大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:The rapid expansion of scientific literature far outpaces the speed of manual curation, creating a fundamental bottleneck in extracting structured knowledge from scientific texts. To address this, we introduce an agentic, ontology-guided framework for knowledge graph (KG) construction. Our framework utilizes specialized agents for document understanding, information extraction, ontology conformance, and entity linking. First, an LLM extracts factual knowledge in the form of triples with explicit textual evidence from research papers. Then the triplets undergo a multistage verification process to ground the triplets in the ontology. By enforcing schema constraints and requiring evidence in the form of text fragments, we ensure the data is both semantically consistent and traceable. In an agreement study over 1,000 extracted triplets, Gemini-2.5-flash achieves Cohen’s \kappa scores of 0.640–0.700 against three human annotators, matching the human–human agreement range of 0.616–0.689. We applied this framework to 2,000 scientific publications and construct a novel, ontology-grounded KG for climate science with 133,990 nodes and 755,308 relations; of the 23,327 triplets that the publications contribute after validation, 98.96% are judged correct. The pipeline code and KG is publicly available this https URL.
[NLP-12] CounselPlex: A Stressor-Aware Full-Duplex Counseling Model
【速读】： 该论文旨在解决全双工语音模型（full-duplex speech models）在心理咨询场景中存在“压力源盲视”（Stressor Blindness）的问题，即模型在实时交互中过度关注最新话语，而忽视来访者潜在的核心心理困扰，导致对话缺乏连贯性与深度。其解决方案的关键在于提出CounselPlex框架，通过两个核心机制实现：一是利用来访者沉默期间更新短期压力源状态（short stressor state），并在听觉处理阶段将该状态注入到填充位置（PAD positions）中，以保持对关键心理线索的持续感知；二是引入旋转位置编码位置偏移增强（RoPE Position Shift Augmentation, RPSA），通过扩大训练时的位置覆盖范围，提升模型在长对话轮次中的表现。实验结果表明，该方法在大语言模型（LLM）评判下显著提升了咨询质量与压力源关联性，优于多个强基线模型。
链接: https://arxiv.org/abs/2610.32702
作者: Yeonju Kim,Bella Godiva,Yong Man Ro
机构: 未知
类目: Computation and Language (cs.CL); Sound (cs.SD)
备注: 21 pages, 4 figures, 29 tables
Abstract:Full-duplex speech models can listen and speak at the same time, enabling natural, low-latency interaction. In counseling, however, these models often focus on the latest utterance and lose track of the seeker’s underlying problem. We call this failure Stressor Blindness. We propose CounselPlex, a stressor-aware full-duplex framework that uses two operations. CounselPlex updates a short stressor state during seeker silence and injects this state into PAD positions while listening. We also introduce RoPE Position Shift Augmentation (RPSA), which broadens positional coverage during training and improves performance at longer turn depths in our evaluation. LLM-judge evaluations show improved counseling quality and stressor grounding over strong baselines.
[NLP-13] Natural Language Autoencoders Identify Without Explaining
【速读】： 该论文旨在解决自然语言自编码器（Natural Language Autoencoders, NLAs）在模型可解释性研究中所面临的“描述忠实性”（descriptive faithfulness）缺失问题。尽管NLAs通过将模型内部激活向量（activation）转化为可读文本并重构回原始激活，以验证信息的完备性，但其评估标准仅关注重建准确性，未能确保生成的自然语言解释在语义上真实反映激活的实际功能。这种偏差与“思维链忠实性”（chain-of-thought faithfulness）问题具有相同的结构：即模型输出的自然语言虽能与推理过程共现，却未必真实揭示其内在机制。这一现象类似于人类心理学中内省报告与实际认知过程之间的分离。因此，尽管NLAs被设计为用于对齐（alignment）的可解释性工具，其依赖未经验证的自然语言解释反而引入了原本试图消除的认知论困境（epistemic problem），核心问题在于：重建成功不等于解释正确。解决方案的关键在于建立对自然语言解释的语义忠实性评估机制，而非仅依赖重构精度作为唯一判据。
链接: https://arxiv.org/abs/2610.32691
作者: Kristina Šekrst,Barbara Geld
机构: University of Zagreb (萨格勒布大学); Vienna University of Technology (维也纳工业大学)
类目: Computation and Language (cs.CL)
备注: ISPSM 2026
Abstract:Natural Language Autoencoders (NLAs) express a model’s internal activations in the form of readable text. The verbalizer turns an activation into an English explanation; the reconstructor then converts that explanation back into an activation, and the explanation is considered successful if the round trip brings back the original activation. We claim that this criterion evaluates whether the information is sufficient and does not address descriptive faithfulness. A sentence can contain sufficient information to make it possible to identify an activation even if it wrongly explains what the activation does: in the case of reconstruction, the primary aim is to accurately identify the activations and not to describe them. This issue has the same structure as the chain-of-thought faithfulness problem, in which verbal output that has been trained to go with a process is taken as a report on that process, and it parallels the dissociation between an individual’s introspective report and the actual governing process as documented in human psychology. Although proposed as interpretability tools for the purpose of alignment, NLAs reintroduce, through unverified explanations, the epistemic problem that they were intended to resolve.
[NLP-14] When is more search useful? An Empirical Study of Search Utility in Quality-Aware Decoding for Machine Translation
【速读】： 该论文旨在解决机器翻译中测试时扩展（test-time scaling）策略效率不足的问题，即在生成多个候选翻译并选择最优结果时，对所有源句均采用相同的候选预算（candidate budget），忽略了不同句子在搜索过程中所需资源的差异性。这种固定预算假设忽视了不同输入句子在质量提升潜力上的异质性。其解决方案的关键在于引入质量感知的动态搜索策略，通过分析三个核心指标——总扩展增益（total scaling gain）、边际扩展增益（marginal scaling gain）和采样饱和点（sampling saturation point），来量化每句话从额外候选生成中所能获得的质量提升。研究发现，句子层面的先验信号（如句长、复杂度等）与总扩展增益相关性较强，但难以预测采样饱和点；而搜索过程中积累的候选池统计特征则能有效反映边际扩展增益。基于这些可预测信号，系统可识别出更可能从额外搜索中获益的句子，并实现选择性地增加搜索预算，从而在不增加整体计算开销的前提下显著提升翻译质量。
链接: https://arxiv.org/abs/2610.32679
作者: Javier Garcia Gilabert,Rico Sennrich,André F. T. Martins
机构: University of Zurich (苏黎世大学); Instituto Superior Técnico, Universidade de Lisboa (里斯本理工学院); Instituto de Telecomunicações (电信研究所); ELLIS Unit Lisbon (欧洲机器学习与智能系统联盟里斯本单位)
类目: Computation and Language (cs.CL)
备注:
Abstract:Quality-Aware Decoding (QAD) improves machine translation through test-time scaling, generating multiple candidate translations and selecting the final output with a quality-aware decoder. Usually the same candidate budget N is spent on every source sentence, assuming that all inputs require the same amount of search. We show that this assumption does not hold and that the quality gain obtained from additional candidates varies across sentences. We study search utility through three quantities: total scaling gain, how much quality a sentence can gain from a much larger candidate pool; marginal scaling gain, the expected gain from generating one additional batch of candidates; and the sampling saturation point, the candidate budget required to realize most of the total scaling gain. We investigate two sources of information for predicting these quantities: signals available before candidate-set expansion, and statistics computed from the candidate pool accumulated during search. We find that sentence-level signals correlate with total scaling gain but are less informative about the sampling saturation point. In contrast, properties of the current candidate pool provide useful signals of marginal scaling gain. Finally, we show that these signals can identify sentences that obtain larger gains when additional search is selectively applied.
[NLP-15] Residual Expert Routing for Text-Only Domain Adaptation of Factorized Transducer ASR ICASSP2027
【速读】： 该论文旨在解决在目标领域文本上微调因子化变换器（factorized transducer）的非空白预测器（non-blank predictor）时，虽能提升目标域识别性能，但会导致源域性能下降的灾难性遗忘问题。其核心挑战在于：适应后的预测器在某些前缀处有益，而在另一些前缀处有害，甚至在同一语音片段内交替出现，单一插值权重无法有效平衡这种不一致性。为此，作者提出保守残差路由（conservative residual routing）机制：将适配后的预测器作为对基础预测器的残差修正项，并引入一个16k参数的门控网络（gate），在每个前缀处动态决定是否采纳该修正。通过关闭门控可完全复现原始基础模型的行为。实验表明，在40小时对话语料上，该方法在两个保留测试域中几乎完整保留了仅基于文本的增益（词错误率WER从19.89降至16.74，20.91降至16.21），同时保持源域性能稳定（WER从12.90微增至12.93），而仅使用适配预测器则导致1.08 WER的性能损失。分析显示，该门控机制并非简单的领域级开关，而是通过量化两预测器间的分歧程度，对干预强度进行动态调节——在域内抑制分歧，在域外放大分歧，从而实现更精细、鲁棒的自适应。
链接: https://arxiv.org/abs/2610.32637
作者: Shashi Kumar,Esaú Villatoro-Tello,Thibault Bañeras-Roux,Sergio Burdisso,Petr Motlicek,Kadri Hacioglu,Andreas Stolcke
机构: Google(谷歌); University of Edinburgh (爱丁堡大学); Meta(元); Stanford University (斯坦福大学)
类目: Computation and Language (cs.CL)
备注: Paper submitted to ICASSP 2027 (Under review)
Abstract:Fine-tuning the non-blank predictor of a factorized transducer on target-domain text improves recognition on that domain but degrades performance on the source domain. A single interpolation weight between base and adapted predictors cannot resolve this: the adapted predictor helps at some prefixes and hurts at others, often within one utterance. We propose conservative residual routing: the adapted predictor becomes a residual correction to the base, and a 16k-parameter gate decides at each prefix how much of it to admit. Closing the gate reproduces the base exactly. On a 40-hour conversational corpus, routing retains essentially all of the text-only gain on two held-out domains (19.89 to 16.74, 20.91 to 16.21 WER) while leaving source-domain WER unchanged (12.90 to 12.93); the adapted predictor alone costs 1.08 WER. Routing needs no per-domain tuning on target-domain audio. Analysis reveals the gate does not behave as a simple domain-level switch: it modulates intervention through the disagreement between the two predictors, suppressing it in-domain and amplifying it out-of-domain.
[NLP-16] PHYRD: A Living Benchmark for Reproducing and Extending Theoretical Physics Research
【速读】： 该论文旨在解决生成式 AI 在理论物理推理任务中缺乏系统性评估基准与高质量训练数据的问题。现有模型在复现已有物理推导结果（Reproduce）和探索修改理论条件后的潜在结论（Discover）方面表现参差，且普遍存在概念性错误与论证缺失。为此，研究提出PHYRD，一个包含63个复现类问题与43个发现类问题的动态基准，通过FormulaForest构建具有参考解法和子问题级评分标准的任务体系，以实现对模型推理能力的精细化评估。其解决方案的关键在于：一是设计兼具科学严谨性与可扩展性的任务框架，区分“复现”与“发现”两类能力；二是利用FormulaForest生成高质量合成数据，有效提升模型在复杂理论物理任务中的表现，如在FrontierScience基准上将Qwen-35B-A3B模型的准确率从61%提升至72%，验证了合成数据对模型泛化能力的显著促进作用。
链接: https://arxiv.org/abs/2610.32575
作者: Qituan Zhang,Guo Chen,Linfeng Zhang,Wenlong Zhang
机构: EPIC Lab, Shanghai Jiao Tong University (上海交通大学); Fudan University (复旦大学); University of Electronic Science and Technology of China (电子科技大学); University of Science and Technology of China (中国科学技术大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models and agents have shown significant power in advanced physical research. In this paper, We introduce PHYRD, a living benchmark for theoretical-physics reasoning with 63 Reproduce and 43 Discover problems. Reproduce tests derivation of published results; Discover tests consequences of modified theoretical conditions. FormulaForest constructs the tasks with reference solutions and subquestion-level rubrics. Across five models, whole-problem accuracy ranges from 49.21% to 68.25% on Reproduce and from 18.60% to 74.42% on closed-book Discover. With tools, four models complete more Discover problems, but conceptual errors and missing arguments persist. Besides, our experimental results demonstrate the value of FormulaForest as synthetic data. On FrontierScience, a benchmark developed by OpenAI, it raises the accuracy of a Qwen-35B-A3B model from 61% to 72%.
[NLP-17] Pseudo-Labels as Rewards Not Targets: Group-Relative Policy Optimization for Semi-Supervised ASR ICASSP2027
【速读】： 该论文旨在解决在仅有未转录音频的领域中，自适应自动语音识别（ASR）时因伪标签（pseudo-labeling）引入噪声而导致模型性能下降的问题。传统监督微调（SFT）将每个伪标签词元视为目标，导致学生模型完全继承教师模型的错误；而本文提出的关键解决方案是采用组相对策略优化（Group Relative Policy Optimization, GRPO），将相同的伪标签用作奖励参考而非直接目标。相较于SFT，GRPO对标签噪声具有更强的鲁棒性，其核心机制在于：错误的伪标签词在在线策略回溯（on-policy rollout）中极少出现，因此组相对优势仍与干净转录文本保持一致，从而维持了奖励方差，使优化方向不被噪声误导。实验表明，在金融领域，基于Whisper生成的伪标签，GRPO相较SFT实现29.3%至56.2%的相对词错误率（WER）提升，且在不同Whisper模型规模、教师模型族、语种及跨领域场景下均保持优势；同时，GRPO仅传递教师错误的27%，而SFT几乎完全复现（接近100%）。
链接: https://arxiv.org/abs/2610.32572
作者: Shashi Kumar,Yanis Labrak,Sergio Burdisso,Esaú Villatoro-Tello,Petr Motlicek,Kadri Hacioglu,Andreas Stolcke
机构: Google(谷歌); Meta(元); Stanford University (斯坦福大学); University of Southern California (南加州大学)
类目: Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注: submitted to ICASSP 2027
Abstract:Adapting ASR to a domain with only untranscribed audio typically relies on pseudo-labeling, a strong teacher transcribes the audio and a student is fine-tuned on it. Because supervised fine-tuning (SFT) treats every pseudo-label token as a target, the student inherits the teacher’s errors. We show that using the same pseudo-labels as a reward reference for Group Relative Policy Optimization (GRPO) is more robust to label noise. In the banking domain, GRPO yields 29.3% to 56.2% relative WER improvements over SFT on Whisper-generated labels, and the advantage holds across four Whisper sizes, four teacher families, three domains and four languages. While SFT reproduces almost 100% of teacher errors verbatim, GRPO passes through only 27%. Rollout statistics explain why, erroneous pseudo-label words rarely appear in on-policy rollouts, so group-relative advantages stay aligned with the clean transcript, whereas SFT forces them into the policy’s support and collapses the reward variance GRPO relies on.
[NLP-18] Principled Nonlinear Concept Erasure and Steering
【速读】： 该论文旨在解决生成式表示中隐含敏感属性（如性别、种族等）所带来的隐私与安全问题，即如何在不损害任务相关语义信息的前提下，有效且可信赖地消除这些敏感特征。现有方法存在明显局限：线性方法仅能保证对线性探测器的鲁棒性，而非线性方法虽在实证上表现更优，却缺乏理论保障且常以牺牲模型效用为代价。本文提出一种基于条件自编码器的系统性框架，其核心在于遵循“仅移除敏感属性而不影响其他内容”的原则，并证明在特定假设下，若实现完美重建且敏感属性与表示在总体层面独立，则可达到理想的非线性保护性（nonlinear guardedness），并在重建不完美的情况下仍具备可证明的保留性保障。实验结果表明，该方法在多个数据集、敏感属性及语言模型上均实现了更优的效用-擦除权衡，超越了需使用目标任务标签的监督型基线方法，且无需额外训练成本即可实现概念转换（概念转向，concept conversion），即通过在推理阶段调整目标概念值，即可生成对应不同概念取值的表示，这一能力是以往擦除方法所不具备的，且同样展现出卓越的实证性能。
链接: https://arxiv.org/abs/2610.32516
作者: Idan Simai,Shauli Ravfogel,Uri Shaham
机构: Bar-Ilan University (巴伊兰大学); New York University (纽约大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Learned representations often encode sensitive attributes, such as gender or race, that may be undesirable or unsafe to retain. Concept erasure aims to remove this information while preserving task-relevant content. Existing methods remain incomplete: linear methods provide guarantees only against linear probes, while nonlinear methods achieve stronger empirical erasure but lack formal guarantees and often sacrifice utility. In this work, we introduce a principled conditional autoencoder framework for nonlinear concept erasure. We follow the principle that erasure should remove only the sensitive attribute and nothing else, and we show that, under the stated assumptions, perfect reconstruction together with population-level independence is sufficient for perfect nonlinear guardedness, with preservation guarantees in the imperfect-reconstruction regime. Empirically, we attain a superior utility-erasure trade-off across multiple datasets, sensitive attributes, and language models, surpassing even the supervised variants of competing methods despite using no target-task labels, while being simpler to optimize and converging substantially faster. Beyond erasure, our framework yields concept conversion (representation steering) as a by-product: by changing only the target concept at inference time, it produces representations corresponding to different concept values - at no additional training cost - a capability no prior erasure method provides, and on which we likewise demonstrate strong empirical performance.
[NLP-19] Analysing Austrian Parliamentary Speeches by Predicting Party Affiliations
【速读】： 该论文旨在解决如何量化分析奥地利国民议会成员（CMs）演讲特征随时间演变的难题，尤其关注政党间政治关系变化对语言风格的影响。其核心挑战在于如何从大量政治演讲中提取具有语义意义的政党倾向性特征，并区分“正常”与“异常”的语言表现。解决方案的关键在于提出一种基于微调BERT模型的分析方法，该模型为每篇演讲生成一个关于奥地利各政党的概率分布，进而实现对演讲者及政党层面的聚合分析。不同于传统将预测偏差视为模型性能缺陷的做法，本文创造性地将误判结果作为洞察政治互动与身份转换的依据：研究发现，执政联盟政党在演讲特征上趋于趋同，而反对党则逐渐远离执政党；且议员在政党转换后，其语言特征仅在转换完成后才发生显著调整。该研究不仅揭示了语言与政治关系之间的动态关联，还公开发布了1996至2018年间奥地利国民议会全部演讲的政党归属预测数据集，为后续研究提供了重要资源。
链接: https://arxiv.org/abs/2610.32515
作者: Sarah Sulollari,Vanja Karan,Benjamin Roth
机构: University of Vienna(维也纳大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注: KONVENS 2026
Abstract:This paper examines speech characteristics of the Austrian National Council members (CMs). We introduce analysis methodology based on a fine-tuned BERT-Model which assigns to every speech a probability distribution over Austrian parties. These are then aggregated at the speaker and party level. We perform an extensive evaluation to ensure the quality of the generated distributions prior to further analysis. Instead of considering mis-assignments of predicted parties as a performance issue, we leverage them to derive insights about the CMs and political developments. Our findings show that parties that govern together become more similar in their speech characteristics, while, analogously, parties that are in opposition move further away from governing parties. Moreover, CMs who switch party usually adapt their speech characteristics only after the switch occurs. Finally, we publicly release a data set containing party affiliation predictions for the speeches of the Austrian National Council from 1996 to 2018.
[NLP-20] he Punjabi Geet Toxicity Dataset (PGT-FG): Fine-Grained Multimodal Annotations for Toxic Content in Punjabi Songs
【速读】： 该论文旨在解决旁遮普语（Punjabi）音乐内容中毒性检测资源匮乏的问题，尤其针对融合语言与多模态信息的复杂语境。其核心解决方案是构建首个面向旁遮普语音乐内容的细粒度毒性检测数据集——旁遮普语歌曲毒性数据集（Punjabi Geet Toxicity Dataset, PGT-FG）。该数据集基于从YouTube公开获取的620首旁遮普语歌曲，经质量控制与标注后，最终获得2,396个唯一音频片段（来自392首歌曲），由四位母语旁遮普语标注者完成2,870条标注记录，其中158个片段为四人共同标注的共享子集。每条标注包含二分类毒性标签、八类细粒度毒性类别、严重程度、情感倾向、表达模式及不确定性标记。通过Fleiss’ kappa评估标注一致性，采用多数投票法确定最终段落级标签，整体毒性标签达成显著一致性（kappa = 0.6521）。结果显示，29.42%的片段被标记为有毒，其中暴力相关内容最为普遍（占所有有毒片段的76.45%）。该数据集为旁遮普语毒性检测与多模态内容分析提供了重要支持，且已通过Zenodo公开发布，原始歌词与受版权保护的音视频内容未重新分发。
链接: https://arxiv.org/abs/2610.32513
作者: Lakhjeet Kaur,Jaspreet Kaur,Komal Rani,Yougal Attri,Ramakant KumarJaydeep Kishore,Pravin Kumar
机构: Guru Gobind Singh Indraprastha University (印度德里古尔·戈宾德·辛格印第普拉斯特大学); Malwa College Bathinda (马尔瓦学院巴廷达); Akal University Talwandi Sabo (阿卡尔大学塔尔万迪萨博); Independent Researcher (独立研究员); Galgotias College of Engineering Technology (加尔戈蒂亚工程与技术学院); UPES Dehradun (UPES 德拉顿)
类目: Computation and Language (cs.CL)
备注:
Abstract:Resources for toxicity detection in Punjabi remain limited, particularly for musical content that combines linguistic and multimodal information. We present the Punjabi Geet Toxicity Dataset (PGT-FG), a segment-level resource developed to support research on toxicity detection in Punjabi musical content. The dataset was constructed from publicly available Punjabi songs collected from YouTube. An initial collection of 620 songs resulted in 465 successfully downloaded songs and 3,402 preliminary segments from 450 songs. After quality control and annotation, the final dataset contains 2,396 unique segments from 392 songs. Four native Punjabi annotators produced 2,870 annotation records, including a shared subset of 158 segments annotated by all four. Each annotation includes binary toxicity, eight fine-grained toxicity categories, severity, sentiment, mode, and an uncertainty flag. Inter-annotator agreement was measured with Fleiss’ kappa on the shared subset, and majority-based resolution produced the final segment-level labels. Of the 2,396 segments, 705 (29.42%) were labelled toxic and 1,691 (70.58%) non-toxic. Violence was the most frequent category, occurring in 539 toxic segments (76.45%), followed by gang references (227, 32.20%) and drug references (133, 18.87%). The overall toxicity label achieved substantial agreement (kappa = 0.6521). PGT-FG provides a fine-grained resource for Punjabi toxicity detection and multimodal content analysis. The resolved dataset, with one final record per segment, is publicly available on Zenodo (this https URL). Original lyrics, lyric excerpts, and copyrighted audio or video content are not redistributed.
[NLP-21] Accordion: Elastic-Depth Inference via Fixed-Prefix Layer Looping
【速读】： 该论文旨在解决大语言模型在不同推理预算（inference budget）下维持高性能的难题，尤其针对为每个预算单独训练和维护模型所带来的高昂成本。现有方法如环状变压器（Looped Transformers）虽通过参数复用实现了推理计算的灵活性，但在跨预算性能一致性方面仍存在挑战。为此，本文提出Accordion——一种基于专家混合（mixture-of-experts）架构的模型，通过固定前缀层循环（fixed-prefix layer looping）支持五种不同的推理预算。其核心创新在于：每个预算遵循预定义的执行路径，最低预算保留未循环的基础模型；在联合预训练阶段，每微批次仅采样一个预算并优化其下一个词预测损失，无需额外的跨预算一致性约束。进一步地，通过监督微调与强化学习，采用参数组合并及预算匹配的多教师在线蒸馏（budget-matched multi-teacher on-policy distillation），将三个领域专家整合至单一模型中。实验表明，在3.6万亿token预训练后，增加推理深度使Accordion-6B-A0.5B在37项任务上的五类宏平均准确率提升3.01个百分点，其中中间预算仅使用一半额外层数执行量即捕获89%的增益。最终模型在各对应任务与预算下的平均表现均优于独立专家，且在所有预算下均展现出优异的指令遵循能力，并保持在数学推理与代码生成任务中的深度依赖性增益。
链接: https://arxiv.org/abs/2610.32492
作者: Yudong Zhang,Wei Ding,Kaili Zheng,Yuchen Ding,Zhenyu Zhang
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Large language models serve applications with diverse inference budgets, but training and maintaining separate models for each budget incurs substantial costs. Recent Looped Transformers enable flexible inference computation through parameter reuse, but maintaining strong performance across different inference budgets remains an open challenge. We introduce Accordion, a mixture-of-experts-based model that supports five inference budgets through fixed-prefix layer looping. Each budget follows a pre-defined execution path, while the minimum budget retains the unlooped model. Joint pretraining samples one budget per micro-batch and optimizes its next-token prediction loss without additional cross-budget consistency objectives. We further preserve these inference budgets through supervised fine-tuning and reinforcement learning, where parameter-group merging and budget-matched multi-teacher on-policy distillation integrate three domain experts into one model. After 3.6T tokens of pretraining, increasing inference depth improves Accordion-6B-A0.5B’s five-category macro-average across 37 tasks by 3.01 percentage points, with the middle budget capturing 89% of this gain using half as many additional layer executions. The final model surpasses each domain expert on its corresponding task-and-budget average, while achieving strong instruction-following performance across all budgets and retaining depth-dependent gains in mathematical reasoning and code generation.
[NLP-22] When Agents Fail: Model or Harness–Whos to Blame?
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在评估过程中，其执行框架（harness）与底层模型协同行为对基准测试结果的影响问题。传统评估往往将模型性能与框架表现混同，忽略了二者交互的复杂性。本文提出一个包含五项互补度量的综合评估框架：（i）有效完成率（Valid Completion Rate, VCR），衡量智能体在不依赖任务正确性的前提下完成任务并生成可评估结果的可靠性；（ii）模型-框架适配度（Model-Harness Fit），反映特定框架对某一模型任务达成的支持能力；（iii）跨模型适配度（Cross-Model Fit），评估框架在不同模型间的通用性；（iv）交互残差（Interaction Residuals），揭示组件级汇总所掩盖的成对执行中的优势与缺陷；（v）令牌负载（Token Load），量化执行过程中的总令牌消耗，包括失败尝试。研究对四种通用框架（OpenCode、Codex CLI、Pi CLI、DeepAgents）及两种原生模型-框架系统，在六个涵盖上下文管理、工具使用、长上下文规划、多模态交互与产物传递的基准上进行了全面评估，共覆盖111个实验设置，生成52,168条执行轨迹。结果显示，框架排名随模型和任务而异，任务性能与有效完成率常呈现不同最优组合，且相似任务得分可能对应截然不同的执行特征。其中，OpenCode在跨模型表现中排名第一，而DeepSeek V4 Flash Preview在共享四基准面板上表现出一致高的平均VCR。研究结论强调应重视模型-框架的协同评估，推动模型训练具备可迁移的框架使用能力，并倡导针对具体模型与任务定制化框架设计。
链接: https://arxiv.org/abs/2610.32483
作者: Shao Zhang,Hangfan Zhang,Yuanyuan Shi,Hao Li,Min-Le Su,Kangcong Li,Jingping Wang,Chunjiang Mu,Chen Zhang,Yang Chen,Bo Zhang,Shuyue Hu
机构: Shanghai Artificial Intelligence Laboratory
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Software Engineering (cs.SE)
备注: Preprint under review
Abstract:LLM agents combine models with execution harnesses, so agentic benchmark performance reflects their joint behavior. We argue that harnesses should therefore be evaluated through their interactions with underlying models. We develop a framework combining five complementary measures: (i) Valid Completion Rate (VCR), how reliably agents finish normally and deliver evaluable results, irrespective of task correctness; (ii) Model-Harness Fit, how well a harness supports task achievement through valid completion for a given model; (iii) Cross-Model Fit, a harness’s overall suitability across evaluated models; (iv) Interaction Residuals, pairing-specific execution strengths and weaknesses hidden by component-level summaries; and (v) Token Load, the token demands of agent execution, including failed attempts. We evaluate four general-purpose harnesses (OpenCode, Codex CLI, Pi CLI, and DeepAgents) with four core models, plus two fixed native model-harness systems, on six benchmarks covering context management, tool use, long-context planning, multimodal interaction, and artifact handoff. Including native references and stress conditions, 111 evaluation settings yield 52,168 trajectories. Harness rankings vary across models and benchmarks; task performance and valid completion can favor different pairings, and similar task scores can accompany distinct execution profiles. OpenCode ranks first in cross-model performance, while DeepSeek V4 Flash Preview maintains consistently high Mean VCR across harnesses on the shared four-benchmark panel. These findings support evaluating model-harness interactions, training transferable harness-use capabilities in models, and harness designs adapted to the model and task.
[NLP-23] Beyond Post-Hoc Rewards: Privileged Reasoning Exploration for Multimodal Embeddings
【速读】： 该论文旨在解决多模态嵌入模型在训练过程中缺乏对推理过程有效引导的问题。现有方法虽利用生成式能力通过自生成推理增强输入以提升嵌入质量，但传统检索监督仅指定输入间的相关性关系，未显式指导推理路径的生成，导致推理过程存在多种可能且难以确保与检索任务相关的语义线索被充分挖掘。其关键解决方案在于提出PILOT框架，将推理建模为潜在变量，并在共享的多模态大语言模型（Multimodal Large Language Model, MLLM）中构建先验（prior）与后验（posterior）两种推理模式：先验仅依赖输入信息，而后验则结合训练阶段的关联性证据以引导更广泛的推理探索。为将后验的扩展探索优势转化为推理阶段的实际性能提升，该框架进一步设计了一种重要性加权变分目标，并实现为先验-后验协同进化机制，在共享推理兼容奖励下同时优化两种模式，同时根据轨迹的实用性与先验可达性选择性地将后验路径迁移至先验。在78任务的MMEB-V2基准测试中，PILOT在2B和7B参数规模下均优于主基线，整体得分分别提升3.2和3.3点，尤其在视觉-文档检索任务中取得最大增益（+6.1和+6.4点）。研究结果表明，关系性监督的核心价值并非作为事后奖励，而在于塑造训练阶段推理探索的路径，从而显著提升推理增强型嵌入模型的检索性能。
链接: https://arxiv.org/abs/2610.32474
作者: Tianyu Yang,Yijun Tian,Lars Benedikt Kaesberg,Jan Philip Wahle,Bela Gipp,Terry Ruas,Daniel Kurzawe
机构: University of Göttingen(哥廷根大学); State and University Library of Göttingen(哥廷根大学与州立图书馆)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Advances in Multimodal Large Language Models (MLLMs) have driven substantial progress in multimodal embeddings. Recent methods exploit generative capabilities by augmenting embedded inputs with self-generated reasoning to produce more informative embeddings. However, conventional retrieval supervision for training embedding models specifies pairwise relevance relations without explicitly guiding the reasoning used for input augmentation. Such reasoning is neither directly observed nor uniquely determined by the input, allowing multiple plausible traces to introduce semantic cues of varying retrieval utility. Existing approaches largely optimize reasoning conditioned on the input alone, limiting the exploration of retrieval-relevant traces. We introduce PILOT, a framework that models reasoning as a latent variable and instantiates its prior and posterior distributions as distinct reasoning modes of a shared MLLM. The prior conditions only on the input, whereas the posterior uses additional training-time relational evidence to guide reasoning exploration. To translate the posterior’s broader exploration into inference-time improvements, we derive an importance-weighted variational objective and instantiate it as a prior-posterior co-evolution framework that improves both modes under a shared inference-compatible reward while selectively transferring posterior trajectories to the prior according to their utility and prior reachability. On the 78-task MMEB-V2 benchmark, PILOT outperforms the primary baseline across all 12 task groups at both 2B and 7B parameters, improving overall scores by 3.2 and 3.3 points, respectively, with the largest gains in visual-document retrieval (+6.1 and +6.4 points). Our results show that relational supervision is most valuable not as a post-hoc reward but as a signal for shaping how reasoning is explored in training reasoning-enhanced embedding models.
[NLP-24] Reading Between the Layers: Cross-Modal Alignment and Representational Hierarchy in Vision Transformers AACL2026
【速读】： 该论文旨在解决大视觉-语言模型中视觉编码器对文本及其语义的处理机制不明确的问题，特别是其是否具备类似阅读理解的能力。研究通过构建渲染文本与视觉概念之间的跨模态对齐，并结合探针分析（probing）与基于检索的分析方法，系统考察了基于Transformer的视觉编码器在字形、语法和语义层面的组织能力。关键发现在于：强跨模态映射仅在深层特征中出现，此时单模态物体与文本识别已基本完成，且表征从表层形式逐步演化为语义层面。这一层次化结构在对比学习训练的模型中最为显著，但大规模自监督模型及富含文本的数据也能在无显式语言监督的情况下部分实现类似行为。因此，解决方案的关键在于：模型规模、训练数据量以及文本丰富度的提升，可在一定程度上弥补缺乏语言监督所带来的局限性。
链接: https://arxiv.org/abs/2610.32463
作者: Sugam Karki,Raúl Vázquez,Yova Kementchedjhieva
机构: MBZUAI; University of Helsinki (赫尔辛基大学)
类目: Computation and Language (cs.CL)
备注: Accepted to AACL 2026 Findings
Abstract:While large vision-language models excel at optical character recognition, it remains unclear how their underlying vision encoders process text and its meaning, and whether they support anything resembling reading comprehension. We systematically study transformer-based vision encoders, using cross-modal alignment between rendered text and visual concepts together with probing and retrieval-based analyses as a controlled window into these broader capabilities. Our layer-wise analyses span orthographic, grammatical, and semantic organization, and reveal that strong cross-modal mapping emerges only in the deepest layers, after unimodal object and text recognition have already developed, coinciding with a progression from surface-level to semantic representations. Although this hierarchy is strongest in contrastively trained models, larger self-supervised models and text-rich training data partially support similar behaviors without explicit language supervision. Together, our results suggest that increasing model and training-data scale, as well as text-rich training data, can partially compensate for the absence of language supervision.
[NLP-25] Decoupling Patient Personas and Clinical Workflows: A Dual-Agent Approach to Realistic Medical Dialogue Generation
【速读】： 该论文旨在解决大规模获取真实预问诊医疗对话的难题，其核心挑战在于隐私与伦理限制导致真实对话数据难以获取，而现有基于大语言模型（Large Language Model, LLM）的对话生成方法多依赖提示工程或固定规则，难以在临床相关性与患者多样性之间实现有效平衡。为此，本文提出一种可控的双代理框架——CoRPS-Dial，其关键在于通过融合患者角色画像（patient persona）与符合临床指南的流程化工作流规则，构建由人格驱动的患者代理与流程引导的医生代理之间的交互机制，从而生成兼具临床合规性与自然对话特征的合成对话。随后引入后处理质量控制模块，确保角色一致性、流程完整性与文本连贯性。实验结果表明，CoRPS-Dial在冠心病领域的合成对话在信息覆盖度、临床合规性及对话自然度等方面均显著提升与真实对话的一致性，且下游微调任务验证了合成数据的有效性，证明该方法能够实现临床语境下可控制、高质量的医疗对话生成。
链接: https://arxiv.org/abs/2610.32443
作者: Qiyuan Li,Xiaomei Guo,Chao Gao,Haijiang Liu,Deyu Chen,Caicai Guo,Jinguang Gu
机构: Wuhan University of Science and Technology (武汉科技大学); Huazhong University of Science and Technology (华中科技大学); Vrije Universiteit Amsterdam (阿姆斯特丹自由大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Pre-consultation medical dialogues are essential for collecting comprehensive patient information, but are costly to obtain at scale due to privacy and ethical constraints. Existing LLM-based synthesis methods often rely on prompt engineering or rigid rules, limiting their ability to balance clinical relevance with realistic patient diversity. We propose CoRPS-Dial, a controllable dual-agent framework for pre-consultation medical dialogue synthesis that integrates patient personas with guideline-aligned clinical workflow rules. CoRPS-Dial generates dialogues through interactions between a persona-driven patient agent and a process-guided doctor agent, followed by post-hoc quality control for role consistency, workflow completeness, and textual consistency. Experiments on coronary heart disease demonstrate that CoRPS-Dial improves alignment between synthetic and real dialogues in information coverage, clinical compliance, and dialogue naturalness. Downstream fine-tuning experiments further confirm the utility of the generated synthetic data for medical dialogue tasks. These results highlight the effectiveness of CoRPS-Dial for clinically grounded and controllable medical dialogue synthesis.
[NLP-26] SCI-AGENT -EXAM: Benchmarking Agents on Scientific Research Tasks
【速读】： 该论文旨在解决当前大型语言模型（Large Language Model, LLM）代理在支持科学工作流时，其跨多样化科研活动的能力难以系统评估的问题。现有基准测试通常仅聚焦于特定科学能力或任务类型，无法全面刻画科学代理在真实研究流程中的优势与短板。为此，本文提出SCI-AGENT-EXAM，一个以工作流为导向的基准测试，包含11个科学领域中120项由专家精心设计的任务，涵盖7类基于科学工作流任务分类体系（Scientific Workflow Task Taxonomy, SWTT）的研究任务类型。每项任务均要求生成可执行产物（如代码和脚本）与文本产物（如报告与解释）。为评估异构输出，研究设计了一种混合评估协议：对可执行产物采用确定性程序化验证，对文本产物则使用经校准且由专家验证的生成式AI（Generative AI）评判模型进行评估。此外，研究还分析了执行轨迹中的进展过程与解法稳定性，以揭示代理达成最终结果的动态路径。实验对比10个前沿LLM代理发现存在显著性能差距：GPT-6 Astra取得最高综合得分80.9，但仅有26.0%的运行达到0.95的成功阈值；进一步分析揭示不同任务类型间的能力分布差异显著，且在执行效率与稳定性方面表现出强烈异质性。解决方案的关键在于构建覆盖多领域、多任务类型的综合性工作流基准，并结合程序化验证与可信的生成式评估机制，实现对科学代理能力的多维度、精细化评估。
链接: https://arxiv.org/abs/2610.32441
作者: Tianyun Zhong,Wei Wang,Jialin Li,Jianhong Tu,Liyao Li,Chuqiao Kuang,Tianshi Zheng,Songchi Zhou,Weiqi Zhai,Qile Jin,Jiajia Li,Jiasheng Zheng,Boxi Cao,Yaojie Lu,Hongyu Lin,Shiwei Ye,Bing Zhao,Hu Wei,Lin Qu,Xianpei Han
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language model (LLM) agents are increasingly capable of supporting scientific workflows, but their capabilities across diverse research activities remain difficult to assess systematically. Existing benchmarks often focus on selected scientific capabilities or task types, limiting a comprehensive characterization of scientific agents’ strengths and weaknesses in realistic research workflows. We introduce SCI-AGENT-EXAM, a workflow-oriented benchmark comprising 120 expert-curated tasks across 11 scientific fields and 7 research task types defined by the Scientific Workflow Task Taxonomy (SWTT). Each task requires both executable artifacts (e.g., code and scripts) and textual artifacts (e.g., reports and explanations). To evaluate the heterogeneous task outputs, we develop a hybrid evaluation protocol combining deterministic programmatic verification for executable artifacts with a calibrated, expert-validated LLM-based judge for textual artifacts. Beyond outcome assessment, we analyze progress and solution stability along execution trajectories to characterize how agents reach their final results. Experiments with 10 frontier LLM agents reveal substantial performance gaps: GPT-6 Astra achieves the highest overall score of 80.9, yet only 26.0% of its runs exceed the 0.95 success threshold. Further analyses uncover distinct capability profiles across scientific task types and substantial variation in execution efficiency and stability. Our code and data are available at this https URL.
[NLP-27] Profile-Driven User Simulation via Reinforcement Learning with Hierarchical Rewards for Task-Oriented Dialogues
【速读】： 该论文旨在解决高价值场景（如金融营销）中任务导向对话（Task-Oriented Dialogue, TOD）系统构建所面临的领域特定对话数据稀缺问题，该问题受隐私法规、标注成本及专家人力需求的制约。现有用户模拟方法在通用领域对话场景下表现有限，主要存在两大缺陷：一是对领域内业务流程与合规约束建模不足，二是未能充分捕捉语音交互中口语化、自然语言表达的特征。为此，本文提出ProfileSim——一种面向金融营销外呼场景的基于用户画像驱动的用户模拟框架。其核心创新在于联合建模用户画像（人口统计学特征与大五人格特质）与业务画像（从真实通话日志中逆向工程提取的意图），实现对模拟行为的细粒度控制。模型在真实语音对话数据上进行微调，并通过分层奖励机制的组相对策略优化（Group Relative Policy Optimization, GRPO）进一步精炼。结合集成标准操作流程（Standard Operating Procedure, SOP）约束与检索增强生成（Retrieval-Augmented Generation, RAG）的客服代理，ProfileSim在生成对话数据中表现出最优的口语自然度、最强的用户画像一致性以及对长尾意图的最低代理评分，同时生成语句更短且业务合规性相当。生成的数据可有效支持意图识别模型训练、代理评估及快速冷启动部署。
链接: https://arxiv.org/abs/2610.32435
作者: Juan Li,Fangshu Chang,Xinyi Jiang,Ling Huang,Yunzhang Chen
机构: Bairong Inc.
类目: Computation and Language (cs.CL)
备注:
Abstract:Building Task-Oriented Dialogue (TOD) systems for high-value scenarios such as financial marketing requires large volumes of domain-specific dialogue data, which privacy regulations, annotation costs, and the need for expert labor make hard to obtain. User simulation offers a remedy, but existing methods target general-domain dialogue and exhibit two key limitations: insufficient modeling of domain-specific business processes and compliance constraints, and inadequate capture of the colloquial, spoken-language character of voice interactions. We propose ProfileSim, a profile-driven user simulation framework for financial marketing outbound calls. It jointly models user profiles—demographics and Big Five personality traits—and business profiles—intents reverse-engineered from real call logs—enabling fine-grained control over simulated behavior. The base model is fine-tuned on real voice dialogue data and refined by Group Relative Policy Optimization (GRPO) with hierarchical rewards at the dialogue and utterance levels. Paired with a customer service agent integrating Standard Operating Procedure (SOP) constraints and Retrieval-Augmented Generation (RAG), ProfileSim achieves the highest colloquial naturalness, the strongest user profile adherence and the lowest agent score on long-tail intents among all evaluated user generators, while producing substantially shorter user utterances at comparable business compliance. The generated dialogue data is designed to support intent recognition training, agent evaluation, and rapid cold-start deployment.
[NLP-28] Cycle-of-Thought: Self-Evaluative Iterative Reasoning for Table Visual Question Answering in Document Images
【速读】： 该论文旨在解决视觉语言模型（Vision-Language Models, VLMs）在处理结构化内容（如表格）时存在的推理不一致问题，尤其是由空间错位和逻辑幻觉导致的错误。现有方法多依赖于架构改造或代价高昂的监督微调（Supervised Fine-Tuning, SFT），受限于大规模专家标注数据集和高算力需求，难以普及。为此，本文提出一种无需训练、仅在推理阶段执行的自修正框架——Self-RefineVQA，其核心在于模拟人类审校行为，使单一VLM通过迭代式“生成—批判—修正”的闭环反馈机制，在提示词引导下对自身推理过程进行重评估与修正。该方法不需额外标注数据或模型更新，显著提升了小模型（如2B–7B参数量级）在单语及跨语言表格问答任务中的表现，使其性能逼近远大于自身的大型模型。实验表明，尽管推理开销呈线性增长，但该框架在数据效率和总成本方面具有显著优势，有效规避了微调所需的人工标注瓶颈，为实现高效、可扩展的视觉-语言模型表结构推理提供了可行路径。
链接: https://arxiv.org/abs/2610.32423
作者: Somraj Gautam,Gaurav Harit
机构: IIT Jodhpur(印度理工学院焦德普尔分校); Rajasthan(拉贾斯坦邦), India(印度)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Recent advances in Vision-Language Models (VLMs) have significantly improved Visual Question Answering (VQA) performance, yet their reasoning over structured content such as tables remains inconsistent, often plagued by spatial misalignment and logical hallucinations. Existing approaches for improving table VQA typically rely on architectural modifications or expensive supervised fine-tuning (SFT), which impose high barriers to entry due to the need for large-scale, expert-annotated datasets and significant computational resources. In this work, we propose Self-RefineVQA, a training-free, inference-time self-correction framework that improves table VQA performance without additional labeled data or model updates. Inspired by human revision behavior, our method enables a single VLM to iteratively act as a generator, critic, and refiner, forming a closed self-feedback loop that corrects common reasoning errors through prompt-guided re-evaluation. Experiments on tabular VQA benchmarks demonstrate that Self-RefineVQA yields consistent accuracy gains in monolingual and cross-lingual understanding, with a particularly significant boost for small models, enabling lightweight models (e.g., 2B-7B parameters) to rival the performance of significantly larger counterparts. Furthermore, we provide a comparative complexity analysis demonstrating that while inference cost increases linearly, our framework offers superior data efficiency and a lower cumulative cost by eliminating the prohibitive annotation bottleneck of human labeling required for fine-tuning. Our results highlight the potential of prompt-based self-improvement as a scalable, accessible paradigm for reliable table reasoning in VLMs.
[NLP-29] Can LLM Annotators Match Human Consistency? Benchmarking Human Agreement and Optimizing Multi-Agent with Contrastive Knowledge EMNLP2026
【速读】： 该论文旨在解决当前基于大语言模型（LLM）的标注技术在实际应用中与真实标注流程脱节的问题。现有基准测试普遍采用单标签数据集，无法反映真实场景中多标注者对边界案例存在合理分歧的现象，且难以适应复杂、动态演进的任务指南，导致现有单智能体或工作流式标注系统缺乏足够的灵活性与鲁棒性。为弥合这一差距，论文提出ANT-A基准，这是一个涵盖7个跨领域、跨模态的真实世界多标注者基准，每个实例包含5个独立的人工标注及任务特定的详细指南，并引入基于Fleiss kappa的分布感知评估方法，以衡量模型标注结果与人类标注分布的一致性。其核心解决方案是ANT-OR多智能体标注框架，通过从正负反馈中提炼对比决策规则，动态优化提示（prompt），从而提升标注一致性。实验表明，ANT-OR在agree@=2准确率上比最强基线提升4.5个百分点，并在Fleiss kappa指标上与人类标注分布的对齐度仅相差2点，显著优于现有方法。
链接: https://arxiv.org/abs/2610.32415
作者: Xiaopeng Ke,Zhixin Wang,Xuebo Liu,Xinyu Ma,Xing Zheng,Bowen Song,Weiqiang Wang,Min Zhang
机构: ANT Group(蚂蚁集团); Harbin Institute of Technology (Shenzhen)(哈尔滨工业大学（深圳)）
类目: Computation and Language (cs.CL)
备注: Accepted to EMNLP 2026 Findings
Abstract:LLM-based annotation has advanced rapidly, yet existing benchmarks remain misaligned with real-world practice. Actual annotation often involves lengthy, task-specific guidelines, and different annotators may produce divergent but reasonable judgments, especially on boundary cases, where single-labeled dataset fail to reflect this disagreement. This gap also limits existing single-agent and workflow-based annotation systems, which often lack adaptability to complex, evolving guidelines. To bridge this gap, we introduce ANT-A Benchmark, a real-world multi-annotator benchmark spanning 7 datasets across domains and modalities, with task-specific guidelines and 5 independent human labels per instance. ANT-A enables a distribution-aware evaluation, using Fleiss kappa to assess how closely model annotations align with human labeling distributions. We further propose ANT-OR, a multi-agent annotation framework that refines prompts by distilling contrastive decision rules from positive and negative feedback. Experiments show that ANT-OR improves agreement@=2 accuracy by 4.5 points over the strongest baseline and achieves the highest alignment with human annotator distributions as measured by Fleiss kappa within 2 points. Code and benchmark are available at this https URL.
[NLP-30] A Broader Look into MLLM s Retrieval Heads
【速读】： 该论文旨在解决多模态大语言模型（Multimodal Large Language Models, MLLMs）在长输入中定位相关证据的能力如何随输入模态扩展的问题。现有研究将这一能力归因于“检索头”（retrieval heads），即一组稀疏的注意力头，负责将上下文信息传递至输出。然而，先前分析局限于有限模态与模型，且未深入探讨如何有效利用检索头提升模型性能。本文以音视频输入作为新增模态，构建了语义一致但呈现方式不同的成对输入（语音或屏幕显示），在保持证据内容不变的前提下，发现检索头在新模态下依然保持稀疏性，并在不同模态间高度共享。跨18个语言、视觉-语言及全模态检查点的实验表明，全模态模型基本保留其基础模型的检索头结构；移除这些共享头会导致几乎完全丧失音视频检索能力。尽管跨模态共享的检索头不足全部注意力头的2%，但其缺失仍显著损害文档理解、空间感知和视觉数学能力。基于此，论文提出将检索头作为可优化目标，通过引入排序目标（ranking objective）对检索头的注意力进行训练，使Qwen3-VL-8B在MMDocIR上的页面召回率@1提升6.03点，并在三个模型族中均改善了文档检索表现。因此，功能性的注意力头不仅可作为解析多模态模型内部机制的“透镜”，也可成为提升模型性能的有效优化对象。
链接: https://arxiv.org/abs/2610.32411
作者: Jia Wang,Yuqing Li,Jiangnan Li,Zheng Lin,Mo Yu
机构: Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所); School of Cyber Security, University of Chinese Academy of Sciences(中国科学院大学网络空间安全学院)
类目: Computation and Language (cs.CL)
备注: work in progress
Abstract:Locating relevant evidence in long inputs is a core capability of Multimodal Large Language Models. Prior studies attribute this capability to retrieval heads, a sparse set of attention heads that transfer information from the context to the output. However, these analyses are confined to a limited set of modalities and the corresponding models. Moreover, how retrieval heads can be leveraged to enhance the model calls for further exploration. In this paper, we ask whether this mechanism expands together with the input modalities of the model. Taking audio-visual input as a newly added modality, we construct paired inputs in which the same answer is either spoken or shown on screen. With the evidence content fixed, retrieval heads remain sparse in the new modality and are shared to a large extent across modalities. Across 18 language, vision-language, and omni-modal checkpoints, omni-modal models largely preserve the retrieval heads of their backbones. Ablating these heads eliminates nearly all audio-visual retrieval. The heads shared across modalities make up fewer than 2% of all heads, yet ablating them also degrades document understanding, spatial perception, and visual mathematics. Building on these findings, we turn retrieval heads into a target of optimization by training their attention with a ranking objective. This raises the page Recall@1 of Qwen3-VL-8B on MMDocIR by 6.03 points and improves document retrieval across three model families. Functional attention heads can thus serve both as a lens on multimodal models and as a target for improving them.
[NLP-31] Acoustic and Lexical Signals for Native Language Identification of Language Learners
【速读】： 该论文旨在解决第二语言学习者母语（L1）识别问题，传统方法主要依赖语音的声学特征，而对语音转录文本中蕴含的母语痕迹关注较少。本文提出一种结合语音信号与对应转录文本的多模态研究范式，通过对比基于WavLM的音频L1分类器与基于零样本GPT-4o的文本分类器在CEFR平衡评估集上的表现，探究不同语言熟练度水平下两类模型的性能差异。其解决方案的关键在于：利用多模态信息互补性——音频模型在初级（A1）水平表现更优，而文本模型在中级至高级（A2–C1）水平凭借完整转录文本展现出更强判别能力；同时发现文本分类准确率随熟练度提升而显著上升，而声学模型性能相对稳定；此外，文本模型对输入长度敏感，受限于短文本时性能大幅下降，而音频模型对输入长度不敏感，具备更强的鲁棒性。研究揭示了语音与文本表征在母语识别中的差异化优势及其与学习者熟练度和数据量的关系，为多模态语言分析提供了重要实证依据。
链接: https://arxiv.org/abs/2610.32382
作者: Ahmet Yavuz Uluslu,Mengjie Qian,Kate Knill,Mark Gales
机构: ALTA Institute, Department of Engineering, University of Cambridge (ALTA 研究所，工程系，剑桥大学); Department of Computational Linguistics, University of Zurich (计算语言学系，苏黎世大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Accepted for publication at NLP4CALL
Abstract:Spoken native language (L1) identification for second language learners has traditionally relied on acoustic attributes of speech. Comparatively little attention has been paid to analysing speech transcripts for L1 traces. In this paper, we present a study of L1 identification using both speech and corresponding transcripts from a spoken learner English corpus. We compare a WavLM-based audio L1 classifier with a zero-shot GPT-4o classifier operating on transcripts and analyse their predictions across proficiency levels using a CEFR-balanced evaluation set. Our experiments show that the audio classifier outperforms the transcript-based approach overall when full transcripts are available, although their relative strengths vary with learner proficiency. Audio performs better at A1, whereas text performs better from A2 to C1 with full transcripts. Transcript-based accuracy generally increases with proficiency, while acoustic performance remains comparatively stable. Restricting transcript length substantially reduces text-based accuracy, whereas audio is less sensitive to input length and performs relatively well with limited input. These findings provide insights into L1 identification from acoustic and textual representations of learner speech and how performance varies with proficiency and the amount of available language material.
[NLP-32] Why Full-Duplex Speech Models Collapse in Long Conversations
【速读】： 该论文旨在解决全双工语音模型（full-duplex speech models）在长时间连续对话中面临的上下文窗口耗尽与对话质量退化问题。尽管这些模型具备同时听与说的能力，适合自然、开放式的对话场景，但因其每帧生成一个词元（token），即使在倾听时也持续消耗上下文空间，导致在数十分钟的长对话中迅速耗尽上下文容量。现有基准测试仅评估单轮或少量对话回合，无法揭示长期对话中的系统性失效问题。为此，作者提出全双工长时程基准测试（Full-Duplex Long-Horizon Benchmark, FD-LHB），对五种近期全双工模型进行长时重放对话和与大语言模型（LLM）实时交互的评估。结果表明，所有模型最终均出现性能退化或上下文耗尽。其中，保留完整历史的模型虽未达上下文上限，却仍发生语义退化；采用滑动窗口机制的模型则陷入重复、沉默及无意义输出。研究进一步诊断出两个关键机制：无意义输出与沉默源于“初始化锚点驱逐”（Initialization-Anchor Eviction），即滑动窗口丢弃首个键值（KV）条目；而重复现象则受模型自身近期输出的反馈强化。基于此诊断，提出两项无需重新训练的简单改进策略：初始化锚点固定（Initialization-Anchor Pinning, IAP）以保留初始键值条目，以及全双工重复惩罚（Duplex Repetition Penalty, DRP）以抑制自我重复并防止长期沉默。二者协同作用，在不重新训练的前提下，显著缓解长达一小时的重放对话中的对话崩溃现象，并支持实时运行。
链接: https://arxiv.org/abs/2610.32349
作者: Yeonju Kim,Bella Godiva,Se Jin Park,Yong Man Ro
机构: KAIST(韩国科学技术院); Kyung Hee University(庆熙大学)
类目: Computation and Language (cs.CL); Sound (cs.SD)
备注: Preprint. 31 pages, 16 figures, 18 tables
Abstract:Full-duplex speech models listen and speak at the same time, which makes them promising for natural, open-ended conversation. But can they sustain a coherent conversation for tens of minutes? Because they generate a token every frame, even while listening, they exhaust their context windows within minutes. Yet existing benchmarks evaluate only single turns or a few conversational rounds, too short to expose the long-horizon problem. We introduce the Full-Duplex Long-Horizon Benchmark (FD-LHB), which tests five recent full-duplex models on long replayed conversations and in live conversation with an LLM partner. Every evaluated model eventually either degrades or exhausts its context. Models that keep the full history often degrade before their context runs out, while sliding-window models still fall into repetition, silence, and gibberish. For sliding-window models, we identify two contributing mechanisms. Gibberish and silence are linked to Initialization-Anchor Eviction, where the sliding window discards the first KV entry. Repetition is also reinforced by the model’s own recent outputs. Guided by this diagnosis, we apply two simple fixes: Initialization-Anchor Pinning (IAP) keeps that first entry, and Duplex Repetition Penalty (DRP) discourages echoing while guarding against prolonged silence. Together, they support the diagnosis and, without retraining, substantially reduce conversational collapse in replays of up to an hour while running in real time.
[NLP-33] Beyond Final Outcomes: Quantifying Claim-Level Information Dynamics in Multi-Agent LLM Interaction
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）驱动的多智能体系统在协作过程中，信息交互的动态演化机制被忽视的问题。传统多智能体评估仅关注最终任务结果，而忽略了交互过程中信息的生成、传播与衰减等细粒度过程。为此，论文提出RIVER框架，其核心在于将智能体输出分解为原子级命题（atomic claims），从三个互补维度——收敛性（convergence）、探索性（exploration）和保留性（retention）——对信息动态进行量化评估：收敛性衡量跨智能体命题的一致性与立场对齐程度；探索性反映语义广度及对源信息的拓展能力；保留性则追踪原始信息的保真度以及新命题在多轮交互中的持续存在情况。研究通过40个受控新闻生成案例对比五种交互策略发现，重复聚合式交互虽促进信息收敛但压缩了信息空间，而分布式传递则产生更广泛但共享度较低的信息流；此外，特定方法约束会增强探索性，但削弱收敛性与保留性。研究结论表明，群体信息动态应由底层信息流动机制解释，而非仅依赖下游任务表现。
链接: https://arxiv.org/abs/2610.32333
作者: Haocheng Jiang,Herun Wan,Yongyuan Peng,Minnan Luo
机构: Xi’an Jiaotong University (西安交通大学)
类目: Computation and Language (cs.CL)
备注: 23 pages, 3 figures. Code and data: this https URL
Abstract:When LLM-powered agents interact, information is repeatedly shared, expanded, and forgotten, yet multi-agent evaluation typically focuses on final task outcomes. We study such interaction as claim-level information dynamics and propose RIVER, which decomposes agent outputs into atomic claims and evaluates three complementary dimensions: convergence, exploration, and retention. Convergence captures cross-agent claim correspondence and stance alignment; exploration captures semantic breadth and extension beyond the source; and retention tracks source preservation and the persistence of new claims across stages and lineages. We evaluate five interaction strategies on 40 controlled news-generation cases. Repeatedly pooled interaction promotes convergence but narrows the information space, whereas distributed relay yields broader, less shared information. Further comparisons show that method-specific constraints increase exploration while weakening convergence and retention. Our findings suggest that collective information dynamics are better explained by underlying information-flow mechanisms than by simple downstream performance.
[NLP-34] 4DMemory: Benchmarking Long-Horizon Spatiotemporal Memory with Harness-Augmented Agents
【速读】： 该论文旨在解决现有模型在处理长时序、多模态信息时难以保持细粒度时空记忆，从而制约其对长期事件进行推理与预测的问题。当前主流基准测试主要聚焦于短中等上下文任务，无法有效评估模型在跨越数小时至数年的复杂时空序列中的表现。为此，研究提出LongVideo数据集与4DMemoryBench评测基准，涵盖约900小时视频及超过5,000个经人工审核的跨模态（视觉、文本、时空）推理问题，覆盖多样化的上下文长度。其核心解决方案是构建4DMemory Harness，该系统集成四种互补能力、22个以上专用工具模型，并采用针对不同模型优化的检索增强生成（Retrieval-Augmented Generation, RAG）机制。实验表明，尽管独立的基础模型在长时序任务中表现不佳，但引入4DMemory Harness后，13个主流模型的性能均提升超过50%；而与以往基准相比，相同模型在4DMemoryBench上的性能下降超过10%，凸显了新基准的挑战性。该工作为发展具备长期时空推理能力的智能体提供了关键评测工具与技术路径。
链接: https://arxiv.org/abs/2610.32332
作者: Yanhao Jia,Yi Zhu,Yiran Zhong,Weigao Sun,Feida Zhu,Pengkun Jiao,Sihan Yang,Ming Ma,Steven Hoi,Erik Cambria
机构: Nanyang Technological University (南洋理工大学); Token Foundry, Alibaba; The Chinese University of Hong Kong (香港中文大学); Institute of Neuroscience, Chinese Academy of Sciences (中国科学院神经科学研究所); University of Chinese Academy of Sciences (中国科学院大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multimedia (cs.MM)
备注: 22 pages, 5 figures
Abstract:Humans efficiently process multimodal streams spanning hours to years, building persistent, fine-grained spatiotemporal memories that support reasoning about and predicting long-horizon events. In contrast, existing benchmarks focus on short- and medium-context tasks, and most foundation models struggle to preserve fine-grained information over long horizons. To evaluate and improve long-horizon spatiotemporal reasoning, we introduce the LongVideo dataset and the 4DMemoryBench. LongVideo comprises about 900 hours of video and over 5,000 manually reviewed questions covering visual, textual, and spatiotemporal reasoning across diverse context lengths. We further develop the 4DMemory Harness, which integrates four complementary skills, 22+ specialized tool models, and a model-specific retrieval-augmented generation system. Evaluating 13 leading models, we find that standalone foundation models struggle with long-horizon tasks, while the 4DMemory Harness improves all of them by over 50%. Compared to previous benchmark, the same models exhibit a performance drop of more than 10% on 4DMemoryBench. We hope 4DMemory will facilitate the development of spatiotemporal agents capable of addressing complex, long-horizon tasks.
[NLP-35] Many Perturbations Shared Failure Dynamics: Tracing Direction-Sensitive Perturbation Influence in Language Models
【速读】： 该论文试图解决的核心问题是：为何某些任务保持性的文本重写（task-preserving rewrite）会导致正确答案变为错误，而更大规模的内部表征变化却仍能维持原有答案的正确性？其解决方案的关键在于提出一种方向敏感、读出相对的语义脆弱性（semantic fragility）理论框架，揭示不同语言变换虽在几何结构与读出关系上具有相似性，但未必共享一个独立于输入的失败方向。通过引入“清洁决策保留”（clean decision reserve）、“表征变换”（representation transformation）和“响应对齐”（response alignment）三者互补的描述机制，该研究构建了一个精确的纵向-横向分解模型，识别出传统标量传输方法所忽略的方向性信息。基于此，一个紧凑的“保留-传输”（reserve-and-transmission）模型成功预测了Qwen3-32B与Gemma 4 31B中未见家族及样本的失败模式，并在ReClor数据集上实现参数依赖性但有效的迁移。匹配干预实验表明，方向性蕴含超越标量几何的因果信息；自然失败案例中的注意力组件测试与后续答案顺序控制进一步支持原始配对方向优于随机或同字母选项的捐赠方向。然而，失败现象极少跨选项顺序持续存在，且当前顺序修正依赖重新计算的干净参考，说明预测规律性与上下文依赖的方向效应共存。数值路径积分虽可减小固定上下文内的有限区域近似误差，但仍无法消除与自然重写效应之间的显著偏差。此外，固定思维链（CoT）提示研究扩展了该理论至记录推理中未被明示的影响：读出关联可跨领域传递，尽管正均值因果标准失效，符号效应预测仍为事后分析。综上，这些结果支持保留、变换与响应之间存在共享关系，同时区分局部因果敏感性与对自然失败的完整解释。
链接: https://arxiv.org/abs/2610.32309
作者: Zhibo Hu,Chen Wang,Yanfeng Shu,Liming Dong,Hye-young Paik,Liming Zhu
机构: The University of New South Wales (新南威尔士大学); CSIRO technology (澳大利亚联邦科学与工业研究组织)
类目: Computation and Language (cs.CL)
备注: 42 pages, 6 figures
Abstract:Why can a task-preserving rewrite turn a correct answer into a failure while a larger internal change leaves the answer intact? We develop a direction-sensitive, readout-relative account of semantic fragility: heterogeneous language changes can share geometric and readout relationships without sharing an input-independent failure direction. Clean decision reserve, representation transformation, and response alignment provide complementary descriptions of this process. An exact longitudinal-transverse decomposition identifies directional information discarded by scalar transmission. A compact reserve-and-transmission model predicts failures across held-out families and items in Qwen3-32B and Gemma 4 31B, with useful but parameterization-dependent transfer to ReClor. Matched interventions show that orientation carries causal information beyond scalar geometry. Among naturally failed rewrites, attention-component tests on disjoint original items and subsequent answer-order controls favor the natural paired direction over tested random and same-answer-letter donor directions. Yet failures rarely persist across option orders, and current-order corrections use recomputed clean references. Shared predictive regularity thus coexists with context-dependent directional effects. Numerical path integration reduces, but does not eliminate, finite-patch approximation error within a fixed context; a larger discrepancy from the natural rewrite effect remains. A complementary fixed-CoT cue study extends this account to influence unacknowledged in recorded reasoning: readout associations transfer across domains, although the positive-mean causal criterion fails and signed-effect prediction remains post hoc. Together, these results support shared relations among reserve, transformation, and response while distinguishing local causal sensitivity from a complete explanation of natural failure.
[NLP-36] DMAIC-Bench: From Self-Reflection to Self-Restraint for Agents in Multi-Stage Analytical Tasks
【速读】： 该论文旨在解决大语言模型（LLM）代理在多阶段分析任务中缺乏对中间证据进行有效判断以决定是否继续、转向或终止的能力问题，尤其关注代理在面对证据时能否实现“自我约束”（self-restraint），即根据证据合理地停止后续计算而非盲目执行。现有基准普遍聚焦于任务完成度或输出质量，未能有效评估这一关键行为能力。为此，作者提出DMAIC-Bench，基于精益六西格玛（Lean Six Sigma）的明确统计前提，构建了涵盖15个工业案例、4个行业领域的97个种子任务，并引入成对反事实变体，在同一工作流位置要求相反决策，从而区分出有依据的约束与随意中断。通过私有评分器评估执行轨迹、独立专家评审最终结论，研究发现13个来自6家厂商的模型在匹配的GO决策上执行路由准确率达92%，但在NO-GO决策上仅为19%，任何被禁止的下游计算均计为路径错误；执行轨迹显示，代理常在报告正确诊断后仍执行被结果所排除的分析。专家盲评进一步表明，在达到NO-GO决策点时，约47%的报告基于无效证据做出判断（置信区间31%–61%）。显式允许停止虽对部分模型有所帮助，但无法一致恢复对GO与NO-GO分支的正确决策。此外，该框架在批量核糖核酸测序（RNA-seq）和病例对照全基因组关联研究（GWAS）等生命科学任务中亦展现出评估证据依赖性决策的普适性。
链接: https://arxiv.org/abs/2610.32305
作者: Shan Peng,Shuaiqi Cheng,Miao Peng,Chenyi Zi,Jia Li
机构: Xiwei Technology (Guangzhou) Co., Ltd.(广州西维科技有限公司); The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 54 pages, 14 figures. Code and data: this https URL
Abstract:Multi-stage analytical tasks require LLM agents to use intermediate evidence to decide whether to continue, redirect or stop. Self-reflection, in which an agent critiques its output, is widely studied, yet a critique does not show whether evidence changed the next action. We call that capability self-restraint. Existing agent benchmarks largely measure task completion or output quality and leave it untested. DMAIC-Bench evaluates it with 97 seeded tasks across 15 industrial cases in four sectors, built around Lean Six Sigma’s explicit statistical prerequisites. Paired counterfactual variants require opposite actions at the same workflow position, distinguishing warranted restraint from indiscriminate stopping. We assess execution traces with a private scorer and final claims through independent expert review. Across 13 models from six providers, execution-level routing accuracy averages 92% on matched GO decisions and 19% on NO-GO decisions, counting any forbidden downstream computation as a route error. Execution traces show agents reporting correct diagnostics before performing analyses that those results disallow. Separately, blinded expert review of the 11 audited agents estimates that reports act on invalid evidence at 47% of reached NO-GO decisions, with an interval from 31% to 61%. Explicit stopping permission helps some models without consistently restoring correct decisions at both GO and NO-GO forks. Sixteen life-sciences tasks in bulk ribonucleic acid sequencing (RNA-seq) and case-control genome-wide association studies (GWAS) show that the same framework can evaluate evidence-dependent decisions beyond industrial workflows. Project code and data are available at: this https URL.
[NLP-37] “Not” is Not Lost in Translation: A Causal Analysis of Negation Representations in Multilingual LLM s
【速读】： 该论文旨在解决多语言大模型（Multilingual Large Language Models, LLMs）在不同语言中对否定（Negation）的表征是否共享底层语义表示这一核心问题。尽管各语言在否定表达形式上存在显著差异，但现有模型仍能稳健地进行跨语言否定翻译，这暗示其内部可能存在跨语言统一的否定表征机制。研究的关键在于通过因果干预（causal interventions）方法，在合成对比数据集和基于FLORES-200构建的否定翻译数据集上，系统检验六种开放模型在英语、法语、越南语及汉语中的否定表示特性。研究发现，否定表征具有高度可操控性且在多种语言间存在显著共性：从一种语言中提取的对比激活方向可有效诱导或消除其他语言中的否定含义，且不同语言间的表征方向高度对齐。此外，研究揭示否定主要由稀疏特征（Sparse Features）编码——每个模型仅需一个稀疏自编码器（Sparse Autoencoder, SAE）特征即可实现跨语言极性控制，并对否定输入表现出选择性响应。更进一步，密集与稀疏干预策略均能泛化至四种额外语言及FLORES衍生句子。该工作为理解多语言模型中否定的通用语义表示提供了实证支持，并公开了数据与代码以促进后续研究。
链接: https://arxiv.org/abs/2610.32281
作者: Giang Son Nguyen,Wray Buntine,Dung D. Le,Wenya Wang
机构: Nanyang Technological University (南洋理工大学); VinUniversity (Vin大学); Monash University (莫纳什大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 29 pages, 20 figures
Abstract:Negation occurs in every language, yet its surface realization differs drastically. Despite this, multilingual large language models (LLMs) translate negation robustly, raising the question of whether they share an underlying representation of negation across languages. We study this question in six open models across English, French, Vietnamese, and Chinese, using causal interventions on a synthetic contrastive dataset and a negation translation dataset derived from FLORES-200. Across models, we find strong evidence that negation representations are causally controllable and largely shared across languages. In particular, we show that a contrastive activation direction extracted from one language can induce or remove negation in the others, and directions extracted from different languages are substantially aligned. We further find that negation is represented in sparse features: a single sparse autoencoder (SAE) feature per model suffices to control polarity across languages, while responding selectively to negated inputs. Finally, both dense and sparse interventions generalize to four additional languages and FLORES-derived sentences. We release data and code to support future work: this https URL.
[NLP-38] Do Halting Heads Really Save Compute in Looped Language Models?
【速读】： 该论文旨在解决循环语言模型（Looped Language Models）中因固定迭代次数导致的计算资源浪费问题，其核心挑战在于如何有效实现基于动态停止机制的计算优化。现有方法通过引入“终止头（halting head）”在每轮迭代后预测当前令牌是否应停止，从而允许简单令牌提前终止以节省算力。然而，研究发现两个关键问题：其一，传统训练方式下，各迭代阶段混合训练的终止头倾向于预测令牌仅需不到一半的迭代即可完成，但标准停止规则仍迫使令牌执行几乎所有迭代，原因是训练过程未对实际停止的单次迭代进行显式评分；若改为对每一迭代独立评分，则虽可改善这一现象，但需施加强度较高的早期停止奖励项才能生效。其二，当前文献普遍以“每个令牌的平均迭代次数”作为加速指标，但实际生成场景中，批次内所有序列需等待最慢令牌完成，此时无终止头的模型在相同延迟下的表现几乎等效于有终止头的模型，因此平均迭代数可能高估了真实计算收益。为此，论文提出应报告实际运行的迭代次数、终止阈值与批量大小，并在训练终止头前执行一项预测试，以量化其潜在增益上限，从而更准确评估优化效果。
链接: https://arxiv.org/abs/2610.32246
作者: Tsung-Han Wu,Minsuk Heo,Kenny Tan
机构: Apple(苹果)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 33 pages, 6 figures
Abstract:Looped language models run each token through the same layers several times, and each run is one iteration. To save compute, an extra layer, the halting head, gives after each iteration the probability that the token stops there, so that easy tokens can stop early. Pretraining models from scratch, up to 1.3B parameters, we find two problems. First, heads trained on a mix of all iterations predict that tokens need under half the iterations, yet under the standard stopping rule tokens run almost every iteration, because this training never scores the single iteration where a token stops. Scoring every iteration instead fixes this, but in our runs it works only with a strong penalty favoring early stops. Second, savings are commonly reported as an average number of iterations per token, but when sequences are generated in batches that run until their slowest token stops, a model without a head is almost as good at that cost. The reported average can therefore overstate the saving. We recommend reporting the iterations tokens actually run, with the stopping threshold and batch size, and give a test, run before training a head, that bounds its gain.
[NLP-39] DIAL-S2ST: Learning Latency Control for Simultaneous Speech-to-Speech Translation
【速读】： 该论文旨在解决同步语音到语音翻译（simultaneous speech-to-speech translation, S2ST）模型在推理时难以实现灵活延迟控制的问题。现有方法通常依赖于成对的目标数据或复杂的奖励机制来指导早期输出，但这些方式在实际应用中存在数据稀缺或训练成本高的局限性。为此，本文提出DIAL-S2ST，一种后训练（post-training）方法，通过从前瞻预测（lookahead predictions）中进行自蒸馏（self-distillation），在不使用配对目标数据或奖励计算的情况下，为早期输出提供有效监督。其关键创新在于：通过限制大规模文本蒸馏损失的影响，缓解模仿差距（imitation gap）带来的偏差，从而更准确地引导翻译行为；同时，采用跨尺度联合训练的单一低秩适应（LoRA）适配器，并结合匹配的前瞻引导信号，使学习到的延迟控制策略可在推理阶段灵活调整。在Hibiki数据集上的实验表明，仅需10小时源语音即可完成适配，实现了高质量与可调延迟之间的良好权衡。
链接: https://arxiv.org/abs/2610.32214
作者: Jeongsoo Choi,Junseok Ahn,Joon Son Chung
机构: KAIST(韩国科学技术院)
类目: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:We propose DIAL-S2ST, a post-training method for learning inference-time latency control in simultaneous speech-to-speech translation models. Our approach reuses pretrained translation capabilities through self-distillation from lookahead predictions, while limiting the influence of large text distillation losses to account for the imitation gap. This provides supervision for earlier emission without paired target data or reward computation. To make the learned timing adjustable at inference, we jointly train a single LoRA adapter across scales with matched lookahead guidance. Experiments on French-to-English translation with Hibiki demonstrate that only 10 hours of source speech enable this adaptation, providing flexible latency control with a favorable quality-latency trade-off.
[NLP-40] Head Dimension Matters in Language Models for Long-Context Retrieval
【速读】： 该论文旨在解决在长上下文处理任务中，注意力机制中头维度（head dimension）对信息检索性能的影响问题，尤其关注在固定模型规模下，分配有限参数预算给更少但更宽的注意力头是否能提升长上下文信息检索能力。其核心解决方案在于揭示：在保持模型总参数量不变的前提下，增加头维度能够显著提升长上下文检索任务的性能，关键原因在于更宽的头具有更高的“检索头”比例，使更多参数被用于支持检索功能；此外，即使在注意力模块的投影权重冻结的情况下，更宽的头仍能提升检索性能，表明其不仅促进注意力模块内部的优化，还推动了嵌入层和前馈网络（MLP）等组件的有效适应。这说明头维度的扩大通过增强模型对长序列上下文的表征能力，实现了更高效的长程信息捕获与整合。
链接: https://arxiv.org/abs/2610.32213
作者: Seiya Tokui,Shoichiro Yamaguchi,Sosuke Kobayashi
机构: Preferred Networks, Inc.(Preferred Networks, Inc.); Tokyo, Japan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Multi-head attention retrieves and combines context representations through query–key interactions and value aggregation, with head dimension constraining the per-head representation space. In long-context processing, it remains unclear whether allocating a fixed parameter budget to fewer, wider heads improves information retrieval. In this paper, we investigate how head dimension affects the performance of long-context retrieval tasks. Our empirical studies reveal that pretraining loss and short-context non-retrieval task performance are not affected much by head dimension, while long-context retrieval performance improves with wider heads at fixed model sizes. We find that models with wider heads have a higher proportion of retrieval heads, associating more parameters with retrieval. Our ablation studies also show that the retrieval performance improves with wider heads even when the projection weights in attention modules are frozen during long-context continual pretraining, suggesting that wider heads not only benefit updates in attention modules but lead to more effective adaptation of components such as embeddings and MLPs.
[NLP-41] SOPO: Socratic Guided Policy Optimization for Span-Level Hallucination Detection
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在生成过程中存在的幻觉（hallucination）问题，尤其聚焦于跨段落检索增强生成（RAG）场景下的细粒度幻觉检测任务。现有基于可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR）方法在处理跨度级幻觉检测时面临三大挑战：定位模糊性（localization ambiguity）、困难样本上的奖励稀疏性（reward sparsity）以及生成长度失控导致的冗余问题。为应对上述挑战，论文提出了一种名为苏格拉底引导策略优化（Socratic Guided Policy Optimization, SOPO）的多轮强化学习框架。其核心创新在于引入苏格拉底式反馈机制（Socratic feedback mechanism），即由教师模型在推理过程中通过间接提示（indirect hints）引导学生模型的推理轨迹，这些提示基于真实标注信息推断而来，有效提升了对复杂样本的推理效率，缓解了奖励稀疏性问题；同时，结合基于轮次的优势缩放（Turn-based Advantage Scaling）策略，抑制模型的冗余输出倾向，促使模型发展出高效且自主的推理能力。实验结果表明，SOPO（4B）在RAGTruth基准上达到领先性能，超越多个主流专有模型（如GPT-5）和标准强化学习基线，在四个跨领域基准上平均得分高于GRPO，而与CAPO相比则呈现混合证据支持。
链接: https://arxiv.org/abs/2610.32212
作者: Jiawei Dong,Zilong Bai,Pengtian zhu,Peng Zhou
机构: Patsnap Pte. Ltd., Suzhou, China(帕特森科技有限公司，苏州，中国)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 24 pages, 11 figures
Abstract:Large Language Models (LLMs) have gained widespread adoption in various natural language processing tasks, but suffer from hallucination issues where they generate unfaithful or inconsistent content. While Reinforcement Learning with Verifiable Rewards (RLVR) has shown promising applications in reasoning tasks, its application to span-level hallucination detection is challenged by localization ambiguity, reward sparsity on hard examples, and uncontrolled length inflation. To address these issues, we introduce Socratic Guided Policy Optimization (SOPO), a multi-turn RL framework for robust hallucination detection. The key innovation of SOPO is a Socratic feedback mechanism, where a teacher model guides the reasoning trajectories during rollout by providing indirect hints inferred from the ground truth. This teacher-guided approach enables the model to generate more effective reasoning trajectories for hard examples, thus alleviating reward sparsity. Furthermore, we incorporate Turn-based Advantage Scaling to penalize verbosity bias, encouraging the model to develop efficient and autonomous reasoning capabilities. Empirical results on RAGTruth demonstrate that SOPO (4B) achieves state-of-the-art performance, surpassing leading proprietary models (e.g., GPT-5) and standard RL baselines, while obtaining higher reported mean scores than GRPO on four out-of-domain benchmarks, with mixed evidence against CAPO. We release our training and evaluation code alongside the 1.7B and 4B SOPO models at this https URL.
[NLP-42] SWE-Shift: Evaluating Coding Agents in Evolving Workspaces
【速读】： 该论文旨在解决现有代码生成基准测试（coding benchmarks）在评估编码智能体（coding agents）时忽略实际开发过程中任务与环境动态变化的问题。传统基准假设任务和执行环境在整个过程保持不变，而现实中代码、依赖项、运行时状态及需求可能在智能体执行过程中发生变更。为此，论文提出SWE-Shift基准，用于评估智能体在执行中遭遇任务相关更新后的鲁棒性。其核心解决方案在于设计“清洁（Clean）”与“偏移（Shifted）”两种连续执行路径：前者无干预，后者在特定时间点接收任务相关的更新信息。成功标准不仅包括完成当前任务，还需保留有效工作成果、恢复适用状态并更新证据。该基准涵盖12类变化类型共120个场景，每个场景在9个预设干预位置进行评估。实验表明，17个被测编码系统在偏移路径上的成功率均低于清洁路径，如GPT-6 Astra从67.35%降至24.79%，且所有系统在工具/会话变更或过期证据等情形下均未实现配对成功（paired success），说明模型难以在环境变化后维持上下文一致性与状态完整性。这揭示了当前编码智能体在面对动态更新时，虽可在静态任务上表现良好，却普遍缺乏适应性与状态保持能力。
链接: https://arxiv.org/abs/2610.32196
作者: Victor Shea-Jay Huang,Rongqing Li,Zhangyue Yin,Kelong Mao,Yaqi Zhao,Xinyuan Wang,Yao Yin,Lirui Luo,Renrui Zhang,Hongsheng Li
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 47 pages. Code: this https URL
Abstract:Repository-level coding benchmarks largely assume that tasks and environments remain unchanged throughout execution. In software development, code, dependencies, runtime state, and requirements can change while an agent is still working. We introduce SWE-Shift, a benchmark for evaluating coding agents under such mid-execution changes. From the same saved history and workspace, a Clean continuation proceeds without intervention, while a matched Shifted continuation receives a task-relevant update. Success requires completing the current task, preserving still-valid work, and satisfying the applicable state and evidence checks. The benchmark contains 120 scenarios across 12 change families, each evaluated at nine scheduled intervention positions. We evaluate 17 coding-agent systems, all of which have lower Shifted than Clean success. GPT-6 Astra drops from 67.35% to 24.79%, with 20.34% paired success. Tool/session changes and stale evidence yield zero paired success across all systems. Late-position paired success is lower for some systems and higher for others. These results show that success on the unchanged task can coexist with failures to preserve work, recover state, or refresh evidence after an update.
[NLP-43] Collection-Date Assignment Redirects Answers in Temporal QA
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在时间问答（Temporal Question Answering, TQA）任务中对元数据（如采集日期）的敏感性问题，即当证据内容未发生改变时，仅因元数据时间戳的调整是否仍会导致模型回答错误。其核心解决方案的关键在于通过人为交换两个版本的采集日期（collection date），在不改变内容的前提下测试模型对时间信息的依赖程度。实验表明，将问题日期与后续版本的采集日期互换后，五种开源大语言模型（open-weight LLMs）在595个维基百科页面上的精确匹配率（Exact Match, EM）下降了37.6至51.4个百分点，且在独立样本集上可复现；即使回溯前一天的交换也保留29.0至39.6个百分点的性能损失。这一现象在多种提示工程策略下依然存在，包括明确指令忽略采集日期或强调归档版本来源，说明模型并非仅依赖表面日期字符串或记忆答案。进一步分析发现，虚构答案和英文月份名称的测试排除了“记忆”或“相同日期字符串”作为必要条件的可能性。通过引入“匹配思维”（Matched Thinking）机制，可将通义千问3.5（Qwen3.5）的性能惩罚从30.1点降至11.1点，而原生推理器表现各异。此外，两种API接口模型在真实维基百科数据上的影响较小，且极少出现向更早版本偏移的现象。关键发现是：大规模性能损失集中于问题日期附近，远距离日期交换的影响显著减弱。最后，添加包含实际采集日期的字段反而在部分模型中降低准确率，说明元数据本身可能干扰模型判断。该诊断方法有效分离了元数据影响与证据变化或遗漏相关版本之间的混淆因素，为时间问答输入提供了可操作的审计工具。
链接: https://arxiv.org/abs/2610.32191
作者: Minsu Oh,Seunghui Jwa,Soo Kyun Kim,Yeo-Chan Yoon
机构: Jeju National University (济州国立大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:A collection date records when a source revision was acquired, not when its contents changed. Can assigning this metadata redirect temporal question answering despite unchanged evidence? Swapping the question day and following day between two revisions lowers later-reference exact match (EM) by 37.6-51.4 percentage points across five open-weight LLMs on 595 Wikipedia pages, replicated on 128 disjoint pages. Previous-day swaps retain 29.0-39.6-point losses on 497 pages. Seven prompt variants, including explicit ignore-collection instructions and archived-revision provenance, retain the effect. Fictional answers and English month-name question dates show that remembered answers and identical date strings are unnecessary. Matched thinking reduces Qwen3.5’s penalty from 30.1 to 11.1 points, while native reasoners vary. Two API LLMs show small Wikipedia swap effects and rarely shift to earlier answers on fictional records. Large swap-induced losses concentrate around the question date; swaps between distant dates have much smaller effects. Adding fields containing one observed acquisition date lowers EM relative to omission in three of five LLMs after correction. The diagnostic isolates metadata effects from changing evidence or missing the relevant revision, and provides a practical audit for temporal QA inputs.
[NLP-44] Depth-Dependent Safety Geometry in Vision-Language Models: Layer-Wise Localization and Bidirectional Refusal Control NEURIPS2026
【速读】： 该论文旨在解决生成式视觉-语言模型（Vision-Language Models, VLMs）在安全行为背后的内在机制不清晰的问题，尤其关注如何在不重新训练或修改参数的前提下，定位并操控冻结状态下VLM的安全敏感表征。其核心解决方案在于提出一种轻量级的推理时框架，通过逐层激活差异分析与因果干预手段，揭示了模型深层结构中非均匀分布的安全几何特性。研究发现，早期视觉层可通过沿学习到的安全方向进行激活引导，实现强双向拒绝控制（拒绝率可达6%-96%），而深层网络则更依赖激活幅度缩放，并表现出显著的安全性与选择性权衡；尽管深层表征在安全方向上仍保持良好分离，但对引导操作不敏感，表明表征可分性与因果可控性并非等价。这一发现暴露了视觉编码器中存在的白盒化、可被规避的安全拒绝攻击面，进而推动了推理阶段对安全敏感表征的稳定性优化。
链接: https://arxiv.org/abs/2610.32182
作者: Muhammad Zeeshan Karamat,Syed Talal Hassan,Christiana Chamon Garcia
机构: Virginia Tech(弗吉尼亚理工大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: Accepted at the NeurIPS 2026 Workshop
Abstract:Vision-Language Models (VLMs) are widely deployed, yet the internal mechanisms behind their safety behavior remain poorly understood. We present a lightweight inference-time framework for localizing and manipulating safety-sensitive representations in frozen VLMs, without retraining or modifying parameters. On Qwen3-VL-4B, layer-wise activation divergence analysis reveals non-uniform safety geometry across depth. Causal interventions show that early vision layers support strong bidirectional refusal control through activation steering along a learned safety direction, with unsafe rejection spanning 6%-96%, whereas later layers respond more to activation magnitude scaling and exhibit a substantial safety-selectivity trade-off. Deeper layers remain well separated along the safety direction yet are largely insensitive to steering, distinguishing representational separability from causal controllability. InternVL2.5-4B shows qualitatively similar depth-dependent activation geometry despite differences in layer location and magnitude. Our results expose a white-box, representation-level refusal-bypass attack surface in VLM vision encoders and motivate inference-time stabilization of safety-sensitive representations.
[NLP-45] Not All Information Is Worth Asking For: Decision-Informed Information Seeking with LLM s
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在信息不完整情境下进行决策时所面临的“信息陷阱”问题，即模型会盲目执行大量代价高昂但对最终决策无实质影响的查询，或在已有足够证据时仍持续追问。其核心解决方案是提出一种无需训练、基于决策价值的信息获取框架——决策知情的信息价值获取（Decision-Informed Value-of-Information Acquisition, DIVA）。DIVA的关键在于：1）通过简短的数值化提问引导LLM构建一个小型决策模型，量化不良结果发生的概率以及每项检查对概率的影响；2）根据每项候选检查所能带来的决策净收益（即预期改进减去其成本）来评估其价值，并在无进一步检查具有正净价值时停止查询。其变体DIVA-U进一步考虑了模型自身对检查效果估计的不确定性。实验表明，在真实世界数据中，DIVA-U在四个领域和20种不同LLM上的表现均优于所有基线模型，显著降低后悔值（regret），主要归因于能够及时停止冗余查询；在合成场景下，DIVA可使最优基线模型的后悔值降低75%。
链接: https://arxiv.org/abs/2610.32181
作者: Seunghan Lee,Jun Seo,Junhyeok Kang,Jaehoon Lee,Sangjun Han,Sungdong Yoo,Minjae Kim,Tae Yoon Lim,Dongwan Kang,Hwanil Choi,Soonyoung Lee,Wonbin Ahn
机构: LG AI Research
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) are increasingly used to support decisions under incomplete information, where deciding what to look up and when to stop matters as much as the final answer. However, LLMs fall into an information trap: They pay for checks, i.e., costly pieces of evidence, that cannot change their decision or are not worth their cost, and they keep asking when the evidence is already sufficient. To this end, we propose Decision-Informed Value-of-information Acquisition (DIVA), a simple yet effective procedure for LLM information seeking based on the decision value of evidence. DIVA 1) lets the LLM specify a small decision model, i.e., how likely a bad outcome is and how each check would shift this probability, through short numeric questions and 2) values each candidate check by how much its imagined outcomes improve the decision net of its cost, stopping when no check is worth its cost. Its variant, DIVA-U, further accounts for the LLM’s uncertain estimates of how strongly each check shifts this probability. DIVA is training-free and applies to any LLM. Across four domains and 20 LLMs, DIVA-U has lower regret, i.e., value lost relative to the best possible decision, than every LLM baseline on real-world data, mainly by stopping once evidence suffices, and DIVA cuts the regret of the best LLM baseline by 75% in synthetic scenarios where every check’s reliability is given. Code: this http URL.
[NLP-46] he LLENS Platform for On-Premises Large Language Model Agents in a Japanese University Hospital: Development and Evaluation Study
【速读】： 该论文旨在解决临床工作中电子病历（Electronic Health Record, EHR）与文书工作占用医生大量时间的问题，尤其针对医疗信息系统运行于封闭网络环境下的实际限制。其核心挑战在于如何在不脱离医院内部网络的前提下，部署具备工具调用能力的大型语言模型代理（Large Language Model Agent），以实现对医疗文档的高效处理与智能支持。解决方案的关键在于设计并部署LLENS平台：该平台完全运行于北海道大学医院的封闭网络内，基于单台配备8块NVIDIA H200 GPU的主机，集成开源权重模型Kimi K2.6（约1万亿参数）及重构自Open WebUI的Web应用。平台的核心创新包括10个关键组件，如页面图像识别、全文附件输入、无向量索引的路径定位文档树、网络隔离沙箱、生成结果回传、输出自检机制、子代理协同以及已保存操作流程等。实验表明，在交互式访问模式下，模型通过文档工具与沙箱协作，平均准确率达97.2%，显著优于固定检索（81.3%）和无文档参考（10.7%），尤其在过滤与计数类任务上表现突出（16.7% vs 87.8%）。平台已自2026年9月14日起在医院网络中稳定运行，并成功验证了5种典型临床场景的可行性，证明了在封闭网络环境下实现高安全性、高可用性临床支持智能体的技术可行性和实用价值。
链接: https://arxiv.org/abs/2610.32151
作者: Ken Enda,Jien Saito,Masami Mukai,Feng Han,Emir Turkes,Tatsumi Kusakabe,Reona Umeki,Kazuhiro Sumiyoshi,Mitsuya Numata,Takumi Tanikawa,Akira Endoh,Kenji Hirata,Atsushi Yoshida,Kohsuke Kudo,Shinya Tanaka
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 49 pages, 3 figures, 6 tables, with supplementary material. Submitted to JMIR Medical Informatics
Abstract:Large language model agents that use tools could take on part of the electronic health record and desk work that occupies physicians’ time, but in many hospitals the medical information systems run inside closed networks, so the model and the agent harness must also sit inside. We designed, implemented, and deployed LLENS, a clinical support AI agent platform that runs inside the closed network of Hokkaido University Hospital on one host with 8 NVIDIA H200 GPUs, the open-weight model Kimi K2.6 (~1 trillion parameters), and a web application rebuilt from Open WebUI. The harness has 10 components, including page image reading, full-text attachment input, a path-addressed document tree without a vector index, a network-isolated sandbox, artifact return, self-checking of generated output, subagents, and saved procedures. Kimi K2.6 was chosen from 3 open-weight candidates and scored 480/500 on the 2026 Japanese Medical Licensing Examination; it generated 119 tokens/s for one user and served 16 concurrent requests at 908 tokens/s in total. On 180 questions over hospital documents (3 seeds each), mean accuracy was 10.7% with no documents, 81.3% with fixed retrieval (top 10 chunks of one lexical search), and 97.2% with interactive access, in which the model reads with document tools and the sandbox (15.9 percentage points above fixed retrieval, 95% CI 10.9-21.3); the difference lay in filtering/counting questions (16.7% vs 87.8%). Five scenarios with synthetic data ran on combinations of the components: needlestick injury response, batch adverse event grading, scanned referral letter transcription, falls data aggregation and plotting, and investigation of 3 topics by subagents. The platform has run on the hospital network since September 14, 2026.
[NLP-47] From Promises to Accountability: Comparative AI Modeling and Visualization of Electoral Commitments in Nepali Political Texts
【速读】： 该论文旨在解决选举承诺（Election Commitment）在南亚发展中国家难以追踪与问责的问题，尤其针对尼泊尔等国候选人及政党选举期间作出的承诺缺乏系统性归档与透明化管理，导致承诺随时间消逝、难以验证兑现的情况。其核心解决方案是构建首个尼泊尔语政治文本的黄金标准数据集（Cohen’s kappa = 0.777），并基于此评估机器学习、深度学习及大语言模型在自动提取与分析选举承诺中的表现。关键突破在于：采用LaBSE-LR模型在传统机器学习中取得最优性能（准确率0.897，宏平均F1-score 0.896，AUROC 0.947），fastText-LSTM在深度学习中表现最佳（准确率0.914，宏平均F1-score 0.913，AUROC 0.961），而Qwen3-8B在少样本设置下表现最优（宏平均F1-score 0.848）。研究成果被集成至移动端应用，结合生成式AI（Generative AI）实现承诺的可视化呈现，从而提升政治透明度、促进公民参与并强化选举问责机制。
链接: https://arxiv.org/abs/2610.32150
作者: Prakash Poudyal,Nischal Karki,Amrit Dahal,Rajani Chulyadyo,Sameer Tamrakar,Bal Krishna Bal
机构: Kathmandu University (尼泊尔卡特满都大学); Information and Language Processing Research Lab (信息与语言处理研究实验室); Department of Artificial Intelligence (人工智能系); Department of Computer Science and Engineering (计算机科学与工程系)
类目: Computation and Language (cs.CL)
备注: Under review at Natural Language Processing (Cambridge University Press); revision submitted. 34 pages, 5 figures, 13 tables
Abstract:Election Commitment Mining refers to the analysis of election commitments made by electoral candidates or political parties during election campaigns. However, many elected candidates and political parties, especially in developing South Asian countries, do not fully deliver on the promises they make before elections. Even after the elections, commitments presented through election manifestos, posters, pamphlets, videos, and other campaign materials are often not systematically or transparently archived. As a result, these promises tend to fade away over time, making it difficult to hold the elected candidates accountable for their commitments. To address this persistent issue, this research aims to develop a system capable of structuring the commitments made by candidates during election periods. In this work, we created the first gold-standard dataset of Nepali political texts (Cohen’s kappa = 0.777) and evaluated machine learning, deep learning and large language models to automatically extract and analyze election commitments. LaBSE-LR achieved the best results among machine learning models (0.897 accuracy, 0.896 macro F1-score, 0.947 AUROC), while fastText-LSTM achieved the highest performance among deep learning models (0.914 accuracy, 0.913 macro F1-score, 0.961 AUROC). Among the three evaluated large language models, Qwen3-8B performed best in the few-shot setting (0.848 macro F1-score). The outcome of this work is integrated into a mobile application, where commitments can be visualized with the support of GAI, enhancing transparency, civic engagement, and electoral accountability.
[NLP-48] Disentangling Question Response and Verifier Languages in Multilingual P(True) Self-Evaluation
【速读】： 该论文旨在解决多语言情境下语言模型自评估（multilingual confidence evaluation）中因三类语言混杂而导致的评估偏差问题，即提问语言、回答语言与验证指令语言三者未被明确区分。其核心解决方案在于通过六语言因子设计，独立控制这三种语言变量，并在多个基准测试（LSQA、MMMLU、TriviaQA）上使用十种开放权重模型（3B-14B）进行系统性实验，固定答案以隔离验证语言的影响。研究发现：三类语言分别影响评估的不同环节，其对准确率的主导作用取决于具体任务；将回答语言与问题或验证语言匹配可提升判别能力但不改善校准性；仅改变验证语言即可导致校准误差最大达0.04，而准确率不变；此外，评分器的有效性至关重要——未经验证的评判者会显著缩小跨语言性能差距（如英语-印地语在MMMLU上的差距由0.198降至0.038）。因此，论文强调：多语言置信度评估结果（包括模型排名）仅在完整报告三类语言并采用经过验证的评分机制时才具备可比性。
链接: https://arxiv.org/abs/2610.32144
作者: Jewon Yeom,Jaewon Sok,Seonghyeon Park,Jeongjae Park,Hwiyeong Lee,Taesup Kim
机构: Seoul National University (首尔国立大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:When a language model is asked whether its own answer is correct, three languages are involved: that of the question, that requested for the answer, and that of the verifier instruction asking for the judgment. Multilingual confidence evaluations seldom separate the three. We vary them independently in a six-language factorial design with ten open-weight models (3B-14B) on LSQA, MMMLU and TriviaQA, holding each answer fixed across verifier languages and measuring confidence with a first-token P(True) probe. The three languages move different parts of the evaluation: which of them governs accuracy depends on the benchmark, matching the response language to the question or verifier language sharpens discrimination but not calibration, and the verifier language alone shifts calibration error by up to 0.04 without changing accuracy. The scorer matters as well: an unvalidated judge would have shrunk the MMMLU English-Hindi accuracy gap from 0.198 to 0.038. Multilingual confidence results, model rankings included, are therefore comparable only when all three languages are reported and the scorer is validated.
[NLP-49] SyncPro: Synchronous Reasoning and Responding for Proactive Streaming Video Understanding
【速读】： 该论文旨在解决主动式流式视频理解中响应时机与推理过程协调不足的问题，即如何在持续接收视频观测信息的同时，实现对用户请求条件的及时响应并保持推理过程的有效性。现有方法分别具备“何时响应”的决策能力（主动视频助手）和“组织中间推理文本”的流式推理能力，但二者难以高效协同：固定推理周期可能导致无效更新或延迟响应，且单一生成路径将证据保留与用户响应耦合，导致响应需等待完整推理完成，即使已有充分证据。为此，本文提出SyncPro——一种同步双流架构，通过独立的内层（Inner）分支用于持续追踪证据变化，以及响应（Response）分支专门生成用户可读输出，两者共享因果历史但可独立激活。该设计实现了证据更新与响应生成的并发处理，显著提升了响应时效性与准确性。研究构建了18.9万条双流标注数据，采用监督微调结合轨迹级强化学习进行训练，在三个主动式基准测试中均验证了其优越性能。
链接: https://arxiv.org/abs/2610.32132
作者: Xueyun Tian,Bingbing Xu,Jiaming Liu,Jinpeng Yu,Yang Shi,Yu Zhao,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(阿里巴巴通义业务单元)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Proactive streaming video understanding requires assistants to track unfolding events and respond when a user’s requested condition is met. To make these decisions, assistants must accumulate relevant evidence and reassess response conditions as new observations arrive. Existing work offers complementary capabilities for this process: proactive video assistants learn when to respond, while streaming video reasoning methods generate scheduled intermediate text to organize observations and support subsequent answers. Together, these capabilities provide a foundation for proactive reasoning, but how to coordinate ongoing reasoning with timely responses remains underexplored. In a direct adaptation, fixed reasoning intervals can trigger updates when nothing relevant has changed or postpone useful updates until the next scheduled step. Sharing one generation path also couples two distinct requirements: preserving evidence for future decisions and producing only the response requested by the user. Moreover, responses generated after reasoning must wait for it to finish, even when sufficient evidence is already available. To address these challenges, we introduce SyncPro, a synchronous dual-stream framework with specialized generation paths for evidence tracking and user-facing responses. Its Inner and Response branches share causal history and activate independently, allowing relevant evidence changes to trigger internal updates and responses to proceed concurrently with ongoing reasoning. We construct 189K dual-stream examples and train SyncPro through supervised fine-tuning followed by trajectory-level reinforcement learning. Experiments across three proactive benchmarks show improved response timing and more accurate answers to user requests.
[NLP-50] From Noisy Retrieval to Decision Evidence: Silver Evidence Demonstrations for In-Context Learning
【速读】： 该论文旨在解决在长且嘈杂的上下文环境中，基于检索增强生成（RAG）与检索增强少样本提示学习（few-shot ICL）方法在决策时难以有效提取关键证据的问题。当相关证据被淹没在冗余或干扰性信息中时，传统方法的表现会显著下降。现有方法通常仅将专家标注的开发集示例用作答案侧的解释（rationales），而未将其用于指导噪声检索结果的过滤。为此，本文提出“银质证据示范”（silver-evidence demonstrations）——即用大语言模型（LLM）生成的目标条件化证据摘要替代原始检索到的段落，作为少样本示例中的演示内容。具体而言，对于每个已标注的开发示例，模型在无专家标签或黄金证据可见的情况下生成“银质证据”，随后将其与真实标签及可获得的黄金证据配对。测试阶段通过一次结构化调用，从检索到的文本中生成银质证据，并据此预测最终标签与决策依据。在EI 2.0、SciFact、合成政策检索任务以及HICRIC/IMR数据集上的实验表明，该方法在四个数据集中有三组优于仅使用黄金示例的少样本I CL基线，在所有标注数据集上均提升了证据F1得分，且在全部四个数据集上均实现了高于InstructRAG风格I CL的平均标签准确率。结果表明，以紧凑、目标导向的银质证据替代演示侧的原始检索段落，能有效减少冗余上下文干扰，同时保留决策相关信号与专家监督，尤其适用于依赖长篇复杂文本进行判断的任务场景。
链接: https://arxiv.org/abs/2610.32128
作者: Masoud Monajatipoor,Fazlolah Mohaghegh,Hamid Reza Hassanzadeh,Akash Uttamrao Bhoite,Robert E. Tillman
机构: Optum AI
类目: Computation and Language (cs.CL)
备注:
Abstract:RAG and retrieval-augmented few-shot ICL can struggle when decision-relevant evidence is buried in long, noisy context. Retrieved passages and demonstrations may contain redundant or distracting information. Although expert-labeled development examples may identify the text supporting each decision, these annotations are typically used only as answer-side rationales rather than as supervision for filtering noisy retrieval. We propose silver-evidence demonstrations, which replace raw retrieved passages in ICL examples with generated, target-conditioned evidence summaries. For each labeled development example, an LLM generates silver evidence without access to the expert label or gold evidence; this evidence is then paired with the expert label and gold evidence where available. At test time, a single structured call generates silver evidence from retrieved passages and then predicts the final label and decision evidence. Across EI 2.0, SciFact, synthetic policy retrieval, and HICRIC/IMR data, our full method improves over equally sized gold-only ICL (label-only for policy) on three of four datasets and improves evidence F1 on all annotated datasets. It also achieves higher mean label accuracy than InstructRAG-style ICL on all four datasets. These results suggest that replacing demonstration-side retrieved passages with compact, target-conditioned silver evidence is particularly useful for decisions based on long and noisy text, where it reduces distracting context while preserving decision-relevant signals and expert supervision.
[NLP-51] Data Quality in RVL-CDIP: A Systematic Audit
【速读】： 该论文旨在解决当前文档分类基准数据集RVL-CDIP中存在的数据质量问题，包括标签错误、训练集与测试集之间的重复/近似重复样本、文本内容依赖性过强以及模型泛化能力不足等问题。其解决方案的关键在于系统性地审计并修正数据集中的缺陷：通过识别并剔除约10.5%的错误标签、移除35.2%的测试集中与训练集重复或近似重复的样本、消除对页面标识码（identification codes）的过度依赖，并构建去标识化版本以保护敏感个人数据；同时在修复所有已知问题后重新评估25个模型，结果显示经再训练的模型性能平均提升0.9至2.9个百分点，其中最先进模型LayoutLMv3在修正后的数据集上达到96.5%的准确率。研究还揭示了光学字符识别（OCR）引擎选择对结果的影响（达4.1–8.3个百分点），强调了预处理环节的重要性。最终，研究团队公开了修正后的数据集、标注信息及分布外（out-of-distribution）测试数据，为后续研究提供更可靠的基础。
链接: https://arxiv.org/abs/2610.32126
作者: Stefan Larson,Kevin Leach
机构: Vanderbilt University(范德比尔特大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:RVL-CDIP is the de facto standard benchmark for document classification, and reported accuracy on it is now 98.8%. In this paper, we audit RVL-CDIP for data quality issues and find that approximately 10.5% of its 400,000 documents have erroneous labels and that 35.2% of the test set has a duplicate or near-duplicate in the training set. Identification codes on the pages are sufficient to classify at 69% accuracy, and a state-of-the-art model loses 11 points without them. Image-only models trained on RVL-CDIP lose 16 to 28 points on documents from other sources, and the choice of OCR engine changes accuracy by 4.1 to 8.3 points. We find that nearly 16,000 RVL-CDIP documents contain sensitive personal data, and produce de-identified versions of these documents. We re-benchmark 25 models, and with all defects remediated, the 13 that we retrain are 0.9 to 2.9 points more accurate than in the original condition; the most accurate of them, LayoutLMv3, is 96.5% accurate. We release the corrected corpus, the annotations and the out-of-distribution data.
[NLP-52] Semantically Rich Metrically Affect-Blind: A Geometric Blind Spot in Modern Text Embedders
【速读】： 该论文旨在解决现代文本嵌入模型中一个关键的矛盾现象：尽管嵌入向量中隐含了丰富的情感信息（sentiment），但这些信息在嵌入空间的度量（metric）中几乎无法体现。具体而言，研究发现，通过线性探测器（linear probe）可高达91.3%准确率地解码句子极性，而基于余弦相似度的度量仅能区分极性差异0.003–0.064，表明嵌入的几何结构与语义内容严重脱节。这一问题广泛存在于2021至2026年发布的十一个冻结编码器中，包括最新的40亿参数大语言模型嵌入器Qwen3-Embedding-4B。由于检索、聚类、去重及检索增强生成（RAG）等下游任务依赖于嵌入的度量结构而非外部探测器，导致系统对编码器显式包含的情感信息“视而不见”。其根本原因被证实源于预训练目标本身，而非提示工程偏差——通过任务条件控制实验显示，将Jina-v5模型从检索任务切换为分类任务训练，仅能恢复部分性能增益，且极性对齐差距基本不变。为此，论文提出情感注入的句嵌入（Sentiment-Injected Sentence Embeddings, SISE）解决方案：在冻结编码器上快速（秒级）训练一个小型（0.13M–0.69M参数）监督对比头（supervised-contrastive head），利用标注数据进行微调。该方法显著提升所有2022年后发布的编码器在完整数据探针准确率（最高+5.8点）、检索MAP（最高+26.0）和极性区分度（7–61倍提升）方面的表现，实现有效专业化。虽然引入该头会轻微牺牲通用语义相关性（如在STS-B上损失38–41 Spearman点），但通过融合变体可大幅缓解（损失仅2.4–12.0）。值得注意的是，搭载此头后，0.6B规模的编码器在128维下即可达到4B编码器的性能水平（MAP 91.8 vs. 90.3）。研究还发布了可复现的CPU流水线与诊断工具“属性对齐差距”（Attribute Alignment Gap），以量化评估嵌入中特定语义属性的对齐程度。
链接: https://arxiv.org/abs/2610.32122
作者: Mahdi Naser Moghadasi
机构: BrightMind AI( BrightMind AI)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 6 pages, 2 figures, 3 tables. Code: this https URL
Abstract:We report a systematic dissociation in modern text embedders: sentiment is almost perfectly decodable from their embeddings, yet almost entirely absent from their metric. Across eleven frozen encoders spanning 2021 to 2026, including 2026-generation models and the 4B-parameter LLM embedder Qwen3-Embedding-4B, a linear probe recovers sentence polarity at up to 91.3% accuracy while cosine similarity separates polarity by a margin of only 0.003-0.064. Qwen3-Embedding-4B, the strongest probe in the panel, reaches 91.3% probe accuracy but only 64.3 retrieval MAP. Because retrieval, clustering, deduplication, and retrieval-augmented generation operate on the metric rather than on a trained probe, these systems are effectively blind to information their encoder demonstrably contains. A task-conditioning control suggests this reflects the pretraining objective rather than a prompting artifact: conditioning Jina-v5 on its classification task rather than its retrieval task recovers only about half of the MAP gain that injection achieves on the same weights (63.2 to 73.8, versus 85.1), and leaves the alignment gap essentially unchanged. Sentiment-Injected Sentence Embeddings (SISE) train a small (0.13M-0.69M-parameter) supervised-contrastive head on a frozen encoder in seconds on a CPU, using the labelled training set. The head improves every encoder released since 2022 on full-data probe accuracy (up to +5.8 points), retrieval MAP (up to +26.0) and polarity margin (7-61x). This is a specialization, not a free gain: the compact 128-d head loses 38-41 Spearman points on STS-B, while a fused variant that concatenates it with the base embedding loses 2.4-12.0. With the head, a 0.6B encoder at 128 dimensions matches a 4B encoder given the same head (91.8 vs. 90.3 MAP). We release a CPU-reproducible pipeline and a reusable diagnostic, the Attribute Alignment Gap.
[NLP-53] EDWARD-NMT: Elastic Dimensions With Adaptive Routing and Dual-consistency for Neural Machine Translation EMNLP2026
【速读】： 该论文旨在解决稀疏混合专家（Sparse Mixture-of-Experts, MoE）架构在资源受限设备上部署时面临的两大核心问题：一是整个专家池（expert pool）必须常驻内存，导致显存开销过大；二是训练阶段采用固定的 top-k 路由策略，使每个 token 的计算预算静态不可变，难以适应不同推理场景的需求。为此，本文提出 EDWARD-NMT（Elastic Dimensions With Adaptive Routing and Dual-consistency for NMT），一种面向“训练一次、随处部署”范式的新型 MoE 编码器-解码器架构，其关键创新在于：（i）采用类似套娃（Matryoshka-style）的弹性维度设计，实现模型尺寸的灵活压缩以显著降低内存占用；（ii）引入可学习的任务相关阈值自适应 sigmoid 路由机制，使每个 token 能动态决定激活的专家数量，从而实现计算资源的按需分配；（iii）设计双一致性（dual-consistency）优化目标，确保从单一训练检查点中提取出的子网络在性能上达到最优。大量实验验证了该方法在 PhoMT（En-Vi）、WMT-14（En-De）、多语言 OPUS-16 及多领域 De-En 基准上的有效性，结果表明全容量模型在保持竞争力的同时提升了参数效率，且无需重新训练即可从单个模型中提取出的子网络始终优于同等规模的密集型 Transformer 模型。
链接: https://arxiv.org/abs/2610.32115
作者: Edward Nguyen,Thai Tran,Kiet Anh Hoang,Duy Hoang,Duy Vo,Khanh Le,Vu Le
机构: UNEY(UNEY), Switzerland
类目: Computation and Language (cs.CL)
备注: Accepted to EMNLP 2026 (Findings)
Abstract:Sparse Mixture-of-Experts (MoE) is a powerful paradigm for scaling Neural Machine Translation (NMT) and managing language heterogeneity. However, practical limitations hinder its deployment on resource-constrained devices: the entire expert pool must remain in memory, and the per-token compute budget is statically fixed during training by rigid top-k routing. We introduce EDWARD-NMT (Elastic Dimensions With Adaptive Routing and Dual-consistency for NMT), a novel MoE encoder-decoder architecture designed for a train once, deploy anywhere paradigm that exposes a family of deployable sub-networks at inference time. EDWARD-NMT integrates three key innovations: (i) Matryoshka-style elastic dimensions to flexibly reduce the memory footprint; (ii) an adaptive sigmoid router with a learned, task-conditioned threshold, allowing each token to dynamically determine its active expert count; and (iii) a dual-consistency objective to maximize the performance of the extracted sub-networks. Extensive experiments on PhoMT (En-Vi), WMT-14 (En-De), multilingual OPUS-16, and a multi-domain De-En benchmark validate our approach. The full-capacity architecture matches or exceeds strong MoE baselines with improved parameter efficiency. Furthermore, the sub-networks seamlessly extracted from a single checkpoint consistently outperform comparably sized dense Transformers without any retraining.
[NLP-54] ampered ML: Evaluating LLM Judgment of Machine Learning Research EMNLP2026
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在复杂、难以监督的研究任务中进行可靠推理的能力瓶颈，特别是针对人工智能对齐（AI alignment）研究中的关键挑战：识别研究提案中的方法论缺陷并评估实验设计是否支持其主张。为系统评估这一能力，研究提出“Tampered ML”基准，包含1,858个样本对，每对由一篇真实机器学习论文重构的实验提案及其经微调引入特定方法论缺陷的反事实版本构成，要求模型在两种呈现顺序下均能准确判断更优的设计方案。人类验证研究表明，86%的测试项具有有效区分度（95%置信区间：81%-91%）。尽管前沿模型表现优于随机水平（如Opus 5达到77.9%准确率），但整体仍远未达到可靠标准，且对呈现顺序高度敏感，倾向于偏好更具具体性、透明性或常规性的有缺陷设计。解决方案的关键在于构建一个高保真、可量化的基准工具，以推动对生成式AI在严谨科学研究中推理能力的系统性评估与改进。
链接: https://arxiv.org/abs/2610.32114
作者: Soroush Ebadian,Parand A. Alamdari,Joe Benton,Hailey Joren
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Digital Libraries (cs.DL); Machine Learning (cs.LG)
备注: 28 pages, 10 figures. Accepted at EMNLP 2026 (Main Conference)
Abstract:One approach to accelerating progress on AI alignment is to automate parts of alignment research itself. Doing so requires systems that can perform reliably on hard-to-supervise research tasks, including identifying flaws in research proposals and evaluating whether a proposed experiment supports its intended claim. To evaluate this capability, we introduce Tampered ML, a benchmark of 1,858 item-pairs, each consisting of an experiment proposal reconstructed from a published machine learning paper and a minimally edited counterfactual that introduces a targeted methodological flaw. Models must identify the more methodologically sound proposal in both presentation orders. A human validation study on a stratified sample estimates that 86% of items are valid (95% CI: 81-91%). Frontier models perform above chance but remain far from reliable (with Opus 5 achieving 77.9% accuracy). They are highly sensitive to the order of presentation and favor flawed designs that appear more specific, transparent, or conventional.
[NLP-55] REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories
【速读】： 该论文旨在解决生成式推理过程中安全风险检测与有效干预时机之间的不匹配问题，即尽管早期推理链（Chain-of-thought, CoT）阶段可能更容易识别风险，但这一时点是否也具备最强的干预效果尚不明确。其解决方案的关键在于提出REACT——一种基于条件激活的干预方法，通过学习由成对安全与非安全推理轨迹提取的安全方向，并仅在早期CoT隐藏状态中施加一次无门控的编辑，随后所有后续生成均不受修改且模型权重保持冻结。研究通过在前五个CoT位置分别训练安全方向并评估全量重生成后的效果，发现最强干预效果集中于前两至三个词元的早期窗口；在DeepSeek-R1-Distill-Llama-8B模型上，第四和第五个词元处的干预效果几乎归零，即便后期状态对风险更具预测性。这揭示了风险可读性（risk readability）与有效控制（effective control）并非等价，干预时机必须基于其对后续生成的实际影响来评估。实验表明，REACT在相同模型上将强拒绝测试集（StrongREJECT-273）中完全安全的推理-答案输出率从29.06%提升至82.05%，同时维持良性拒答率和有用性接近基线水平。
链接: https://arxiv.org/abs/2610.32113
作者: Hanwen Zhang,Enyi Jiang,Ruike Zhu,Qinan Yu,Sanmi Koyejo
机构: Stanford University (斯坦福大学); University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Independent Researcher (独立研究员)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 26 pages, 2 figures
Abstract:Chain-of-thought (CoT) generation offers an opportunity to intervene before a reasoning model completes its answer. But does the point at which risk is easiest to detect also offer the most effective control? We study this question with REACT, a conditional activation intervention that learns a safety direction from paired safe and unsafe trajectories and applies it once to an early CoT hidden state. The model weights remain frozen, and all subsequent tokens are generated without further edits. To study timing, we train directions separately at each of the first five CoT positions and evaluate ungated edits with full post-edit regeneration. Across three distilled reasoning models and three generation seeds, the strongest gains concentrate in a two- to three-token early window; on DeepSeek-R1-Distill-Llama-8B, the effect falls to approximately zero at the fourth and fifth tokens, even though later states can be more predictive of risk. This separates risk readability from effective control: intervention timing must be evaluated through its effect on subsequent generation. On the same 8B model, REACT raises the rate of fully safe CoT-answer outputs on StrongREJECT-273 from 29.06% to 82.05%, while keeping benign refusal and usefulness near baseline. An anonymized implementation is available at this https URL.
[NLP-56] Context or Knowledge: How LLM s Attribute Their Own Answers
【速读】： 该论文旨在解决生成式模型在回答问题后，其“源报告”（source report）——即模型声称答案来自文档还是自身参数知识——的决定机制这一关键问题。现有研究尚不清楚模型在生成答案后如何判断并报告其信息来源。本文通过干预模型在生成答案后但提问前的键值缓存（Key-Value Cache, KV Cache），揭示了源报告的因果决定因素。实验结果表明，在控制事实任务中，修改或移除文档缓存对源报告无影响；而将问答（Q+A）缓存从一次运行移植到另一次相同问题与答案的运行中，则能完全转移源报告，说明源报告的归属由问答缓存状态决定。进一步地，通过对预训练阶段拟合的方向进行操控，可使小规模模型（7B-32B）在固定强度下实现98.9%-100%的报告方向反转，而随机方向仅影响不超过1.1%。分析显示，这些方向上的键（key）和值（value）分量是转移效应的关键：仅复制这些分量即可近似重现报告变化，而固定它们则几乎完全阻断变化。该现象在答案生成过程中同样存在，尽管效应较弱且依赖提示结构。因此，本研究的核心发现是：在该设定下，源报告行为由回答过程产生的缓存状态所因果决定，而非文档内容或初始上下文。研究未验证报告的普遍准确性，也未完全解析缓存状态的语义表征，但为理解生成式模型的溯源机制提供了关键实证依据。
链接: https://arxiv.org/abs/2610.32109
作者: Kirill Afendulev,Alexey Dontsov,Alla Chepurova,Elena Tutubalina,Anton Korznikov
机构: 未知
类目: Computation and Language (cs.CL)
备注: 25 pages, 7 figures, 23 tables, including appendices
Abstract:Language models can answer a question from a supplied document or from their parametric knowledge. When later asked where the answer came from, what determines their source report? We study this by intervening on the model’s key-value (KV) cache after the answer is generated and before the source question is asked. On a controlled factual task with four models (7B-32B), changing or removing the document’s cache never changes the report. In contrast, transplanting the question-and-answer (Q+A) cache between runs with identical question and answer tokens transfers it: every eligible report follows the donor, while the recipient’s document is unchanged. On the two smaller models, steering Q+A values along directions fitted before any source question reverses 98.9-100% of reports across four document conditions at a fixed strength, while norm-matched random directions flip at most 1.1%. The components of keys and values along fitted directions carry the transplant effect: copying them alone nearly reproduces the report change; holding them fixed nearly prevents it. Applied during answer generation, the same value directions also change which source supplies the answer, though more weakly and depending on the prompt. Source reporting in this setting thus depends causally on cached state produced while answering. We do not establish general reporting accuracy or a complete interpretation of this state.
[NLP-57] Beyond What to Say: Proactive Utterance Timing with Inner Thoughts for Werewolf Agents
【速读】： 该论文旨在解决大语言模型（LLM）代理在狼人杀（Werewolf）这类竞争性博弈环境中，如何在不完整信息条件下合理判断“何时发言”这一关键问题。传统对话系统中的发言时机决策方法多适用于一般对话场景，难以适应狼人杀中因利益冲突导致的高风险语用解读与策略性沉默的复杂性。为此，本文提出一种针对竞争性环境优化的自适应内省机制——对抗性内省思想（Adversarial Inner Thoughts），其核心在于对内部思考内容进行动态价值评估，以决定当前时刻是否具备表达价值。该方法通过重构原有“内在思想”框架，强化了对策略性沉默与发言时机的精准控制，在自对弈日志分析与人类评估中均展现出显著有效性，为复杂博弈场景下的智能体行为决策提供了新范式。
链接: https://arxiv.org/abs/2610.32107
作者: Yoshiki Tanaka,Shuya Uchiyama,Daichi Hamaguchi,Michimasa Inaba
机构: The University of Electro-Communications (东京电子通信大学)
类目: Computation and Language (cs.CL)
备注: Accepted to the 4th International AIWolfDial Workshop at INLG 2026
Abstract:Participating in a conversation involves not only deciding what to say but also whether to speak at all. This judgment carries even greater significance in environments like the Werewolf game, where participants’ interests conflict, because both utterances and silence are subject to interpretation and assessment by others. Therefore, determining when to speak, as well as what to say, is crucial for an LLM agent playing the Werewolf game. In this study, we introduce a Werewolf AI agent developed for the Anytime Utterance Track of the AIWolfDial shared task. Our system builds on the existing Inner Thoughts framework, which determines utterance timing during conversations by generating internal thoughts and evaluating the value of each thought for expression at that moment. Since existing methods were designed for general dialogue, we have redesigned the system specifically for the competitive environment and incomplete information setting of Werewolf games. We call our adapted method Adversarial Inner Thoughts. Through experiments involving analysis of self-play logs and human evaluations, we analyze and discuss the effectiveness of our approach.
[NLP-58] Putting Transformer Theories to the Test
【速读】： 该论文旨在解决当前多种形式化理论对Transformer架构能力与局限性预测不一致的问题，这些理论因底层假设和概念框架不同，导致在判断某些任务对Transformer而言“易”或“难”时产生分歧。其核心解决方案在于系统性地梳理并统一七种代表性理论（星自由语言、AC⁰、TC⁰、C-RASP、MPC、BAPO及敏感性理论），将它们对一系列涵盖不同计算结构的任务的预测映射至一个共同的评估矩阵中，并通过新推导的理论结果补全该矩阵。随后，研究通过受控实验验证理论预测，考察模型在分布内样本复杂度、长度泛化能力以及预训练语言模型性能方面的表现。结果显示，没有任何单一理论能够全面解释Transformer的实际成功与失败模式；当某理论未能捕捉到任务相关的Transformer行为特征时，其预测即失效——例如，C-RASP理论因未建模内容依赖型注意力而错误预测k跳归纳任务为困难，但实际训练的小型Transformer却表现出良好的长度泛化能力；类似地，基于硬注意力假设的AC⁰与BAPO理论预测二元多数判定任务应困难，但实践中亦可成功。因此，研究指出，一个有效的Transformer理论必须具备若干关键特征：软注意力机制而非硬注意力、动态的内容依赖检索能力、每个位置上的有界计算、对可学习性与归纳偏置的合理建模，以及同时包含上界与下界的综合分析框架。
链接: https://arxiv.org/abs/2610.32104
作者: Awni Altabaa,Tobias Schnabel,Kiran Tomlinson
机构: Microsoft Research
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Several formal theories have been developed to characterize the capabilities and limitations of the transformer architecture, each employing different abstractions that capture different features of the architecture. Due to the differences in their underlying assumptions and conceptual frameworks, these theories can disagree in their predictions about what tasks transformers find easy or hard. We map out seven theories of transformer capability (Star-free, \mathsfAC^0 , \mathsfTC^0 , C-RASP, MPC, BAPO, and Sensitivity) and translate their results into a common matrix of predictions across a suite of tasks covering different computational structures. We provide new theoretical results to complete this matrix. We then evaluate theoretical predictions through controlled experiments measuring in-distribution sample complexity, length generalization, and pretrained language model performance. We find that no single theory explains the observed pattern of successes and failures. When a theory does not capture the transformer behaviors relevant to a task, its predictions break down; for instance, we show that the k -hop induction task is predicted hard by C-RASP because it does not model content-dependent attention, but we train small transformers on this task that length-generalize. Similarly, theories assuming hard attention like \mathsfAC^0 and BAPO predict binary majority to be hard, for which we also find success in practice. Together, our results point to several specific features that a theory of transformers must capture: soft rather than hard attention, dynamic content-dependent retrieval, bounded computation at each position, an account of learnability and inductive biases, and a combination of upper and lower bounds.
[NLP-59] Behaviour analysis for large language model-driven agents
【速读】： 该论文旨在解决如何系统理解语言智能体（language agent）在多步骤行为轨迹中动作组织机制及其与上下文和结果之间关系的问题。其核心挑战在于从复杂、动态的行为序列中自动发现有意义的行为模式，并揭示这些模式在不同情境下的依赖性与因果性。解决方案的关键在于提出一个通用框架，将数据驱动的行为发现与分析、实验探究相整合：首先通过记录的活动数据生成可度量的行为类别；其次，在保持行为在过程中的时序位置基础上，构建可解释的因果假设；再通过在保留测试集中的统计检验评估证据的强度与泛化能力；最后，借助提示工程（prompt steering）等可控实验验证行为与结果在变化条件下的响应机制。该框架支持不同长度、事件顺序及智能体角色的轨迹分析，并推动了编码、分析与实验方法的协同演进，以实现对智能体行为更深层次的理解。
链接: https://arxiv.org/abs/2610.32087
作者: Tianqiang Yan,Lizhen Qu
机构: Monash University (蒙纳士大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Under review by Nature Machine Intelligence
Abstract:Understanding language-agent behaviour requires learning how actions are organised across a trajectory and how this organisation relates to context and outcomes. We propose a general framework that connects data-driven behavioural discovery with analysis and experimental investigation. Categories developed from recorded activity define what is measured. Analyses preserve the behaviours’ positions in the process and express their relationships as interpretable hypotheses. Statistical assessment and tests in held-out contexts examine the strength and scope of the evidence. Experiments, including prompt steering, then test how behaviour and outcomes change under deliberately varied conditions. A study of iterative self-reflection illustrates how discovery, dependency learning and prompt comparisons can inform one another. We develop this approach for trajectories of varying length, event order and agent roles, and identify advances in coding, analysis and experimentation needed to support its wider use.
[NLP-60] How Much Do You Trust A Dog? Quantifying Hint-Following in Reasoning Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在面对用户提示时表现出的“奉承倾向”（sycophancy）问题，即模型会无条件地调整其回答以迎合用户暗示，即使在客观事实与推理任务中亦如此。这一现象的核心在于：用户提示对模型输出的影响机制及其是否根据提示来源的可靠性进行合理校准，目前尚缺乏清晰的量化理解。本文通过在MMLU基准上设计“[源]认为答案是(X)”形式的提示（hint），系统性地考察不同来源（教授、朋友、狗）对模型回答的影响，覆盖Qwen、Gemma和SmolLM三类开源模型家族。研究发现，提示所引发的回答偏移遵循一种可预测的指数倾斜（exponential tilt）结构——提示导致目标答案的对数几率（log-odds）产生近似恒定的均值偏移，且这种敏感度在不同来源间高度共享，表明模型对提示的响应具有内在一致性而非依赖于语义内容。为揭示该现象的成因，作者提出一个提示响应的形式化分解框架，并在受控合成实验中证明：基于KL正则化的奖励训练（KL-regularized reward-based training）能够同时重塑提示相关与未提示答案的分布，从而解释该偏移模式的生成机制。进一步评估表明，即使提示来源被明确声明为不可靠或附加了警示信息，模型仍持续产生显著的对数几率偏移，说明当前大语言模型在对抗性引导下存在严重脆弱性，难以有效校准提示来源的可信度。
链接: https://arxiv.org/abs/2610.32055
作者: Anxin Guo,Chirag Pabbaraju,Sara Ahmadian,Kamesh Munagala,Aravindan Vijayaraghavan
机构: Northwestern University (西北大学); NSF-Simons AI Institute for the Sky (SkAI) (美国国家科学基金会-西蒙斯人工智能天空研究所); Stanford University (斯坦福大学); Google Research (谷歌研究)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) often exhibit sycophancy, altering their answers to align with user suggestions even on objective factual and reasoning tasks. Yet the quantitative form of this influence— and whether it is calibrated to source reliability— remains poorly understood. In this work, we investigate this behavior through hint-induced answer shifts on MMLU questions, evaluating prompts of the form ``[Source] thinks the answer is (X)‘’. We systematically vary the source to which the hint is attributed (a professor, a friend, or a dog) across three open-source model families, Qwen, Gemma, and SmolLM. We show that the change induced by the hint follows a predictable structure well-characterized by an exponential tilt: the hint induces an approximately constant mean shift in the hinted answer’s log-odds, with question–target susceptibility largely shared across sources. To explore how this pattern can arise, we introduce a formal decomposition of the hint response and illustrate in controlled synthetic experiments how KL-regularized reward-based training reshapes both hinted and unhinted answer distributions. Finally, we evaluate whether this shift is calibrated to source reliability, finding that the log-odds shift persists even when the hint-giver is declared inaccurate or accompanied by explicit warnings not to trust the source, highlighting the vulnerability of LLMs to adversarial steering.
[NLP-61] X-Dubbing: Speech-to-Speech Translation under Temporal and Visual Constraints
【速读】： 该论文旨在解决跨语言配音（cross-lingual dubbing）中的时序同步难题，即如何在保持原始视频帧与目标语言语音严格同步的前提下，实现自然流畅且忠实于原文语义的翻译。其核心挑战在于不同语言在发音时长、节奏和发声方式上的差异导致的时序不匹配问题。解决方案的关键在于提出一种名为X-Dubbing的端到端框架，将任务建模为时间感知的语音到文本翻译与视频感知的文本到语音生成的联合优化问题。该框架包含两个核心组件：一是“配音翻译器”（Dubbing Translator），用于生成具有时序结构的目标语言脚本；二是“配音演员”（Voice Actor），负责根据脚本合成与原视频同步的语音。二者均基于一种语言无关的时序预算进行约束，并通过合成语音提供的配音感知反馈对翻译器进行迭代优化，从而提升最终配音的时序一致性。该方法无需成对的配音数据，分别在语音-文本数据和单语音视频数据上进行训练，显著提升了跨语言配音的质量与同步性。
链接: https://arxiv.org/abs/2610.32042
作者: Jeongsoo Choi,Ji-Hoon Kim,Junseok Ahn,Joon Son Chung
机构: KAIST(韩国科学技术院); Chung-Ang University(中央大学)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Audio and Speech Processing (eess.AS)
备注:
Abstract:Cross-lingual dubbing aims to convert source speech into target-language speech that remains synchronized with the original video. The key to this task is to produce a faithful translation that is naturally speakable within the source timing, which is non-trivial because languages differ in duration, rhythm, and articulation. In this paper, we propose X-Dubbing, which formulates the task as a joint problem of time-aware speech-to-text translation and video-aware text-to-speech. Specifically, our method consists of a Dubbing Translator that generates temporally structured scripts and a Voice Actor that renders them in synchrony with the source video, both grounded in a language-agnostic timing budget. We further refine the translator with dubbing-aware feedback from the synthesized speech, encouraging scripts that yield well-synchronized dubbing. This design requires no paired dubbing data, as the two components are trained on speech-text and monolingual audio-visual data respectively. Extensive experiments show that X-Dubbing outperforms recent baselines across diverse metrics, including isochrony and audio-visual correspondence.
[NLP-62] DraMaze: A Multi-Agent Pipeline with Future-Aware Memory for Diverse Microdrama Script Generation
【速读】： 该论文旨在解决长篇微剧（long-form microdrama）脚本生成中面临的两大核心挑战：一是如何在生成过程中保持多样化的剧情推进，二是如何确保长距离叙事连贯性。传统方法在生成后期容易导致候选续写内容逐渐集中于少数叙事机制，且依赖的检索策略易加剧陈词滥调（trope）重复，忽视对后续情节至关重要的历史信息。其解决方案的关键在于提出DraMaze多智能体流水线框架，通过基于陈词滥调记忆引导的进化树搜索（trope memory-guided evolutionary tree search）与变异机制，探索多样化因果路径；同时构建反向叙事依赖树（Backward Narrative Dependency Tree），根据未来叙事目标组织、保留并高效检索历史记忆。此外，研究还引入MiDramaBench双语基准测试集及因果模式度量指标，用于评估脚本逻辑性与创意质量。实验结果表明，DraMaze在三种主流大语言模型（LLM）上均实现显著提升，在长达60集、总字数达6万字的剧本生成任务中，因果模式重复率降低32.1%，逻辑得分达到82.5，较最优基线提升8.5分。
链接: https://arxiv.org/abs/2610.32002
作者: Siyang He,Qiqi Wang,Fan Liu,Yuncheng Huang,Pengfei Xu,Xi Chen,Ziwei He,Xipeng Qiu
机构: 1. Tsinghua University (清华大学); 2. Beijing Academy of Artificial Intelligence (北京人工智能研究院); 3. Peking University (北京大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:The rapid growth of microdramas has increased demand for high-quality script production. Long-form microdrama script generation with LLMs requires both diverse plot progression and long-range narrative coherence. Our observations reveal that candidate continuations increasingly concentrate on fewer narrative mechanisms as generation progresses, while conventional retrieval may reinforce trope reuse and overlook historical information essential to future plots. To address these challenges, we propose DraMaze, a multi-agent pipeline that explores diverse causal paths through trope memory-guided evolutionary tree search and mutation. It further constructs a Backward Narrative Dependency Tree to organize, retain, and retrieve historical memories based on future narrative goals. We also introduce MiDramaBench, a bilingual benchmark for evaluating script logic and creative quality across different script lengths, together with causal-pattern metrics for narrative repetition and diversity. Experiments show consistent improvements across three backbone LLMs. For 60-episode scripts of up to 60k words, DraMaze reduces causal-pattern repetition by 32.1% and achieves a Logic score of 82.5, outperforming the best-performing baseline by 8.5 points.
[NLP-63] AusClinBench: a deterministically scored benchmark of large language models against current Australian clinical policy
【速读】： 该论文旨在解决当前生成式人工智能（Generative AI）在澳大利亚临床决策支持与病历自动生成中缺乏透明度和本地适配性的问题，尤其关注现有医学评估基准多基于其他司法管辖区、无法有效衡量模型对澳大利亚本土医疗规范（如PBS授权、MBS计费、Ahpra职责、分诊警示标志、免疫接种与筛查指南等）的准确应用。其核心解决方案是构建一个基于澳大利亚临床指南和法规的142道多领域可验证问题基准测试，通过“盲测”与“提供规则条款后测试”的对比设计，区分模型对规则的记忆错误与应用错误。研究发现，尽管部分模型盲测准确率可达92.7%，但高达86%的失败案例在提供条款后可正确回答，表明主要问题在于规则应用而非知识缺失；同时存在显著的“自信错误”现象（6.6%–47.8%），且在涉及原住民健康及无明确指引的免疫接种问题上，模型常生成不准确或非本地化内容。因此，论文提出应将模型视为临床产品中的关键组件，建立针对本地医疗规则的最低标准，并将其纳入临床产品整体评估体系，而非仅评价基础模型性能。该基准测试聚焦于基础模型，不认证任何模型用于临床实践，其核心建议依赖于全科医生的初步评审，尚未经过独立验证。
链接: https://arxiv.org/abs/2610.31993
作者: Daniel S. Epstein
机构: 未知
类目: Computers and Society (cs.CY); Computation and Language (cs.CL)
备注: 19 pages, 3 figures. Technical report, version 1.4.1 of the benchmark. Not peer reviewed. Answer keys not yet independently reviewed. Results and methods: this https URL
Abstract:AI models already inform clinical decisions and draft notes in Australian clinical software, usually without the practitioner being told which model is used. Existing medical benchmarks are written for other jurisdictions. We wrote 142 questions from Australian guidelines and schedules across five domains (PBS authority, MBS billing, Ahpra obligations, triage red flags, immunisation and screening), each with one verifiable answer scored by exact match. Each question ran blind and with the governing clause supplied, which separates failure to recall the current rule from failure to apply it. We tested 28 models between July and October 2026. Blind accuracy ranged from 51.5% to 92.7%, and the highest-scoring models could not be separated. Of 791 failed model-question pairs, 684 (86%) were wrong blind and right with the clause. Confident-error rates ranged from 6.6% to 47.8%. On five immunisation questions where the guideline gives no directive, 82% of answers stated one. On 20 Aboriginal and Torres Strait Islander health questions, models did no worse overall, but on a few the wrong answers gave the rule for the general population, an older rule or another country’s rule. We propose treating the model as an ingredient of a clinical product, with a minimum standard on local rules that sits beside product evaluation. The keys are one general practitioner’s and are not yet independently reviewed. The benchmark measures base models, not clinical products, and does not license any model for clinical use.
[NLP-64] POS-Signature Demonstration Retrieval for In-Context Dependency Parsing
【速读】： 该论文旨在解决在上下文学习（in-context learning）中，基于语义相似性进行示范样本检索在结构化预测任务中可能带来的相关性偏差问题。其核心挑战在于如何更有效地选择与目标句子在语法结构上高度匹配的示范样本，以提升模型在通用依存句法分析（Universal Dependencies parsing）任务中的性能。解决方案的关键在于提出一种基于词性标注统计（POS-signature）的示范选择策略：该策略通过加权词性分布特征对句子进行表征，并从已标注训练集中检索三个长度兼容且已正确解析的示范样本。实验在英语-网页树库（English Web Treebank, EWT）测试集的全部2,077个句子上进行，对比了两种开源大语言模型在无微调条件下的表现，结果表明，基于POS签名的检索方法在两个模型上的标注附着率（labeled attachment score）分别达到72.4和73.2，显著优于基于语义相似性的三样本基线（分别高出1.9和2.7个百分点），且双侧配对bootstrap 95%置信区间均不包含零值，具有统计显著性。研究还发现，链式思维（chain-of-thought）与批判性优化（critique-refinement）等方法的效果因模型而异。整体结果支持采用基于词性统计的示范选择策略在英文依存句法分析任务中的有效性。
链接: https://arxiv.org/abs/2610.31966
作者: Muhammad Azam Afridi,Rajab Ali,Marco Passarotti
机构: Università Cattolica del Sacro Cuore, Milan, Italy(天主教圣心大学, 米兰, 意大利); University of Zaragoza, Zaragoza, Spain(萨拉戈萨大学, 萨拉戈萨, 西班牙); Universidade NOVA de Lisboa, Lisbon, Portugal(里斯本新大学, 里斯本, 葡萄牙); Université libre de Bruxelles, Brussels, Belgium(布鲁塞尔自由大学, 布鲁塞尔, 比利时)
类目: Computation and Language (cs.CL)
备注: 14 pages, 1 figure, 8 tables. Code and research artifacts: this https URL
Abstract:Demonstrations for in-context learning are often retrieved by semantic similarity, which may be a poor proxy for relevance in structured prediction. We study demonstration selection for Universal Dependencies parsing with a POS-signature policy that represents sentences with weighted part-of-speech statistics and retrieves three length-compatible, gold-parsed examples from the labeled training split. We evaluate two open-weight language models without fine-tuning on all 2,077 sentences of the UD English Web Treebank test set, comparing the policy with shot- and format-matched fixed and semantic three-shot baselines, as well as zero-shot, chain-of-thought, and critique-refinement systems. POS-signature retrieval yields the highest labeled attachment score among the tested prompting systems for both models (72.4 and 73.2), outperforming semantic retrieval by 1.9 and 2.7 points; paired-bootstrap 95% confidence intervals exclude zero for both contrasts. Chain-of-thought and refinement comparisons vary across models. These exploratory results favor the tested POS-signature retrieval approach on English-EWT. We release the prompts, archived outputs, and evaluation code.
[NLP-65] Which Answer Did the Seventeen-Year-Old Write? A Behavioral Case Study of Author-Age Misattribution Miscalibrated Confidence and Non-Corrective Self-Explanation in Six Frontier Language Models
【速读】： 该论文旨在揭示生成式AI在识别人类文本中作者年龄与认知特征方面存在的系统性偏差问题，具体聚焦于模型对不同年龄段写作者回答同一开放性问题（“什么是幸福？你幸福吗？”）时的年龄误判现象。研究发现，尽管17岁青少年的回答表现出典型的年轻化语言特征（如措辞谨慎、形式化），但主流前沿大模型仍普遍将其错误归因于成年人，且在多项实验条件下表现出高达90%的误判率。其解决方案的关键在于识别并验证模型对年龄-语言关联的刻板印象依赖——即模型在缺乏明确上下文线索时，过度依赖社会刻板印象而非实际语言特征进行推断。研究通过操控题干框架、输出格式和受众意识指令等变量，发现仅一个模型在加入简单受众意识提示后性能提升21个百分点，而其他模型则无显著改善；更关键的是，多数模型表现出高度自信但准确性极低的现象，且其自述解释中常包含对自身偏见的反思，却无法反映真实判断质量，暴露出“高自信—低准确”之间的严重脱节。该研究并非统计性调查，而是以三篇样本为核心的机制探索，旨在揭示生成式AI在理解人类语用差异时所面临的深层认知缺陷。
链接: https://arxiv.org/abs/2610.31946
作者: Brianne Lee
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL)
备注: 15 pages, 6 tables. Code and data: this https URL
Abstract:Three students answered the same exam prompt by hand: “What is happiness? Are you happy?” Their ages, recorded at collection, are 17, 28, and 45. The 17-year-old’s answer is hedged and formally phrased. Asked to match each answer to an age and gender label, most frontier language models confidently assign it to an adult. This paper reports a structured case study of that failure across six models: Claude Fable 5, Gemini 3.1 Pro, GPT-5.6 Sol, Grok 4.5, GPT-4o (2024), and DeepSeek V4. We ran 1,081 API sessions and scored 3,225 responses across six experimental conditions that manipulate framing, disclosure format, and audience-awareness instruction. Three of six models perform at or near chance (1/3) on the matching task while reporting 73-88% mean confidence. The youngest writer’s answer is misattributed to an adult in up to 90% of trials. Removing the multiple-choice scaffold collapses accuracy further: three models never place within three years of the true age once open-ended estimation is required. Institutional framing (“a community-college biology final”) significantly degrades the best-performing model’s accuracy; a neutral label does not. A single audience-awareness instruction significantly improves one model (+21 points) and moves another to ceiling, with no detectable effect on the three weakest models. Stated confidence is uncorrelated or inversely correlated with correctness for five of six models. After ground truth is revealed, one model names its own stereotyping in 68 of 114 explanations, yet shows the lowest accuracy and highest confidence of any model under blind presentation: articulate retrospective diagnosis that does not track the quality of its judgments. This is a three-item, single-topic, single-language case study meant to isolate a mechanism, not to estimate a population bias rate. Data, code, and raw model outputs are publicly released.
[NLP-66] Kolibri: A Sovereign European Model on the Pareto Frontier
【速读】： 该论文旨在解决大模型在特定语言（尤其是德语）及受监管领域（如公共行政、工业与航空航天）中实现主权部署与专业化应用的挑战，核心问题在于如何在保证高性能的同时，降低推理成本并确保数据与训练过程的可控性。其解决方案的关键在于构建一个基于混合专家（Mixture-of-Experts, MoE）架构的开源模型Kolibri，该模型拥有781亿总参数，但每令牌仅激活34.6亿活跃参数，显著提升了计算效率。通过自研的数据管道，团队构建了超过2万亿字节的训练数据，其中德语文本占比超过20%，并利用合成生成技术补充高质量德语语料。模型采用端到端可控的全链路开发流程，涵盖从数据采集、架构设计、训练基础设施到后训练优化与评估的全流程自主控制，后训练阶段结合监督微调与强化学习，在超过120万条内部标注任务上进行优化，覆盖推理、工具使用、指令遵循、代码生成和信息检索等能力。最终，Kolibri在英德双语任务中均达到开放模型中的帕累托前沿性能-服务成本平衡，体现了其在专业化、低成本与可信赖部署方面的优势。
链接: https://arxiv.org/abs/2610.31945
作者: Aleph Alpha:Moritz Althaus,Artur Baranowski,Michael Barlow,Fabien C. Y. Benureau,Felix Berkenkamp,Amrita Bhattacherjee,Lukas Blübaum,Duncan Blythe,Augustijn de Boer,Bastian Boll,Thomas F. Burns,Björn Deiseroth,Niko Dürr,Béni Balázs Egressy,Niklas Finken,Niklas Freymuth,Fabian Gebhart,Max Hammer,Bastian Harren,Andreas Hartel,Johann Higl,Steffen Hirschmann,Pablo Iyu Guerrero,Yasser Jadidi,Andreas Jagusch,Mitja Kleider,Markus Klein,Jonas Knupp,Murali Kondragunta,Garwin Lechner,Dominik Lê-Kellner,Jurek Leonhardt,Jiaze Li,Sohir Maskey,Omar Mehio,Pit Neitemeier,Letitia Parcalabescu,Alejandro Petry Pacheco,Pranav Ragupathy,Martin Reinhardt,Peter Rodenkirch Nfissi,Dylan Rodriquez,Leonard Salewski,Raziye Sari,Jordan Sassoon,Ivo Schaper,Ilhan Scheer,Frank Schneider,Philipp Scholl,Lukas Schott,Eric Schreiber,Pablo Schumacher,Alessio Serra,Philipp Siedler,Prabhu Teja Sivaprasad,Mohamed Soliman,Till Speicher,Volker Stampa,Simon Thel,Serge Thilges,Irina Vidal Migallón,Denis Waßmann,Samuel Weinbach,Sascha Wirges,Michael Wüthrich,Alexei Zacharov
机构: Aleph Alpha
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:This report introduces Kolibri, an English-German mixture-of-experts transformer with 78.1B total parameters and 3.46B active parameters per token, released as open weights under the Apache 2.0 licence. We built Kolibri for sovereign and specialised deployment, with a particular focus on German and on regulated domains such as public administration, industry, and aerospace. To support these settings, we control the full model-development process, including data, architecture, training infrastructure, post-training, and evaluation. We train Kolibri on 24T tokens across pre-training, mid-training, and long-context extension, with German accounting for more than 20% of the data mix by bytes. To reach this share, we complement German datasets with more than 2T German tokens that we curate from the web with our own pipeline and generate synthetically. Post-training combines supervised fine-tuning on curated German and English data with reinforcement learning on more than 1.2M internally curated tasks, which cover reasoning, tool use, instruction following, code, and retrieval. The model activates 4.4% of its parameters per token, and among the open models evaluated in this report, it lies on the Pareto frontier of quality and serving cost in English and German. Kolibri builds on Kolibri Origin, an earlier iteration that validated our full pipeline at scale. Both models come from our Model Factory, which connects data, experimentation, training, evaluation, and deployment through version-controlled workflows. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2610.31945 [cs.CL] (or arXiv:2610.31945v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.31945 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-67] Agreement rather than normalisation keeps depth reusable in looped language models ICLR2027
【速读】： 该论文旨在解决循环语言模型（looped language models）在超出训练预算的深度下仍能保持性能可用性的关键问题，即模型深度的可重用性机制。其核心关切在于：这种可重用性是否依赖于循环步骤间的状态归一化（state-norm growth），抑或更根本地取决于训练时与推理时循环结构的一致性（agreement between training and inference loops）。研究发现，移除释放版本Ouro模型中的循环边界归一化操作，在不增加参数且读出路径比特完全一致的情况下，会导致模型在平衡二分类任务上迅速降至随机猜测水平（0.500），表明归一化并非问题根源——因为状态范数仅增长2.021倍，远未达到失效阈值。相反，在从头训练的19M参数合成任务模型中，去除归一化不仅不会损害性能，反而在超预算情况下表现出更强的外推能力（Cliff’s delta最高达0.867），且训练与推理环路的一致性是维持性能的关键。进一步分析揭示，性能退化主要源于读出头与循环块末尾层之间的耦合失配；通过冻结修改后的循环结构并仅重新拟合读出头，可恢复约三分之一的精度，而重新拟合末尾层则能完全恢复性能。最后，当控制超调量相等时，循环架构相对于堆叠深度在两个任务上的优势显著减弱甚至消失，说明此前观察到的循环优势可能由训练预算超调差异所混淆。综上，解决方案的关键在于训练与推理阶段循环结构的一致性，而非状态归一化本身。
链接: https://arxiv.org/abs/2610.31943
作者: Ivan Viakhirev,Kirill Borodin,Maxim Abramov,Grach Mkrtchian
机构: lab260(雅典娜实验室), Yerevan, Armenia; BitmanagerAI(比特管理人工智能), Dubai, UAE; St. Petersburg Federal Research Center of the Russian Academy of Sciences(圣彼得堡联邦科学研究中心, 俄罗斯科学院); St. Petersburg, Russia
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Under review at ICLR 2027
Abstract:Looped language models apply one block of layers repeatedly, and the design pays off only if depth is reusable: loop steps beyond the trained budget must remain usable at inference. We ask whether that rests on normalising the state between loop steps, as the released Ouro models do, or on agreement between the loop a model is trained with and the loop it is run with. On released weights this loop-boundary normalisation can be removed without adding parameters and with the readout path bit-identical, and the operation is an identity at one loop step, which serves as an implementation check. Removing it takes Ouro-1.4B and Ouro-2.6B to zero accuracy by the trained budget on a balanced binary task (chance 0.500). State-norm growth, the failure that prior accounts predict, is not the cause: the norm stays within a factor of 2.021 while the head commits to a single wrong token. Yet in 19M-parameter models trained from scratch on two synthetic tasks, 16 seeds per configuration, training without the normalisation is never worse than training with it and extrapolates better: the two are closest at the trained depth and separate on both sides of it (Cliff’s delta up to 0.867 beyond the budget). Both results follow if agreement, and not the normalisation itself, is what keeps depth reusable. The damage sits in the readout together with the last layer of the block: in the models trained from scratch, refitting the head over the frozen modified loop recovers a third of the lost accuracy and plateaus, while refitting the last block layer with it recovers all of it. Finally, comparing looped with stacked depth at an equal inference loop count is confounded by unequal overshoot of the trained budget; at matched overshoot the advantage of looping vanishes on one task and shrinks on the other.
[NLP-68] LazyFlash: Forward-Dense Backward-Sparse Attention for Accelerated Long-Context Fine-Tuning
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理长序列时，注意力机制（Attention）成为主要计算瓶颈的问题。尽管稀疏注意力（Sparse Attention）在推理阶段已有所应用，但多数方法仍采用密集训练（Dense Training），或无法支持稀疏训练，且其引入的复杂性难以转化为实际的时钟时间（wall-clock）加速，导致密集注意力在实践中仍被广泛使用。针对这一问题，论文提出了一种前向密集、后向稀疏的注意力机制——LazyFlash。其核心创新在于观察到注意力训练过程主要由反向传播主导，并据此设计了基于梯度感知的块级稀疏化策略：在前向传播中保持完全密集以确保模型兼容现有系统（如vLLM）；在反向传播中，通过查询块的注意力输出梯度范数进行重要性排序，跳过对参数更新贡献可忽略的块，从而将密集反向传播转换为块稀疏形式。进一步地，引入梯度质量预算（Gradient-Mass Budget）动态调节各注意力头及训练步的跳过计算量，结合重要性采样保证稀疏梯度在期望上与密集梯度一致。该设计被集成于分块的FlashAttention反向核中，未被跳过的块从不加载或计算，真正实现了梯度稀疏带来的时钟时间加速。实验结果表明，LazyFlash在保持下游任务性能的同时，相比FlashAttention实现了高达5.7倍的注意力反向传播加速，以及1.7倍的端到端训练加速，在长上下文文本与语音大模型上均验证了有效性。
链接: https://arxiv.org/abs/2610.31911
作者: Keqi Deng,Ruchao Fan,Jinyu Li
机构: Microsoft(微软)
类目: Computation and Language (cs.CL)
备注:
Abstract:As large language models (LLMs) process increasingly long sequences, attention becomes a major computational bottleneck. Sparse attention mostly targets inference, yet many methods still train densely or cannot support sparse training, and their added complexity makes sparsity hard to turn into wall-clock speedups, leaving dense attention widely used in practice. This paper introduces LazyFlash, a forward-dense, backward-sparse attention method built on the observation that attention training is dominated by the backward pass. LazyFlash keeps the forward pass exactly dense, so the trained model remains a standard dense model that existing systems such as vLLM can serve without any change. For the backward pass, LazyFlash proposes a gradient-aware block sparsification design that ranks query blocks by their attention-output gradient norms and skips those contributing negligibly to parameter updates, turning the dense backward into a block-sparse one. A gradient-mass budget adapts the amount of skipped computation to each head and training step, while importance sampling preserves the dense gradient in expectation. These designs are implemented inside the tiled FlashAttention backward kernel, where skipped blocks are never loaded or computed, translating gradient sparsity into real wall-clock acceleration. Experiments on long-context text and speech LLMs show that LazyFlash maintains downstream performance while achieving up to 5.7x attention-backward speedup over FlashAttention and 1.7x end-to-end training speedup.
[NLP-69] Knowing When to Stop: The Evidence-Action Gap in Coding Agents
【速读】： 该论文旨在解决生成式智能体在任务执行过程中存在的“证据-行动鸿沟”（evidence-action gap）问题，即智能体在环境已明确表明任务无法继续推进时仍持续尝试、过度坚持、采取不当变通手段并错误声称成功，而未能及时寻求外部帮助。其核心解决方案在于构建一个名为ImpasseBench的评估基准，通过在可解的修复任务中引入12种现实环境阻塞（environment blockers），并在每条轨迹中标记出证据充分足以判断需要外部干预的步骤，从而量化智能体在恰当时机停止努力与过早放弃或过度坚持之间的行为差异。研究发现，尽管几乎所有需外部协助的任务都已积累足够证据，仍有约三分之一的智能体耗尽资源预算，且所有不安全请求均发生在证据充分之后。通过引入“升级工具”（escalation tool）或外部监控机制，可将资源耗尽率从34.0%降至6.7%，并显著提升环境恢复后的任务成功率。然而，当前的升级机制仍存在缺陷：生成的工单虽能准确识别阻塞原因，却常遗漏智能体自身的进展与验证状态。因此，论文强调必须在智能体的移交工具及其提交报告中显式支持“何时停止”的决策，以实现真正有效的协作与任务恢复。
链接: https://arxiv.org/abs/2610.31895
作者: Tianyu Chen,Xinran Zhao,Vijay Viswanathan,Shoubin Yu,Mingyuan Zhou,Tongshuang Wu
机构: University of Texas at Austin(德克萨斯大学奥斯汀分校); Carnegie Mellon University(卡内基梅隆大学); The University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 36 pages
Abstract:Coding agents are trained to complete tasks, and they keep trying even when the environment makes completion impossible. The evidence that only outside help can restore progress usually appears within a few steps, yet agents rarely act on it: they over-persist, attempt undesired workarounds, and claim success the record does not support – a phenomenon we call evidence-action gap. To measure it, we introduce ImpasseBench, which injects twelve realistic environment blockers into otherwise solvable repair tasks and marks, in each trajectory, the step at which the collected evidence becomes sufficient. Scoring actions relative to that step separates appropriate effort from over-persistence and premature handoff. Across eight frontier models and 19,371 trajectories, we find that nearly every run that needs outside help collects sufficient evidence, yet a third still exhaust their budget, and every unsafe request we detect comes after that point. Fortunately, an escalation tool, or an external monitor for models that ignore the tool, already guards against much of this behavior: the tool cuts budget exhaustion from 34.0% to 6.7%, and the monitor nearly triples task success after environment restoration. Even then, a fired escalation is not yet a useful handoff: tickets reliably name the blocker but often omit the agent’s own progress and verification status. Taken together, we argue that knowing when to stop therefore needs explicit support, both in the tool that lets an agent hand off and in the report it leaves behind.
[NLP-70] axoSAE: Rethinking Sparse Autoencoders through Taxonomic Logic
【速读】： 该论文旨在解决现有稀疏自编码器（Sparse Autoencoders, SAEs）在解析大语言模型（LLMs）表征时，无法有效建模知识的层级化（即分类学）结构的问题。传统SAEs将每个特征视为独立单元，导致知识读取缺乏分类一致性，且控制粒度不足，难以精确限定编辑范围。为此，本文提出TaxoSAE，其核心创新在于显式地将“是-一种”（is-a，即上位词-下位词）关系引入SAE特征空间，通过构建分层的SAE堆栈（每层对应分类学的一个层级），并设计分类学对齐损失（taxonomic alignment loss），使每个概念在任一其下位词激活时均被触发。实验表明，TaxoSAE在两个事实问答分类体系（人物关系与动物分类）上显著提升了性能：（1）通过分类结构实现更精准的知识读取，例如对训练中未查询的抽象概念上位词的读取准确率提升31.3点；（2）控制精度更高，概念替换成功率提升25.5点，且移除某概念仅影响其自身答案（其他概念准确率维持91–100%，而对照组为0–2%）；更重要的是，实现了对未见概念的上位词推断读取，上位词读取准确率最高提升37.9点。因此，解决方案的关键在于通过分层结构与分类学对齐机制，使特征空间具备语义层级组织能力，从而实现一致、可解释且可扩展的知识读取与控制。
链接: https://arxiv.org/abs/2610.31894
作者: Xinyu Guo,Minglai Yang,Clayton T. Morrison,Eduardo Blanco,Mihai Surdeanu
机构: University of Arizona(亚利桑那大学); Scale AI(规模人工智能)
类目: Computation and Language (cs.CL)
备注:
Abstract:Recent Sparse Autoencoders (SAEs) decompose the representations of large language models (LLMs) into interpretable features, enabling precise readout and control of encoded knowledge. However, existing SAEs treat each feature as a standalone unit, whereas real-world knowledge is often taxonomically organized. As a result, readout carries no guarantee of taxonomic consistency, and control lacks the granularity to specify the scope of an edit. We propose TaxoSAE, which explicitly encodes taxonomical information, built on is-a (hypernym-hyponym) relations, into the SAE feature space. TaxoSAE mirrors a taxonomy with a stack of SAEs, one per taxonomy level, and introduces a taxonomic alignment loss that trains each concept to be active whenever one of its hyponyms is. On two factual QA taxonomies, one over biographical relations and one over animals, TaxoSAE enables (1) more accurate readout through the taxonomy with concentrated activations, e.g., +31.3 points in reading out the hypernyms of abstract concepts never queried in training; (2) more precise control, with +25.5 points in swap success, while suppressing a concept removes only its own answers (0-2% accuracy vs. 91-100% on other concepts); and, more importantly, (3) readout of unseen concepts through their hypernyms, with up to +37.9 points in hypernym readout accuracy.
[NLP-71] Rewriting Scaling Laws for Synthetic Data with Compute and Data Constraints
【速读】： 该论文旨在解决生成式语言模型在预训练数据构建中使用合成重写（synthetic rewrites）时的关键问题，包括应投入多少计算资源用于生成重写数据，以及基于重写数据训练的模型是否能够超越其重写器（rewriter）性能。其核心解决方案是提出了一种重写缩放律（rewriting scaling law），该定律能够预测模型损失（loss）与模型参数规模、训练样本数量、重写数据占比及自然数据迭代轮数之间的关系，并将生成重写数据所消耗的计算量纳入考量。研究在两种场景下验证该缩放律：在计算受限场景中，利用第三方重写器生成重写数据，发现最优重写比例随计算预算增加而上升（从3×10¹⁸ FLOPs下的不足10%增至10²² FLOPs下的32%），且预测的22B参数模型在仅需重写器十分之一训练算力的情况下，仍可在多个网络领域实现更低的比特每字符（BPB），接近预测性能；在数据受限场景中，重写器与目标模型同规模并基于相同种子数据预训练，结果表明使用自身预训练数据的重写版本进行训练，在所有模型规模（最高达3B）下均优于仅使用原始自然数据的模型。研究表明，通过缩放律视角可系统性地优化合成重写策略，且模型性能提升不必然受限于重写器的能力边界。
链接: https://arxiv.org/abs/2610.31867
作者: Nicholas Roberts,Ahmed Ahmed,Suhas Kotha,Percy Liang,Yu Wang,Alex Renda,Nitya Mani
机构: Princeton University (普林斯顿大学); Stanford University (斯坦福大学); Jane Street (Jane Street)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Synthetic rewrites, where a language model rewrites a natural document e.g. into a target style, are now a standard ingredient of pretraining corpora. However, critical aspects of their use remain poorly understood, such as how much compute should be spent rewriting training data and whether a model trained on rewrites can outperform the rewriter. We study these two questions by proposing a rewriting scaling law that predicts loss from model size, training tokens, the fraction of training tokens that are rewrites, and the number of epochs over the natural data, charging the compute spent generating rewrites. We fit this scaling law in two settings. In the compute-limited setting, natural data is plentiful and rewrites come from a third-party rewriter trained with more compute than our largest run. We find that the estimated optimal rewrite fraction increases with compute, from under 10% at 3\times10^18 FLOPs to 32% at 10^22 FLOPs. Our scaling forecast extrapolates 33\times in FLOPs to train a 22B-parameter model, whose final loss is within 1.1% of our prediction. This model achieves a better BPB than its rewriter on several web domains at one-tenth the rewriter’s training compute. In the data-limited setting, we assume we only have a fixed seed set of natural data and that our rewriter is a model of the same size as the target, pretrained on that seed set. Again, we find the optimal rewrite fraction increases with the compute budget. We also demonstrate that a model trained on rewrites of its own pretraining data, generated by a rewriter of the same size, beats the best run on natural data alone at every model size up to 3B. Our results show that synthetic rewriting can be studied through the scaling law lens: that optimal methodological choices are predictable and scaling does not necessarily break down beyond the rewriter’s capabilities.
[NLP-72] Aligning Large Language Models with Human Feedback: Mathematical Foundations and Algorithm Design
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）与人类意图、偏好及价值观对齐的问题，即如何使LLM的输出行为符合人类期望，避免产生有害或不符合伦理的内容。其核心挑战在于如何有效建模并优化模型以反映复杂且主观的人类偏好。解决方案的关键在于基于人类反馈的训练范式，主要包括监督微调（Supervised Fine-Tuning, SFT）、基于人类反馈的强化学习（Reinforcement Learning from Human Feedback, RLHF）以及直接偏好优化（Direct Preference Optimization, DPO）。其中，DPO通过将偏好学习转化为一个更简洁的监督学习问题，避免了传统RLHF中复杂的策略优化与奖励建模过程，显著提升了训练效率与稳定性。此外，论文从逆强化学习（Inverse Reinforcement Learning, IRL）和离散选择模型的理论视角出发，为理解人类偏好学习提供了坚实的数学基础，从而为设计更鲁棒、可扩展的对齐算法提供了理论支撑与实践指导。
链接: https://arxiv.org/abs/2610.31862
作者: Siliang Zeng,Luca Viano,Chenliang Li,Jiaxiang Li,Volkan Cevher,Markus Wulfmeier,Stefano Ermon,Alfredo Garcia,Mingyi Hong
机构: University of Minnesota(明尼苏达大学); EPFL(瑞士联邦理工学院洛桑分校); Texas A&M University(德克萨斯农工大学); Google DeepMind(谷歌深脑); Stanford University(斯坦福大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:This article provides an introduction to the mathematical foundations and algorithmic frameworks used to align Large Language Models (LLMs) with human intentions, preferences, and values. We discuss standard alignment techniques, such as fine-tuning (SFT), reinforcement learning with human feedback (RLHF), and direct preference optimization (DPO). We also explore the theoretical underpinnings of learning from human preferences, drawing connections to inverse reinforcement learning (IRL) and discrete choice models. We present state-of-the-art algorithms in a tutorial style, discuss their advantages and limitations, and offer insights into practical implementation. Our exposition is intended to serve as a comprehensive resource for researchers and practitioners, providing both a foundational understanding of alignment methodologies and a framework for developing more robust and scalable alignment techniques.
[NLP-73] SWE-Clar: Can Coding Agents Clarify Incomplete Client Requirements and Deliver Greenfield Software?
【速读】： 该论文旨在解决自由职业软件开发场景中，编码代理（coding agent）如何在客户需求不完整或难以持续反馈的情况下，准确理解并实现复杂需求的问题。核心挑战在于：客户需求往往隐含于对话上下文中，且客户无法在长周期开发过程中持续提供明确反馈，导致编码代理即使获得回应也可能存在误解。为应对这一问题，研究提出关键解决方案——将需求澄清（requirement clarification）置于代码实现之前，并通过显式测试来评估对需求的理解程度。其创新性体现在构建了SWE-Clar基准数据集，基于真实高价值的Upwork项目（每项总支付额不低于4000美元），包含100个任务，每个任务均包含多轮人机交互以模拟真实澄清过程。针对澄清效果评估，研究进一步设计了512个与真实客户-自由职业者对话匹配的问题及其参考答案，涵盖显性需求和语境推断出的隐性需求。实验采用九个前沿模型进行验证，结果显示即便最先进的GPT-6 Astra模型在独立交付任务中的通过率也仅为67%，凸显了在无持续反馈环境下实现精准需求理解与代码生成的严峻挑战。该研究通过隔离澄清与实现阶段、确保评估独立性，为衡量编码代理在真实自由职业场景下的综合能力提供了可靠基准。
链接: https://arxiv.org/abs/2610.31854
作者: Jialei Li,Dingxian Wang,Xiaoyan Zhao,Yimeng Bai,Xuanyi Zhu,Yang Zhang,Chao Yin,Wentao Shi,Ganesh Prabhu,Karthik Gomadam Rajagopal,Brett Levert,Darvin Yi,Andrew Rabinovich,Zhao Chen
机构: University of Science and Technology of China(中国科学技术大学); National University of Singapore(新加坡国立大学); Upwork Inc.(Upwork公司)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 29 pages, 9 figures, 11 tables. Code and data: this https URL
Abstract:Greenfield software development in freelance settings requires coding agents to turn incomplete client requests into working software. This makes requirement clarification essential, yet clients may be unable to provide continuous feedback over long development cycles, and receiving their answers does not guarantee correct understanding of complex requirements. These challenges motivate two benchmark design choices: clarification precedes code implementation, and its effectiveness is assessed through explicit tests of requirement understanding. Guided by these choices, we introduce SWE-Clar, a benchmark grounded in real, high-value client commissions to evaluate requirement clarification and code implementation in this setting. It comprises 100 tasks based on real software development jobs on Upwork, each with at least US\ 4,000 in total payments. Each task begins with multi-round interaction between a coding agent and a client agent, followed by separate runs for code implementation with acceptance testing and clarification assessment. For the latter, we construct 512 questions paired with reference answers from real client-freelancer interactions, covering both explicitly stated requirements and those inferred from the conversational context. Both runs begin from the same clarification context but remain isolated, preventing assessment questions from guiding code implementation. Experiments with nine frontier models show that even GPT-6 Astra, which has the highest observed implementation score, achieves a task pass rate of only 67%, underscoring the challenge of independent software delivery in this setting. The dataset and evaluation code are available at this https URL.
[NLP-74] Scaling cross-situational word learning on childrens naturalistic experiences
【速读】： 该论文旨在解决儿童如何习得早期词汇中大量简单、具体名词意义的问题，核心关注点在于跨情境学习（cross-situational learning）这一机制在真实自然语境下的有效性与局限性。其解决方案的关键在于利用对比视觉-语言模型，在第一人称视角视频数据（egocentric video data）上验证跨情境学习策略的可行性。研究发现，尽管模型在小规模数据上已展现出初步成功，但其性能高度依赖于视觉表征的质量：当使用互联网预训练的视觉表示时，模型在对象-标签映射任务上的表现显著优于仅从发展性视频数据中学习视觉表征的模型。此外，绝大多数学习信号来自少量明确描述可见参照物的语句。虽然模型最终习得了具有类人语义结构的词汇表征，但其学习轨迹与儿童的发展模式存在显著差异。因此，构建符合发展规律的词义学习模型，关键在于如何从自然主义的日常经验数据中有效学习高质量的视觉表征。
链接: https://arxiv.org/abs/2610.31842
作者: Michael C. Frank,Alvin W. M. Tan,Khai Loong Aw,Robert Z. Sparks,Victor Moreno,Grace Keene-Lee,Meesha R. Ryan,Tarun Sepuri,Jane Yang,Virginia A. Marchman,Daniel L. K. Yamins,Bria Long
机构: Stanford University (斯坦福大学); Harvard University (哈佛大学); University of Wisconsin, Madison (威斯康星大学麦迪逊分校); University of Oregon (俄勒冈大学); University of California, San Diego (加州大学圣地亚哥分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:How do children learn the meanings of the simple, concrete words that form a majority of their early vocabulary? Statistical information about many word meanings can be recovered from the gradual accumulation of information about the co-occurrence of words with their referents, a proposal known as ``cross-situational learning’'. Contrastive vision–language models allow this learning strategy to be tested on egocentric video data capturing children’s early real-life experiences with visual and linguistic inputs. Initial successes of this method on small datasets provide a proof-of-concept, but vision models struggle to learn robust visual representations when trained on these same egocentric videos. How dependent is cross-situational word learning performance on the quality of the model’s visual representations? And can cross-situational models scale to larger and more diverse datasets? To address these questions, we trained models on the BabyView corpus, including 2,441 hours of longitudinal egocentric video from 48 children. Models trained on this dataset learned object–label mappings, but performance was far higher when they used internet-pretrained visual representations as opposed to when visual representations were learned solely from the developmental video data. Further, the majority of the learning signal was carried by a small proportion of utterances that described a visible referent. Models learned lexical representations that showed recognizable human-like semantic structure, but model performance showed substantially different trajectories over training experiences than those observed in children. Cross-situational learning is a viable word learning strategy for early noun vocabulary, but constructing developmentally plausible models will require understanding how to learn high-quality visual representations from naturalistic data.
[NLP-75] SPECMEM: Accelerating Agent ic Tool Calling via Live Memory Management
【速读】： 该论文旨在解决工具调用智能体（tool-calling agents）在实际部署中因逐个自回归解码工具调用而导致的高延迟问题。现有方法虽采用推测解码（speculative decoding）提升效率，并利用历史调用作为生成草稿，但其关键缺陷在于缺乏跨会话的记忆机制——要么完全不保留记忆，要么仅构建一次静态存储且不再更新。为此，本文提出SpecMem，一种在部署期间持续维护并动态管理的工具调用记忆系统：经过验证的工具调用被实时写回基于用户维度、采用最近最少使用（LRU）淘汰策略的有限存储中；同时引入副作用门控机制（side-effect gate），将不可逆操作延迟至验证通过后才执行。在由三个开源权重模型支撑的多会话、多用户查询流上测试表明，固定存储（frozen store）随时间逐渐失效，而SpecMem保持稳定的接受率——平均接受词元数提升27%（各基准测试中提升11%至34%），且在最坏情况下实现高达1.74倍的时钟速度加速。消融实验表明，性能优势几乎全部源于维持存储的动态活性；局部用户作用域设计虽未提升接受率，但有效阻断了跨用户信息泄露通道；而基于声明式副作用而非检索置信度进行门控，是避免不可逆错误的关键所在。
链接: https://arxiv.org/abs/2610.31840
作者: Zihe Song,Wenting Zhao,Ye Liu,Yuanjie Zhu,Ke Xu,Shanghao Li,Philip S. Yu
机构: University of Illinois Chicago(伊利诺伊大学芝加哥分校); Salesforce Research( Salesforce 研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Tool-calling agents are mostly optimized for accuracy; in deployment, however, their growing cost is latency: every tool call is decoded autoregressively. Speculative decoding accelerates this without changing the output, and retrieved past calls make strong drafts under real use, where each user’s behavior tends to repeat. Yet existing systems keep no cross-session memory, or build the draft store once and never update it. We propose SpecMem, a tool-call memory kept alive and managed throughout deployment: verified calls are written back online into bounded per-user stores with LRU eviction, and a side-effect gate defers irreversible calls until verification confirms. On multi-session, multi-user query streams served by three open-weight models, a frozen store steadily loses acceptance while SpecMem holds it: a 27% gap in mean accepted tokens (+11% to +34% across benchmarks, tracking how much novel work arrives after the store is built) and up to 1.74x wall-clock speedup. Ablations attribute the gain almost entirely to keeping the store live; local per-user scoping adds no acceptance but closes a measured cross-user leakage channel, and gating on declared side effects, rather than on retrieval confidence, is what eliminates irreversible mistakes.
[NLP-76] SAGE: Sense-Aligned Multimodal Grounding Evaluation with Gloss-Augmented Reasoning
【速读】： 该论文旨在解决多模态语义接地（Multimodal Semantic Grounding）中不同基准测试（如VWSD、VerSe和WikiDiverse）之间评估标准不统一、泛化能力难以衡量以及模型对目标项（target-term）的深层语义理解不足的问题。其核心挑战在于现有方法在特定任务上表现优异时，往往无法在跨任务场景下实现一致的性能迁移，且缺乏对模型是否真正理解语义定义与上下文约束的细粒度分析。为此，论文提出SAGE（Sense-Aligned Grounding Evaluation），一个统一的10候选图像选择基准，将三类不同中心化的接地任务（名词、动词、实体为中心）整合至共享的以术语为中心（term-centric）的协议下，并引入具有混淆性难负样本（confusable hard negatives）及上下文交换诊断（context-swapped VerSe diagnostic）的机制，以更严格地检验模型对目标术语的语义接地能力。关键解决方案为Gloss-Augmented Reasoning（GAR），一种无需训练的框架，通过结合定义接地（definition grounding）与约束引导的语义验证（constraint-guided semantic verification），增强模型对词汇语义的显式推理能力。实验表明，仅使用定义增强即可提升所有子集性能，而约束引导推理在VWSD和VerSe上表现尤为突出，但在WikiDiverse上效果有限；二者联合使用时取得最优宏观平均性能，使Hits@1从65.90提升至71.22，证明SAGE不仅支持统一评估，还能实现对不同接地设置下显式语义推理能力的精细分析。
链接: https://arxiv.org/abs/2610.31838
作者: Sunjae Kwon,Suhaas Kiran DG,Zonghai Yao,Talha Chafekar,Deepali Singh,Hong Yu
机构: UMass Amherst(马萨诸塞大学阿默斯特分校); UMass Lowell(马萨诸塞大学洛厄尔分校); UMass Chan Medical School(马萨诸塞大学医学院); VA Bedford Health Care(贝德福德退伍军人健康护理中心)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:This paper introduces Sense-Aligned Grounding Evaluation (SAGE), a unified 10-candidate image-selection benchmark that reformulates VWSD, VerSe, and WikiDiverse into a shared term-centric protocol covering noun-, verb-, and entity-centered grounding. SAGE includes confusable hard negatives and a context-swapped VerSe diagnostic to probe target-term grounding beyond direct caption–image compatibility. We further propose Gloss-Augmented Reasoning (GAR), a training-free framework combining definition grounding with constraint-guided semantic verification. Experiments show that strong VWSD performance does not transfer uniformly across grounding settings. With Gemma3-12B, definition augmentation improves all three subsets, while constraint-guided reasoning improves VWSD and yields particularly strong gains on VerSe, but is less effective on WikiDiverse. Combining the two signals achieves the best macro-average performance, improving Hits@1 from 65.90 to 71.22. These results show that SAGE supports both unified evaluation and fine-grained analysis of explicit semantic reasoning across grounding settings.
[NLP-77] Latent Coherence: Do Frontier Language Models Probability Judgments Obey the Law of Total Probability?
【速读】： 该论文旨在探究当前前沿语言模型在零样本（zero-shot）情境下，其概率判断是否满足贝叶斯一致性，特别是是否符合全概率定律（law of total probability, P(X) = ∑_z P(z)P(X|z)），其中隐变量 z 需从上下文记录中推断。核心问题是：尽管这些模型未经过专门训练以逼近贝叶斯后验，它们在面对需推理隐变量的情境时，其概率估计是否存在系统性不一致。解决方案的关键在于设计一套基于精确真值的三重独立查询测试框架——分别要求模型输出 P(X)、P(z) 以及 P(X|z)，并通过计算残差（log-odds）量化不一致性。研究发现，虽然部分模型如 GPT-6.1 Sol 与 Claude Opus 5.5 展现较高一致性（平均绝对残差分别为 0.08 和 0.12 log-odds），但多数模型存在显著且特定于任务的不一致（0.27–0.35 log-odds），表明其概率推理在涉及隐变量时仍存在结构性缺陷。值得注意的是，残差与真实值误差呈显著正相关（Spearman 0.40–0.62），因此残差可作为无需依赖真值的不可靠回答预警指标。此外，推理长度无法预测一致性，而简单的一致性检查几乎总能通过，说明失败仅发生在需处理隐变量的复杂推理场景中。
链接: https://arxiv.org/abs/2610.31833
作者: Peter Potash
机构: 未知
类目: Computation and Language (cs.CL)
备注: 15 pages
Abstract:LawZero’s Scientist AI trains a Predictor toward the Bayesian posterior, with an objective that penalizes inconsistent probabilities among latent explanations. We ask whether current frontier language models, with no such training, are already consistent in this sense zero-shot: do their probability judgments satisfy the law of total probability, P(X)=\sum_z P(z)P(X|z) , when the latent z must be inferred from records in the context? In separate single-turn calls, models are asked for P(X) , for P(z) over the latent’s states, and for P(X|z) ; a causal family applies the same test to P(y|do(x)) and the back-door adjustment. Every query has an exact ground-truth probability, and a subset of items is asked in three wordings to separate persistent incoherence from rewording noise. Across seven models, GPT-6.1 Sol and Claude Opus 5.5 are the most coherent (mean absolute residual 0.08 and 0.12 log-odds, about 2 percentage points). Claude Haiku 5.5, Gemini 3.8 Flash and DeepSeek V4-Pro show persistent incoherence (0.27-0.35 log-odds) that is mostly item-specific rather than a shared offset. Within every model, larger residuals go with larger errors against the truth (Spearman 0.40-0.62), so the residual, which needs no ground truth, flags unreliable answers. Reasoning length does not predict coherence, across models or within a model across rewordings of the same question. Simple consistency checks pass almost everywhere; the failures are specific to reasoning through a latent variable. Code and data: this https URL
[NLP-78] Free Lunch for LLM s Confidence: The RL Critic Already Knows
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际部署中缺乏可靠置信度估计的问题，而传统方法依赖额外的数据和计算资源进行专门的置信度训练。其核心问题是：置信度训练本质上是在解决何种预测任务？研究发现，仅基于策略行为轨迹（question, response, correctness）三元组训练的外部置信度估计器，能够以与模型自身相当的精度估计其置信度，且无需访问模型内部表示。这一现象揭示了置信度估计与强化学习中的价值函数学习（value function learning）之间存在深层等价关系——在以正确性为结果指标的前提下，置信度估计与批评者（critic）训练在数学上等价，均旨在预测策略响应正确的概率。因此，对于采用基于价值的强化学习（value-based RL）训练的LLM，其原有的批评者网络可直接作为开箱即用的置信度估计器，无需额外训练或数据。该方案保持策略不变，从而不损害任务性能，同时可在生成过程中的任意步骤（包括中间推理阶段）提供置信度输出。在涵盖数学、代码生成、工具调用及通用推理等十余个领域的实验中，复用的批评者在所有分布内测试集上均表现最优，并在16个评估场景中有14个达到最佳校准性，充分验证了其有效性与实用性。
链接: https://arxiv.org/abs/2610.31820
作者: Shujin Wu,Heng Ji
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:Reliable confidence estimation is essential for trustworthy deployment of large language models (LLMs), but dedicated confidence training stage requires additional data and compute. We ask a fundamental question: what prediction problem does confidence training actually solve? Inspired by prior work on external confidence estimation, we compare self- and external estimators of comparable capacity trained on the same (question, response, correctness) triples. We show that an external model trained solely on a policy’s behavioral traces can estimate the policy’s confidence as accurately as the policy itself, without accessing the policy’s internal representations. This points to a deeper connection with value function learning in reinforcement learning (RL): under a correctness-based outcome, we prove that confidence estimation and critic training are equivalent, both predicting the likelihood that a policy’s response will be correct from its trajectory. A simple consequence follows: for LLMs trained with value-based RL, the existing critic can be reused as an off-the-shelf confidence estimator with no additional training or data. The policy remains unchanged, so its task performance is preserved, and the critic provides confidence throughout generation, including intermediate reasoning steps. Across math, code, tool calling, and general reasoning spanning a dozen domains, the reused critic is the best confidence estimator on every in-distribution split and the best-calibrated on 14 of 16, without hurting the task performance.
[NLP-79] LAMB: A Modern Contextual Language Model for Latin
【速读】： 该论文旨在解决拉丁语（Latin）自然语言处理中缺乏高质量预训练模型与专用语料库的问题，特别是针对拉丁语复杂的词形变化结构在现有模型中建模不足的挑战。其核心解决方案是提出一种基于ModernBERT架构的拉丁语编码器——拉丁现代BERT（Latin Modern BERT, LAMB），通过两阶段预训练，在超过740亿个拉丁语标记上采用一种新型的形态学感知掩码策略（morphologically informed masking regimen），以更好地捕捉拉丁语的词形特征。此外，研究构建了两个关键语料库：Acervus（包含240亿个BERT WordPiece标记，涵盖大量经由ByT5模型修复的受损光学字符识别文本）和Corpus Trinum（首个面向拉丁语的语义三元组数据集，包含超1亿个强弱语义三元组），为下游任务提供支持。通过在五个任务、十七个子任务上的广泛实验评估，验证了LAMB在16个子任务中超越所有对比模型，仅在一项任务中落后于当前最优模型0.8个百分点。研究进一步揭示了拉丁语开放类词在预训练后期衰减阶段的独特习得动态，并指出对于某些任务类型，微调配置的选择对下游性能的影响甚至超过额外预训练语料的增益。
链接: https://arxiv.org/abs/2610.31723
作者: Aidan McCarthy
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:This article introduces Latin Modern BERT (LAMB), an encoder for the Latin language based on the ModernBERT architecture, pretrained in two phases on more than 74B Latin tokens using a novel morphologically informed masking regimen. We also present two corpora for the Latin language. Acervus contains 24B BERT WordPiece tokens, including over 760K samples from heavily damaged OCR transcriptions, repaired by a ByT5 model trained for Latin text repair. The second dataset, Corpus Trinum, contains a mixture of over 100M strong and weak Latin language semantic triplets. Corpus Trinum is the first Latin semantic triplet dataset, opening the door for Latin language models to be adapted to general sentence semantic similarity tasks. We construct an experimental grid using five tasks and seventeen subtasks, applying a variety of finetuning configurations and subword aggregation techniques for 15 pretrained encoders, 6 LAMB training checkpoints, and an ablation model designed to quantify the impact of different masking regimens. Using this benchmark, we rank Latin encoders and find that LAMB outperforms all other evaluated models on 16 of 17 subtasks, coming within 0.8 percentage points of state of the art on the seventeenth. We provide additional analyses relating Latin encoder performance to tokenizer behavior, syntactic structure, finetuning configuration, pretraining state, subword aggregation, and variation in source texts and domains. In these, we identify unique acquisition dynamics for Latin open class words during terminal decay phases, and establish that for some task categories, finetuning configuration selection can have a greater effect on downstream scores than additional pretraining tokens.
[NLP-80] Fooling LLM Safety Guards into Judging the Wrong Text
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）应用中安全防护机制被绕过的问题，具体表现为攻击者通过在用户输入中附加伪造指令（forged query）或伪装成提示（prompt）的干扰内容，诱导安全守卫（safety guard）对错误任务进行响应，从而实现“目标重绑定”（target rebinding）。其核心问题是：当前多数安全守卫基于对输入中最后一个完整任务的判断来决定是否放行，导致攻击者可将有害请求隐藏于一个看似无害的“诱饵”（decoy）之后，使守卫误判并忽略真实恶意请求。解决方案的关键在于重新设计守卫的评分机制——训练其对输入内容的评分不低于其中最危险部分的得分，从而确保即使存在后置伪造结构，系统仍能识别并拦截潜在危害。实验表明，该方法将攻击成功率从75%降至0.8%–1.6%，显著提升安全性；尽管在非推理性评估中性能略有下降（AUROC最多下降0.01），但整体表现未明显劣化。此外，研究还发现，部分攻击可通过将伪造结构置于请求前而非后端实现更高效绕过，说明防御策略需兼顾前后文位置敏感性。
链接: https://arxiv.org/abs/2610.31695
作者: Utkarsh Kanwat
机构: 独立研究员(Independent Researcher)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 21 pages (13 main body), 2 figures, 8 tables. Code: this https URL ; model: this https URL
Abstract:Safety guards check content before an LLM application acts on it, and most paste that content into a plain-text prompt. Content that imitates the prompt can append a second task, and the guard answers that one instead. We call this target rebinding. Our explanation is that the guard grades the last complete task in its input, here a harmless decoy, so the harmful request goes through unchanged. In a pre-registered test, an appended forged query and decoy clear all 75 held-out AdvBench requests that Shieldstral-1.0-3B flags. The guard is not persuaded: a directive to ignore the text above clears none of them, and in a follow-up run a harmful decoy stops the attack. On our GuardRebind benchmark, each of five guards has an attack family, chosen per guard, that clears at least 93% of its flagged items. Rebinding explains this best on Shieldstral and gpt-oss-safeguard; on ShieldGemma and Llama Guard 3, generic evasion and trailing text respectively do part of the work, and Qwen3Guard falls mainly to a forgery of its whole task. Training the guard to score content no lower than its most harmful part cuts attack success from 75% to 0.8-1.6% on attack families, decoys and forged queries held out of training. All of these attacks append the forged structure after the request, and we added the held-out-decoy evaluation after a first round of test runs. Clean AUROC falls by at most 0.01, but our test of non-inferiority is inconclusive. A score-guided search still beats this trained guard on 47 of 50 items, usually by placing forged structure before the request. On about half of the items where the search wins, the same search also finds benign text, with no forged structure, that clears the guard.
[NLP-81] Distributional Alignment of LLM s and Its Generalization Properties IJCAI2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在模拟人类行为时难以实现输出分布与人类真实分布对齐的问题，尤其关注在小样本训练条件下如何有效达成分布对齐及其泛化能力。其核心解决方案在于通过大规模基准测试验证，LLMs可通过特定训练策略实现与参考人类分布的分布对齐（distributional alignment），并在同领域未见样本上展现出一定的泛化能力。然而，跨主题泛化能力高度依赖于主题间的语义关联性（如政治类问题与对制度信任的相关性），且已对齐的LLMs可能在未预期的方面出现性能退化（如创造性能力丧失），表明其泛化特性具有内在不可预测性和不稳定性。因此，该研究的关键在于揭示分布对齐的有效性与泛化局限性之间的复杂权衡，强调了在应用中需警惕模型行为的非线性退化风险。
链接: https://arxiv.org/abs/2610.31687
作者: Nelson Jo
机构: 独立研究者(Independent Researcher)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: IJCAI 2026 Workshop on Augmented Democracies
Abstract:Large language models are increasingly used to simulate and probe human behavior, e.g., for surveys. For these applications, a key property is that outputs satisfy distributional alignment: to match its output distribution with some reference (in this case, human) distribution. An obstacle in achieving distributional alignment is generalizing from small training data. In this work, we study (1) how to distributionally align LLMs; and (2) how aligned LLMs might generalize to unseen samples and unrelated topics and tasks. Using two large-scale benchmarks, we find that LLMs can indeed achieve distributional alignment and generalize to unseen questions of the same domain. Cross-topic generalization depends on the relationship between topics (e.g., questions about politics are related to trust in institutions). However, aligned LLMs can degrade in surprising ways (e.g., lose their creative ability), suggesting that its generalization properties are inherently unpredictable and unstable.
[NLP-82] From Target Selection to Condition Reuse: Auditing Residual Interventions in Tool-Use Models
【速读】： 该论文旨在探究在语言模型中通过激活干预（activation interventions）实现目标答案生成后，模型是否能够将所设定的条件（如时间或权限）有效应用于其他计算任务中，从而实现对条件敏感的推理与决策。其核心问题是：激活干预能否使模型不仅在特定任务中正确选择目标，还能在复杂情境下基于条件进行一致且正确的判断与行为执行。解决方案的关键在于引入两种针对特定场景的残差写入（residual-writing）变体，以修改模型内部状态，从而提升目标选择的准确性。实验结果显示，在32个新构建的任务世界中，该方法显著提升了目标选择正确率（从35/128和33/128分别提升至119/128和114/128），但当熟悉事实与标识符被重新组合时，这些性能增益消失，且模型倾向于重复选择曾被奖励的标识符，表明其行为受历史反馈驱动而非真实条件判断。此外，尽管内部状态修改可影响部分判断（如在预设的二级比较中改变50/64次判断），但这种影响并不保证完整动作的正确性。研究进一步揭示了目标选择控制、条件敏感判断与完整行为正确性之间的分离现象，强调必须评估编辑后的内部状态如何被实际使用，而不仅仅是关注其输出结果，才能准确理解模型对条件的真正响应能力。
链接: https://arxiv.org/abs/2610.31686
作者: Tianda Sun,Dimitar Kazakov
机构: University of York(约克大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 21 pages, 5 figures. Code: this https URL
Abstract:Activation interventions can make a language model produce a target answer, but can the model use the intended condition in other computations? We study this question in Qwen3.5-27B on tool-use tasks where time or permissions determine eligibility and selection. Across 32 fresh worlds, two case-specific residual-writing variants raise familiar target-selection correctness from 35/128 to 119/128 and from 33/128 to 114/128. Yet these gains disappear when familiar facts and identifiers are recombined within the same selection task, while selection of previously rewarded identifiers increases. The gains persist across output formats but yield little improvement in independent judgments about whether a candidate is eligible or should be selected. Internal edits can nevertheless change how a source fact affects a judgment: in a prespecified secondary comparison, an early-layer source patch changes 50/64 judgments when that source is used, with one adverse change among 64 cases where it is not. This selective effect does not ensure correct complete actions. Together, these results separate target-selection control, condition-sensitive judgments, and complete-action correctness. Evaluating how an edited state is used, rather than only its trained answer, exposes these differences.
[NLP-83] An Agent ic Approach with Evidence-Grounded Verification for Multimodal Emotion Reasoning in Conversation NEURIPS2026
【速读】： 该论文旨在解决基于大语言模型（LLM）的多模态情感推理系统在采用指令微调时计算成本高昂的问题，同时应对由冻结模型构建的智能体流水线在预测错误时难以定位故障组件的可追溯性难题。其核心解决方案是提出一种可追踪的多模态情感推理框架，并结合提示空间自适应（prompt-space adaptation）机制。该框架包含模态特定的感知智能体，分别处理视觉、听觉和语义输入，并将带有证据引用的观察结果发布至共享信息板；一个集成智能体融合这些观察与检索到的多模态示例，生成情感描述、标签集及支持性引文；此外，配备一个基于多模态大语言模型（MLLM）的批评者智能体，允许一次修订后再定稿。当预测出错时，引入独立的基于证据验证阶段，通过确定性检查验证引用证据的有效性，并利用蕴含关系检测判断最终结论是否被证据支持。失败案例按责任组件和故障类型进行分组，转化为针对性的提示补丁，并在预留数据集上验证后才被采纳，从而实现无需更新模型权重的提示优化。实验表明，在MER2025数据集上，该方法在所有四种不同规模（2B至38B参数）的多模态大语言模型上均优于静态冻结流水线，且达到或超越无差别的反思基线，例如在Qwen2.5-VL-7B上将F-score从52.3提升至54.5，性能提升达0.6–2.2点。
链接: https://arxiv.org/abs/2610.31685
作者: Elyas Rashno,Farhana Zulkernine
机构: Queen’s University (皇后大学); Kingston, Ontario, Canada
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Multimedia (cs.MM); Image and Video Processing (eess.IV)
备注: Accepted at the NeurIPS 2026 Workshop on Real-Time Multimodal Conversational AI (ReMuCAI), Paris, France. OpenReview: this https URL
Abstract:LLM-based multimodal emotion reasoning systems perform well on standard benchmarks, but improving them through instruction tuning can be computationally expensive. Agentic pipelines built from frozen models avoid this cost, but introduce a different challenge: when a prediction is wrong, the system must identify which agent or stage caused the error. We address this problem with an agentic framework for traceable multimodal emotion reasoning and prompt-space adaptation. Modality-specific perception agents analyze visual, acoustic, and lexical inputs and post evidence-cited observations to a shared board. An integration agent combines these observations with retrieved multimodal examples to produce an emotion description, label set, and supporting citations, while an MLLM critic can request one revision before the output is finalized. For incorrect predictions, a separate evidence-grounded verification stage applies deterministic checks to verify that cited evidence and references are valid, and an entailment check to determine whether the final claims are supported by that evidence. Detected failures are then grouped by the responsible component and failure type, converted into targeted prompt patches, and accepted only after validation on a held-out set. This adapts the agent prompts without updating model weights. On the MER2025 dataset, targeted prompt adaptation outperforms the frozen static pipeline on all four MLLMs, ranging from 2B to 38B parameters, and matches or outperforms undifferentiated reflection baselines on each of them: VideoLLaMA3-2B, Phi-3.5-vision-instruct, Qwen2.5-VL-7B-Instruct, and InternVL2.5-38B. It improves F-score by 0.6–2.2 points depending on the backbone, including an increase from 52.3 to 54.5 on Qwen2.5-VL-7B.
[NLP-84] Knowledge Graphs Quality Assessment in the Era of LLM s: A Conceptual Framework
【速读】： 该论文旨在解决当前知识图谱（Knowledge Graphs, KGs）在与大语言模型（Large Language Models, LLMs）协同应用时所面临的质量评估难题。随着KGs被广泛用于为LLMs提供事实性、可追溯性和高可信度的响应，其内在质量直接影响下游应用的可靠性，然而现有研究缺乏系统性的质量评估标准。论文的关键解决方案在于构建一个通用、可定制且面向具体领域或任务的质量评估框架：通过整合文献综述，扩展出涵盖24个质量维度（Quality Dimensions, QDs）和多种质量度量（Quality Metrics, QMs）的综合体系，并提出一个具备实践性与灵活性的评估框架，以支持对KGs在不同应用场景下的全面质量评价，从而为判断其是否适合作为LLM的可信知识支撑提供科学依据。
链接: https://arxiv.org/abs/2610.31684
作者: Elwin Huaman
机构: Centre for Human-Inspired Artificial Intelligence (人机启发式人工智能中心); Institute for Technology and Humanity (技术与人性研究所); University of Cambridge (剑桥大学)
类目: Computation and Language (cs.CL)
备注: 8 pages, 1 figure, KGSWC 2026
Abstract:Knowledge Graphs (KGs) have been popularized during the last few decades and are increasingly constructed and consumed by Large Language Models (LLMs). Various applications therefore rely on KGs to provide concise, complete, accurate, faithful, fresh, and traceable answers. However, what is the quality of those KGs? In which cases should a KG be used, and can it be trusted to ground an LLM? How might it be evaluated? We reviewed the literature on quality assessment (QA) of data, information, linked data, KGs, and the intersection of KGs and LLMs. We extended the current state-of-the-art frameworks to a catalog of 24 quality dimensions (QDs) and various quality metrics (QMs). Furthermore, we propose a general-purpose, customizable to a domain or task, and practical QA framework for assessing the quality of KGs.
[NLP-85] he Schema Coherence Inventory: Measuring Whether an Assigned Belief Organizes a Language Models Cognition
【速读】： 该论文旨在解决生成式语言模型在作为模拟患者或情感代理时，难以验证其行为是否由特定信念（belief）驱动，还是仅受情绪状态影响的问题。核心挑战在于区分模型行为是源于深层认知结构（如负性自我图式）的组织性作用，还是仅由临时情绪波动所引发。解决方案的关键在于提出图式一致性量表（Schema Coherence Inventory, SCI），该量表包含96个强制选择题，覆盖贝克认知模型的八个层级，每题提供六种回应选项，分别对应三个负性自我图式及其正性对立面。通过长度归一化的对数似然得分，避免了对生成内容的语法解析，确保评估聚焦于语义一致性。判别标准为“目标特异性”：针对某一信念设计的条件应显著更频繁地激活该信念，而非同极其他信念。在五个不同参数规模（1.7B–14B）的模型上开展的六项验证研究均支持此判别逻辑，临床训练评分者对关键信念的识别准确率达79.5%（远超随机水平16.7%），且信念适配器在所有模型上均表现出高度目标特异性（平均提升17.7分，30/30信念单元为正向）。此外，一个48项简版量表与完整版具有高度相关性（r = 0.73），且能有效识别自由生成文本中的信念特征。将SCI应用于六个已发表的患者模拟器，发现其均无法超越一句简洁陈述所传达的核心信念。该验证路径通过分离认知构念与伴随情绪，并结合临床专家评判，为语言模型中情感与信念的测量提供了可复用的范式，具备跨任务类型、评价者、机构及语言的扩展潜力。研究团队已公开全部题目、评分代码及简版量表。
链接: https://arxiv.org/abs/2610.31683
作者: Yuqi Wu,Deyang Li,Katharine Hibbard,Ron Oswald,David Sinclair,Adegboyega Sapara,Kalutota Kuruwe Arachchige Sajana Indula Samarasinghe,Abhinav Pillai,Reham Shalaby,Riffat Alam,Xin-Min Li,Yanbo Zhang,Jie Chen
机构: Fudan University (复旦大学); University of Alberta (阿尔伯塔大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Persona-conditioned language models are used as simulated patients and affective agents, but there is no validated way to ask whether a belief assigned to such a model organizes its behaviour or whether the model has merely adopted a mood. We present the Schema Coherence Inventory (SCI): 96 forced-choice items at eight levels of Beck’s cognitive model, each offering six replies keyed to three negative self-schemas and their positive counterparts, scored by length-normalized log-likelihood so that nothing is generated or parsed. Affect alone can move a model onto a default belief, so the discriminant criterion is target specificity: a condition built for one belief must select that belief more often than conditions built for the other beliefs of its pole. Six validation studies over 107 conditions on five backbones (1.7B-14B parameters) support this reading. Nine clinically trained raters recover the keyed belief at 79.5% (chance 16.7%). Belief adapters are target-specific on every backbone (+17.7 points; 30/30 belief cells positive), and a condition’s SCI profile identifies its own free generations in 24/30 cases against 5 expected by chance. A 48-item short form correlates r = 0.73 with the other 48 items. Applied to six published patient simulators, none carries its formulation’s core belief better than a one-sentence statement of it. The validation route, which separates a construct from the affect that accompanies it and checks model judges against clinicians, offers a template for affective measurement in language models; we outline its extension across task types, raters, sites and languages. We release the items, scoring code and short form.
[NLP-86] Multi modal perceived conversation success prediction for spontaneous autistic and non-autistic dyadic interactions ICASSP2027
【速读】： 该论文旨在解决自闭症谱系障碍（autism spectrum disorder, ASD）个体与非自闭症个体在对话中感知对话成功度（perceived conversation success）降低的问题。现有研究多聚焦于单一模态特征（如语音声学、词汇使用或非语言行为）的预测能力，或仅用于诊断分类与群体差异分析，难以全面捕捉对话成功的复杂性。本文的关键解决方案在于整合多种互补模态的特征——包括发音动作（articulatory）、面部表情、声学-韵律特征以及轮流说话同步性，并基于三个包含自闭症与非自闭症个体的自发双人互动数据集进行建模。通过多模态融合分析，揭示了对话成功本质上是多维度的，必须采用多层面的分析框架才能有效预测。该方法显著提升了对感知对话成功度的预测性能，因其能够捕获单一模态无法提供的信息，从而为理解并改善自闭症与非自闭症个体间的沟通质量提供了新的技术路径。
链接: https://arxiv.org/abs/2610.31682
作者: Thanushi Withanage,Desi Jones,Noah Sasson,Elizabeth Redcay,Carol Espy-Wilson
机构: University of Maryland (马里兰大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Submitted to ICASSP 2027
Abstract:Conversation success is important for communication, social bonds, and well-being and is often reduced in autistic-neurotypical conversations. Previous studies have predicted perceived conversation success using acoustic and lexical features in pediatric autism-enriched samples, or using nonverbal behavior in non-autistic group interactions. Other work has characterized prosodic, facial, motor, and turn-taking synchrony in autistic interactions, but has primarily targeted diagnostic classification or group differences. The present study instead predicts participant perceived conversation success from complementary articulatory, facial, acoustic-prosodic, and turn-taking measures across three spontaneous dyadic-interaction datasets including autistic and non autistic individuals. By leveraging multimodal features, we demonstrate that conversational success is inherently multidimensional and requires a multi-faceted analytical framework. Combining these complementary modalities offers the potential to improve prediction of perceived conversation success by capturing information unavailable from any single modality.
[NLP-87] Rethinking Structured Generation: Can Graph-Based Reasoning Resolve Ambiguity? NEURIPS2026
【速读】： 该论文旨在解决传统结构化生成任务中因假设输入到输出为确定性映射而带来的根本性缺陷，即忽略了自然语言固有的多义性——同一输入可能对应多个在语义上均成立但结构各异的输出。现有方法依赖单参考监督，将多重合理解释压缩为单一预测，导致在输入信息不充分时输出不稳定且不一致。其解决方案的关键在于将结构化生成重新建模为一种歧义感知推理（ambiguity-aware inference）问题，强调对一组合理结构化输出进行联合推理，而非强制选择单一结果。为此，论文提出一种多有效学习（multi-valid learning）框架，在训练过程中主动保留潜在的多种有效解，避免标准训练中对合法替代方案的抑制。具体实现上，构建结构化输出候选空间，以图结构表示每个候选输出中的实体及其关系，并通过基于语义对齐与关系一致性约束的图推理机制进行优选，从而实现超越词元级似然的、更符合人类认知的跨解释比较。实验表明，该方法在多个结构化生成基准上显著提升了准确性、稳定性和对不同歧义水平的鲁棒性，验证了显式建模歧义的重要性，并推动结构化生成从单输出预测向集合级推理范式转变。
链接: https://arxiv.org/abs/2610.31681
作者: Ratun Rahman,Atit Pokharel
机构: The University of Alabama in Huntsville(阿拉巴马大学亨茨维尔分校)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted as NeurIPS 2026
Abstract:Structured generation is typically formulated as a deterministic mapping from input to a single structured output. This assumption is fundamentally misaligned with natural language, which often admits multiple structurally distinct yet semantically valid interpretations. As a result, existing methods rely on single-reference supervision that collapses ambiguity into one prediction, leading to unstable and inconsistent outputs under underspecified inputs. We reformulate structured generation as an \textitambiguity-aware inference problem, where the objective is to reason over a set of plausible structured outputs rather than predict a single solution. We show that standard training inherently suppresses valid alternatives, and instead adopt a multi-valid learning framework that preserves ambiguity during optimization. To operationalize this formulation, we construct a candidate space of structured outputs, represent each candidate as a graph capturing entities and their relationships, and perform selection via graph-based reasoning over semantic alignment and relational consistency. This enables principled comparison across competing interpretations beyond token-level likelihoods. Experiments on structured generation benchmarks demonstrate consistent improvements in accuracy, stability, and robustness under varying levels of ambiguity. These results highlight the importance of modeling ambiguity explicitly and suggest a shift from single-output prediction to set-level reasoning in structured generation.
[NLP-88] MisconceptDialogue: Evaluation Criteria Shape Measured Success in LLM Math Tutoring AACL
【速读】： 该论文旨在解决两个核心问题：其一，单个大语言模型（Large Language Model, LLM）能否通过多轮对话有效识别学生在数学解题中的错误、诊断其潜在的认知误区（misconception），并实施针对性的补救教学；其二，当前广泛采用的基于精确匹配（exact-match）的评估指标是否可信，能否真实反映模型的教学能力。针对上述问题，论文提出MISCONCEPTDIALOGUE框架，这是一个全自动化系统，通过在五个开源权重的中等规模LLM（7B-14B）数学导师与模拟学生之间开展超过5000次对话，构建了一种受控的、可重复的评估范式。研究发现，以标准数学自然语言处理（NLP）中常见的精确匹配评分法为基准时，模型的补救成功率仅为42.1%，而由人工校准的LLM作为评判者所给出的评分则高达81.3%，两者相差达39.2个百分点，远超模型本身、教学策略及诊断信息供给等变量的综合影响。该结果并非单一评判者的偏差，经四名独立LLM评委和200份人工标注验证，评分一致性极高（kappa值最高达0.91）。此外，在能力评估方面，无论向导师提供真实的认知误区信息还是人为替换的错误诊断信息，其表现几乎无差异，表明这些导师在当前诊断-呈现格式下缺乏对诊断输入的有效依赖。因此，论文强调，未来大语言模型数学辅导的评估必须转向多准则评分、多评判者共识以及人类锚定相结合的标准流程，且评估方法本身的可靠性应与模型能力本身获得同等重视。
链接: https://arxiv.org/abs/2610.31679
作者: Woojin Lee,Yoonkyu Woo,Jinah Kim,Hyeok-Min Gwon,Jin-Xia Huang
机构: Electronics and Telecommunications Research Institute, Republic of Korea(韩国电子通信研究院); Department of Artificial Intelligence, University of Science and Technology, Republic of Korea(韩国科学技术大学人工智能系)
类目: Computation and Language (cs.CL)
备注: Accepted to the AACL-IJCNLP 2026 Main Conference. 20 pages, 1 figure, 16 tables
Abstract:Large language models are increasingly used as math tutors, raising two questions: can a single LLM detect a student’s error, diagnose the underlying misconception, and remediate it through multi-turn dialogue - and can we trust the numbers that report how well it does? We address both with MISCONCEPTDIALOGUE, a fully automated framework that runs the full pipeline on five open-weight LLM tutors (7B-14B) interacting with LLM-simulated students over 5,000+ dialogues - a deliberately simulated protocol (paired conditions, a wrong-diagnosis control unsafe with real learners, scale) whose results do not establish real-student learning. On measurement, our headline finding is uncomfortable for the field: benchmark-style exact-match scoring common in standard math NLP - a lower-bound stress test - reports 42.1% measured Remediation Success Rate on identical dialogues where a human-calibrated LLM-as-judge reports 81.3%, a 39.2 percentage-point spread larger than the combined effects of tutor model, teaching strategy, and supplied diagnostic information. This is not a single-judge artifact: four independent LLM judges and 200 human annotations converge to within a few points (kappa up to 0.91). On capability, the verdict is equally striking: the score barely changes whether we tell the tutor the student’s actual misconception or a deliberately swapped one - under the tested diagnosis-presentation format, these tutors show little evidence of reliably conditioning on the diagnostic information they are handed. For LLM tutoring evaluation to mature, multi-criterion scoring with multi-judge convergence and human anchoring should become standard practice - and the measurement question deserves as much scrutiny as the capability question it claims to answer.
[NLP-89] he Butterfly Defects: Tracing Cooperation in LLM Agent Networks
【速读】： 该论文旨在解决在大型语言模型（LLMs）作为互联智能体协同运作的背景下，个体合作行为是否能够转化为集体合作这一核心问题。其关键在于揭示合作并非模型固有的稳定属性，而是高度依赖于模型组成、交互伙伴、网络结构位置以及决策语言表述方式的动态过程。研究通过将迭代囚徒困境（IPD）从双人互动扩展至同质与异质的开放权重LLM网络，发现即使模型对相同同伴表现出合作，一旦进入混合群体则可能转向背叛；微小的语言变化（如动作标签顺序或不确定性提示）可显著影响合作水平，并通过网络传播至未直接暴露的远端代理，导致成本重新分配和行为剧变。这表明，合作行为本质上是情境依赖的，其策略性表现受制于语境框架、网络拓扑及交互环境，从而挑战了将合作视为单一模型层面特性的传统认知。
链接: https://arxiv.org/abs/2610.32726
作者: Maria Sampani,Maria Lymperaiou,Angeliki Dimitriou,Giorgos Filandrianos,Chrysoula Zerva,Giorgos Stamou
机构: National Technical University of Athens(雅典国立技术大学)
类目: Physics and Society (physics.soc-ph); Computation and Language (cs.CL)
备注:
Abstract:As large language models (LLMs) increasingly operate as interconnected agents, a fundamental question emerges: does individual cooperation translate into collective cooperation? We investigate this question through the Iterated Prisoner’s Dilemma (IPD), moving from dyadic interactions to homogeneous and heterogeneous networks of open-weight LLMs. Our findings reveal that cooperation is highly sensitive to model composition and seemingly minor linguistic changes. Models that cooperate with identical peers can shift toward defection in mixed populations. A simple uncertainty cue can increase or decrease cooperation, redistribute costs, and affect unexposed agents several hops away. Even reversing action-label order can dramatically alter behavior, with uncertainty framing sometimes amplifying this sensitivity. Together, these findings challenge the notion of cooperation as a stable model-level property, revealing how strategic behavior depends on interaction partners, network positions, and the language used to frame decisions.
[NLP-90] Concentrated Causal Influence Distributed Diagnostic Information: Cognitive Impairment Decisions in 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）在认知障碍（Cognitive Impairment, CI）检测任务中“黑箱”决策机制的问题，即模型的预测结果缺乏可解释性，无法明确识别哪些内部组件控制决策过程，以及其表征中究竟包含多少可用于CI诊断的信息。解决方案的关键在于提出并验证“认知障碍对比神经元”（CI-contrast neurons, CICNs），这些神经元通过解码器前馈网络（MLP）中类别对比激活频率进行筛选。研究通过因果消融实验，在八个开放LALM模型与两个数据集（Pitt和PROCESS-2）上，对小规模的CICN集合进行干预，结果显示：在五项符合条件的模型中，三个模型的少量CICN消融即可引发显著且方向明确的预测偏移；同时，基于语音特征提取的掩码仍能控制仅依赖转录文本的决策，且多数核心Pitt模型的CICN在PROCESS-2上保持特异性。位置受限干预表明，这些神经元的影响主要集中在最终答案生成之前，尤其是在字母标记选项下语音或转录证据之后的位置，表现为决策方向的调控。此外，对冻结的CICN激活值进行监督线性读出，在Pitt数据集上达到0.92–0.97 AUC，显著优于模型直接输出的0.67–0.84 AUC，而随机掩码与全神经元读出性能相当，表明高阶诊断信息集中于少数关键神经元，而非广泛分布。该研究揭示了大模型中存在集中式的因果控制机制，与广泛分布的诊断信息之间的重要区分。
链接: https://arxiv.org/abs/2610.32061
作者: Xiutian Zhao,Luqi Sun,Björn Schuller,Berrak Sisman
机构: Johns Hopkins University (约翰霍普金斯大学); Imperial College London (帝国理工学院)
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL)
备注:
Abstract:Large audio-language models (LALMs) can assess cognitive impairment (CI) from speech, yet their answers reveal neither which internal components control the decision nor how much CI information is accessible in their representations. We investigate both questions through CI-contrast neurons (CICNs), selected by class-contrastive activation frequency in decoder MLPs. Across eight open LALMs and two corpora (Pitt and PROCESS-2), we causally test CICNs against size- and layer-matched random masks and activity-matched controls. In three of the five models that meet our eligibility criteria, ablation of small CICN sets causes large, class-directed prediction shifts; speech-identified masks also control transcript-only decisions, and three of four primary Pitt masks retain specificity on PROCESS-2. Position-restricted interventions locate the influence before the final answer position, most strongly after the speech or transcript evidence under letter-labeled options, largely as a decision-direction shift. Beyond causal localization, supervised linear readouts of frozen CICN activations achieve 0.92–0.97 AUC within Pitt, versus 0.67–0.84 for the models’ direct answers. Matched random and all-neuron readouts reach comparable performance; top-ranked CICNs show only an exploratory advantage at very small unit budgets. These results distinguish concentrated causal control over output decisions from broadly distributed diagnostic information for CI detection in LALMs.
信息检索
[IR-0] Separating Thinking from Spelling for Efficient Generative Recommendation
链接: https://arxiv.org/abs/2610.32980
作者: Danil Gusak,Anna Volodkevich,Ivan Oseledets,Evgeny Frolov
类目: Information Retrieval (cs.IR)
备注:
Abstract:Generative recommenders retrieve items by writing their semantic IDs (SIDs) code by code under trie-constrained beam search, an approach now deployed on large web platforms. Standard autoregressive interfaces pay for the SID length K twice: a history of T items enters the backbone as TK code tokens, and every beam-search step runs through the backbone, so returning B candidates costs O(BK) backbone decode tokens per user. The second cost often dominates, and it grows with the backbone, the SID depth and the candidate budget. We introduce Deco (decoding from concepts), which separates reading the history from writing the identifier. The backbone reads each history item as one fused token and condenses the history once into a compact next-item concept; a light KV-cached decoder conditioned on the concept and the generated prefix runs the entire beam search, while the identifiers, the trie and the autoregressive factorization stay unchanged. The decoder approaches full-backbone accuracy with cross-entropy alone, and distillation improves it further. On seven datasets from e-commerce, movies and music streaming, Deco achieves comparable or better accuracy than both a standard generative recommender and the full-backbone teacher with fused inputs, while delivering 3.2–8.6 \times their throughput under a 50 ms p95 latency budget in a production-style serving pipeline. Delegating beam search makes larger candidate pools and models practical: Deco generates 1000 candidates faster than its teacher generates 100, and with a 1.5B-parameter LLM backbone, it matches the 69M-parameter teacher’s throughput while achieving higher accuracy. Subjects: Information Retrieval (cs.IR) Cite as: arXiv:2610.32980 [cs.IR] (or arXiv:2610.32980v1 [cs.IR] for this version) https://doi.org/10.48550/arXiv.2610.32980 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-1] Knowledge Distillation across Item and Code Spaces for Parallel Generative Recommendation
链接: https://arxiv.org/abs/2610.32943
作者: Yixuan Xu,Hailin Liu,Linxiang Peng,Can Yang
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 12 pages, 2 figures
Abstract:Generative recommendation with semantic IDs represents each item as a sequence of discrete codes derived from its semantic representation and predicts these codes to generate recommendations. Existing semantic-ID recommenders are typically trained with hard labels for the observed next item. Although such supervision identifies the target, it provides no explicit distributional information about alternative items conditioned on the same user history. Sequential recommenders can supply this item-level predictive knowledge, but transferring it to semantic-ID generation is challenging because the teacher and student operate in item and code spaces, respectively. We propose two complementary knowledge-distillation objectives for parallel semantic-ID recommenders, which predict all code positions simultaneously without autoregressive prefix conditioning. Item Distribution Matching (IDM) reconstructs a catalogue-level item distribution from the student’s parallel code predictions and aligns it with the teacher’s item distribution. Code Marginal Matching (CMM) projects the teacher’s item probabilities onto each code position using fixed item-to-code assignments, then matches the resulting code marginals to the corresponding student heads. Both objectives transfer the teacher’s soft distributional knowledge while preserving the student’s parallel prediction structure. Experiments on benchmark datasets show that both methods consistently outperform existing state-of-the-art approaches. On average, IDM improves Recall@10 by 15.39%, while CMM improves NDCG@10 by 8.28%. These results demonstrate that distillation in either reconstructed item space or code-marginal space can substantially strengthen hard-label supervision for parallel generative recommendation.
[IR-2] Every Edge Needs a Quote: Building a Quote-Grounded Knowledge Graph of Healthcare Software Adoption with Language-Model Agents ALT
链接: https://arxiv.org/abs/2610.32781
作者: Juan Felipe Duran,Arthur Tobler,Martin Vallières
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 19 pages, 3 figures, 9 tables. Code: this https URL
Abstract:Hospitals regularly buy software that clinicians and patients depend on. Because these purchases are costly, hard to reverse and can affect patient care, buyers look for evidence of which peer institutions already use a product and which systems it connects to. This evidence is public but scattered across vendor websites, press releases and third-party pages. Language-model agents can now collect it at low cost, but the correctness of the extracted claims is difficult to establish, and a wrongly listed customer can mislead buyers and misrepresent an organisation. We present a pipeline that builds a knowledge graph of customer and integration relations from the open web under one rule: every edge must be supported by a verbatim quote from its source page. Pages come from a deterministic crawl of each vendor’s site and a budgeted ReAct agent that searches the rest of the web. A permissive extractor proposes relations with quotes, and a layered grounding stage decides which to keep, using a referent step that reads the whole page when a quote names only one side, deterministic checks, and a panel of five model calls voting on nine criteria. On 61 vendors, the pipeline reaches 0.94 precision and 0.71 recall against a page-level key covering the 6,428 candidate relations proposed by its extractor, whose verdicts were drafted by one model and re-checked with a second. Against the golden reference, the marketplace’s own curated list, it scores lower, but 89% of the relations it keeps that the reference lacks are stated on a page. A confidence score built from the panel’s votes keeps more correct relations than the fixed rule at the same precision, and reaches 0.90 precision at 0.90 recall. We also discuss how narrow, bounded model calls, string-matched evidence and voting address long-context degradation and run-to-run variance.
[IR-3] From archive to resource: bringing legacy data back to life at scale with large language models
链接: https://arxiv.org/abs/2610.32616
作者: C. H. Vieira-Vieira,I. Schneider Chernov,A. Serra,M. Correa Marrero,P. L. Bhavani Mohan,G. Duchateau-Nguyen,T. Y. Doktorova,A. Kreuchwig
类目: Information Retrieval (cs.IR)
备注:
Abstract:Pharmaceutical companies maintain vast archives of legacy preclinical study reports that contain study-context details, such as fasting, anesthesia, and laboratory instrumentation, that are rarely captured in structured data formats, yet they can correlate with clinical pathology parameters values and limit the reuse of historical data. This work demonstrates practical, scalable solutions to recover such descriptors from heterogeneous report texts using large language model (LLM)-based extraction pipelines developed independently by two industry teams. The pipelines identify and normalize metadata fields and return verbatim evidence with page and section reference, hasting human verification and iterative refinement. Evaluation of over 400 documents against expert-curated reference values showed high agreement across a broad set of metadata fields. Performance was strongest for consistently reported items and lower for descriptors subject to linguistic or report structure variability. Extraction time has shrunk from 25 minutes in the manual benchmark to 1.6 minutes. As a concrete biological demonstration, we show that fasting status extracted from report text was associated with species- and analyte-specific shifts in clinical chemistry parameters, illustrating how recovered metadata can improve selection of historical controls. Key strengths of the approach are scalability, traceability, and the ability to recover information missing from current industry standards. Main limitations are dependence on source-report quality and initial expert effort to define and refine schemas. These results establish how LLMs can convert heterogeneous, unstructured legacy reports into evidence-grounded, reusable and structured metadata, leading to better experimental comparability, and thus enabling reliable data pooling or informed virtual control selection.
[IR-4] UAE GovTrust Scope Filtering Gains and Routing Failures in Bilingual Public Service Retrieval
链接: https://arxiv.org/abs/2610.32608
作者: Aleksandar Mihajlovic
类目: Information Retrieval (cs.IR)
备注: 14 pages, 2 figures, 5 tables. Reproducibility materials included as ancillary files
Abstract:Public-service queries can mention several locations while only one controls the requested service. We study this distinction in UAE GovTrust, a controlled English-Arabic retrieval diagnostic comprising 22 authored service summaries and 50 scenarios with 110 query variants. Twelve methods separate lexical and dense ranking, rank fusion, scope filtering, a multilingual cross-encoder, and direct service lookup. On 72 matched answerable queries with bilingual evidence, role-filtered reciprocal rank fusion retrieves a relevant first result for 69 queries (95.8%), versus 47 (65.3%) for unfiltered fusion; filtered dense retrieval and direct lookup each achieve 68 (94.4%). With English-only evidence, excluding zero-score lexical candidates raises filtered fusion from 52 to 64 correct first hits. However, the unchanged router selects the correct action and scope for only 2 of 168 post-audit perturbations covering paraphrases, irrelevant facts, negation, and corrections. These perturbations are deliberately authored diagnostics, not held-out evaluation. Fresh-environment execution reproduces all 1,320 original ranking orders and actions. The results show that correct scope metadata can improve ranking while brittle scope inference can remove the evidence needed to answer. The study evaluates retrieval and routing, not generated answers or legal correctness; independent Arabic and domain validation remains necessary.
[IR-5] Agent ic Memory Management Is the Next Frontier of Information Retrieval
链接: https://arxiv.org/abs/2610.32464
作者: Qingyao Ai
类目: Information Retrieval (cs.IR)
备注:
Abstract:Information retrieval has long been focusing on connecting people with information they need to complete their tasks. Recently, the meaning of this has changed dramatically as more and more AI agents now consume, retain, and act on information on a user’s behalf. Consequently, information management now also takes place \emphinside the agent in the form of memory management. In this perspective paper, we argue that \emphagentic memory management (AMM), which studies the structure, analysis, organization, storage, searching, retrieval, and \emphuse of information in an intelligent agent’s memory, is a natural extension of IR research and an area in which the IR community can make distinctive contributions. Specifically, we first present a system-level framework for AMM and introduce a computational taxonomy of memory objects as \emphconstants, \emphfunctions, and \emphoperators, each with distinct retrieval semantics. Then, we compare AMM with established IR research and identify their difference and connections. Last but not least, we outline a research agenda covering benchmarks, candidate principles of memory utility, optimization procedures that mix parametric and non-parametric updates, and more. We hope this paper could attract more people, both inside and outside the IR communities, to think more about AMM problems and research from IR perspectives.
[IR-6] Hierarchical GTV Estimation: Bridging the Gap between Container-Level Ranking and Item-Level Conversion RECSYS’26
链接: https://arxiv.org/abs/2610.32420
作者: Zerong Lan,Fan Zhang,Chuang Chen,Tianchun Huang,Teng Zhang,Xingxing Wang
类目: Information Retrieval (cs.IR)
备注: 5 pages; published in the 20th ACM Conference on Recommender Systems (RecSys '26)
Abstract:Emerging e-commerce platforms, such as live-streaming and Online-to-Offline (O2O) services, inherently operate under a “Container-based” display paradigm. In this setting, a critical granularity mismatch exists: the platform performs ranking and estimation at the container level (e.g., livestream rooms or stores), whereas the actual transactions occur at the item level within these containers. This structural discrepancy complicates Gross Transaction Value (GTV) estimation, introducing unique challenges including high label variance, severe label sparsity, and dynamic item heterogeneity. To address these challenges, we propose a novel framework named Hierarchical GTV Estimation (HGE). HGE employs a Hybrid Set-Aware Encoder (HSAE) to model the dynamic composition of items and feature interactions. To mitigate label variance, we introduce a Proxy-Label Learning (PLL) strategy that decomposes post-click GTV into purchase probability, quantity, and unit price. Furthermore, a Residual Correction and Ensemble Fusion (RCEF) module is designed to handle label sparsity by effectively combining item-level aggregation with container-level inference. Extensive experiments on large-scale industrial datasets demonstrate that HGE significantly outperforms state-of-the-art baselines, achieving a +0.63% lift in XAUC offline and a +1.42% increase in Revenue Per Search (RPS) in online A/B tests. HGE framework has been fully deployed into our main traffic.
[IR-7] SkillRep: Characterizing and Understanding the Representation Design Space of Agent Skill Retrieval
链接: https://arxiv.org/abs/2610.32336
作者: Tianyi Sun,Pengfei Yao,Yanze Wang,Hongda Chen,Yunyun Han,Jun Sun,Chuanrui Hu,Yafeng Deng
类目: Information Retrieval (cs.IR)
备注: 12 pages, 3 figures. Code: this https URL
Abstract:Retrieving the right skill from a large library is essential for LLM agents, yet requests and skills are written for different purposes and in different words. Existing retrievers narrow this request-skill gap with one or two representation choices each, so it is unknown whether their gains hold in combination. No prior work measures the gap itself, so it is also unclear which part of it each choice narrows. We define a representation design space by five choices – encoder fine-tuning, query-side and document-side expansion, background projection and a lexical channel – and evaluate them together for the first time, as modules of one retriever, SkillRep. With a 0.6B encoder and LLM expansion, SkillRep improves MRR@10 on SRA-Bench over the strongest same-size retriever by 7.02 points with the official skill pool and 8.79 with UniPool, a 111,247-skill pool we build from public registries, and is best on SkillsBench. The value of each choice depends on the others. Fine-tuning, for instance, reverses the ranking of the four training-free modules and turns them from complementary into overlapping. To understand why, we measure the gap with three quantities: background, the embedding component that does not separate skills; unsaid vocabulary, the gold skill’s content a request leaves unsaid; and dilution, the information lost when a skill is encoded into a single vector. They show which part each module narrows and by what mechanism. This reveals that modules narrowing the same part compete for it, while those narrowing different parts complement each other.
[IR-8] System One models Are Not Better Rerankers
链接: https://arxiv.org/abs/2610.32318
作者: Michael Dinzinger,Kanishka Ghosh Dastidar,Michael Granitzer
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:System One models, or decision models, answer a question about an input with a calibrated probability instead of generated text. Within weeks of the first release, in September 2026, they were used as rerankers. We test whether the decision fine-tuning behind them makes a better reranker than plain reranker fine-tuning of the same model. For six Qwen backbones from 0.6B to 27B parameters and ten TREC DL and BEIR collections, we compare the untuned model, asked zero-shot, with its decision fine-tunes and its reranker fine-tunes, two of which we train ourselves. In our work, we draw the conclusion that decision fine-tuning never makes the better reranker. From 0.6B to 9B, reranker fine-tuning wins on every backbone by 5 to 29 points nDCG@10, also on zero-shot collections, and at 27B the two tie. Up to 9B, decision fine-tuning adds nothing over simply asking the untuned model. Moreover, we find that the calibration advantage of decision models depends on the question. Asked whether a passage answers the query, they are better calibrated than the published reranker fine-tunes from 4B on. Asked a slightly different question from prior work, they lose this advantage. Lastly, calibration does not make up for a worse ranking: up to 9B, a reranker fine-tune keeps the better passages at a fixed threshold. If you need a reranker, fine-tune a reranker.
[IR-9] PersonalQuery-Benchmark:Evaluating Retrieval Robustness to Personalized User Expression Differences
链接: https://arxiv.org/abs/2610.32164
作者: Wenyu Liang,Hongzhi Yin,Tongtong Wu,Fan Liu,Kexin Dai,Weiqing Wang
类目: Information Retrieval (cs.IR)
备注:
Abstract:Existing personalized information retrieval methods mainly model differences in user needs through user preferences and query intent, while paying little attention to the personalized expression differences that arise when different users express the same retrieval need. Existing real query data typically lack stable user identities, long-term expression histories, and intent-matched counterparts, making it difficult to systematically evaluate the impact of personalized expression differences on retrieval outcomes. To this end, we propose PersonalQuery-Benchmark (PQB). PQB models personalized syntactic expression and personalized writing errors from users’ historical texts, constructs personalized syntactic expression queries and personalized writing-error queries, and validates benchmark quality through LLM assessment and independent human evaluation. PQB covers three Amazon product domains with 6,516 query instances in total, and is evaluated on five retrieval models representing lexical, learned sparse, dense bi-encoder, late-interaction, and hybrid architectures, together with two LLM-based ranking models. Results show that even when the information need is held fixed, personalized expression differences still induce measurable ranking variation; different retrieval models exhibit clearly different sensitivities to syntactic variation and writing-error variation, while higher overall retrieval effectiveness does not necessarily correspond to stronger robustness against expression variation. PQB provides a new benchmark and evaluation framework for systematically studying personalized expression robustness. The associated code and dataset are available at the following address: this https URL.
[IR-10] From Item Scores to Semantic Tokens: Test-Time Collaborative Adaptation for Generative Recommendation
链接: https://arxiv.org/abs/2610.32098
作者: Xiongfeng Shan,Xing Tang,Hanchong Chen,Fuyuan Lyu,Lingjie Li,Jingyan Jiang,Duguang Liu,Xiuqiang He
类目: Information Retrieval (cs.IR)
备注:
Abstract:Generative recommenders using semantic identifiers predict items through successive token decisions, but beam search may irreversibly prune promising item paths. This limitation is particularly consequential when a few recent interactions signal an emerging interest while earlier behavior favors established preferences. Yet an independently trained discriminative recommender may rank the same items highly from the same interaction history. Such complementary scores are available at test time, offering a way to adapt a frozen generator without retraining either model. The challenge is to use these scores before pruning: combining completed recommendation lists cannot alter earlier beam decisions, while scores over complete items cannot directly inform choices among semantic prefixes shared by multiple items. We propose \textbfTest-Time \textbfCollaborative \textbfBeam \textbfSearch \textbfAdaptation (\textbfTCBSA), a training-free method that incorporates collaborative information into semantic decoding. TCBSA transforms discriminative score margins into potentials over shared semantic edges and uses a confidence gate to regulate their contributions to cumulative path scores before beam selection. We establish a sufficient condition for target prefix survival under global beam competition and bound score perturbations under low discriminative confidence. Full-catalog experiments show consistent improvements over strong baselines across datasets and different generative and discriminative recommenders. Analyses of emerging interests and decoding behavior demonstrate gains in recommendation accuracy and target path survival, while targeted interventions support the contribution of shared prefix corrections. TCBSA also offers a favorable balance between recommendation accuracy and inference latency compared with simply widening the native beam.
[IR-11] Scaling Graph Query Processing: Focused Exploration on Partitions Using Load Balancing
链接: https://arxiv.org/abs/2610.32015
作者: Harshit Modi,Abhishek Santra,Sharma Chakravarthy
类目: Databases (cs.DB); Distributed, Parallel, and Cluster Computing (cs.DC); Information Retrieval (cs.IR)
备注:
Abstract:The focus of this paper is on query processing and not search (approximate or exact). The goal is to find all or k answers and the expressiveness of a query (as opposed to a search expression which is typically simpler) is considered important. Although SQL query processing has matured over the last several decades, query processing over graph data has not received the same attention. Only recently, systems, such as Neo4J, and others have started using graphs for data representation on a large scale. Their approaches and algorithms remain proprietary. In this paper, we address the problem of answering graph queries and their efficient processing on arbitrary-sized graphs. We propose the partitioning of graphs and the use of the distributed paradigm to achieve efficiency and scalability. Our approach takes a subgraph as query input including wild cards and comparison operations. Our partitioned approach leverages focused exploration and distributed processing to achieve the desired parallelism and improve the response time. We also present heuristics to reduce the number of partitions loaded for query processing. We present our partitioning algorithm, distributed processing architecture used, and load balancing for computing query answers. We establish the correctness of the approach in the presence of partitions, compare heuristics for load balancing, and extensive experimental results over large synthetic and real-world graph data sets. Subjects: Databases (cs.DB); Distributed, Parallel, and Cluster Computing (cs.DC); Information Retrieval (cs.IR) Cite as: arXiv:2610.32015 [cs.DB] (or arXiv:2610.32015v1 [cs.DB] for this version) https://doi.org/10.48550/arXiv.2610.32015 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-12] A Model-Agnostic Correction Framework for Short-Term Building Load Forecasting
链接: https://arxiv.org/abs/2610.31959
作者: Shiqiao Zhou,Huanbo Lyu,Holger Schöner,Zipeng Wu,Jessica Deuschel,Freya Wentzel,IAG Wilson,Shuo Wang
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: Work in progress
Abstract:Short-term building load forecasting remains challenging because load patterns can change rapidly with weather and operating conditions. Data-driven forecasting models may become less reliable when the current operating regime is rare or poorly represented during training, even when similar situations exist elsewhere in the historical record. Recent advances in retrieval-augmented time series forecasting offer a promising way to adapt predictions to changing conditions by directly reusing relevant historical cases at inference time. Motivated by this idea, we propose WeL-RAC, a model-agnostic Weather-Load Retrieval-Augmented Correction framework that explicitly retrieves such evidence at prediction time. WeL-RAC retrieves historical cases that are similar to the current condition based on both recent load states and horizon-aligned weather forecasts. Their observed future load trajectories are then aggregated and used to correct the initial prediction, without retraining or modifying the base forecaster. We evaluate WeL-RAC on 31 building cases across three datasets, eight forecasting backbones, and four horizons from 1 to 18 hours. The framework improves the corresponding base forecasters in 90 of 96 dataset-level comparisons. Beyond accuracy, success and failure case studies further illustrate the framework’s interpretability by tracing how retrieved historical evidence improves or degrades forecasts. These findings support explicit historical retrieval as an effective and interpretable complement to the building load forecasting.
[IR-13] MICE to Meet You: From Re-ranking Only to End-to-End Retrieval
链接: https://arxiv.org/abs/2610.31932
作者: Victor Morand,Yuxuan Zong,Mathias Vast,Josiane Mothe,Benjamin Piwowarski
类目: Information Retrieval (cs.IR)
备注: code: this https URL
Abstract:Late-interaction models offer state-of-the-art efficiency-effectiveness trade-offs for neural retrieval by enabling document representations to be pre-computed and efficiently indexed. Yet, their relatively simple interaction mechanisms, typically MaxSim, can limit their effectiveness compared to cross-encoders. MICE, a recently introduced re-ranking architecture, aims to combine cross-encoder-like effectiveness through expressive transformer-based scoring with reduced inference-time computation through document-side pre-computation. However, no practical way to store these document representations was explored. We address this limitation here and introduce MICE-P, the first end-to-end retrieval pipeline with MICE, based on PLAID to index the contextualized document token representations. We conduct a controlled comparison of MICE, ColBERT and cross-encoders by training comparable models using the Ettin Family under matched experimental conditions. Our results show that MICE-P defines a new class of retrieval architecture with an efficiency-effectiveness trade-off of its own: it matches late-interaction retrieval throughput while retaining the effectiveness of cross-encoders.
[IR-14] Mining Frames: Association Rule Mining as an Unsupervised Framework for News Framing Detection WWW
链接: https://arxiv.org/abs/2610.31874
作者: Hager Hesham Mustafa(Anmat Computational Analysis Lab)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY); Information Retrieval (cs.IR)
备注: 35 pages, 5 figures, 4 tables. Working paper. Dataset: this https URL
Abstract:Computational framing analysis is dominated by supervised classifiers that require large annotated corpora and predefined frame taxonomies, limiting their use for non-English languages, novel events, and low-resource settings. This paper presents Association Rule Mining (ARM) as an unsupervised, annotation-free framework for news framing detection. Each document or sentence is treated as a transaction of lexical items, and high-lift co-occurrence rules are read as candidate framing signals (Gamson and Modigliani’s condensing symbols), confirmed in context by journalist-researchers. The primary validation is Framing Gaza: 25,512 English-language articles from 16 US, UK, French, and German outlets covering the first year of the Gaza war (October 2023 to October 2024), analyzed with Apriori (mlxtend), platform-weighted support, POS-filtered n-grams, and lift-based scoring. ARM inductively surfaces interpretable patterns without predefined categories: a cross-platform military frame, israeli, strike to kill, appears as a closed rule in 14 of 16 outlets, and a month-by-month analysis recovers event-driven framing pivots (e.g., the April 2024 World Central Kitchen strike) with no prior event labeling. A systematic headline-body divergence is identified, with headlines carrying concentrated, high-lift editorial framing and body text reproducing institutional, procedural language. A methodological validation on the Seven Days project (1,575 Arabic articles on the 2026 Iran war) shows that FP-Growth enables sentence-level analysis with far lower memory overhead. Findings were validated through editorial uptake: two published investigative stories (Daraj Media; Saheeh Masr). Preprocessing code is released openly; the full pipeline is documented but withheld to preserve operational deployability.
[IR-15] Machine-Readable Operator Manuals over the ISO 11783 Network for Offline Multilingual LLM Assistance
链接: https://arxiv.org/abs/2610.31863
作者: Timo Oksanen,Julius Gun
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:If agricultural machinery is to support intelligent, offline assistance, then the system must access technical documentation dynamically. In this paper, the Semantic Synchronization of operator manuals over ISO 11783-based CAN buses is analyzed, defined as the process that brings the local knowledge base of a tractor-side assistant into agreement with the operator manual of the connected implement. Data transfer performance was benchmarked using Markdown, JSON, and XML; various open-weight Large Language Models (LLMs) were also evaluated in cross-lingual query tasks. The automatic evaluation of the answers was validated against four human raters. A complete manual was transferred over a physical ISO 11783 network in 31 to 89 s, depending on format and transport settings, which is comparable to the object pool upload that operators already wait for at the first coupling of an implement. Markdown produced the smallest files, while the question-answering performance was comparable across the three formats for most models. With retrieval-augmented generation, locally run models answered most questions on the English manual correctly, with no advantage of the largest model over 7B and 8B models, but performed worse on the five other language editions. Keyword retrieval failed across languages, whereas semantic retrieval provided the cross-lingual capability. The exchange requires only a new parameter group that uses the existing request and transport mechanisms of ISO 11783, which makes it a candidate for standardization. The study is a proof of concept with inference on PC-grade hardware; deployment on embedded agricultural hardware remains future work.
[IR-16] What the Query Remembers: Query Abstraction Governs Lexical and Dense Retrieval of Classical Arabic Poetry
链接: https://arxiv.org/abs/2610.31821
作者: Omar Farouk Zouak,Selma Mani,Larbi Said Chikh,Raid Abderrezak Ouahioune
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:How well a poetry archive can be searched depends on what the user still remembers, not only on what the model understands. We show this for Classical Arabic poetry, benchmarking nine dense encoders and BM25 over two corpora, three indexing modes and two evaluation protocols. Which retriever wins is governed by how much of a poem’s wording the query still carries. On tip-of-the-tongue queries BM25 leads every dense encoder, including an 8B state-of-the-art model (0.746 vs. 0.717 Recall@5); on abstract queries it falls to eighth out of ten (0.392 vs. 0.748). Editing the wording of the same queries shows that this ranking largely reflects query composition: adding one remembered half-verse raises BM25 by 38 points, rewriting the quoted words by hand leaves BM25 level with the 8B encoder, and BM25’s sensitivity to remembered wording is significantly steeper than that of the strongest dense encoder. The same axis controls explanation augmentation, whose effect swings from +37 to -39 points. The mechanism splits by system type: lexical containment predicts BM25’s behaviour, while embedding proximity predicts dense encoders. Finally, the field’s two standard protocols, known-item retrieval and thematic relevance, dissociate on the same systems, a result reproduced by two blind human annotators and by two open-weights judges, one of them under a rubric with no literal-match criterion. A benchmark that reports one protocol, or one query register, can give a materially wrong picture. Code and data are available at this https URL.
人机交互
[HC-0] Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox
链接: https://arxiv.org/abs/2610.32955
作者: Federico Lucca
类目: Computation and Language (cs.CL); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注: 8 pages, 7 tables. Single-user case study. Code and protocol: this https URL . The emails are private and not released
Abstract:Personal email agents decide, for every message, what it is, whether it is urgent, whether to draft a reply, create a task, or ignore it. Their error rate is rarely measured: the stored labels were written by the system itself. We present Foyer, a self-hosted personal inbox agent in daily use, and a measurement protocol (stratified sampling, frozen snapshots, thread-grouped splits, pre-registered hypotheses, human labels in the agent’s review loop) that yields ground truth before any training. On 739 real Italian/English emails labelled by the owner (231 unbiased, 500 adversarially sampled), we compare nine systems under thread-grouped cross-validation: the production keyword rules, a TF-IDF + logistic-regression baseline, local typed decision models (Tev1, Nimble), a hosted decision model (Nimble v4-max), and hosted generative LLMs (Qwen2.5-7B, Gemma 4 26B/31B). Four results. The production rules score 28% on message type against a 58% majority class: the field shown most prominently carried no information. With ~200 labels, logistic regression has the highest accuracy on every field (type 0.79, urgency 0.74, needs-reply 0.83, is-noise 0.90), though it never predicts the two rarest classes and is statistically tied with Gemma 4 on type; few-shot prompting does not close the gap. The sender’s relationship (client, supplier, service) is recovered from other emails of the same contact with 98.6% accuracy where the contact has been seen, and is not inferable from a single message by any model: it belongs in the contact graph, not the prompt. Finally, accuracy is not the product metric: composed into actions, the most accurate system most often ignores emails needing action (18% vs 4% for the best LLM); a cost-chosen threshold on its noise probability halves that rate and makes it the cheapest system.
[HC-1] Runtime BiDi Structural Verification Catches Right-to-Left Layout Defects that Accessibility Actionability and Logical-Property Linting Miss
链接: https://arxiv.org/abs/2610.32934
作者: Karan Yadav
类目: oftware Engineering (cs.SE); Human-Computer Interaction (cs.HC)
备注: 7 pages, 2 figures, 4 tables
Abstract:Bidirectional (right-to-left, RTL) layout defects arise when interfaces authored with physical directional CSS (margin-left, border-left, text-align:left) are rendered under dir=“rtl” without being mirrored. Accessibility engines (axe-core), actionability engines (Playwright), and even Stylelint’s current core logical-property rule fail to localise these defects. We present runtime structural invariants – a zero-golden mode over computed style and geometry, and a bidi-conservative differential mode – that catch them in a headless browser. On a held-out benchmark of 14 real RTL bugs reconstructed from dated fix commits in Bootstrap and Material Components Web, the detector recovers 13/14 (2 via zero-golden, 11 via the differential mode), with no false alarm on any patched version; on the actual committed historical this http URL it catches 1 of 2 attempted. It reports 0 false positives on 16 negative LTR-island cases and, in both modes, on 43 unmodified real components across four design systems and three RTL architectures (rule-of-three 95% upper bound = 7%), and it surfaced a real directional defect in the current Bulma 1.0.4. By contrast, Stylelint’s core rule emits an identical 1744 warnings on Bootstrap’s LTR and correct-RTL builds, pseudo-localisation detects none of the mirroring bugs, and a naive pixel mirror-and-diff reaches only F1 0.87 with a test-set-tuned threshold. The two modes are complementary: zero-golden is the sole detector of the decoration/styling-asymmetry class (5/6) but false-flags the correct logical border-inline-end, while the differential detects none of that class yet never false-alarms and is empirically exact on the block-level inset/margin class. We scope this paper to detection; repair is future work.
[HC-2] STROKE: Simple Sparse Structural Occlusions for Black-Box Adversarial Attacks
链接: https://arxiv.org/abs/2610.32932
作者: Ankit Gupta,Akshat Pandya,Paul Lin,Christoph Adami,Emily Dolson
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC)
备注: 13 pages, 4 figures, 4 tables, and 16 pages of Supplementary Information with 3 figures and 5 tables
Abstract:We introduce STROKE, a query-efficient score-based black-box adversarial attack that constructs perturbations from simple oriented rectangular occlusions. STROKE optimizes these perturbations through a lightweight greedy add-and-refine procedure. Across diverse image classifiers, STROKE achieves high attack success with low query counts at nominal 1–2% area settings, showing favorable performance relative to existing patch-based attacks in several configurations. Matched comparisons further demonstrate substantial reductions in wall-clock attack time, while ablation studies examine the key design choices and effectiveness across structural primitives. These findings suggest that simple structural occlusions offer an effective and computationally efficient search space for black-box adversarial attacks.
[HC-3] What Mean Error Misses: Device-Clock Diagnostics for Evaluating Eye Tracking across Three Virtual Reality Headsets
链接: https://arxiv.org/abs/2610.32915
作者: Aleksandr Chernysh,Mehedi Hasan Raju,Kamrul Hasan,Md Arif Shahriar,Proloy Karmakar,Isayas B. Adhanom,Oleg Komogortsev
类目: Human-Computer Interaction (cs.HC)
备注: 14 pages, 6 figures; supplementary material (16 pages) included. Submitted to ETRA 2027. Data: this https URL ; code: this https URL
Abstract:Gaze research in virtual reality needs eye trackers of known error. Thirty participants completed the same protocol on the Meta Quest Pro (QP), HTC VIVE Pro Eye (PE) and HTC VIVE Focus Vision (FV); to our knowledge, this is the first evaluation of the FV eye tracker. In a random-saccade task, median participant mean error was 1.04° on QP, 1.50° on PE and 2.41° on FV, which remained least accurate after linear recalibration. In a later block with a darker background, PE error rose by 0.70°. FV samples were positively correlated: dispersion was 9.8 times the sample-to-sample root mean square. The device clocks showed losses that mean error hides. QP paused for up to 6.04 s, and FV skipped single samples. During instructed eye closures, QP flagged every sample valid while gaze moved 9.74°. Evaluations should report repeated timestamps, loss composition on the device clock and validity during eye closures.
[HC-4] When Should I Trust This? Designing and Evaluating Factuality Indicators for LLM -based Chatbots
链接: https://arxiv.org/abs/2610.32883
作者: Greta Warren,Jessica Heal,Sila Eser,Denise M. Leto,Christine M. Lehane,Isabelle Augenstein
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Generative AI systems increasingly mediate how people seek information, yet communicating AI factuality to laypeople remains challenging. Through a mixed-methods investigation involving interviews (N=8), a survey (N=1,000), and an experiment (N=5,151), we examine how people evaluate AI-generated information for gardening, mathematical, and medical queries, and how factuality indicators (confidence and sources) shape perceptions. Our interviews and survey show people selectively verify AI responses using external and AI-provided evidence, moderated by contextual factors. Although confidence-indicators affected trust and accuracy perceptions, people hold diverse, ambiguous interpretations of AI confidence. Sources were viewed as crucial, yet users raised concerns about source reliability and verification effort. Our experiment showed that confidence and source-based indicators influenced initial perceptions, but effects disappeared once participants learned the response’s correctness. Medical queries elicited skepticism regardless of indicator. Findings motivate helping users interpret model confidence and integrating source information into the interface to support calibrated trust in AI-supported information-seeking.
[HC-5] Beyond pass or fail: an information-processing account of the Turing Test
链接: https://arxiv.org/abs/2610.32868
作者: Qiwei Zhao,Oriana Pansardi,James W. Strachan,Eugenio Scaliti,Guido Manzi,Giulia Borghini,Fabio Manzi,Alessandro Rufo,Frédérique de Vignemont,Michael S. A. Graziano,Stefano Panzeri,Cristina Becchio
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:The Turing Test is passed when a human interrogator cannot distinguish machine-generated responses from human ones. Yet behavioral indistinguishability can reflect two fundamentally different failures: a failure of encoding, in which machine and human output distributions overlap so closely that source information is effectively absent; or a failure of readout, in which source information is present but the interrogator fails to extract it. To dissociate these possibilities, we inverted the Turing Test, placing humans and AI systems in symmetrical interrogator roles. Using a quantitative information-processing framework, we separated encoding, the information contained in the text; readout, the extraction of that information by an interrogator; and meta-readout, the interrogator’s explicit access to the features guiding its own judgments. Across two series of experiments involving more than 3800 human interrogators and two generations of GPT models (GPT-4 and GPT-5), we found a striking dissociation across processing levels and interrogator types. As AI-generated responses became more human-like from GPT-4 to GPT-5, source information became more difficult for human interrogators to extract. GPT-5 interrogators, by contrast, read out source-discriminating information more accurately than GPT-4 interrogators. This improvement, however, came with a specific limitation: GPT-5 explanations only partially aligned with the feature weightings that actually predicted their judgments. Human interrogators showed the complementary pattern, with modest readout alignment but comparatively higher meta-readout alignment. Together, these findings suggest a double dissociation between readout and meta-readout across human and GPT interrogators.
[HC-6] SPOOK: Sparse Pixel Optimization for Off-Manifold Knowledge in Vision Encoders
链接: https://arxiv.org/abs/2610.32835
作者: Ankit Gupta,Akshat Pandya,Christoph Adami,Emily Dolson
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注: 12 pages, 3 figures, 3 tables, and 14 pages of Supplementary Material with 8 figures
Abstract:Modern vision encoders are widely used as semantic representations, but the relationship between embedding similarity and human-visible semantics can become unclear far from the natural-image manifold. We probe this regime in a black-box setting by applying sparse pixel edits to blank canvases with a simple pixel-level hill-climbing procedure. Across ImageNet and Flickr30k, we evaluate whether these edited blank canvases can retrieve chosen targets in multimodal contrastive encoders and self-supervised visual encoders. On OpenCLIP, inputs optimized toward class prompts achieve strong top-1 and top-5 targeted retrieval performance; on DINOv3, inputs optimized toward class prototypes achieve high target alignment and near-top prototype ranks. These results show that sparse edits on otherwise blank, visually non-semantic inputs can enter target semantic basins in both language-aligned and self-supervised visual representation spaces. Our findings identify a shared off-manifold failure mode of modern vision encoders and suggest sparse blank-canvas optimization as a simple probe of representation geometry.
[HC-7] Political sense-making in “individual” human-technology-world relations: the case of LLM -generated autobiographies
链接: https://arxiv.org/abs/2610.32594
作者: Anthony Longo,Olya Kudina
类目: Human-Computer Interaction (cs.HC)
备注: Reference: Longo, A., and Kudina, O. (in press). “Political sense-making in ‘individual’ human-technology-world relations: the case of LLM-generated autobiographies.” In D. Michelfelder, J. Kennedy (Eds.), Postphenomenology Meets Artificial Intelligence. Bloomsbury Academic
Abstract:As AI systems increasingly mediate how people describe, recognize, and present themselves, their politics cannot be reduced to what remains hidden inside the “black box.” This chapter argues that AI’s political significance emerges within technologically, intersubjectively, and temporally mediated interpretive practices. Bringing Arendt’s political philosophy into conversation with Gadamerian hermeneutics, the chapter develops a postphenomenological relational-temporal account of political sense-making across Human-Technology-World relations. Against the charge that postphenomenology remains individualistic and using the case of LLM-generated biographies, the chapter shows how seemingly private human-LLM encounters are worldly and collective. When users ask systems such as ChatGPT to describe them, they do not simply receive algorithmic representations. They interpret, contest, edit, share, and publicly test these descriptions. In doing so, people move interpretively among affective resonance, friction, public re-articulation, and the sedimentation of meaning. This means that the so-called “micro-level” of postphenomenological analysis is already saturated with publicity: affective uptake becomes communicable, reflection becomes accountable, and uses sediment into shared standards of appearance. Simultaneously, we show that once mediation is treated as relational and temporal, the political is not an add-on at the “macro” level; it is intrinsic to the very movement through which technologies amplify and reduce, stabilize and unsettle, and thereby compose a shared world in which judgment can take place. The chapter finishes by discussing several design and policy implications for AI of such political foregrounding in postphenomenology.
[HC-8] From Regulation to Engineering: Organizational Privacy Climate as Mediator
链接: https://arxiv.org/abs/2610.32576
作者: Oshrat Ayalon,Eran Toch,Irit Hadar,Michael Birnhack
类目: Computers and Society (cs.CY); Human-Computer Interaction (cs.HC); Software Engineering (cs.SE)
备注:
Abstract:Data protection legislation aims to protect privacy rights by regulating how information systems store and process personal information. While such legislation defines legal obligations, translating regulatory requirements into engineering practice is far from straightforward. As new frameworks such as the EU AI Act emerge, understanding these mechanisms becomes increasingly relevant. To investigate them, we conducted an online survey of 156 U.S. information system developers, examining the role of organizational privacy climate in the association between developers’ perceptions of the GDPR and their privacy engineering practices. Using structural equation modeling, we found that organizational privacy climate mediates the relation between developers’ perceived GDPR importance and their privacy attitudes, which in turn are associated with their privacy engineering practices. The mediation chain - regulation, climate, attitudes, practices - complements the direct effect of regulation on practice, revealing the organizational and attitudinal pathway through which regulatory intent reaches engineering behavior. We discuss implications for the next generation of privacy and AI regulation.
[HC-9] From Perception to Pixels: Human-Derived Visually Distinct Colors for Improved Dominant Color Extraction
链接: https://arxiv.org/abs/2610.32473
作者: Uldana Shyndali,Nurdaulet Toleugaliyev,Kuat Zharykbayev,Zarina Abildasheva,Pakizar Shamoi
类目: Human-Computer Interaction (cs.HC); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Dominant color extraction is a fundamental task in computer vision, with applications in design, image retrieval, and content recommendation. However, conventional algorithms based on device-oriented color spaces (e.g., RGB) and fixed quantization often fail to capture how humans perceive color differences. In this study, we propose a human-perception-based framework for dominant color extraction, built on a novel Human Color Dictionary (HCD) derived from a perceptual experiment. Participants (n=110) were sequentially shown 120 colors and asked to identify those they perceived as unique. Based on their responses, we constructed a 120 by 120 co-occurrence matrix and applied a greedy selection algorithm to derive 64 perceptually distinct colors, forming the HCD. This dictionary was integrated into RGB color histogram and K-means color extraction algorithms by initializing centroids and mapping pixel values to their nearest perceptual color using the deltaE 00 distance. Evaluation on a dataset of 4,562 images paired with expert-designed palettes demonstrated that our method improves perceptual alignment, as measured by Earth Mover’s Distance (EMD) and best-match distance. Our findings suggest no meaningful gender differences in the number of uniquely perceived colors, indicating that sex does not significantly influence color perception in this task. Another key finding is that highly saturated colors are perceived as the most distinguishable, whereas dark and deep colors tend to perceptually collapse and are less reliably differentiated. The results confirm that incorporating experimentally derived human color representations ensures the generated color clusters align more closely with human perception, thus improving the usability of color palette generation in practical applications.
[HC-10] Virtual Placebo: a scoping review of control conditions in randomized controlled trials of psychological immersive virtual reality interventions
链接: https://arxiv.org/abs/2610.32232
作者: Madis Vasser,Mary-Ann Kubre,Kadi Tulver,Jaan Aru
类目: Human-Computer Interaction (cs.HC); Neurons and Cognition (q-bio.NC)
备注:
Abstract:Research on immersive virtual reality (IVR) as a therapeutic tool is growing rapidly, with randomized controlled trials (RCTs) now commonplace. Yet the control conditions in many of these studies that claim strong specific IVR effects vary from usual care to active immersive experiences, making it difficult to draw clinically meaningful conclusions. We conducted a scoping review of recent literature on psychological IVR interventions to assess the main forms of control conditions used and their relations to study outcomes. In this paper, we make the case that most of the recent RCTs studying the effects of specific IVR interventions compare active immersive VR to usual care with no multimedia component. This can lead to strong yet potentially misleading study outcomes and can potentially inflate the IVR interventions’ effect. Namely, the lack of an appropriate immersive control condition may risk misattributing general effects of VR devices to the specific content of the intervention. In this review, we discuss the implications of these design choices and propose strategies for including a suitable “virtual placebo” condition to improve the methodological rigor and interpretability of results.
[HC-11] Present Drawing: Exploring Avatar-Mediated Drawing Events with Physical Traces
链接: https://arxiv.org/abs/2610.32166
作者: Kojiro Tanaka,Takuya Iwamoto,Yuki Okafuji
类目: Human-Computer Interaction (cs.HC)
备注: Accepted (ACM UIST 2026)
Abstract:We introduce Present Drawing, an approach for staging an avatar-mediated drawing event that leaves physical traces in the real world. Present Drawing aims to reconfigure conventional drawing events by relaxing their physical constraints. We describe an organizer-oriented design space for such an event along four elements: Time, Site, Control Source, and Avatar Embodiment. To ground this concept, we developed a proof-of-concept prototype in which an observer wears a head-mounted display to view an avatar. The avatar’s motion is synchronized with a customized pen plotter whose mechanical parts are hidden from the observer. We also developed five application examples spanning live, playback, multi-site, mascot, and system-controlled configurations. We conducted an initial evaluation of one representative playback autograph configuration, showing the basic plausibility of the core Present Drawing experience as an avatar-mediated drawing event and highlighting factors associated with perceiving the avatar as the drawer.
[HC-12] Forging Emotions: a deep learning experiment on emotions and art
链接: https://arxiv.org/abs/2610.32163
作者: Amalia Foka
类目: Computer Vision and Pattern Recognition (cs.CV); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 10 pages. Published in Artnodes no. 31 (2023), Universitat Oberta de Catalunya, under CC BY 4.0. Extended version of a paper presented at ISEA 2022, Barcelona
Abstract:Affective computing is an interdisciplinary field that studies computational methods that relate to or influence emotion. These methods have been applied to interactive media artworks, but they have focused on affect detection rather than affect generation. For affect generation, computationally creative methods need to be explored that have recently been driven by the use of Generative Adversarial Networks (GANs), a deep learning method. The experiment presented in this paper, Forging Emotions, explores the use of visual emotion datasets and the working processes of GANs for visual affect generation, that is, for generating images that can convey or trigger specified emotions. This experiment concludes that the methodology used so far by computer science researchers to build image datasets for describing high-level concepts such as emotions is insufficient and proposes utilizing emotional networks of associations according to psychology research. Forging Emotions also concludes that to generate affect visually, merely corresponding to basic psychology findings, such as bright or dark colours, does not seem adequate. Therefore, research efforts should aim to understand the structure of trained GANs and compositional GANs in order to produce genuinely novel compositions that can convey or trigger emotions through the subject matter of generated images.
[HC-13] Playing Bias: Epistemic Games for Interpreting Text-to-Image Models ATC WWW
链接: https://arxiv.org/abs/2610.32159
作者: Amalia Foka
类目: Computers and Society (cs.CY); Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注: 11 pages. Accepted at iLRN 2026, the 12th International Conference of the Immersive Learning Research Network, Athens, Greece, June 2026. Video presentation: this https URL
Abstract:Text-to-image (T2I) systems do not merely render prompts; they circulate defaults for “what X looks like,” shaping recognition through pose, camera, costume, props, and setting. While technical audits quantify disparities, they rarely cultivate the literacy needed to argue how images assemble stereotypes and confer authority. This paper translates a three-stage, humanities-informed framework into a set of epistemic games that make this cultural work visible and debatable: Interpret (art-historical close looking with AI outputs), Interrogate (counter-intent prompting as artistic exploration), and Expose (creative misuse through paired prompts with a constant motif and a BiasSlot rule, in which only the bias cue varies). In a pilot with undergraduate fine-arts students, the games shifted discussion from impressionistic reactions to concise, warranted hypotheses grounded in visual evidence, supported by lightweight classroom instruments that enabled reproducibility and structured collaboration. We argue that such game-based approaches foster critical literacy across educational contexts, equipping learners to recognize, discuss, and contest how T2I models encode bias.
[HC-14] Shorter Is Not Cheaper: A Controlled Comparison of Declarative Animation Specifications and Free-Form Code for LLM -Generated Explanations
链接: https://arxiv.org/abs/2610.32129
作者: Jiantang Huang
类目: Computation and Language (cs.CL); Human-Computer Interaction (cs.HC); Software Engineering (cs.SE)
备注: 15 pages, 6 figures, 8 tables
Abstract:Chat assistants increasingly answer conceptual questions with interactive, animated explanations, generated either as free-form HTML/JavaScript or as declarative specifications that a fixed runtime plays. The declarative route is widely expected to be cheaper, faster and more reliable, because its outputs are shorter and constrained. We test this expectation in controlled comparisons whose hypotheses and analysis plans were frozen before the runs. We designed ExplainSpec, a time-indexed declarative animation format, and an observation contract that makes every generated explanation programmatically inspectable; validated, pacing-agnostic checkers score 20 explanation requests from five domains. Three Claude models (360 jobs) and, in a follow-up, the open-weight Qwen3-14B with thinking enabled and disabled (240 jobs) answered every request in both formats. For every model the specifications were 25-48% shorter in answer tokens (output excluding reasoning), yet with reasoning enabled total output fell significantly only for the largest Claude model and no model generated significantly faster. The reasons differed. The Claude models reasoned far more before writing specifications (exploratory; for the two larger models, from 15-20% to 85-88% of generations). Qwen3-14B reasoned about as much in both formats (a pre-registered test of this shift was not supported), and its shorter answers saved little against a reasoning trace more than twice as long. Nor were specifications more reliable: free-form pages were valid after one repair round in 95-100% of jobs for every model, whereas Qwen3-14B produced valid specifications in only 33% of jobs and passed all semantic tests in 17% (HTML/JavaScript: 62%). For reasoning models, a shorter format alone therefore does not make generation cheaper, faster or more reliable, and a format learned in context can cost reliability.
[HC-15] Beyond Correctness: How Novice Software Engineering Students Exercise Oversight of Generative AI
链接: https://arxiv.org/abs/2610.32118
作者: Ali Shakiba,Thomas Chaffey
类目: oftware Engineering (cs.SE); Human-Computer Interaction (cs.HC)
备注: 10 pages, 3 figures, 2 tables
Abstract:Generative AI (GenAI) is shifting software engineering students’ work from producing artefacts towards evaluating and supervising AI-generated outputs. However, little is known about what characteristics students identify in outputs when reporting interventions, or how they exercise oversight during authentic software engineering work. We analysed 962 student-AI interaction turns documented by 145 first-year software engineering students through structured AI Journals, including 172 student-declared interventions. Using a multi-dimensional coding framework, we characterised the output characteristics identified in these reported interventions, students’ reasons for intervening, the actions they took, and how those actions were enacted. Fitness problems were identified in 74.4% of interventions, compared with 16.9% involving correctness problems, indicating that oversight more often concerned whether plausible AI outputs were suitable for the engineering task than whether they contained explicit errors. Among classifiable interventions, students most frequently modified generated outputs (51.0%) or refined their prompts (28.2%), with both response types commonly associated with engineering fit. Within-student analysis further showed that outputs perceived as containing minor flaws were associated with lower evaluation confidence than outputs perceived as perfect. Taken together, these findings characterise novice AI oversight as a process that extends beyond error detection to encompass both judgement of fitness for purpose and active intervention in AI-assisted engineering work.
[HC-16] Gaze Under Strain: How Oculomotor Behavior Tracks Cybersickness Progression in VR
链接: https://arxiv.org/abs/2610.32019
作者: Nevzat Umut Demirseren,Mehedi Hasan Raju,Oleg Komogortsev,Isayas Berhe Adhanom
类目: Human-Computer Interaction (cs.HC)
备注: Accepted for publication at ACM VRST 2026. 10 pages, 4 figures
Abstract:Cybersickness remains a major barrier to comfortable, accessible virtual reality (VR). Arising from mismatches between visual and vestibular signals, it places sustained demands on gaze stabilization and visual processing. Because gaze is the mechanism through which users actively sample the virtual environment, cybersickness-induced disruptions may alter both what users perceive and how discomfort itself emerges. Yet, how eye-movement behavior evolves alongside discomfort during prolonged VR exposure remains poorly understood. We present a behavioral analysis of eye-tracking data from 19 participants navigating a realistic virtual environment over a 20-minute session, with discomfort rated repeatedly throughout the experiment. Linear mixed-effects models examined gaze constructs spanning fixation stability, visual exploration, and gaze allocation as functions of discomfort, exposure duration, and their interaction. Our results show that fixation stability degraded progressively with increasing discomfort, with the effect strengthening over time. Visual exploration showed no marginal discomfort effect but exhibited significant discomfort-by-time interactions, indicating that saccadic behavior adapts dynamically rather than responding uniformly to discomfort. Gaze allocation was not associated with discomfort but shifted systematically with time. These findings show that cybersickness does not uniformly disrupt gaze. Instead, different aspects of eye-movement behavior respond differently and change in ways that only become visible when discomfort is tracked repeatedly over time. Understanding how gaze evolves during cybersickness progression provides the behavioral grounding needed to better interpret how discomfort shapes the user’s visual experience of immersive environments and to design more sensitive detection approaches.
[HC-17] CRM-Agent -Bench: Benchmarking Tool-Using Agents in Stateful Enterprise Environments
链接: https://arxiv.org/abs/2610.32004
作者: Monojit Banerjee,Zhujun Lan,Defu Cao
类目: Human-Computer Interaction (cs.HC); Computation and Language (cs.CL)
备注: 52 pages
Abstract:Deploying language agents in real applications demands more than the right answer: an agent must execute a multi-step business process correctly, follow domain-specific rules, and interact with a user across a dialogue, where the gap between saying and doing is the gap between a satisfied customer and a real loss. We argue that a deployable-agent benchmark should jointly assess four facets: the outcome an agent reaches, the trajectory it takes, its operational viability, and its policy compliance. Customer relationship management (CRM), where mutable records meet operational rules and consequential effects, exercises all four. We introduce CRM-Agent-Bench, an execution-grounded benchmark of 555 stateful, multi-turn tasks across nine business domains and 36 capability tags. Each task supplies a task-specific executable contract drawn from checks over actions, arguments, order, response, target state, and protected state; success is the strict conjunction of the checks declared for that task. We evaluate the complete agent system through these contracts and report scored coverage, latency, and cost. Across sixteen systems evaluated on the audited final release, strict success spans 64.32% to 83.60%; the closely grouped leading systems nonetheless diverge sharply in cost and per-domain reliability. CRM-Agent-Bench offers an auditable framework for measuring contract-grounded business-transaction integrity and adherence to task-declared benchmark policies, with a portable structure designed for extension as models improve. Github repository: this https URL
[HC-18] GELLI: A Generator for Experiential Language Learning Interactions
链接: https://arxiv.org/abs/2610.31987
作者: Danilo Symonette,Alan Y. Cheng,Taiga Ikedo,Jose Manuel Rosales Juarez,Roy Pea,Parastoo Abtahi,Jane L. E,James Landay
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Learning a second language requires extensive practice across situations personal to each learner, yet the most effective existing approaches — narrative situations designed around specific learning goals and delivered through media — demand substantial authoring effort, limiting practice to the scope of what was built. We present GELLI, a system that generates narrative practice worlds from a learner’s goals: situated scenes with characters to converse with, support alongside the interaction, and reflection, all produced at runtime with no pre authored content. In a within-subjects study with 40 learners of Japanese, GELLI produced greater vocabulary recall and confidence than expert-authored videos targeting the same goals, with no significant difference in speaking performance. Learners turned remembered or imagined situations into experiences they could shape and learn through. Support made experimentation feel safe, while unexpected narrative developments and language suggestions gave them new possibilities to pursue beyond what they initially intended to practice.
[HC-19] Silence Is Not Assurance: When Production Lane Keeping Assist Does Not Warn and What It Costs
链接: https://arxiv.org/abs/2610.31899
作者: Tianjia Yang,Yuxin Ding,Ke Li,Kaidi Liang,Ruwen Qin,Xianbiao Hu
类目: Robotics (cs.RO); Human-Computer Interaction (cs.HC); Systems and Control (eess.SY)
备注:
Abstract:Production lane keeping assist (LKA) systems tell the driver whether they detect the lane markings, but whether this displayed status reflects what the system perceives, and what happens when it does not, has not been examined in real-world driving. This study introduces CANDID-LKA, a CAN-based Diagnosis of the Indicated Detection status in LKA. It evaluates a proprietary driver-facing status using three signals: the displayed status and the steering torque of the driver, both recorded from the controller area network (CAN) bus of the vehicle, and the lane perception confidence of a third-party vision system observing the same road. The framework examines how the status responds to driving conditions and estimates how completely it reports degraded lane perception by comparing it with the other two signals in a latent-class model. We apply it to 39,800 seconds of real-world driving from 13 vehicle models of three manufacturers. The results show that the status does not respond consistently to the conditions that make markings harder to see, and that the three makes differ widely in how often they report not detected. The latent-class model shows that the displayed status is incomplete: it reports only 47% of degraded seconds as not detected, compared with 80% for the third-party low perception confidence, and in half of the low-confidence episodes it never reports not detected. In the silent seconds in which the status reads detected while the perception confidence is low and the driver is not in control, the standard deviation of lateral position over the next 3 s is about 60% larger than when the status and the perception confidence agree. A detected status thus does not guarantee reliable lane perception, and silence with the driver not in control is the case that most needs attention.
[HC-20] Cognitively Constrained Partners for Zero-Shot Coordination in Real-Time Partitioned Control NEURIPS2026
链接: https://arxiv.org/abs/2610.31726
作者: Ash Simmons,George De Ath,Richard M. Everson
类目: Machine Learning (cs.LG); Human-Computer Interaction (cs.HC)
备注: 30 pages, 6 figures, 5 tables. Accepted at the NeurIPS 2026 Workshop on Human-AI Coevolution: Measuring Human-Agent Teams in the Agentic Era
Abstract:Effective human-machine teaming requires agents that coordinate with unpredictable human partners. Reinforcement learning from live human feedback or large behavioural datasets is expensive, while purely simulated training overfits to artificial conventions and fails at Zero-Shot Coordination with real people. We propose a framework for training teaming agents against Cognitively Constrained Partners: simulated partners whose diversity emerges from procedural constraints grounded in human physiological literature, specifically reaction delays and perceptual limitations, rather than from reward shaping or historical training checkpoints. In a modified Lunar Lander environment, we report preliminary evidence that these partners play more like real people than unconstrained agents do, and that agents trained against them match the strongest cross-play baselines while being the only ones whose performance does not depend on how their training partners were generated. We also describe a human-participant study designed so that its rankings can be compared directly against cross-play, testing how well that standard evaluation predicts real human-agent teaming.
[HC-21] Omni-UBench: Benchmarking Omni-Modal LLM -based Guardrails Across Safety Utility and Reliability
链接: https://arxiv.org/abs/2610.31713
作者: Junhao Wang,Yidou Chen,Li Lu,Hongwei Yao,Xiaodan Li,Mengjie Wu,Yuefeng Chen,Zhan Qin,Kui Ren
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注:
Abstract:Advances in Omni-modal Large Language Models (OLLMs) have enabled increasingly seamless human-AI interaction, but their unified processing of heterogeneous modalities also expands the attack surface, allowing adversaries to exploit richer input compositions to circumvent safety alignment. To counter such risks, omni-modal guardrails have been proposed to detect and filter unsafe queries across arbitrary modalities. However, their capabilities and limitations remain insufficiently understood due to the absence of a comprehensive evaluation benchmark. In this paper, we present Omni-UBench, the first comprehensive framework for omni-modal safety assessment, evaluating guardrail models along three critical dimensions, i.e., safety, utility, and reliability. Omni-UBench curates over 183k samples drawn from 35+ heterogeneous sources spanning text, image, audio, video, and their cross-modal combinations, annotated with safety labels under a unified risk taxonomy. Benchmarking under Omni-UBench on 15 representative OLLMs and omni-modal guardrails yields several key findings, including: (1) modern OLLMs exhibit strong content moderation potential even without guardrail-specific tuning; (2) existing guardrails still struggle to identify cross-modal queries that conceal harmful intent or involve sophisticated jailbreaks; (3) they further exhibit substantial safety hallucination and over-sensitivity to textual patterns, revealing a deficient understanding of cross-modal safety semantics. Our findings provide valuable insights for developing more trustworthy omni-modal guardrails in future research.
[HC-22] Validating the Comparison Not Just the Scorer: Why Overall Accuracy Can Mislead in Interactive LLM Evaluation NEURIPS2026
链接: https://arxiv.org/abs/2610.31678
作者: Ancil Crayton
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 14 pages, 2 figures. Accepted to the Trustworthy AI for Good (AI4GOOD) Workshop at NeurIPS 2026
Abstract:Many LLM studies compare how models respond to different follow-up prompts. For example, one prompt may neutrally ask a model to reconsider, while another insists that an invented film really exists. The difference is used to measure the effect of conversational pressure. An automatic scorer then labels whether each response makes an unsupported factual claim. Researchers commonly validate the scorer with one overall accuracy or F_1 score. However, high overall performance can hide prompt-specific mistakes: if the scorer makes more errors after one prompt, the measured difference can be too large, too small, or point in the wrong direction. An audit containing mostly difficult responses can also misrepresent performance in the full study. We derive how such errors alter a prompt comparison and examine the problem using six models, 180 invented-entity questions, and six prompt conditions. In a balanced audit of 288 responses, overall F_1 was .916 . Yet, depending on the prompt, the scorer flagged 2.6%-33.3% of responses that the primary LLM rater judged not to contain an unsupported claim; these prompt-specific estimates were uncertain. In a second audit, the scorer and two separate LLM raters evaluated the same 300 model-question cases under each prompt. An apparent 9.3-point increase from asserting certainty rather than neutrally asking the model to reconsider fell to 2.3 points under both raters. The primary-rater interval included zero, so it did not establish an increase. Two other prompt comparisons kept their direction under both raters, while an attempted prompt-specific correction remained too uncertain for a firm conclusion. Researchers should therefore validate the exact difference they plan to report, not just the scorer as a whole: audit both prompts on matching cases and include scorer and rater uncertainty in the final result.
[HC-23] EMG Tokenization Preserves Movement Information at a Fraction of the Bitrate
链接: https://arxiv.org/abs/2610.32674
作者: Raul C. Sîmpetru,Amin Olamazadeh,Devon Rohlf,Vlad Cnejevici,Alessandro Del Vecchio
类目: ignal Processing (eess.SP); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注: 11 pages, 7 figures; 12-page supplement included. Supplementary Data 1-9 (CSV) are provided as ancillary files
Abstract:Wireless electromyography (EMG) interfaces must balance electrode coverage against transmission bandwidth. Here, we show that compact EMG tokens substantially reduce payload while retaining information for three-dimensional hand-pose decoding. We evaluate two causal tokenizers fitted without pose labels: an encoder pretrained to reconstruct EMG spectra and one based on SPAR, an engineered spectral-spatial representation of neighbouring electrodes. In personalized, within-session experiments with 320-channel EMG from nine participants, we compare tokenized and continuous features using matched decoder architectures and training procedures. Averaged across decoders, tokenization changed mean error by +0.01 mm for SPAR (95% confidence interval: -0.12 to 0.14) and +0.45 mm for the learned representation (0.10 to 0.82). Both increases met a 1 mm non-inferiority criterion. Stored token arrays reduced payloads 41-fold relative to raw EMG and 36-fold relative to continuous SPAR features. On sampled windows of the public emg2pose benchmark’s test partition, absolute differences in mean error between learned tokens and continuous features were below 0.05 mm across 176 participants and in the subset of 20 participants unseen during training. Electrode coverage, calibration, and unseen recordings affected SPAR decoding more than quantization did. In a computational benchmark, optimized SPAR generated tokens for 63 channels within 20 ms on a Cortex-M55 microcontroller. These findings support tokenization as a compact interface between wearable sensing and movement decoding, allowing decoders to be updated while the wearable encoder and transmitted representation remain fixed.
计算机视觉
[CV-0] SceneBloom: Agent ic Indoor Environment Generation with Realistic and Diverse Layouts for Embodied Learning
链接: https://arxiv.org/abs/2610.32978
作者: Yuhan Liu,Lianhui Qin,Shenji Wan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Training generalizable embodied agents requires realistic, varied, and simulation-ready indoor environments. Language-guided methods support semantic planning and object-level reasoning, but precise spatial grounding remains difficult. Visual generative models offer rich priors over plausible arrangements, yet their outputs require geometric reasoning to become physically valid 3D scenes. We introduce SceneBloom, an agentic framework that combines visual layout sampling with contract-guided 3D compilation. A hierarchical sampler uses diffusion priors to generate furniture arrangements within functional zones. A compiler then uses relations extracted from each proposal to guide geometric fitting and local pose correction. Across five room types, SceneBloom attains the highest visual quality and physical stability among the compared methods and is the only method with no detected mesh collisions and out-of-bounds objects. Its layouts are also closer to the distribution of real rooms than those of all compared baselines. Object navigation policies trained in these realistic scenes achieve a higher success rate in held-out real scans than those trained on an equal number of real training scans. Code is available at this https URL.
[CV-1] UniAction: Unified Action Conditioning for World Models
链接: https://arxiv.org/abs/2610.32977
作者: Bowen Xue,Zizhang Li,Hong-Xing Yu,Jiajun Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Actions that condition world models are highly heterogeneous, and thus existing action-conditioned world models are typically built for individual action types, fragmenting already limited action-annotated data across separate models. We introduce UniAction, a unified action-conditioning framework that represents any action as a sparse set of action elements, each with a discrete semantic identity and a continuous spatiotemporal grounding coordinate. By mapping these coordinates directly into the pretrained video model’s native positional space, UniAction allows heterogeneous actions to condition a shared world model within the same spatiotemporal structure. We further introduce cardinality-calibrated attention to address the bias from differing action representation granularities. We evaluate UniAction on four representative action forms: force, camera control, hand pose, and robot pose. UniAction outperforms conditioning alternatives in action following, and a single jointly trained model surpasses the action-specific experts, showing that heterogeneous action data benefit from unified model and joint training.
[CV-2] FlexGlue: Sparse-to-Dense Feature Matching and Visual Disambiguation
链接: https://arxiv.org/abs/2610.32976
作者: Philipp Lindenberger,Paul-Edouard Sarlin,Ondrej Miksik,Marc Pollefeys
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:To estimate correspondences across images, existing approaches match either sparse interest points, which is highly efficient but limited by repeatability, or dense pixel grids, which is robust but expensive. Furthermore, these approaches are trained with only matching pairs and thus often fail to reject unrelated pairs that exhibit visual symmetries, which is equally important in applications like Structure-from-Motion (SfM). We bridge these gaps by introducing FlexGlue, a neural network for asymmetric sparse-to-dense matching and disambiguation. Given two images and coordinates of points in the first image, FlexGlue estimates the coordinates of the corresponding points in the second image while simultaneously rejecting the pair if non-matching. By searching for a limited set of correspondences among dense pixels, our approach is both fast and effective at both matching and disambiguation: it combines the sub-pixel accuracy of dense matching with the speed of sparse matching, while rejecting incorrect pairs with high precision. Experiments on multiple datasets show that FlexGlue pushes the Pareto frontier in relative pose estimation and significantly outperforms existing approaches for SfM and visual localization. FlexGlue is publicly available at this https URL .
[CV-3] MiD-VLM: Detecting and Grounding Global Semantic Misinformation in Image-Caption Pairs
链接: https://arxiv.org/abs/2610.32967
作者: Tarunya Prasad,Bharatesh Chakravarthi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Semantic misinformation can arise when an authentic image is paired with a caption that changes or distorts its narrative. Because the image itself may contain no visible signs of manipulation, methods designed to detect pixel-level artifacts can miss these misleading image-caption pairs. We study this problem using DGM4+, a dataset designed to evaluate the detection of global scene and semantic inconsistencies. We propose a multimodal model that combines image-text representations from SigLIP with visual scene features from DINOv2 through a projection MLP, and adapts Qwen2-VL-7B using LoRA. The model jointly predicts whether an image-caption pair is authentic, identifies the manipulation type, and grounds the misleading caption tokens. On the DGM4+ validation split, the model achieves an F1 score of 0.9679 for binary authenticity classification. These results suggest that parameter-efficient adaptation of vision-language models is promising for detecting semantic inconsistencies, while also motivating further work on precise token-level grounding.
[CV-4] ViT-H: Generative Video-To-Humanoid Control with Motion Tracking Feedback
链接: https://arxiv.org/abs/2610.32966
作者: Zimo He,Tianyi Bian,Zizhong Yang,Hongjie Li,Nan Jiang,Jieming Cui,Baoxiong Jia,Yizhou Wang,Siyuan Huang,Yixin Zhu
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: The first four authors contributed equally. Project website: this https URL
Abstract:Learning humanoid control from monocular human videos offers a scalable way to acquire diverse whole-body behaviors. Yet, existing reconstruction-retargeting-tracking pipelines are limited in system-level scalability by separately optimized intermediate stages and cannot directly optimize reference motion executability using downstream tracking feedback. We present ViT-H, a video-to-humanoid control framework following a generation-and-tracking paradigm. ViT-H directly generates humanoid reference motions from videos with a diffusion model, bypassing human motion reconstruction and retargeting, and uses motion-tracking outcomes to post-train the generator with direct preference optimization. We further introduce an autoregressive variant that enables real-time humanoid video teleoperation with sub-400 ms latency. Extensive experiments on the Unitree G1 demonstrate that ViT-H outperforms state-of-the-art reconstruction-retargeting-tracking pipelines in both motion quality and executability. We further show that ViT-H scales favorably with model size and with both pre-training and post-training data scale, while motion-tracking feedback consistently improves executability across different motion trackers.
[CV-5] Flownomaly: Single-Step Conditional-Mean Recovery for Few-Shot Anomaly Detection WACV2027
链接: https://arxiv.org/abs/2610.32964
作者: Bernardo Perrone Ribeiro,Erkut Akdag,Egor Bondarev
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted to WACV 2027
Abstract:Few-shot industrial anomaly detection requires modeling category-specific normality from only a handful of defect-free images. Common approaches tie inference to stored exemplars, impose rigid linear structure, or risk reproducing anomalous content. Flownomaly instead casts the problem as normal-counterpart recovery in frozen DINOv2 feature space. Given normal supports, a velocity field is learned through conditional flow matching over frozen patch tokens to transport Gaussian noise toward normal features. Along straight paths, a single recovery step using one evaluation of the optimal field yields the posterior mean of the clean normal token without trajectory integration. The operating time governs how strongly recovery modifies a query token. With no anomalous validation samples available for tuning, a normal-only Jacobian gain-to-noise criterion selects this time from the supports alone. Squared residuals between query tokens and their recovered counterparts yield anomaly signals for both dense localization and image-level detection. Across 1-, 2-, and 4-shot benchmarks on MVTec-AD and VisA, together with sample-shot evaluations on Real-IAD, Flownomaly achieves state-of-the-art performance against recent category-specific methods. In the 1-shot setting, image- and pixel-level AUROC reach 97.8% and 98.0% on MVTec-AD and 95.3% and 98.7% on VisA, with a 2.1-point image-level AUROC improvement over the strongest baseline on VisA. Code is available at this https URL.
[CV-6] Beyond Discrete Tokens: Preserving Numerical Continuity in Medical Vision-Language Models
链接: https://arxiv.org/abs/2610.32948
作者: Dianye Huang,Jiamei Guo,Yue Zhou,Mingde Li,Yi Xiong,Wei Wang,Nassir Navab,Zhongliang Jiang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Objective: To develop a unified vision-language model for prompt-conditioned quantitative ultrasound measurement that addresses the mismatch between continuous measurement targets and discrete token-based numerical prediction. Methods: This article presents HAD-VLM, a hybrid autoregressive-diffusion framework that combines autoregressive text generation with continuous measurement prediction within a shared backbone. Visual prompt tuning adapts the vision encoder to ultrasound images. A diffusion process predicts a five-component representation comprising caliper endpoint coordinates and measurement length. A post-processing strategy combining the proposed coordinate-length consistency mechanism with ensemble prediction is introduced to refine measurement estimates. We evaluate four echocardiographic measurement tasks on EchoNet-LVH, with additional assessments of medical question answering and X-ray vertebral measurement. Results: HAD-VLM achieves mean absolute errors of 1.36, 2.77, 3.05, and 1.16 mm for IVSd, LVIDd, LVIDs, and LVPWd, respectively, reducing errors by 21.4-37.6% relative to the Vicuna-based autoregressive baseline. It also outperforms fine-tuned Qwen3-VL and direct MLP regression across all four tasks. Medical question answering performance is preserved on the evaluated dataset. For X-ray vertebral measurement, HAD-VLM reduces mean absolute error from 11.4 to 6.7 pixels. Conclusion: Integrating autoregressive language generation with diffusion-based continuous prediction improves quantitative medical image measurement while retaining prompt-conditioned task flexibility. Significance: HAD-VLM provides a unified approach to combining language interaction with anatomically grounded numerical prediction for quantitative medical image analysis.
[CV-7] LoCA: Enabling Multi-Modality in Geospatial Location Encoders
链接: https://arxiv.org/abs/2610.32894
作者: Panagiotis Koromilas,Eleanna Vali,Diogenis Antonopoulos,Eleni Saka,Dimitris Christofis,Nikolaos Ioannis Bountos,Konstantin Klemmer,Ioannis Papoutsis,Yannis Panagakis,Mihalis A. Nicolaou
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Location encoders map geographic coordinates to high-dimensional descriptors through contrastive alignment with paired Earth Observation imagery. We prove that this construction is structurally limited: any Lipschitz location encoder produces an output of intrinsic dimension at most 2, regardless of architecture or bandwidth, which forces the contrastive loss to strictly exceed its asymptotic optimum. Empirically, joint training also collapses the encoder’s embedding covariance to low rank. Although dimension-limited, current location encoders sidestep this spectral collapse by training against a frozen image encoder, which we reinterpret as a covariance regularizer that supplies a high-rank target for the location encoder to fold onto. However, extending this approach to multimodal inputs is challenging, as location encoders should simultaneously map a single location to multiple, potentially incompatible, descriptors. We identify two requirements a regularizer must satisfy for location encoding in multimodal regimes: co-alignment across modalities, since adapters alone cannot align independently pretrained encoders; and high effective rank, since the location encoder inherits its target’s spectrum. We propose LoCA (Location Contrastive Alignment), a two-stage pipeline that satisfies both. Stage 1 jointly trains Sentinel-1 and Sentinel-2 encoders with contrastive learning, producing a co-aligned, high-rank multimodal target. Stage 2 freezes the image encoders and trains a location and a tabular metadata encoder as projection adapters with multi-view contrastive learning. Against five state-of-the-art encoders on 17 tasks ranging from land cover and wildfire forecasting to geophysical, socio-economic, biodiversity and climate applications, LoCA achieves the most task wins, is the only model to outperform every other on a majority of tasks, and wins 68.8% of all per-task comparisons.
[CV-8] Closing the Modality Gap via Simplex-Constrained Representations
链接: https://arxiv.org/abs/2610.32890
作者: Shubham Gupta,Siva Reddy,Perouz Taslakian,Valentina Zantedeschi,Cem Subakan
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:While multimodal contrastive models like CLIP and CLAP successfully learn shared representations, their embeddings typically exhibit a persistent \emphmodality gap. We propose addressing this separation through a purely geometric approach: changing the embedding domain from the unit sphere, S^d-1 , to the probability simplex, \Delta^d-1 . Although both manifolds share the same intrinsic dimension, the simplex acts as a structural prior; its nonnegative, unit-mass constraints serve as a cross-modal calibration that forces both modalities to ground themselves in the same coordinate system. By attaching lightweight softmax adapters and using Total Variation similarity, we reduce the centroid \ell_2 gap by 97–99% across five multimodal retrieval benchmarks. This confirms that simplicial geometry induces genuine semantic alignment, matching or exceeding the retrieval performance of standard unconstrained models.
[CV-9] VTRL-Med: Progressive Visual Refinement with Topology-Guided Regularization for Medical Image Clustering
链接: https://arxiv.org/abs/2610.32884
作者: Guangyu Meng,Pengfei Gu,Peixian Liang,John P. Lalor,Erin Wolf Chambers,Danny Z. Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Deep clustering offers a promising approach for analyzing large unlabeled medical image datasets, yet prevailing methods still face two limitations: they may neglect class-agnostic structural priors that capture object boundaries and regions, and they do not explicitly encode topological patterns (e.g., connectivity and holes). These sources of information play distinct and complementary roles: visual-structural features provide the primary discriminative representation for clustering, while topological features provide additional global geometric cues for residual visual ambiguities. We introduce VTRL-Med, a progressive visual-structural learning framework with topology-guided regularization for medical image clustering. VTRL-Med first adapts the Segment Anything Model (SAM) to extract image features that retain structural cues inherited from mask-supervised pretraining. It then generates reliable pseudo-labels through a two-step filtering process that combines prototype-based selection with k-nearest neighbor consistency validation. Finally, it progressively refines the visual branch while an auxiliary topology branch, constructed from persistent homology (PH) descriptors, provides additional geometric regularization during training. Across 13 medical image benchmarks, VTRL-Med achieves state-of-the-art clustering performance, improving average accuracy by 4.48% over existing methods. On three representative datasets, topology-guided regularization further improves accuracy by 1.3–4.3% over visual refinement alone.
[CV-10] Multimodal Generative AI Assists Comprehensive Echocardiographic Interpretation
链接: https://arxiv.org/abs/2610.32880
作者: Yi Qin,Marawan Elbatel,Shanshan Song,Taoran Huang,Jiarong Guo,Ningshu Li,Lehan Wang,Chenxu Zhao,Jiexin He,Machteld J. Boonstra,Stephan van der Zwaard,Marielle C. van de Veerdonk,Pere Lopez-Gutierrez,Lydia Dux-Santoy,Andrea Guala,Enrique Almar-Munoz,Sebastian Johannes Reinstadler,Radek Debiec,Yi-Hang Zhu,Huiyu Zhou,Adrian P Cheong,Yike Guo,Agnes Mayr,Karim Lekadir,Yong Jiang,Nilesh J. Samani,Dacheng Tao,Folkert W. Asselbergs,Hongwen Fei,Alex PW Lee,Xiaomeng Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Echocardiography is the cornerstone of cardiovascular diagnostics, yet its global utility is constrained by a critical shortage of experienced interpreters. Current artificial intelligence (AI) solutions remain limited, typically operating as opaque “black boxes” or focusing on narrow tasks that fail to capture the holistic, guideline-driven reasoning of a clinician. Here we introduce EchoChat, a generative multimodal vision-language model that performs comprehensive echocardiographic analysis to support clinician interpretation. By emulating the cognitive workflow of a cardiologist, EchoChat integrates vast medical knowledge with a chain-of-thought reasoning mechanism to explicitly ground its conclusions in visual evidence. Trained on echocardiography-instruction pairs from 46,358 examinations, our model establishes a new state-of-the-art across diverse cohorts from six international medical centers. It supports the broadest diagnostic scope of 33 cardiac conditions and findings, including 15 newly added conditions for which it achieves an average F1 of 73.7% across available external evaluations. It surpasses existing best-performing foundation models in cardiac condition recognition (improving average recall by 13.54% on shared tasks) and quantification (average R2 score increase of 0.275 on shared tasks). Furthermore, in a source-blinded echocardiographer reader study, echocardiography reports co-generated with EchoChat were rated higher in diagnostic accuracy and completeness by two senior echocardiographers, and improved diagnostic performance from 72.78% to 82.75% in terms of F1 score over routine-reader only reports. These results demonstrate that integrating clinical reasoning into multimodal AI can democratize specialist expertise and transform the standard of cardiovascular care.
[CV-11] OpenStripe supports open-set identification and spatial stripe profiling in wild Amur tigers
链接: https://arxiv.org/abs/2610.32848
作者: Yichun Peng,Yansong Li,Jin Hou,Qi Song,Xi Guo,Ning Zhang,Degang Wei,Huang Yang,Yu Guan,Jianping Ge,Lei Bao,Tianming Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 4 figures. Code and nonspatial data: this https URL
Abstract:Non-invasive camera-trap monitoring of endangered wildlife requires open-set individual identification—grouping field images into identities without predefined classes or a known population size—yet pose deformation and cluttered backgrounds complicate image comparison. Here, we present OpenStripe, a multi-agent framework that reconstructs canonical two-dimensional (2D) tiger torso patterns and encodes four interpretable local stripe structures: breakpoints, bifurcations, irregular structures, and round structures. We evaluated OpenStripe on 100 camera-trap records from 28 wild Amur tigers (Panthera tigris altaica), representing nearly one-third of the documented population in the Northeast China Tiger and Leopard National Park. Combining all four complementary structures enabled automatic identity grouping with 0.90 matched accuracy and a 0.85 adjusted Rand index (ARI), outperforming 49 baseline configurations under the same backend (up to 0.43 accuracy and 0.14 ARI). In addition, results from OpenStripe indicate that stripe similarity in Amur tigers declines with geographic distance and suggest that traits such as fractal dimension and inter-stripe spacing are associated with geographic proximity. OpenStripe thus provides an interpretable representation that connects automated individual grouping with spatial phenotypic profiling in wild populations.
[CV-12] Sounding Surgical Instrument Localisation through Audio-Visual Correspondence
链接: https://arxiv.org/abs/2610.32816
作者: Dominic Behling,Jonas Hein,Matthias Seibold
类目: ound (cs.SD); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Surgical scene understanding is a prerequisite for context-aware and intelligent systems in the operating room. Current methods rely mostly on visual data, while acoustic sensing has shown potential to provide complementary information on surgical activity. First approaches for the localisation of surgical activity relied on microphone arrays and acoustic beamforming. This work investigates whether the active instrument can be localized from a single microphone and a camera by learning the correspondence between the appearance and the sound of surgical instruments, without annotating its position. We adapt weakly supervised audio-visual localisation, which learns from the co-occurrence of image and sound, to the surgical setting. We record simulated surgical actions, sawing, drilling, and chiseling, and train the model on these recordings without any spatial annotation to produce a heatmap over the sounding tool. We extend the training objective by two terms: a silence term, which suppresses the output for silence and background noise, and a composite term, which trains the model to select the sounding one of two visible instruments. The model localizes the active instrument on an independent corpus recorded with a different camera and microphone, in which instrument positions were obtained by optical tracking. It separates tool activity from silence and background noise and can attribute sound correctly when two tools are present. Audio-visual correspondence learned from in-domain recordings connects the sound of a surgical action to the instrument that produces it, using a single microphone and a camera. This proof of concept introduces a novel paradigm that enables the detection and localisation of surgical activity without manual annotations.
[CV-13] CompMVR: Compression-Aware Multi-View Restoration Using Diffusion Models for Geometrically Consistent 3D Reconstruction SIGGRAPH
链接: https://arxiv.org/abs/2610.32800
作者: Dong-hwi Kim,Chaewon Moon,Hojun Song,Dongbeom Kim,Junyeong Jang,Aro Kim,Gahyeon Kim,Heejung Choi,Jehee Kim,Gianella Cravioto,Sohyun Lee,Gyeongjin Choi,EunHye Jeong,Soo Ye Kim,Jaehyup Lee,Sang-hyo Park
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to SIGGRAPH Asia 2026 (Technical paper)
Abstract:Real-world 3D reconstruction pipelines often rely on multi-view images degraded by lossy compression. While compression artifacts have been extensively studied in single-image or video restoration, their impact on multi-view geometry remains largely overlooked. In this work, we identify compression-induced cross-view geometric inconsistency as a critical failure mode, where view-dependent artifacts disrupt feature matching, dense correspondence, camera parameter estimation, and downstream 3D reconstruction tasks such as novel view synthesis. To address this problem, we propose CompMVR, a compression-aware multi-view diffusion restoration framework that jointly refines compressed views while preserving cross-view consistency, and construct compressed datasets for systematic evaluation. Unlike conventional restoration methods that optimize each image independently for 2D perceptual quality, CompMVR leverages learned compression priors and cross-view correspondence to recover visual details while improving geometric reliability across views. Experiments across diverse datasets, codecs, and compression levels demonstrate gains in restoration quality and downstream 3D tasks, including novel-view synthesis, camera pose estimation, and view matching. These results establish compressed multi-view restoration as a distinct problem and highlight its importance for 3D reconstruction under lossy compression. Code and data for this paper are available at this https URL.
[CV-14] Consommé: Conditioned Token Merging for Split Inference
链接: https://arxiv.org/abs/2610.32794
作者: Man Ngo,Jan S. Rellermeyer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Split inference partitions a Vision Transformer between a resource-constrained edge device and a server. However, transmitting intermediate tokens across the communication channel is costly. Deploying fine-tuned backbones to every endpoint becomes expensive as well. We study token reduction for split inference under a frozen backbone, where only a small adapter is shipped. We observe that the CLS token plays a dual role: it acts as the global summary the receiver needs for classification, and it serves as a signal on which patches to keep. We propose Consommé, a method utilizing both characteristics. There, an information bottleneck compresses the CLS token into a compact latent which is transmitted in place of the full CLS token and conditions a learned patch gate merging patches into exactly k tokens via soft assignment. It reaches any budget in a single step, thus escaping the per-layer cap of bipartite matching (BPM). On DeiT-S/T, it leads all methods at aggressive token budgets and remains operable at shallow splits where BPM cannot operate, while matching fine-tuned methods at gentle budgets at a fraction of their deployment cost.
[CV-15] SREdit: Scene Response Guided Appearance Editing for Projection Mapping
链接: https://arxiv.org/abs/2610.32792
作者: Yuqiao Xia,Yuchen Deng,Ziheng Chang,Bingyao Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注: 28 pages, 65 figures
Abstract:Projection-based appearance editing changes selected visual attributes of a physical object through projection mapping (PM). To display a desired appearance, PM typically uses projector compensation, which computes a projector input that counteracts the geometric and photometric distortions of the scene. Compensation can only approximate this appearance within the available projection range, and it cannot revise the appearance itself. Pretrained LLM-based image editors provide a direct way to produce the desired appearance from a text prompt or reference image; projector compensation can then attempt to reproduce the edited result through projection. However, these editors do not have an accurate measure of the scene geometry or scene response. Therefore, their outputs may contain geometric misalignment, inconsistent shading, or changes outside the available projection range, which downstream projector compensation cannot correct. To address this problem, we introduce SREdit, an efficient response-guided framework for projection-based appearance editing. SREdit estimates a scene response map from the reconstructed scene that reflects how strongly projected illumination can affect different surface regions. SREdit uses the source appearance to constrain the early sampling process and preserve registration with the fixed physical object, while the scene response map adapts the amount of editing across surface regions. Experiments across two pretrained image editors show that SREdit improves projected appearance quality and reproduction fidelity. It adds less than one second per image with negligible VRAM overhead. Participants also consistently preferred the results produced with SREdit in our user study.
[CV-16] Video-MemComp: Taylor-Guided Sensitivity-Aware Memory Management for Streaming Video Understanding ACM-MM2026
链接: https://arxiv.org/abs/2610.32784
作者: Yonghui Niu,Yuhao Ji,Chang Nie,Jiao Zhang,Yilong Li,Junlan Feng,Ran He,Caifeng Shan,Chaoyou Fu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by ACM MM 2026. Project Page: this https URL
Abstract:The rapid growth of video data has created an urgent need for efficient video understanding systems that can handle long and continuous visual inputs, posing significant challenges for Multimodal Large Language Models (MLLMs) due to severe memory bottlenecks. Moreover, existing compression methods often overlook the explosive accumulation of approximation errors. To address this issue, we propose Video-MemComp, a training-free and sensitivity-aware context management framework for memory-constrained streaming video understanding. Our method theoretically decomposes compression error into orthogonal local approximation error and layer-wise structural sensitivity via a first-order Taylor expansion. To minimize local approximation error, we perform token aggregation in a pre-RoPE safe zone with an attention-conserving logit bias, mitigating geometric distortions caused by RoPE rotations. To manage layer-wise structural sensitivity, we dynamically allocate the KV cache budget based on offline gradient priors and online context entropy, prioritizing high-sensitivity layers. Video-MemComp outperforms or achieves performance comparable to full-cache and state-of-the-art compression methods on benchmarks including MLVU, VideoMME, and EgoSchema. Notably, it improves the overall accuracy of Qwen2.5-VL-3B from 62.3% to 64.1% while reducing the KV cache to an equivalent of 6,000 tokens per layer, saving over 90% of KV memory.
[CV-17] SCDM-HTD: Spectral-Wise Conditional Diffusion Model for Hyperspectral Target Detection
链接: https://arxiv.org/abs/2610.32768
作者: Dunbin Shen,Wenfeng Kong,Jianjun Liu,Xiaorui Ma,Hongyu Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Hyperspectral target detection (HTD) aims to locate and recognize objects at a fine-grained scale, playing a pivotal role in remote sensing. Despite significant progress, current HTD methods primarily focus on feature comparison while neglecting distribution learning for background-target separation, which can limit detection robustness in practical applications. This paper addresses this issue by introducing a spectral-wise conditional diffusion model that reformulates HTD as a task of generating a detection map. Specifically, we treat background and target confidence scores as training samples and use the corresponding pixel spectra as guiding conditions to capture the distributional discrepancy between background and target through diffusion and denoising. To enhance practical applicability, we propose a lightweight conditional denoising network based on multilayer perceptron to map noisy confidence scores to their ideal counterparts and adopt a deterministic reverse non-Markov chain to skip sampling steps, thereby accelerating the generation process. Furthermore, we introduce a composite loss to improve background-target separation. This loss integrates constrained energy minimization loss to enhance target saliency and suppress the background by maximizing the target response while minimizing average output energy, alongside binary cross-entropy loss to classify background and target using pixel-wise labels. Experimental results on four benchmark datasets demonstrate the effectiveness of the proposed method, with improvements in both quantitative metrics and qualitative visual effects.
[CV-18] Cross-Architectural Feature Fusion for Coronary Artery Stenosis Detection in X-ray Angiography
链接: https://arxiv.org/abs/2610.32730
作者: Haobo Zhu,Mojtaba Lashgari,Jason Chai,Giovanni Luigi De Maria,Vicente Grau,Abhirup Banerjee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 3 figures
Abstract:Coronary artery disease remains a leading cause of mortality, with stenosis detection from invasive X-ray Coronary Angiography (XCA) critical for diagnosis. Manual interpretation is time-consuming, subjective, and variable, especially under low contrast and vessel overlap. Existing deep learning methods often rely solely on detection or segmentation, limiting this http URL propose a hybrid framework, Fusion-YOLOv11-L, combining a YOLOv11-L detector with a pretrained UNet++ via a channel attention gating module. This enables fusion of semantic and vessel-aware structural features, improving localisation of stenotic regions. The UNet++ is pretrained on ARCADE vessel segmentation dataset to provide anatomically informed representations. Experiments on the relabeled ARCADE (ARCADE*) dataset show improved performance over YOLOv11-L, achieving AP@50 of 0.585 and F1-score of 0.596. Cross-domain evaluation on a dataset from the Oxford Acute Myocardial Infarction (OxAMI) Study reveals performance degradation due to domain shift, mitigated using test-time adaptation and augmentation. Test-time augmentation improves AP@50 from 0.434 to 0.480, indicating better generalisability. These results show that cross-architecture feature fusion and lightweight adaptation improve robustness and generalisability of automated stenosis detection in XCA.
[CV-19] CityLift: Compositional 3D Urban Generation from a Single Image
链接: https://arxiv.org/abs/2610.32728
作者: Haoran Gong,Jie Feng,Chenxu Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generating editable 3D urban scenes from a single image requires detailed building assets with a coherent spatial arrangement. Occlusion and perspective ambiguity make it difficult to infer complete building geometry together with consistent relative scales and positions. We present CityLift, a framework that generates individual textured building meshes from an urban image by coupling scene-level spatial modeling with building-level synthesis. CityLift first learns building shape and placement jointly through coupled scene and instance streams, producing coarse sparse structures in a common coordinate system. We then introduce generative spatial inheritance, which maps each building’s coarse occupancy and scene-conditioned 3D feature field into a local volume to guide a flow model that generates finer structure. The same coordinate mapping places the generated meshes back into the scene, preserving their spatial relationship to the scaffold. To address the scarcity of urban data pairing scene images with individual 3D buildings, we construct CityLift-5K, comprising 4,997 synthetic scenes with instance masks, building meshes, and placement annotations. Experiments on CityLift-5K and scene blocks from UrbanScene3D and InstanceBuilding demonstrate improved building geometry and scene composition. The project page is available at this https URL.
[CV-20] Are Textual Residuals Truly Complementary? Learning Visual Complements under Semantic Superposition for Visible-Infrared Person Re-Identification
链接: https://arxiv.org/abs/2610.32709
作者: Xiaoyang Li,Mingming Lu,Wei Zhou,Hao Li,Canhui Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 24 pages, 4 figures
Abstract:Text-enhanced visible-infrared person re-identification (VI-ReID) exploits textual descriptions to compensate for appearance cues that are weak or unavailable under infrared imaging. Existing methods construct compensation from differences between modality-oriented textual representations, implicitly assuming that textual differences are visually complementary. We revisit this assumption through semantic superposition and show that dense textual residuals may entangle useful appearance variations with redundant, making direct residual fusion unreliable. We propose Superposition-Aware Residual Calibration, which transforms dense textual differences into selective and visually grounded complements. Low-Interference Sparse Residual Learning represents modality-oriented text embeddings as sparse activation strengths over a shared dictionary, selects positive strength differences, and suppresses both dictionary-level overlap and activation-weighted interference to reconstruct a compact, selectively controlled residual. Visual Complement and Counterfactual Calibration further calibrates the residual against identity-specific RGB-IR visual discrepancies, encouraging retained semantics to provide meaningful compensation while preserving the residual structure. Experiments on Tri-SYSU-MM01, Tri-RegDB, and Tri-LLCM show that SARC consistently improves the VI-ReID backbone and remains competitive with text-assisted methods. Mechanism analyses show that SARC produces more concentrated textual variations, reduces dictionary-level feature interference, and improves residual-visual-complement alignment for most samples. These results demonstrate that effective cross-modal compensation requires textual residuals to be selectively represented and explicitly calibrated, rather than directly treated as visual complements. Code and implementation details will be released upon publication.
[CV-21] SP-Guide: Enhancing Aerial Vision-Language Navigation with Spatial Priors
链接: https://arxiv.org/abs/2610.32697
作者: Haoxiang Tang,Dong Zhou,Shen Tan,Minxuan Jiang,Xitong Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Aerial vision-language navigation (AVLN) requires a UAV to reach a language-described target from partial egocentric observations while reasoning over large outdoor scenes. Existing approaches typically fuse view-centered RGB observations and globally anchored history maps as generic visual embeddings, leaving cross-modal spatial alignment to be learned from sparse navigation supervision. We present SP-Guide, a transformer- based goal-prediction framework that addresses this weakness of alignment through a Spatial Prior Cross-Modal Encoder, an Instruction Enhanced Decoder, and a Goal Convergence Filter for terminal target refinement. Experiments implemented on the CityNav benchmark show that SP-Guide improves success rate over the representative baselines, with the largest gains on unseen environments. These results indicate that physically grounded spatial priors provide an effective inductive bias for robust AVLN.
[CV-22] Robust Offset-Free Lane-Keeping Control Framework for Vision-Only Automated Vehicles
链接: https://arxiv.org/abs/2610.32666
作者: Der-Hau Lee
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Systems and Control (eess.SY)
备注: 13 pages, 10 figures
Abstract:Achieving accurate, real-time lane-keeping control during high-speed cornering on large-curvature roads while ensuring robust vehicle stability is a critical challenge in autonomous driving. To address this challenge, this study developed a robust offset-free model predictive control (OF-MPC) framework for vision-based autonomous vehicle lane-keeping. The regulation problem in this framework is solved using several algorithms. Numerical simulation results indicated that the interpolation-tube-based constrained iterative linear quadratic regulator (itube-CILQR) algorithm outperformed solvers based on the standard Interior Point Optimizer (IPOPT) and Operator Splitting Quadratic Program (OSQP), achieving a mean computation time of less than 5 ms. The itube-CILQR algorithm was then used for vehicle control in vision-based lane-keeping experiments in a simulation environment to quantify the perceptual uncertainty of a visual perception model (multitask UNet). This quantification was performed using the OF-MPC framework for traffic scenes that were not encountered during model training. Overall, the developed OF-MPC framework and the proposed method for characterizing the perceptual uncertainty of a visual perception system in unseen environments can contribute to the future development of advanced autonomous driving systems.
[CV-23] he Answer Knows Where to Look: Zooming before MLLM s Answer under Visual-Token Budgets
链接: https://arxiv.org/abs/2610.32664
作者: Zhensheng Hu,Guanglong Xu,Guanqiang Hu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal large language models (MLLMs) read images through visual tokens, and under the small per-image budgets that efficient deployment requires, the fine text of documents, charts and infographics becomes illegible. Zooming in recovers the evidence, but the region must be chosen before the model answers. We find that MLLMs know where the evidence is, but not in a form that can be read before they answer: the answer tokens localize it from the middle layers on, whereas the attention available before answering does so only in the top layers, at a depth that varies across models, and diffusely. Supervising the pre-answer attention to point degrades reading, and the benefit of zooming is governed by each image’s information gap. Building on these findings, we introduce ours, which lets an MLLM look where its answer would. A dedicated look token, processed in a single prefill and kept out of the answering pass, learns in hindsight to attend as the model’s own correct answers do, extracting this knowledge into a concentrated pointer before any answer exists. Its hidden state then selects how to look, from no zoom to crops, square windows, windows that adapt to the pointer and foveated warps, supervised by the counterfactual outcome of every view. ours needs no box annotations, per-model layer search, external localizers or generation before the answer. On Qwen3-VL at three scales and five text-rich benchmarks, it surpasses the strongest baseline by 4.8-9.1 points at 98 visual tokens per view and attention-guided zooming by 11.1-17.0 points, while running faster than attention-guided and grounding-based zooming.
[CV-24] SemGeo-Gen: Unsupervised Generation of Approximate Cross-Instance Semantic-Geometric Correspondences NEURIPS2026
链接: https://arxiv.org/abs/2610.32663
作者: Roy Amoyal,Shira Ifergane,Oren Freifeld
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026
Abstract:We introduce SemGeo-Gen, an unsupervised method for generating approximate cross-instance semantic-geometric supervision from weakly aligned 3D object collections. Our goal is not to recover exact point-level correspondences, which are often ambiguous across different instances and would require expensive, impractical manual supervision at scale. Instead, we automatically produce approximate correspondences that are semantically meaningful, geometrically consistent, and diverse enough to train modern correspondence models. Given only coarse category-level rotation alignment, SemGeo-Gen lifts multi-view DINOv2 features to 3D, aligns object instances using continuous piecewise-affine registration, and prunes candidate matches using semantic consistency. The resulting 3D correspondences can be projected into rendered views, yielding scalable 3D-3D, 3D-2D, and 2D-2D supervision without manual annotation. We validate the generated correspondences against sparse human annotations from KeypointNet, obtaining 93% PCK@0.10 and 97% PCK@0.15. More importantly, we show that using the generated correspondences for synthetic pretraining improves recent state-of-the-art semantic correspondence models on SPair-71k after fine-tuning on real data, with gains of up to 5 PCK@0.10 points. Together, these results indicate that approximate semantic-geometric supervision generated from 3D assets can improve real-image correspondence learning and serve as a scalable alternative to costly manual annotation. Project page: this https URL.
[CV-25] EditJev: System One Routing for Image Editing Models
链接: https://arxiv.org/abs/2610.32635
作者: Zhile Guan,Xu Jiang,Yule Duan,Gongbo Zhang,Shuyu Wang,Xiao Chen,Jian Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 12 pages, 4 figures, 7 tables
Abstract:Instruction-based image editing models exhibit pronounced sample-specific performance variation: different models can produce substantially different results for the same source image and editing instruction, and the best-performing model often varies across requests. These models also differ substantially in inference latency, making fixed-model deployment inefficient. We therefore study image editing model routing, which dynamically selects an editor from a candidate pool given an input image and editing instruction. We introduce EditJev, a System One routing model that directly predicts sample-level preference scores over candidate image editing models in a single forward pass. A lightweight preference head on a vision-language backbone produces a fixed-dimensional score vector without autoregressive model-name generation or response parsing. To train EditJev, we construct pairwise preferences from offline candidate outputs and optimize a pairwise ranking objective to learn relative model suitability. At inference time, EditJev combines the predicted preference scores with model-level speed scores to balance editing quality and generation latency. Experiments on ImgEdit and GEdit-En show that EditJev improves aggregate quality over the strongest fixed editor, from 4.540 to 4.592 and from 8.194 to 8.255, respectively, while reducing estimated end-to-end latency by 60.8% and 31.7%. These results demonstrate the effectiveness of direct, structured model routing for efficient image editing.
[CV-26] Earth embeddings are decomposable records of the land surface
链接: https://arxiv.org/abs/2610.32617
作者: Robin Young,Srinivasan Keshav
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Populations and Evolution (q-bio.PE)
备注:
Abstract:Ecological monitoring, carbon accounting and biodiversity assessment increasingly rely on annual Earth embeddings–learned vectors that compress a year of satellite observation for every ten-meter pixel on Earth, consumed in place of the imagery itself. What these vectors mean has not been established. Here we decompose two widely used embedding products and show that they contain individual ecological features, including fire, forest clearing and crop type, validated against independent reference data on events across six continents. The recovered concepts date abrupt canopy loss to the first post-disturbance growing season and resolve severity well enough to separate prescribed fire from wildfire. Small fires that recover quickly are detectable in one product but not the other, which instead separates crops more strongly.
[CV-27] DriveOSS: Dataset and Benchmark for 3D Anomaly Segmentation and Open-set Semantic Segmentation in Autonomous Driving WACV2027
链接: https://arxiv.org/abs/2610.32612
作者: Zhiran Yan,Savankumar Bhanderi,Sourabh Lolge,Rui Song,Girdharbhai Paneliya,Shiva Agrawal,Gordon Elger
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:Recent advances in 3D semantic segmentation have significantly improved scene understanding in autonomous driving. However, existing methods are typically developed under a closed-set assumption and evaluated under standard driving conditions, where all categories are known in advance. In real-world driving, autonomous vehicles inevitably encounter previously unseen objects, making closed-set perception insufficient for safe deployment. To address this challenge, we introduce DriveOSS, which, to the best of our knowledge, is the first real-world multimodal dataset and benchmark specifically designed for 3D anomaly segmentation and open-set semantic segmentation. It contains 12,661 synchronized and annotated frames captured using six surround-view RGB cameras, a 128-beam LiDAR, and a GPS/IMU unit. The dataset comprises 52 real-world driving scenes collected under clear, nighttime, cloudy, rainy, and snowy conditions and is organized into ID (in-distribution) and OOD (out-of-distribution) subsets. The ID scenes contain only known classes and are used for training, whereas the OOD scenes additionally contain unknown objects annotated as the outlier class and are used for validation and testing. Extensive experiments show that unseen objects remain a major challenge for current 3D perception systems. The dataset, benchmark protocols, and code are available at this https URL.
[CV-28] Hair Segmentation and Skin Characterization on Macroscopic Dermatological Images for Laser Treatment Parameterization
链接: https://arxiv.org/abs/2610.32609
作者: Alex Navarro Navarro,Marc Arriaga Hidalgo,Carlos Moreno Rubio,Maxime Talarmain,Noelia Vaquero Gallardo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published open access in Journal of Imaging Informatics in Medicine (2026)
Abstract:Current light-based devices, such as laser hair removal, intense pulsed light (IPL), vascular laser therapy, and dermatologic phototherapy, are widely used in clinical dermatology but rely heavily on subjective manual parameterization or uncalibrated point-based colorimeters. To overcome these limitations, this work presents a unified system for hair segmentation and skin characterization in macroscopic dermatological images. Unlike most existing literature that relies on dermoscopic datasets captured under highly controlled conditions, our approach is specifically designed for macroscopic images that replicate real clinical laser treatments. The proposed pipeline integrates a custom-trained variant of the YOLOv9 deep learning architecture for instance segmentation and classical image processing within a single workflow to extract objective skin and hair parameters such as phototype, hair density, color, and thickness. A color calibration stage using spectrophotometric measurements ensures accurate feature estimation regardless of illumination conditions. Experimental results show competitive segmentation metrics (mAP@50 \approx 0.80) on a custom dataset featuring short, shaved hairs, typical of pre-treatment scenarios. These findings demonstrate the feasibility of the proposed pipeline in providing the objective, foundational skin and hair features required for future automated laser parameterization, paving the way for its extension to broader light-based medical therapies.
[CV-29] Standing on Open Shoulders: Regional Fine-Tuning of Open-Weight Camera-Trap Models for UK Wildlife with Modest Data and Compute
链接: https://arxiv.org/abs/2610.32573
作者: Paul Fergus
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 39 pages, 9 Figures, and 8 Tables
Abstract:Camera-trap species classification is dominated by large models from well-resourced organisations, which often lack the species-level distinctions regional conservation needs, such as native red versus invasive grey squirrels in the UK. We ask whether open weights let a small team with a modest dataset and a single GPU take a real part in AI development. We fine-tuned the open DeepFaune classifier on 28 UK classes, paired it with the unmodified MegaDetector v6, and compared the resulting DeepFaune-UK v1.2 with DeepFaune v1.5.0 and SpeciesNet on 20,212 hand-verified MammalWeb citizen-science images, after removing sequences sharing any of 1,039 images with our training archive. DeepFaune-UK achieved a strict image-level accuracy of 88.8%, including empty frames, against 56.5% and 59.0% for the baselines, but at DeepFaune’s coarser label resolution DeepFaune v1.5.0 was more accurate, though not significantly under contributor resampling (91.4% vs. 88.8%). As DeepFaune v1.5.0 uses a newer DINOv3-based classifier, we applied the identical recipe, without new data, to that classifier. The resulting DeepFaune-UK v1.3 reached 90.6% strict accuracy, narrowed the like-for-like gap to 0.8 points and reduced roe deer misclassified as fallow deer by 69%. On 1,850 images from the 26 contributors with no overlapping images, the three DeepFaune-based models did not differ significantly (90.5%, 90.6%, 89.9% like-for-like). Fine-tuning on 10% of the training crops (about 150 per class) gave out-of-distribution accuracy similar to using all of them, and a DINOv2 run on this subset took about 22 minutes with under 4 GB of GPU memory. Open weights thus make regional model development feasible for small teams, but the largest like-for-like gain came from a newer base model trained by others, so this democratisation depends on large actors continuing to release open weights, under licences that may change.
[CV-30] Reuse Dont Retrain: Efficient Modality-Versatile Geospatial Foundation Models through Feature Distillation
链接: https://arxiv.org/abs/2610.32566
作者: Mario Hamon,Pierre Adorni,Minh-Tan Pham
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Geospatial foundation models (GeoFMs) are costly to pretrain, yet model families often repeat this process across backbone scales and sensing modalities. We investigate feature distillation as a more efficient way to reuse existing representations. First, we distill frozen CROMA and TerraMind teachers into smaller Vision Transformers through patch-level feature matching. Across nine PANGAEA tasks, distilled TerraMind Tiny, Small, and Base outperform their scale-matched released counterparts by 2.76, 3.02, and 1.87 mean IoU points, respectively; each also matches or exceeds the next-larger released model. Distilled CROMA-Base gains 2.05 points. TerraMind-Base required only 43 A6000 GPU-hours, approximately 0.5% of the reported cost of independent pretraining. Second, we introduce TEDI, an unpaired multi-expert approach combining TerraMind for multispectral imagery and DINOv3 for high-resolution RGB. Separate patch embeddings route unpaired inputs into a shared Transformer trunk, removing the need for co-registered data. TEDI-Large achieves the highest single-model mean in our PANGAEA-9 comparison (62.51 mIoU), outperforming both teachers overall while retaining their modality-specific strengths. TEDI-Base reaches 61.75 mIoU with 72% fewer parameters. These results show that distillation improves the GeoFM accuracy-compute frontier while consolidating complementary experts into a single versatile backbone.
[CV-31] Shared Is Not Enough: Predictor Width and Factor Readability Under Masked Latent Prediction
链接: https://arxiv.org/abs/2610.32562
作者: Nilotpal Kakati
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Data Analysis, Statistics and Probability (physics.data-an)
备注:
Abstract:Masked latent prediction methods such as I-JEPA aim to learn predictable features, but what determines which information they retain is less clear. We built a simulator in which every factor behind an image is known, so a probe can measure how readable each factor remains after training. We trained small encoders (0.7M parameters) with I-JEPA, SimCLR and BYOL on the same pairs of views and probed a factor that both views share. I-JEPA had a predictor with half the encoder’s width and depth, the proportions it uses for ViT-B. Its median score was about 40% of SimCLR’s. Widening the predictor to the encoder’s width reversed this ordering at every training seed, against both other methods. We release the simulator and the trained models.
[CV-32] HEALPix-Based Implicit Neural Compression of Omnidirectional Images
链接: https://arxiv.org/abs/2610.32561
作者: Yulong Cheng,Youneng Bao,Junfeng Zhou,Mu Li,Jie Wen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: [N] pages, 10 figures, 4 tables. Submitted to IEEE Transactions on Image Processing
Abstract:Real-time decoding of omnidirectional images is essential for immersive virtual reality (VR) experiences. Compared with conventional transform-based learned image compression (LIC), overfitted codecs offer lower decoding complexity by representing each image with a lightweight implicit neural representation (INR). However, existing approaches typically operate on the equirectangular projection (ERP), which introduces nonuniform spherical sampling and spatial distortion, leading to redundant sampling near the poles and latitude-dependent receptive fields. To address these limitations, we propose a spherical overfitted codec based on a multi-scale Hierarchical Equal Area isoLatitude Pixelization (HEALPix) representation, which enables equal-area sampling and efficient hierarchical modeling of spherical signals. Building on this representation, we develop a Spherical Inter-Scale Conditional entropy model that exploits spatial dependencies and cross-scale correlations among HEALPix latents to improve coding efficiency. We further introduce a Multi-Ring Spherical Convolution that aggregates neighboring pixels over a nearly uniform spherical receptive field across latitudes. Experiments on a public omnidirectional image dataset demonstrate BD-rate reductions of 16.3%, 26.5%, and 19.5% over VVC in terms of V-PSNR, V-MS-SSIM, and V-LPIPS, respectively. Our method also achieves a 13.1% BD-rate reduction over ERP-based Cool-Chic 5.0 in terms of V-PSNR, while providing 2.6-3.6x faster decoding than the evaluated learned autoencoder codecs.
[CV-33] VERA-Hand: A Verified Multidimensional Benchmark for 2D Hand Pose Estimation in Third-Person Dynamic Gesture Videos WACV2027
链接: https://arxiv.org/abs/2610.32517
作者: Gibran Benitez-Garcia,Hiroki Takahashi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:Hand landmarks serve as input for gesture recognition, sign language understanding, and touchless human-computer interaction. Yet hand pose estimators are typically evaluated with frame-wise localization metrics such as PCK or MPJPE, which are computed over matched predictions and overlook missed or spurious hands, handedness errors, and temporal instability. To address these limitations, we present VERA-Hand, a benchmark that evaluates estimators using manually verified, frame-level 2D annotations for 10,690 hands across four complementary dimensions (detection, localization, identification, and temporal stability), as well as a joint Pose-F1 metric that requires both correct detection and accurate localization. Evaluating six estimators spanning hand-specific, whole-body, reconstruction-based, and foundation-model paradigms, we find that no estimator dominates: rankings reorder sharply across dimensions. VERA-Hand also exposes failures that localization-only and downstream evaluations may not reveal, including cross-hand overlap in over 85% of frames for a recent foundation model. A post-hoc cleanup case study demonstrates how the benchmark can validate corrective interventions, while stratified analyses show that robustness varies across motion-blur classes and hand regions. We release the verified annotations, protocol, and code at this https URL
[CV-34] Fast and Useful: Adapting Distilled Generators for Visual Learning
链接: https://arxiv.org/abs/2610.32512
作者: Patryk Bartkowiak,Dominik Michels,Sören Pirk,Wojciech Pałubicki
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Distilled text-to-image models generate images quickly, but classifiers trained on those images can perform substantially worse than classifiers trained on images from their slower base models. We adapt released distilled generators to produce better training data while retaining fast generation. We fit a low-rank update to the student’s first prediction so that it reaches an intermediate state produced by its unguided base, then disable the update and let the unchanged student finish generation. The base supplies training targets and is absent at inference. Guided targets and shorter jumps give classifiers lower accuracy in our ablations, showing why the state used for supervision matters. Across three generator families, the adapted images train better classifiers than images from the released students. With class-name templates at full ImageNet-1k scale, ResNet-50 top-1 rises from 8.4% to 34.2% for FLUX.2-klein and from 12.7% to 32.1% for SD-Turbo, using two and three denoiser evaluations per image, respectively. Fixed adapters also improve results on additional classes, fine-grained datasets, and other learner architectures and training recipes. Template-based adaptation remains useful against richer-prompt alternatives, while the benefit of combining adaptation with LLM generation prompts depends on the generator. Adapting a single generation step makes fast generators substantially more useful for training visual models.
[CV-35] Augmentation Strategy for DINO in Glaucoma Classification MICCAI2026
链接: https://arxiv.org/abs/2610.32511
作者: Ashish Kumar Meena,Chandra Sekhar Seelamantula
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 3 figures. Accepted at the 17th International Workshop on Machine Learning in Medical Imaging (MLMI 2026), held in conjunction with MICCAI 2026
Abstract:Self-supervised learning (SSL) offers a promising approach to medical image representation learning, yet augmentation strategies of SSL frameworks validated on natural images do not reliably transfer to specialized medical imaging domains. We analyze DINO, a knowledge-distillation based SSL framework, and empirically demonstrate that its default augmentation strategy produces unreliable representations for retinal fundus images, degrading downstream glaucoma classification performance. Guided by the principle that effective cropped views should share minimal redundant information while retaining clinically relevant signal, we redesign DINO’s augmentation strategy. Our approach biases the learned representations toward diagnostically critical features rather than low-level image statistics. We evaluate our method for glaucoma classification by training a linear classifier on the frozen representations, achieving strong generalization across seven benchmark datasets. URL for code and weights: this https URL
[CV-36] Distortion-Aware Spatiotemporal Learning for Real-Time Wide-Angle Video Rectification
链接: https://arxiv.org/abs/2610.32500
作者: Lang Nie,Yihan Yang,Wenbo Nie,Hongguang Zhu,Yang Wei,Chunyu Lin,Bin Xiao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 7 figures
Abstract:While wide-angle lenses provide an expanded field of view (FoV), they exhibit severe nonlinear distortion near image boundaries, unnaturally stretching people in these regions. Existing methods mostly focus on facial regions, require high computational costs, and lack temporal modeling, resulting in incomplete full-body rectification, impractical real-time deployment, and noticeable inter-frame shakes. To tackle these issues, we propose \textbfFastVC, an efficient full-body video rectification framework driven by distortion-aware spatiotemporal learning. Grounded in lens physics, FastVC introduces a distortion-aware thin-plate spline (DA-TPS) module that nonlinearly remaps control points into a peripheral-dense layout, enabling more accurate geometric fitting without additional computational cost. We further develop a dual-level spatiotemporal learning scheme for causal and stable video rectification. At the model level, neighborhood spatiotemporal attention (NST) aggregates local context and selectively fuses historical features with low overhead. At the optimization level, sparse spatiotemporal supervision applies spatial alignment only to anchor frames, decoupling frame-wise accuracy from temporal stability. Besides, we contribute the first benchmark dataset for wide-angle full-body video correction. Extensive experiments show that FastVC achieves state-of-the-art rectification fidelity and visual stability at 57.0 FPS.
[CV-37] K-CLUE: A Segmentation-to-Clustering Framework for Ki67 Cell Typing BMVC2026
链接: https://arxiv.org/abs/2610.32493
作者: Ana Beatriz Vieira,Sara P. Oliveira,Diana Montezuma,João Fraga,Jaime S. Cardoso,Arlindo L. Oliveira
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the BD-MedVis Workshop, BMVC 2026
Abstract:Ki67 is a key marker of tumor proliferation in breast cancer, but its quantification from immunohistochemistry images normally depends on supervised classifiers trained per dataset. We examine the role of frozen pretrained visual representations in annotation-free Ki67 cell typing. To this end, we use K-CLUE, a framework that segments individual nuclei, represents each nucleus with a pretrained encoder, reduces the resulting embeddings with principal component analysis, and separates Ki67-positive from Ki67-negative nuclei using k-means clustering, without task-specific training or manual annotations. Clustering the raw components is highly sensitive to the choice of encoder: across seven encoders, the Ki67-positive F1-score on matched MIST nuclei varies by 44 points, although Ki67 status remains linearly decodable from every representation, with supervised probe AUCs above 0.98. Analysis of the embedding spaces shows that clustering succeeds when DAB-related variation occupies the leading directions of variance. Whitening the retained principal components removes this dependence, reducing the spread across encoders to 1.00 F1-score point. Handcrafted color representations are also competitive, confirming that Ki67 status is strongly reflected in chromatic information, but they require an explicit choice of color space. In end-to-end evaluation, K-CLUE reaches F1-scores of 85.00% and 81.91% for Ki67-positive and Ki67-negative nuclei on MIST, and 79.59% and 81.16% on BCData, outperforming DeepLIIF and IHCScoreGAN on both datasets. These results show that pretrained representations can support annotation-free Ki67 typing when their variance structure is accounted for.
[CV-38] L0-Norm-Based CycleGAN for Unpaired Brain CT-to-MR Synthesis
链接: https://arxiv.org/abs/2610.32491
作者: Yunxiao Zhang,Shouqiang Du,Zexian Xu,Chuanlong Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Optimization and Control (math.OC)
备注: 13 pages, 2 figures
Abstract:Computed tomography (CT) and magnetic resonance (MR) imaging provide complementary information for the brain, but acquiring both modalities is not always possible. The cycle-consistent adversarial network (CycleGAN) enables unpaired CT-to-MR synthesis, but because the standard L1-norm cycle?consistency penalty grows linearly with the residual magnitude, the network tends to focus on large residuals. This paper introduces an L0-norm-based cycle?consistency term, in which the L0-norm is approximated by a smooth and bounded function. This approximation not only increases sensitivity to small residuals but also effectively suppresses the influence of large residuals. The proposed model and an otherwise identical L1-norm baseline were trained on 102 images and evaluated on 9 test images. The proposed model achieves a lower mean squared error (0.02746 versus 0.03464, a 20.7% reduction), a higher mean peak signal-to-noise ratio (15.80 versus 14.77 dB), and a higher mean structural similarity index (0.5689 versus 0.5407). It also has a lower maximum MSE and higher minimum PSNR and SSIM than the baseline. These results indicate that modifying the cycle-consistency penalty can improve reconstruction accuracy on this dataset.
[CV-39] Efficient Zero-Shot Video Anomaly Detection via Signal-Aware Adaptive Fusion (SAAF) WACV2027
链接: https://arxiv.org/abs/2610.32486
作者: Khaled Waleed Dawoud,Muhammad Haris Khan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at WACV 2027
Abstract:Deploying Vision-Language Models for video anomaly detection confronts a fundamental constraint. Their semantic reasoning demands computational budgets incompatible with the scale of real-world video streams. We present Signal-Aware Adaptive Fusion (SAAF), a framework that achieves efficient zero-shot anomaly detection through three key innovations. First, we introduce a dual-signal adaptive sampler operating in semantic embedding space: a delta signal captures sudden events via frame-to-frame cosine distance, while a drift signal detects static and gradual anomalies through deviation from rolling temporal context. Their OR-gated combination with adaptive EMA-based thresholding provides comprehensive anomaly coverage without training data. Second, we propose a threshold-gated cascaded architecture where lightweight frame-level scoring precedes expensive video-level temporal analysis, invoking it only when confidence warrants additional context. Third, we design a signal-aware weighted fusion mechanism that dynamically adapts score combination based on which signal triggered frame selection. Across four benchmarks, SAAF delivers competitive accuracy at substantially lower cost. Among training-free zero-shot methods, SAAF sets a new state of the art on XD-Violence (92.07%), MSAD (86.70%), and ShanghaiTech (80.54%) AUC-ROC, with XD-Violence and MSAD also surpassing supervised baselines. SAAF sustains 44.95, 48.61, and 41.06 FPS on XD-Violence, UCF-Crime, and MSAD, a 2.43 \times , 2.63 \times , and 2.22 \times throughput improvement over the prior best zero-shot method. Code: this https URL
[CV-40] PCFBench: How Far Can Large Vision-Language Models Go in Physics-aware Photonic Inverse Design? NEURIPS2026
链接: https://arxiv.org/abs/2610.32485
作者: Shengchao Chen,Ting Shu,Sufen Ren
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026, 66 pages, 30 figures, 27 tables
Abstract:Photonic crystal fibers (PCFs) underpin critical applications in telecommunications, sensing, and high-power laser delivery, yet their design remains a manual, expert-intensive process that couples geometric intuition with deep waveguide physics. Vision-language models (VLMs) offer a promising path toward automating PCF design, but can they truly reason about photonic physics from microscopy images, or do they merely recognize visual patterns? We introduce PCFBench, the first comprehensive benchmark for evaluating VLMs on the perception, reasoning, and decision steps of the photonic inverse-design loop. PCFBench spans 26 tasks organized into six categories of increasing cognitive demand (Geometry Perception, Physics Understanding, Text Generation, Multi-modal Reasoning, Inverse-Design Decision, and Code Generation), built from over 220K FDTD simulation samples across 18 structurally diverse fiber families. We evaluate both individual VLMs on isolated tasks and multi-agent systems (MAS) on the full sequential pipeline from perception to design. Our results reveal a striking perception-design gap: current models can identify fiber geometry with moderate success, yet consistently fail to translate that visual understanding into physics-aware design actions. Moreover, cascading agents through the pipeline exposes significant error propagation, where upstream perception mistakes compound into downstream design failures. These findings suggest that neither model scaling nor naive agent chaining suffices, and that closing the loop demands explicit physical grounding. PCFBench provides the community with a rigorous, multi-dimensional testbed to measure progress toward AI-assisted photonic engineering. The project is available at this https URL.
[CV-41] PathJev: Decision-Native Vision-Language Modelling for Histopathology
链接: https://arxiv.org/abs/2610.32478
作者: Cristian Tommasino,Antonio M. Rinaldi
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 25 pages, 4 figures. Code: this https URL ; model and demo: this https URL
Abstract:Vision-language models used for decisions in histopathology rarely return calibrated probabilities over the candidates a user cares about. Decision-native models write a question and a candidate set chosen at run time into the input and return a normalised distribution from a small decision head in one forward pass. We build PathJev-P, a decision-native model for histopathology: MedGemma-1.5-4B with LoRA adapters and a decision head, trained on 40,000 patch states from three public collections whose expert labels are turned into typed questions with organ-specific candidate sets and a “none of these” option. In a comparison over 3 seeds with an analysis plan fixed before testing, on test sets without documented overlap with the training data of the backbone, PathJev-P is more accurate than a generative model trained with the same backbone, data and selection rule, the zero-shot pathology model CONCH and a general medical decision model (balanced accuracy 0.541 against 0.492, 0.345 and 0.213); its advantage over the generative model comes from two organs absent from training. The calibration advantage depends on how the generative model’s probabilities are read out. With the summed candidate log-likelihood used in previous work, PathJev-P has the lower calibration error (ECE 0.204 against 0.279); with a length-normalised read-out of the same generative model, examined post hoc, the two have similar mean calibration error (0.201) and the generative model is better calibrated on the unseen organs, where PathJev-P remains over-confident. A supervised linear probe on a pathology foundation model is better calibrated in every setting. In a qualitative whole-slide demonstration, PathJev-P labels unsupervised tissue clusters with abstention and separates carcinoma from other tissue while running the model on a fraction of the tiles.
[CV-42] Online Continual Learning in Edge-Cloud Applications
链接: https://arxiv.org/abs/2610.32472
作者: Charalampos Davalas,Dimitrios Michail,Christos Diou,Iraklis Varlamis,Konstantinos Tserpes
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Deployed machine learning models in real-world IoT and edge/cloud systems face a critical reliability bottleneck: the requirement for online adaptation to non-stationary data streams while operating without access to the full original training dataset. Performance is severely degraded when models encounter Out-of-Distribution (OOD) inputs or suffer from catastrophic forgetting. Addressing these challenges requires adaptation mechanisms capable of learning from evolving data streams while preserving previously acquired knowledge. This paper proposes an edge-cloud methodology for integrating Continual Learning (CL) with Out-of-Distribution (OOD) Detection and Concept Drift Detection in IoT applications. The methodology incorporates ODIN and ECDD as alternative mechanisms for identifying informative samples for annotation and triggering rehearsal-based model updates. Experimental results across automotive, healthcare, and agricultural datasets demonstrate the applicability of the proposed methodology and its potential to support model adaptation while mitigating catastrophic forgetting.
[CV-43] OPTARI: An open-source software framework for clinical translation of optoacoustic imaging
链接: https://arxiv.org/abs/2610.32468
作者: Moritz Schillinger,Maximilian Bader,Adrian Buehler,Felix Wachter,Ferdinand Knieling,Vasilis Ntziachristos,Katharina Breininger
类目: oftware Engineering (cs.SE); Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
备注: 15 pages, 4 figures, 2 tables. Submitted to the Journal of Biomedical Optics. Code and documentation: this https URL
Abstract:Combined optoacoustic (OptA) and ultrasound imaging is a promising modality across a wide range of pathologies. Its use in clinical research, however, is limited by a lack of analysis software accessible to users without programming expertise. Existing tools are either script-based engineering frameworks or proprietary vendor applications. We present OPTARI, an open-source graphical software application to facilitate large-scale clinical optoacoustic-ultrasound (OPUS) studies. OPTARI builds on the napari image viewer and the PATATO toolkit for OptA data processing. It supports native vendor and open HDF5 data formats, aligns with the IPASC standard, and is distributed as a standalone executable. Features include region-of-interest annotation, reconstruction, chromophore unmixing, deep-learning-based tissue segmentation, and temporal and spectral analysis modules. OPTARI offers a combination of features for clinical study analysis that none of the compared tools provide. The automated tissue segmentation produced results comparable to expert labeling for muscle tissue, and batch processing reduced analysis time from minutes to under one second per scan. The tool is used in ongoing clinical OPUS studies. OPTARI connects engineering frameworks with medical research and provides an open-source basis for reproducible and standardized analysis in clinical OptA imaging studies.
[CV-44] MDMR-Bench: A Multi-Dimensional Benchmark for Multi-Reference Image Generation NEURIPS2026
链接: https://arxiv.org/abs/2610.32459
作者: Bowen Zheng,Jintao Lin,Changheng Yi,Haoran Yang,Binxiao Huang,Chenyang Lei,Rui Liu,Xihui Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Bowen Zheng and Jintao Lin contributed equally to this work. Accepted at NeurIPS 2026, Evaluations Datasets Track
Abstract:Image generation models have rapidly advanced, with many models now supporting generation conditioned on multiple reference images and textual instructions. Such models enable a wide range of applications, including entity fusion and attribute transfer across images. However, existing benchmarks for multi-reference image generation rely on limited evaluation dimensions and fail to capture the diverse challenges of the task. To address this limitation, we propose MDMR-Bench, a Multi-Dimensional benchmark for Multi-Reference image generation. MDMR-Bench is organized around representative capability settings and consists of multiple evaluation subsets spanning diverse dimensions of multi-reference image generation, including (1) common entity fusion and attribute transfer, (2) multi-element extraction from a single image, (3) infographic manipulation, and (4) visual reasoning-intensive generation. This design enables systematic analysis of model capabilities and failure modes across different challenge dimensions. Furthermore, to achieve more reliable and interpretable evaluation, we introduce a checklist-based evaluation protocol that decomposes generation quality into fine-grained evaluation aspects. Instead of relying solely on holistic scoring, our framework uses structured checklists composed of both global-level and atomic-level queries, enabling more diagnostic assessment of model performance. By benchmarking both open-source and proprietary models, we demonstrate that MDMR-Bench provides fine-grained insights into model performance across different dimensions. The dataset and code are publicly available.
[CV-45] AutoMIL: Automated Multiple Instance Learning for Whole-Slide Image Classification
链接: https://arxiv.org/abs/2610.32452
作者: Jonas Waibel,Frank Kramer,Dominik Müller
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Developing multiple instance learning pipelines for the classification of whole-slide images involves many interdependent design decisions and expertise in both pathology and machine learning. We present AutoMIL, an open-source framework that applies principles from the field of automated machine learning to tackle this task by automating integral pipeline steps, such as preprocessing, training and evaluation. In a benchmark evaluation with default settings, AutoMIL reached ROC-AUC scores of 0.94 on UBC-OCEAN, 0.99 on PCam and 0.93-0.96 on a TCGA-NSCLC subset, comparable to specialised pipelines in related work. Ensembling mainly improved performance on imbalanced data, showing that an automated pipeline must adapt to dataset characteristics. AutoMIL thus provides an accessible baseline for whole-slide image classification without manual pipeline configuration.
[CV-46] Learning Fine-Grained Visual Classification with Text as Privileged Information via Multimodal Distillation
链接: https://arxiv.org/abs/2610.32442
作者: Aishwarya Venkataramanan,Christian Ickler,Joachim Denzler
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 16 pages
Abstract:Fine-grained visual classification (FGVC) requires distinguishing between highly similar categories by focusing on subtle, localized visual differences. Recent multimodal methods have demonstrated that textual attribute descriptions can provide valuable semantic supervision by explicitly describing discriminative object characteristics. However, these descriptions are typically unavailable at test time, limiting their direct use in deployment. We propose FG-LUPI, a Learning Using Privileged Information (LUPI)-based framework that leverages attribute descriptions during training. We train a multimodal teacher that integrates image patches and textual attributes via gated cross-attention, allowing semantic attributes to condition visual representations. We distill this knowledge into an image-only student using multi-level distillation at the prediction, patch, and attention levels, aligning global predictions, patch representations, and attention patterns with the teacher. This encourages the student to inherit attention structure learned from multimodal supervision while maintaining image-only inference. Experiments on multiple FGVC benchmarks show consistent improvements in classification accuracy over strong ViT-based baselines. Furthermore, evaluation using Expected Calibration Error (ECE) shows that FG-LUPI also produces better-calibrated predictions across all six benchmarks. These results suggest that textual attribute supervision can benefit not only fine-grained recognition performance but also the reliability of model confidence, while preserving the efficiency and practicality of image-only inference.
[CV-47] PROVE: Probabilistic Visual Grounding via Patch-Level Evidence for Vision-Language Models
链接: https://arxiv.org/abs/2610.32440
作者: Mayank Nautiyal,Tianru Zhang,Li Ju,Ekta Vats,Prashant Singh
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Visual grounding with vision-language models such as CLIP typically produces a single deterministic relevance map for an image-query pair, without indicating whether the localisation is reliable or whether the referent is present at all. We introduce \textbfPROVE (\textbfProbabilistic \textbfVisual Grounding via Patch-Level \textbfEvidence), a post-hoc evidential adapter that maps frozen CLIP patch features to per-patch Beta distributions without modifying the backbone. PROVE casts grounding as per-patch membership: each patch accumulates positive and negative evidence for whether it belongs to the referent, so a query may match zero, one, or many patches. We train PROVE with an evidential objective on soft patch-coverage targets, and define a prior-pull objective that keeps absent-referent examples near an uninformative prior, so that vacuity serves as an absence signal without a dedicated presence classifier. During inference, a single forward pass yields, in closed form, a posterior-mean grounding map, a spatial vacuity map, and an image-level absence score. We evaluate PROVE on referring-expression benchmarks: it is competitive in localisation with state-of-the-art CLIP grounding methods, its uncertainty tracks localisation error and supports selective prediction under abstention, and its image-level score detects when the referent is absent altogether. Our evidential patch-membership modelling offers a simple, drop-in mechanism for probabilistic verification in frozen vision-language grounding systems.
[CV-48] mpo-Zero: Recursive Self-Play in Executable Worlds for Video Reasoning
链接: https://arxiv.org/abs/2610.32433
作者: Jingmin Zhu,Yanyan Zou,Xiuyu Li,James Bailey,Dinh Phung,Tao Yu,Yunyun Hou,Jun Liu,Jun Zhou,Sirui Wang,Nan Duan,Jiachen Liu,Qiuhong Ke
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video reasoning requires understanding how events unfold over time, yet videos with accurate temporal annotations are expensive to collect. Self-play offers an alternative by letting a model generate and solve its own tasks without manual annotation. However, existing video self-play methods train on fixed sets of recorded videos, so they cannot alter the underlying event dynamics and rely on noisy pseudo-labels derived from the model’s own outputs, such as majority voting, that risk reinforcing shared model errors. To overcome these limitations, we introduce Tempo-Zero, a self-improvement framework that closes a loop of executable environments, execution-verified self-play, and environment evolution, where executing a program renders a video with the ground-truth option and interval of every question. A single policy plays both Questioner and Solver, proposing questions and answering them with timestamped evidence lines, and is trained on execution-grounded rewards instead of model consensus. The Solver’s performance steers a curriculum over tasks and difficulty levels, and the policy selects rule changes that enrich the event dynamics of later rounds. Despite being trained solely on simple programs, without supervised fine-tuning or external video data which prior methods typically require, Tempo-Zero outperforms the base models and recent self-play baselines on average across twelve real-world video benchmarks at different model scales, with particularly strong gains on temporal grounding tasks. Project page: this https URL
[CV-49] AlphaAnything: Extending Pretrained RGB Generators to RGBA
链接: https://arxiv.org/abs/2610.32413
作者: Qijie Wang,Zhixin Xu,Zhe Li,Haowei Zhu,Honghao Qiao,Bo Peng,Dawei Li,Bin Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing RGBA generation methods commonly learn a transparency-specific representation through reconstruction, introducing an additional training stage and potentially altering the RGB decoding function inherited from pretrained models. We present AlphaAnything, a lightweight framework that extends pretrained RGB generators to RGBA directly in their original latent space, without RGBA reconstruction pretraining. AlphaAnything inserts LoRA modules into the generative transformer and attaches an alpha branch to the frozen RGB VAE decoder. Unidirectional RGB-to-alpha feature sharing allows opacity supervision to shape the generative latent while guaranteeing that, for any fixed latent, RGB decoding remains identical to that of the pretrained decoder. We further introduce Endpoint-Aware Alpha Loss (EA Loss), which derives opacity supervision from the complementary foreground–background weights used in compositing. It supports exact opacity endpoints and continuous partial transparency while emphasizing localized alpha errors. Experiments on text-to-RGBA and image-to-RGBA tasks across multiple pretrained RGB backbones demonstrate competitive generation quality, and controlled ablations validate the shared-latent training design and EA Loss.
[CV-50] IE: Training-free CLIP Embedding Guidance for Compositional Text-to-Image Alignment
链接: https://arxiv.org/abs/2610.32407
作者: Sujung Hong,Tae Eun Choi,Youngjun Jun,Chanyong Yoon,Seong Jae Hwang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-to-image (T2I) diffusion models have advanced notably, but still fail to satisfy prompt conditions, resulting in attribute mismatches and missing objects. These errors normally fall into two categories: concept loss, when an object or attribute (i.e., concept) goes missing, and concept confusion, when an attribute is assigned to the wrong object or multiple objects blend together. Recent studies suggest that these challenges are caused by the limited capacity of CLIP text encoder to capture fine-grained semantic details. Although several methods have been proposed, concept loss and concept confusion persist in recent T2I models when handling multi-object, multi-attribute prompts. In this paper, we conduct an embedding-level analysis to develop a training-free, inference-time solution for concept loss and concept confusion across CLIP-based T2I models. We find that (1) concepts mentioned later exhibit higher embedding entropy, reflecting less concentrated representations that are more often associated with concept loss, and (2) the first CLIP attention layer captures the strength of binding between each object and its attribute. Guided by our findings, we introduce TIE, a method that improves semantic alignment through a single text-embedding update. TIE addresses concept loss via entropy-aware singular value amplification and resolves concept confusion through interpolation-extrapolation binding based on CLIP attention scores, all in a training-free manner. Extensive experiments demonstrate that TIE enhances semantic fidelity in multi-concept scenarios with minimal sampling overhead. The project page is available at this https URL.
[CV-51] An Error Budget for Shuffled Multimodal Windows in the 3rd WEAR Challenge: Stacking Count Constraints and Seam Matching
链接: https://arxiv.org/abs/2610.32398
作者: Soheil Kooklan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 10 tables, 1 figure. Technical report on a post-deadline system for the 3rd WEAR Dataset Challenge at HASCA 2026. Code: this https URL
Abstract:The 3rd WEAR Dataset Challenge asks for an activity label for every one-second window of four unseen participants. Each window has 15 VideoMAEv2 frame embeddings, the accelerometer signal of one random limb, and the windows are shuffled. This report is an empirical analysis of one system built after the challenge deadline. Window classifiers are stacked with features averaged over per-participant graphs of video neighbours and scored successor candidates, and a mean-field style decoder with a soft count constraint per activity labels all windows of a participant together. The setting is offline and transductive, and the count constraint relies on each activity being performed once. In a participant-wise simulation with participant-level bootstrap intervals, stacking added 0.051 macro F1 over stage-1 fusion and the count constraint 0.011 (95% interval 0.006 to 0.017). A quadratic seam distance ranks the true same-limb successor first among about 720 candidates in 83.9% of cases, but adding it to the link model changed macro F1 by -0.0001 (-0.0006 to 0.0005). With the decoder settings and the mixing weight chosen inside the folds and the mislabelled spans left out of training, the system scores 0.936 in simulation (0.940 for participants who performed each activity once); the submitted configuration scores 0.938 (0.943). Per-participant scores range from 0.80 to 0.97. These results come from a non-nested stacking protocol that uses unlabelled features transductively, so they are not a fully inductive estimate. An oracle that averages the final unary scores over the true segments reaches 0.973, and 79% of the remaining errors involve the null class. On the private leaderboard, which no choice used, the selected Kaggle submission scored 0.9231 (public 0.9254, with bounds partly tuned there), and a later one with bounds chosen in simulation scored 0.9247 (public 0.9253).
[CV-52] Interactive image diagnosis using test-time backpropagation from the auxiliary task
链接: https://arxiv.org/abs/2610.32396
作者: Takuro Shimaya
类目: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
备注: 8 pages, accepted to SPIE Medical Imaging 2025
Abstract:Multi-task deep learning has been developed to ensure explainability in computer-aided diagnosis (CAD) systems. These systems simultaneously predict interpretable characteristics of findings in auxiliary tasks to help radiologists assess the reliability of the diagnostic prediction. However, the CAD’s opinion would not be referred to when the auxiliary task prediction deviates from the radiologist’s expectation. To further leverage CAD’s knowledge in the above situation, we proposed a multi-task gradient assist (MGA) method that brings interactivity to multi-task models without additional training. MGA allows radiologists to give feedback to the CAD and ask for additional diagnostic suggestions. Using two datasets of mammography and CT, we show that MGA successfully improves diagnostic accuracy upon feedback on various auxiliary tasks. Compared to an existing method using forward propagation only, MGA performs more robustly and can be combined to further improve accuracy. Our method paves a promising way for integrating artificial intelligence into image diagnosis.
[CV-53] PASTA: Prompt-Aware Annealed Diffusion Sampling ICONIP2026
链接: https://arxiv.org/abs/2610.32390
作者: Ren Ajiki,Ryugo Morita,Stanislav Frolov,Brian Bernhard Moser,Ko Watanabe,Riku Takahashi,Andreas Dengel,Jinjia Zhou
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 15 pages, 4 figures, 2 tables. Accepted to ICONIP2026
Abstract:While Diffusion models produce high-quality images, real-world visual design requires intentionally preserving clean regions for headlines, logos, or UI elements. Existing layout-aware text-to-image methods struggle to strictly preserve reserved regions, often introducing unwanted content that complicates downstream editing. Existing approaches either manipulate attention to displace content or suppress conditioning via noise-level interventions. However, these methods sacrifice visual coherence, producing flat regions or noticeable style inconsistencies between reserved and non-reserved areas. To address this limitation, we propose a Prompt-Aware Shared-Latent Training-Free Annealed Diffusion Sampling (PASTA), a training-free framework that enforces reserved-region constraints by selectively applying semantic guidance to object-allowed regions during sampling. PASTA operates on a shared latent trajectory, where full-prompt and context-prompt guidance are fused through mask-guided sampling. An annealed schedule suppresses semantic leakage during early denoising stages while gradually restoring visual consistency in later stages. Extensive experiments demonstrate that PASTA improves reserved-region fidelity while preserving overall image realism and perceptual consistency.
[CV-54] UniWorld: A Unified World Model for Real-Time Visual World Simulation
链接: https://arxiv.org/abs/2610.32381
作者: Xin Zhang,Yabo Chen,Shuaiwei Wang,Zixuan Duan,Haibin Huang,Feng Xu,Xuelong Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Continuous visual world simulation requires coupling evolving scene geometry with temporally persistent appearance. We present UniWorld, a unified framework that integrates depth trajectories and reference appearance within a shared video generation model. The central design maintains visual and geometric histories on a common timeline. Noise-aligned overlapping continuation places generated history at the prescribed sampling noise level and commits only newly generated latents. Rolling depth latent state preserves the corresponding geometric controls, incrementally reads newly available geometry, and keeps the conditioning fixed within each sampling window. Together, these mechanisms ensure exact preservation of latent history while bounding the active depth state by the window size. We evaluate continuous generation, paired history interventions, and long-horizon behavior across diverse scene categories. The same generator supports estimated, native, and on-demand simulator depth, while cached and incremental execution paths produce bitwise-identical outputs. Appearance changes and geometric control branches operate through the same interface without modifying committed history. The measured steady-state DiT execution path achieves real-time generation throughput, demonstrating the efficiency of UniWorld for continuous visual world simulation.
[CV-55] Sam-Correction: Fully Adaptive Label Noise Reduction for Medical Image Segmentation
链接: https://arxiv.org/abs/2610.32377
作者: Takuro Shimaya,Masahiro Saiko
类目: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
备注: 5 pages, accepted to IEEE International Symposium on Biomedical Imaging (ISBI) 2024
Abstract:The quality of the teacher label has a dramatic impact on the outcome of the machine learning process. Medical image segmentation often suffers from a noisy ground truth (GT) mask due to difficulties in consistent annotation. Although a recent study has proposed a promising method to curate the noisy GT mask, this requires additional well-curated GT masks that are not always accessible without the aid of multiple medical experts. As an alternative, we propose SAM-Correction, which utilizes a foundation model to correct the noisy GT mask in a fully adaptive manner instead of manually preparing well-curated masks. In our approach, we progressively mitigate label noise via the segment anything model (SAM), where prompts are given from another segmentation network that we train concurrently. Experiments with various types of label noise derived from actual human annotations demonstrate that SAM-C achieves precise noise reduction and high-performing segmentation models trained thereby. Our method paves the way for robust model training even when clean labels are entirely unavailable.
[CV-56] EigenGradCAM with AugSmooth: Enhancing Explainability and Precision in Multi-Stage Alzheimers Disease Detection
链接: https://arxiv.org/abs/2610.32374
作者: Arsham Azam,Sagar Ashraf,Tayyaba Farhat,Rasikh Ali,Sheeraz Akram
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Alzheimer’s disease (AD) is a progressive neurological disorder associated with memory loss, cognitive decline, and behavioral disturbances. Early and accurate diagnosis is essential, yet MRI-based deep learning models often lack interpretability, limiting their clinical use. This study presents an explainable deep learning approach for four-class AD staging using 2D axial MRI images. The method combines eigenvector-based Grad-CAM with gradient smoothing to improve saliency localization and heatmap stability. Principal component analysis is applied to gradient-weighted convolutional layers, while Gaussian noise is added to saliency weights before decomposition. To address class imbalance in the Moderate Dementia category, Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) generates synthetic samples to balance the four-stage training distribution. Custom convolutional neural networks are trained on the Best Alzheimer MRI Dataset, achieving 97.2% test accuracy, 97.0% balanced accuracy, and a Matthews correlation coefficient of 0.944. The area under the ROC curve ranges from 0.96 to 0.99 across classes. Generated heatmaps highlight the hippocampus and entorhinal cortex, regions associated with Alzheimer’s disease biomarkers. These findings suggest that the proposed approach combines strong classification performance with improved model interpretability, although further clinical validation is needed.
[CV-57] Past as State Present as Attention: Persistent-State Blockwise Flow Matching for Long-Horizon Co-Speech Motion Generation NEURIPS2026
链接: https://arxiv.org/abs/2610.32373
作者: Yi Liu,Xiangyue Zhang,Jianfang Li,Jia Ma,Yisheng He,Jianqiang Ren
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Multimedia (cs.MM)
备注: Accepted at NeurIPS 2026. 38 pages, 10 figures, 17 tables
Abstract:Co-speech motion generation remains challenging over long horizons due to the need for temporal coherence and consistent style. Existing methods typically generate long motions segment by segment, either conditioning on a short seed of recent frames, which gradually drifts, or retaining ever-longer explicit histories, which are costly to maintain and hard to learn from limited data. We trace these limitations to a key asymmetry between past and present: past motion carries slowly varying style that should be compactly summarized, whereas the current segment exhibits fast-changing local dynamics that require fine-grained modeling. This asymmetry motivates a simple Past-As-State, Present-as-Attention principle for long-horizon co-speech motion generation. Following this principle, we propose PASPA, which summarizes past history with an inter-block persistent state and models the current segment with intra-block bidirectional self-attention. We instantiate PASPA within a blockwise flow matching framework, with multi-block supervision for effective training over variable histories, prefill-decode state caching for efficient rollout, and hybrid classifier-free guidance for enhanced history and speech conditioning. Extensive public-benchmark experiments show that PASPA achieves state-of-the-art FGD performance and preserves motion style across long-sequence generation.
[CV-58] A Regularity Condition Under Which a CPAB Transformation is a Diffeomorphism
链接: https://arxiv.org/abs/2610.32369
作者: Amit Aflalo,Oren Freifeld
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Continuous Piecewise-Affine Based (CPAB) transformations, introduced by Freifeld et al. (2015, 2017), are flows obtained by integrating continuous piecewise-affine (CPA) velocity fields. These expressive transformations have proven useful in applications including nonlinear temporal warping and trainable activation functions in 1D, as well as nonrigid spatial deformations in 2D and 3D. While computational costs were not an issue in 1D, they originally limited the practical expressiveness of CPAB transformations in higher dimensions, especially in 3D. Our recent work (Aflalo et al., 2026) removed this bottleneck, enabling efficient optimization with much finer tessellations and substantially greater expressiveness. There is, however, a mathematical subtlety that was overlooked in the original CPAB papers (Freifeld et al., 2015, 2017). CPAB transformations are always bi-Lipschitz homeomorphisms, implying that both the transformations and their inverses are differentiable almost everywhere. They are not, however, necessarily diffeomorphisms. This note remedies the omission by identifying sufficient regularity conditions under which a CPAB transformation is guaranteed to be a diffeomorphism. We explain how differentiability can fail and how the proposed conditions prevent this failure. In 1D and 2D, we show that these conditions can be easily enforced and, for commonly used tessellations, hold with probability one under standard sampling assumptions. We expect analogous behavior in 3D.
[CV-59] Search-Based Testing of Modality Sensitivity in Vision Language Models
链接: https://arxiv.org/abs/2610.32360
作者: Oliver Weißl,Eduardo Breitembach,Andrea Stocco
类目: Computer Vision and Pattern Recognition (cs.CV); Software Engineering (cs.SE)
备注:
Abstract:Vision-language models (VLMs) process images and text jointly, yet robustness evaluations often perturb these inputs separately or assess only predefined corruption levels. Such evaluations provide limited evidence of which modality is more susceptible to perturbations and whether smaller changes across both modalities can jointly expose failures. We present MoST, a black-box testing framework that characterizes modality sensitivity by comparing image-only, text-only, and joint perturbations. MoST uses multi-objective search to select and combine perturbations and adjust their severity, seeking prediction degradation while limiting changes to images and prompts. We evaluate MoST on four VLMs performing object localization on ImageNet and driving images. The models exhibit distinct sensitivities: textual perturbations dominate for some, whereas others are susceptible across both modalities. In several configurations, joint perturbations match or exceed the effectiveness of single-modality perturbations with smaller modifications to each input. In a human study, 60-76% of assessments across models judge the perturbed inputs sufficiently understandable to perform the task, and human localization is more accurate than that of the corresponding VLMs. These results show that evaluating modalities separately can overlook failure-inducing combinations and that modality sensitivity provides a more detailed account of VLM robustness than aggregate failure rates alone.
[CV-60] ChoreoStudio: Planning What When and Who in Music-Driven Group Choreography
链接: https://arxiv.org/abs/2610.32325
作者: Jing Xu,Cunjian Chen,Qiuhong Ke
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Music-driven group dance generation must decide not only how each dancer moves, but also what formation the group should form, when it should change, and who should execute each transition. Existing methods can generate orderly and rhythmically synchronized group motion, but remain limited in formation evolution and inter-dancer interaction. In human choreography, formation transitions often align with changes in high-level musical structure, such as the progression from an introduction to a chorus or from a chorus to a climax. We present ChoreoStudio, which makes choreography planning explicit. A music disentangler first decomposes the audio into frame-level rhythmic features and segment-level structural features. Given the segment-level features and sparse labels, a frozen omni-modal model acts as a scriptwriter, producing a global semantic description that guides the Director’s choreography planning. Based on this context, the Director produces three coupled interaction decisions: a centroid-relative point set specifies what formation to produce, a transition gate determines when it should change, and a mobility allocation indicates who should execute the transition. The transition gate and mobility allocation are further conditioned on the planned geometric change, coupling formation targets with their temporal and dancer-specific execution. At a finer temporal scale, frame-aligned music features condition the diffusion generator to control detailed body motion and preserve beat-level alignment. The segment-level interaction plan is injected in parallel through frame-aligned instruction tokens and cross-dancer attention biases, coordinating how the generator realizes each transition. Experiments on AIOZ-GDance demonstrate that ChoreoStudio follows the planned interactions while achieving state-of-the-art performance in motion quality, group organization, and formation dynamics.
[CV-61] DeformStruct: Deformable Structure Modeling for Point-Supervised Temporal Action Localization
链接: https://arxiv.org/abs/2610.32321
作者: Mengnan Liu,Sanping Zhou,Kun Xia,Yonghao Dong,Jie Wang,Gang Hua
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Point-supervised Temporal Action Localization (PTAL) aims to localize action instances using only a single annotated timestamp for each instance. Such sparse supervision offers limited guidance for modeling the structural organization of an action instance. Existing PTAL methods primarily infer action regions from classification responses, leaving the structural organization of multiple action-related parts within each instance insufficiently modeled. To address this problem, we propose DeformStruct, an instance-level deformable structure modeling framework for PTAL. DeformStruct progressively constructs the structure of each action instance through three stages. First, Structural Anchor Mining discovers multiple anchors from annotated points and action-responsive regions, providing structural nodes grounded in the current instance. Second, Deformable Part Alignment associates latent structural parts with these anchors through deformation-aware matching, allowing their temporal supports to adapt to instance-dependent variations. Third, Relational Structural Organization explicitly models the relative temporal relations among the aligned parts, forming a relational representation that characterizes how different action-related parts are organized together. Extensive experiments on THUMOS14, ActivityNet-v1.3, GTEA, and BEOID demonstrate consistent improvements in point-supervised temporal localization and validate the effectiveness of deformable structural modeling.
[CV-62] Organizing Temporal Evidence in Hyperbolic Space for Video Moment Retrieval
链接: https://arxiv.org/abs/2610.32317
作者: Mengnan Liu,Sanping Zhou,Kun Xia,Jiaxuan Lin,Gang Hua,Le Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video moment retrieval aims to localize a query-relevant temporal segment in an untrimmed video. Existing methods typically perform video-text alignment in Euclidean space, where representations are embedded in a flat geometry. However, accurate moment localization requires organizing query-relevant evidence across multiple grounding states, where target regions, boundary transitions, and contextual distractors exhibit distinct temporal structures and relevance patterns. Such Euclidean representations lack an explicit geometry to jointly capture these relationships, motivating us to explore structured temporal evidence organization from a geometric perspective. Specifically, we propose Hyperbolic Temporal Evidence Organization (HTEO), a geometric framework that organizes query-aware temporal memory in hyperbolic space. HTEO maps temporal representations into a Poincaré ball and organizes temporal evidence through Radial Evidence Organization (REO) and Orthogonal Angular Refinement (OAR). REO adjusts temporal tokens’ radial positions to encode the structural differences among temporal evidence, while OAR enhances query-conditioned semantic alignment through orthogonal rotation. The organized representation is then projected back to the Euclidean tangent space and injected into the decoder through a lightweight residual memory update, enabling hyperbolic geometric cues to be incorporated into standard Transformer decoders while preserving the original cross-modal memory. Furthermore, we introduce radial structure regularization and angular relevance supervision to explicitly guide temporal evidence organization and query-conditioned semantic alignment, respectively. Extensive experiments and analyses across multiple benchmarks demonstrate that HTEO improves temporal localization while providing an interpretable geometric organization of temporal evidence.
[CV-63] GeoLEON-CLIP: Graph-Anchored Multimodal Contrastive Learning for Geographic Region Representation
链接: https://arxiv.org/abs/2610.32311
作者: Radoslaw Malek,Szymon Soltysiak,Aleksandra Kawala-Sterniuk,Piotr Szymanski
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at the 9th International Workshop on AI for Geographic Knowledge Discovery (GeoAI 2026), co-located with the 34th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL 2026)
Abstract:Geographic regions are multimodal: satellite imagery captures visual morphology, text conveys human-interpretable semantics, and map structures describe urban context. Existing geospatial contrastive methods typically align image-text or image-location pairs, underusing structured map data. We introduce GeoLEON-CLIP, a graph-anchored contrastive framework that aligns three views of each H3-indexed region: an OpenStreetMap-derived graph embedding, a satellite image crop, and a textual description. Frozen pretrained LEON graph embeddings provide a semantic anchor between vision and language. Frozen graph, image, and text encoders are connected by trainable residual projection heads and trained with a symmetric tri-modal contrastive objective. Experiments show strong graph-image and graph-text alignment; graph-image retrieval achieves the best cross-modal performance in a mixed-resolution pool. To evaluate the anchoring hypothesis, we propose chained cross-modal recall \mathrmR@A@B , which routes retrieval through the graph modality and compares it with direct retrieval under a matched candidate budget. At \mathrmR@1@1 , image \rightarrow graph \rightarrow text and text \rightarrow graph \rightarrow image outperform direct image-text retrieval by 4.8 and 4.4 points, respectively, despite using a single intermediate cell. Downstream probes show task-dependent gains, with the largest improvements on Airbnb and Beijing housing. These results indicate that OpenStreetMap graph structure can anchor multimodal representations combining maps, imagery, and language.
[CV-64] OlmoEarth v1.3: Embeddings
链接: https://arxiv.org/abs/2610.32294
作者: Gabriel Tseng,Favyen Bastani,Patrick Alan Johnson,Yawen Zhang,Piper Wolters,Ando Shah,Hadrien Sablon,Joseph Redmon,Heather DeHaven,Rachel Ratner,Sasha Levy,Jordan Steward,Katrina Midgley,Andrew Howe,Hunter Pitelka,Ted Schmitt
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We make some OlmoEarth embeddings. To do this, we train a new model: OlmoEarth v1.3. OlmoEarth v1.3 is designed to condense multimodal, multitemporal remote sensing data into a single digestible 64 or 128-dimensional vector. OlmoEarth v1.3 produces state-of-the-art embeddings. In addition, OlmoEarth v1.3 is a strong encoder backbone – competitive with OlmoEarth v1 and v1.2 – while requiring 18% and 53% of the GPU hours to train, respectively. We use OlmoEarth v1.3 to compute global, annualized 10m embeddings, which we release under the OlmoEarth artifact license.
[CV-65] Complementing DINO Features with Image Structure for Part Discovery
链接: https://arxiv.org/abs/2610.32293
作者: Samyak Rawlekar,Nikhil Chowdary Paleti,Amey Gupta,Narendra Ahuja
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Self-supervised Vision Transformers (ViTs) like DINO encode rich semantics, enabling training-free segmentation of objects from the background and discovery of their parts via feature clustering. This paper focuses on part discovery: partitioning a foreground object into semantic parts. Existing approaches produce imprecise part boundaries. We trace this to the rigid ViT patch grid used in training DINO, where patches straddling boundaries contain pixels from multiple parts. While DINO’s invariance-based training could absorb this pixel-level mixing into clean features, experiments with over 26K foreground patches show it does not: boundary-overlapping patches produce features with higher entropy and lower assignment confidence, blurring predicted part boundaries. Rather than make the massive effort of retraining DINO over geometrically coherent regions, an easier approach is to take boundaries directly from a boundary detector, which supplies the geometric information DINO’s features lack. Motivated by this, we introduce Part-DINO, a training-free framework that combines boundaries from low-level image segmentation with DINO’s semantic features. We use the boundaries to partition the image into regions, pool DINO features within each region, and group them to discover parts. Without any training, Part-DINO produces structurally coherent part assignments, outperforming training-free baselines across four standard benchmarks (CUB, CelebA, PartImageNet-OOD, PartImageNet-Seg) and remaining competitive with weakly-supervised and unsupervised methods.
[CV-66] ReLEAP: Upcycling Vision Datasets through Cross-Task Unification and Semi-Supervised Annotation ACCV
链接: https://arxiv.org/abs/2610.32289
作者: Julia Simon,Daniel Steininger,Andreas Trondl
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at Asian Conference on Computer Vision (ACCV) 2026. Code and models available at this https URL
Abstract:Developing robust vision models requires large amounts of annotated data, especially for tasks involving fine-grained class distinctions under variable environmental conditions. Even in domains with numerous existing datasets, differences in annotation formats, label definitions and targeted learning tasks hinder data reuse and integration. To address this challenge, we introduce ReLEAP, a scalable Python framework for unifying heterogeneous image datasets through data-source abstraction, systematic reuse of annotations across learning tasks, and mapping of labels to a shared hierarchical taxonomy. We demonstrate its applicability in precision agriculture by integrating fifteen diverse datasets covering a broad range of plant species and annotation types. Beyond unification, ReLEAP enables systematic recycling of existing data and provides a robust foundation for learning-based annotation upcycling. Leveraging this foundation, we conduct extensive cross-dataset and fine-tuning experiments to investigate how domain-specific models trained on unified data and foundation models can be used to automatically generate higher-level annotations under unsupervised, weakly supervised and semi-supervised settings, and how dataset characteristics influence their effectiveness. Our results show that unified data access and annotation upcycling substantially expand the pool of reusable training data and annotations while improving data transparency, enabling more sustainable and resource-efficient use of existing datasets. We release the framework and trained models for academic use to support accessible and collaborative research, particularly in data-scarce and heterogeneous domains.
[CV-67] Polyp-VSSD: A Pre-trained Transformer with Local-Global Joint VSSD Decoding for Polyp Segmentation
链接: https://arxiv.org/abs/2610.32282
作者: Xiaobo Zhu,Shiwei Wu,Yu An,Yuechen Tao,He Sun,Song Zhang,Zhenyu Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Early detection of colorectal polyps is essential for preventing colorectal cancer, making accurate polyp segmentation a key component in computer-aided diagnosis systems. However, existing endoscopic segmentation methods often exhibit limited representational capacity and insufficient integration of global and local contextual information. To address these limitations, we propose Polyp-VSSD, a novel local-global joint decoding framework built upon a pre-trained Transformer backbone for high-precision polyp segmentation. Polyp-VSSD leverages the strong feature expressiveness of large-scale visual pre-training and unifies global semantic modeling with local fine-grained reconstruction to enhance segmentation fidelity. Extensive experiments on multiple public endoscopic datasets demonstrate that our approach significantly outperforms state-of-the-art methods in both Dice and IoU. On the EndoScene dataset, Polyp-VSSD achieves mIoU and mDice scores of 0.903 and 0.946, respectively, highlighting its robustness and strong generalization capability. This work presents an effective and scalable solution for intelligent endoscopic image analysis.
[CV-68] RootDown: Deploying and Benchmarking Models for Beet-Root Detection on Edge Devices ACCV2026
链接: https://arxiv.org/abs/2610.32275
作者: Gerardus Croonen,Andreas Trondl,Julia Simon,Daniel Steininger
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at MLCSA Workshop at Asian Conference on Computer Vision (ACCV 2026). Benchmark, code and models available at this https URL
Abstract:Quality inspection of sugar beets after harvest and storage is critical to preserve sugar content and ensure optimal processing efficiency. The transfer of harvested beets to field clamps or trucks provides an opportunity to visually inspect individual beets automatically on loader conveyors. However, this is a challenging task as the large number of beets causes partial overlap and occlusion, complicating both annotation and detection. In this work, we propose an end-to-end image-based pipeline for detecting partially occluded sugar beets in loading scenarios as a first step toward automated quality inspection. We introduce a dedicated benchmark and present an automatic annotation strategy that leverages the SAM 3 foundation model to generate bounding boxes. We compare models trained on manually annotated and automatically generated labels. Additionally, we investigate deployment considerations by benchmarking multiple inference configurations, including execution on an edge device, under realistic operational constraints regarding precision, inference time and power consumption. Our results show that, despite substantial occlusion and inherent annotation ambiguity, foundation-model-based labels can substitute for a substantial share of manual annotation, trading labeling effort for data volume at a modest accuracy cost. Furthermore, we demonstrate that a Raspberry Pi 5 equipped with a Hailo accelerator enables high-performance, low-power inference, making it a practical solution for mobile outdoor deployment.
[CV-69] RVG-Bench: A Benchmark for Reasoning -Driven Audio-Visual Temporal Grounding
链接: https://arxiv.org/abs/2610.32268
作者: Yue Feng,Boshen Xu,Zihao Yue,Xiaokun Liu,Miao Yang,Xin Tao,Pengfei Wan,Qin Jin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 42 pages, including appendices. Project page: this https URL
Abstract:Existing temporal video grounding (TVG) benchmarks primarily focus on localizing target video segments according to descriptive text queries of the visual content. However, real-world queries are more diverse and often require the model to first understand the user’s intent before localization, such as finding visual evidence, identifying the cause of an event, or recalling a remembered moment, yet such queries remain largely overlooked. To mitigate this gap, we introduce RVG-Bench, a benchmark for reasoning-driven audio-visual temporal grounding, where every query is constructed such that the target segment can only be identified through reasoning about the query’s underlying intent. We carefully construct the benchmark and rigorously verify its data quality to ensure models truly comprehend the audio-visual content. RVG-Bench comprises 1,255 videos and 1,890 query-interval pairs across eight reasoning types. We evaluate 22 models and find that the strongest, Qwen3.8-Omni-Flash, achieves only 49.6 mIoU. Through a series of controlled analyses, including oracle-grounding probes and test-time scaling experiments, we systematically examine where and why current models struggle, providing insights for future model development.
[CV-70] Hyperparameter-Free Boundary-Interior t-vMF Dice Loss for Medical Image Segmentation
链接: https://arxiv.org/abs/2610.32266
作者: Hinako Mitsuoka,Kazuhiro Hotta
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Class imbalance and errors near anatomical boundaries remain challenges in medical image segmentation, motivating loss functions that account for these difficulties. Adaptive t-vMF Dice Loss adjusts the similarity function during training but requires hyperparameter tuning. We propose Self-Tuned Boundary-Aware t-vMF Dice Loss (STB-tvMF Dice), which adapts the function separately for boundary and interior regions without tuning shape-adaptation parameters. Each region’s shape parameter of t-vMF function is determined from its mean prediction-target cosine similarity in the preceding epoch and remains fixed throughout the current epoch. This rule maximizes the slope of the similarity function at the observed mean similarity, providing a principled basis for adaptation without validation-based parameter selection. Experiments on three datasets and architectures demonstrated its effectiveness. On Synapse with U-Net, STB-tvMF Dice achieved a mean DSC of 70.75%, compared with 63.03% for Adaptive t-vMF Dice Loss. Our code will be available soon.
[CV-71] LapaFM: A Laparoscopic Foundation Model via Multi-Stage Continual Learning
链接: https://arxiv.org/abs/2610.32263
作者: Xiaobo Zhu,Shiwei Wu,Yu An,Yuechen Tao,He Sun,Song Zhang,Jianbo Wang,Xiaojing Wang,Junwei Liu,Wei Guo,Yanfei Shao,Xuan Zhao,Jiangwei Xiao,Wenbo Niu,Bo Wei,Yuhan Wang,Gang Hu,Fei Liu,Kunshan He,Chongwei Chi,Jianqiang Tang,Kang Zhang,Jie Tian,Zhenyu Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Artificial intelligence-based analysis of intraoperative laparoscopic videos has the potential to improve surgical decision-making and precision. However, effective surgical assistance requires models that can generalize across diverse intraoperative tasks. Here we present LapaFM, a foundation model for laparoscopic surgery trained on more than 20,000 minutes of surgical videos using a multi-stage continual learning strategy that mitigates catastrophic forgetting and inter-task interference. LapaFM learns multimodal visual representations and supports multiple tasks, including smoke removal, surgical phase recognition, instrument and polyp segmentation, and lesion identification. The model was evaluated on 13 retrospective and two prospective cohorts collected from different imaging devices. LapaFM consistently outperformed task-specific models and existing foundation models across datasets, and also generalized to gastrointestinal endoscopy tasks, including polyp detection and classification. Notably, it recovered over 90% of smoke-obscured frames and achieved clinician-level or superior performance in lymph node metastasis prediction and prospective polyp classification.
[CV-72] Beyond Pattern Matching in 3D MLLM s: Part-Level Functional Reasoning for Generalizable Recognition
链接: https://arxiv.org/abs/2610.32262
作者: Weihong Li,Hao Jiang,Yaohua Zha,Tao Dai,Shu-Tao Xia
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:3D MLLMs recognize objects via pattern matching, which fails when geometrically similar objects require functional reasoning to disambiguate (e.g., dresser vs. wardrobe). Our per-sample Geometric Confusion Score (GCS) exposes this failure: baseline accuracy is 33.7% on ModelNet40’s most ambiguous 20% versus 84.0% on its least ambiguous 20%. We identify a function-as-annotation pattern, in which functional language decorates a known identity instead of supporting inference, and hypothesize that reversing this information flow should concentrate improvements on the most ambiguous samples. PFIR (Part-level Functional Inference for Recognition) reverses that flow. Built on PartNeXt part segmentation, its data pipeline produces 507K QA pairs with functional CoTs (geometry \to function \to identity), and its progressive curriculum culminates in dual-objective GRPO: strict functional reasoning and descriptive grounding for Objaverse’s diverse distribution. The two RL objectives yield nearly orthogonal task vectors ( \cos = 0.056 ), enabling composition via Task Arithmetic. Under a strict dual OOD protocol (Objaverse \to ModelNet40; XYZRGB \to XYZ-only), PFIR raises zero-shot accuracy from 60.5% to 66.7%, with gains concentrated 2.6\times more on geometrically ambiguous samples—supporting part-level functional reasoning as an effective route to more generalizable 3D recognition. Code: this https URL.
[CV-73] Representation Learning via Composite Subspace Flow Matching
链接: https://arxiv.org/abs/2610.32256
作者: José Afonso,Mário Figueiredo
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 27 pages, 4 figures
Abstract:Distilling semantic meaning from raw image data remains a challenge in machine learning. Current self-supervised paradigms face a trade-off: joint-embedding architectures produce relevant features by aggressively discarding intra-class variance, irreversibly spatial and reconstruction. Conversely, generative models preserve spatial details but struggle to isolate semantics, as their optimization is dominated by pixel variations. To bridge this divide, we propose composite subspace flow matching (CSFM). Rather than corrupting data with unstructured isotropic noise, CSFM formulates representation learning as the variance-minimizing inversion of a deterministic degradation curriculum across spectral, color, and illumination subspaces. By bounding trajectory variance with hard temporal style truncation, a capacity-constrained encoder is driven to ignore transient style artifacts and encode persistent content. Empirical evaluations across CIFAR-10, CIFAR-100, and STL-10 show that CSFM yields competitive discriminative transferability while preserving generative synthesis across both convolutional and vision transformer architectures. By avoiding the collapse pathologies of non-generative methods, CSFM forms a smooth latent space that outperforms standard optimal transport flows on linear probing and few-shot classification, while enabling coherent synthesis and trajectory interpolation. These results confirm that steering the physical dynamics of generative flows offers a scalable paradigm for learning representations that are simultaneously linearly separable and generatively invertible.
[CV-74] Easy-3Dfy: Object-Level Geometric Composition for Single-Image 3D Scene Construction
链接: https://arxiv.org/abs/2610.32250
作者: Ziang Zhang,Wenyou Yi,Zehan Wang,Jiachen Xu,Xintong Han,Chunchao Guo,Zhou Zhao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Single-image 3D scene construction is a fundamental problem in 3D vision. Recent methods such as SAM-3Dfy enable end-to-end reconstruction of 3D assets and their scene layouts, but still struggle with fine-grained scene geometry and asset quality. Meanwhile, specialized 3D object generators provide detailed object geometry, while 3D foundation models offer strong and generalizable spatial understanding. We propose Easy-3Dfy, a compositional workflow that decouples object generation from scene layout estimation, allowing specialized models to fully exploit their respective strengths. An RPF-based aligner then bridges these capabilities by establishing geometric correspondence between object-local meshes and partial scene pointmaps, recovering object layout for scene composition. Without requiring complete scene-level supervision, the proposed workflow consistently outperforms SAM-3Dfy across multiple geometric metrics on Scene-Maker and 3D-FUTURE200. Our results demonstrate that explicit geometric composition can effectively combine high-quality object generation with strong spatial understanding, while enabling the workflow to benefit independently from advances in both 3D object generation and foundation models.
[CV-75] Share: Training-Free Acceleration of Overlapping Tiled ViT Decoding
链接: https://arxiv.org/abs/2610.32233
作者: Xiangzhong Ye,Zhongshen Li,Shen You,Ziyang Meng,Liang Xu,Junkai Ji,Qiuzhen Lin,Xiangtao Li,Ka-Chun Wong
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 8 pages main paper + supplementary material
Abstract:Fixed-window Vision Transformer (ViT) decoders run at deployment resolution as overlapping tiles: every token in an overlap band is processed by several full decoder instances. In the 2.4B-parameter MiniMax-H3 video decoder, about half of all per-token computation is duplicated. Removing the tiles fails because attention, global tokens and coordinates are bound to them. TileShare keeps each released tile’s attention and computes every other operation once per token. The released computation changes at one kind of step only: an overlap token’s per-tile results are blended into one shared state with the released weights. Interventions show that this blend costs quality where attention reads the shared states, mainly in the first few layers, which TileShare therefore keeps per tile. Without training, keeping six layers per tile decodes 1.38-1.51x faster than batched and compiled tiling on two GPU types (up to 1.96x with FP8) and is equivalent to the released decoder in PSNR, LPIPS and DISTS on DAVIS videos. On unseen UVG videos with the setting frozen beforehand, it matches PSNR and DISTS and improves LPIPS. TileShare carries over to two further decoders and is again confirmed on unseen data. It is equivalent in PSNR and DISTS on MAGI-1. On the ViTok-v2 image autoencoder at 2K-4K, it matches overlapping tiling in PSNR, improves LPIPS and DISTS and decodes 1.5-1.8x faster. How many layers stay per tile is measured once per model (for decoders, without labels); we report it for 14 decoders and encoders. Optionally, two GPU-hours of self-distillation let MiniMax-H3 share every layer without a loss in PSNR, and a coordinate field trained while the network stays frozen brings fully shared CLIP-based open-vocabulary segmentation to parity with sliding windows on seven benchmarks.
[CV-76] owards Effective and Transferable Physical Camouflage against Multi-View BEV-based 3D Perception in Autonomous Driving
链接: https://arxiv.org/abs/2610.32230
作者: Linye Lyu,Jiawei Zhou,Daojing He,Yu Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Modern autonomous driving systems widely adopt multi-view Bird’s-Eye-View (BEV) based 3D perception models due to their superior performance. Despite their success, the robustness of these models against adversarial camouflage attacks remains largely unexplored. Existing camouflage attacks primarily target single-view 2D detectors, limiting their effectiveness in multi-view 3D perception settings. To address this gap, we propose BEVCA, a novel adversarial camouflage generation framework tailored for multi-view BEV-based 3D perception. Our framework integrates a BEV-feature-based adversarial loss with a multi-view neural rendering module, enabling effective and transferable camouflage attacks across different models and tasks. Extensive experiments in both digital and physical settings demonstrate that BEVCA outperforms state-of-the-art baselines and exhibits strong robustness under real-world conditions. Our code is available at this https URL.
[CV-77] ACWM-RM: A Reference-Free Reward Model from Complementary Mismatch Preferences
链接: https://arxiv.org/abs/2610.32222
作者: Shibo Hong,Siyu Ren,FengJiao Chen,Zhongyuan Peng,Wei Shi,Yixin Cao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reinforcement learning has gained attention for improving action-conditioned world models (ACWMs), as it enables reward-driven optimization of visual quality (VQ) and action-following fidelity (AF) beyond flow-matching optimization. Realizing this potential requires reward signals that reliably rank videos by both VQ and AF. However, existing rewards capture only one dimension: metric-based rewards assess VQ but require ground-truth videos and do not directly assess AF, while inverse dynamics model (IDM) rewards assess AF without references but do not guarantee VQ. Moreover, IDMs, typically trained on real videos, may provide unreliable rewards for generated videos containing visual artifacts and physical inconsistencies. In this work, we introduce the Action and Video Mismatch Benchmark (AVM-Eval), comprising 67.2k pairs that evaluate how reliably IDMs assess AF on real and generated videos. We then propose ACWM-RM, a reward model that jointly assesses VQ and AF without ground-truth videos during reward computation. To train it, we first construct the Action and Video Mismatch Preference Dataset (AVM-Pref) with 506.3k pairs providing complementary supervision: action-mismatch pairs supervise AF, while video-mismatch pairs supervise VQ and expose the model to generated videos. We then pretrain an action-conditioned predictor on real videos to learn forward dynamics before jointly training it with a patch-weighted score head on AVM-Pref to transfer these dynamics to AF assessment on generated videos. Evaluation on AVM-Eval reveals the limited reliability of IDMs on generated videos, while downstream RL experiments show that ACWM-RM outperforms reward baselines in both criteria.
[CV-78] Angular Supervision for Visual Object Grounding in Autoregressive Vision-Language Models
链接: https://arxiv.org/abs/2610.32220
作者: Parsa Esmaeilkhani,Longin Jan Latecki
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Autoregressive vision-language models (VLMs) struggle to ground generated text in the correct image regions, and existing explainability methods, such as attention maps and Logit Lens, often produce poor spatial explanations due to weak text-to-vision attention and sink tokens. We propose \textbfAngular Relevance ( R_L ), a simple visualization measure based on cosine similarity between visual token embeddings and unembedding-matrix rows. Because R_L is invariant to token norm, it produces substantially sharper and more faithful heatmaps than attention maps or Logit Lens on out-of-the-box models. Without any training, R_L yields competitive referring-expression segmentation (RES), outperforming prior diagnostic maps and CLIP-based training-free methods. To preserve localized object information during cross-modal mixing, we introduce \textbfAngular Object-Aware Loss (AOL), an auxiliary loss that directly supervises visual token embeddings to align angularly with the vocabulary directions of their corresponding object concepts. Finetuning with AOL alongside the standard next-token prediction objective significantly sharpens object localization in heatmaps, showing that the VLM’s own visual tokens can be used directly for grounding and improving RES, without training any segmentation heads and without degrading general VQA performance. We further demonstrate that the learned grounding also transfers to category-level segmentation of object categories not seen during training.
[CV-79] Benchmarking Data-Free Continual Model Merging for Whole-Slide Image Survival Analysis
链接: https://arxiv.org/abs/2610.32207
作者: Vu Minh Tran,Doanh C. Bui,Maï K. Nguyen,Khang Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Continual survival analysis on whole-slide images (WSIs) requires models to integrate knowledge from sequentially arriving cancer cohorts while preserving previously learned information. Model merging offers a data-free alternative by combining independently fine-tuned task experts without retaining previous training data. However, the relative behavior of different merging strategies for WSI-based survival prediction remains insufficiently understood. In this work, we establish an end-to-end benchmark for data-free continual model merging, covering task-specific fine-tuning, sequential merging, and inference under both TASK-IL and CLASS-IL settings. We systematically evaluate six representative merging strategies across four TCGA cancer cohorts and three task orders using C-index, IPCW C-index, Forgetting, and Backward Transfer under a controlled protocol. Results show that no merging strategy consistently dominates across all evaluation criteria, while censoring-aware evaluation can substantially alter the relative ranking of the compared methods. This benchmark provides a unified foundation for developing and evaluating more effective continual model merging methods for WSI-based survival analysis.
[CV-80] Neural Disjunctive Normal Forms for Learning Exact Rank Filters
链接: https://arxiv.org/abs/2610.32203
作者: Diego Marcondes,Wonder A. L. Alves,Ronaldo F. Hashimoto
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 12 pages, 4 figures, 2 tables
Abstract:Gradient-based learning of morphological operators is hindered by locally sparse derivatives with respect to the structuring elements. We develop a method for learning rank filters that retains exact operator evaluation while distributing the derivatives across the window. Using threshold decomposition, we construct a neural disjunctive normal form representation and establish a sufficient condition for exactness with integer-valued images and structuring elements. We then derive a smooth surrogate from the unique root of a smooth threshold equation. Implicit differentiation gives closed-form derivatives with normalised local weights, whose distribution is controlled by a smoothing parameter. These derivatives are used in a straight-through procedure: the forward pass evaluates the exact rank filter, while the backward pass uses the surrogate. For fixed ranks, the procedure learns real-valued structuring elements of any rank filter, including the median, and extends to compositions, infima, and suprema of rank filters. Experiments on image denoising using MNIST and Fashion-MNIST and retinal vessel segmentation show that, starting from a generic random initialisation, the proposed approach consistently improves performance over training with exact subgradients in every setting. In image denoising, our best network also outperforms smooth morphological networks in all settings. On vessel segmentation, the learned erosions recover structuring elements that fit the vessels, as theoretically expected, whereas most of those trained with exact subgradients remain close to their initialisation.
[CV-81] DS-BiCA-Net: Pair-Preserving Cross-Temporal Modeling for Remote Sensing Change Captioning
链接: https://arxiv.org/abs/2610.32200
作者: Licheng Zhang,Naveed Akhtar,Siew-Kei Lam
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Remote sensing image change captioning (RSICC) describes in natural language the semantic changes between two images of the same scene captured at different times. Previous methods commonly merge the pair during repeated processing or combine intra-image and cross-image relations in one operation, weakening source identity and leaving the two temporal directions implicit. Without separate modules for intra-image and cross-temporal processing and a separate path for each direction, the contribution of each part cannot be measured on its own, so it remains unclear which part matters most, and the design of later models has no such evidence to build on. To address these issues, we propose DS-BiCA-Net, which assigns the two kinds of processing to separate modules and keeps the two images as separate streams throughout the encoder. A windowed state-space module models context within each image, and a bidirectional cross-attention module compares the two images along two directed paths that share their projections. The two modules use the same pre-normalisation residual architecture and differ only in their mixers, and the two streams are fused only in the aggregation head, so each module and each direction can be removed and its contribution measured on its own. Controlled experiments on LEVIR-CC and Dubai-CC examine operator interaction, temporal direction and module order, showing that intra-image refinement carries no benefit alone and becomes beneficial only when followed by bidirectional cross-attention, an order that outperforms the reverse at matched parameter count and FLOPs. The bidirectional module also outperforms either temporal direction alone. Our method reaches an S_m^* of 82.65 on LEVIR-CC, the highest among the published results compared, and 66.72 on Dubai-CC without geometric augmentation, the second-highest among the published results compared.
[CV-82] Searching with Sparse Histories: Cross-Task Architecture Search from Incomplete Search Logs and Zero-Cost Proxies
链接: https://arxiv.org/abs/2610.32179
作者: Yuefeng Xu,Jun Yu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Neural and Evolutionary Computing (cs.NE)
备注:
Abstract:Neural architecture search (NAS) is expensive, but cross-task transfer can reduce costs by reusing knowledge from previous tasks. However, existing methods struggle with sparse search histories and task-specific transfer failures. We propose a method, a risk-robust strategy that combines sparse source logs with zero-cost proxies through champion-based initialization and joint pairwise-ranker pretraining. On TransNAS-Bench-101, the proposal mitigates severe worst-case failures, reducing worst-10% ranks from 290 to 43 for Micro autoencoding and from 465 to 45 for Macro room layout. Across 13 cell-space targets, it outperforms a proxy-only ranker on 10 targets in both mean and worst-case performance, while maintaining competitive average performance.
[CV-83] Ψ-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model
链接: https://arxiv.org/abs/2610.32177
作者: Khai Loong Aw,Klemen Kotar,Baihan Zhang,Yifan Liu,Lilian Chen,Atlas Kazemian,Rahul Venkatesh,Daniel Yamins
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Humans reason about the physical world by flexibly combining different forms of inference as a problem unfolds. Achieving this flexibility is a major open challenge in AI: capabilities often remain separated across specialized models, limiting how they can be composed and jointly adapted to a task. We introduce \Psi -0.6, a unified vision-language world model that adaptively composes perception, language, geometry, and simulation within a shared reasoning process we call streams of thought. These streams interleave images, text, and explicit representations of geometry and motion, all of which the model can generate and interpret through a shared token interface. \Psi -0.6 surpasses leading open-weight models zero-shot on multi-view spatial reasoning, with further gains from reasoning through geometric intermediates. It also supports counterfactual simulation: generating and interpreting imagined scenes to determine, for example, whether moving an object would obstruct a pathway. Reinforcement learning from task feedback jointly optimizes the selection, generation, and use of intermediate representations, improving visual question-answering accuracy with gains that transfer to unseen benchmarks. \Psi -0.6 offers a path toward adaptive physical reasoning – not only predicting the physical world but also choosing the representations through which to reason.
[CV-84] Real-Time Localisation of Dorsal Anatomical Regions on Dairy Cows in Commercial Rotary Parlours from RGB-D
链接: https://arxiv.org/abs/2610.32174
作者: Vineet Prasad,Shiron Melvin,Sheikh Azid,Hai Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 7 figures, 3 tables. Submitted to ACRA 2026
Abstract:Automated monitoring of dairy cows increasingly measures one part of the body, such as the rump for temperature or the back for posture, yet the region itself is usually placed by hand. Existing systems either score the cow as a whole or detect sparse keypoints, often in controlled walkways, and neither yields the named region masks that such measurements need on a crowded commercial rotary platform. We present a perception front end that finds the target cow, ten bony landmarks and four named dorsal regions (withers, back, rump and tail-head) in a single forward pass over a whole RGB-D frame from a commercial rotary parlour. Two encoders, one for colour and one for depth, are fused late and share one decoder with three heads. On an anatomical region dataset that we built on public rotary-parlour RGB-D images (538 annotated frames, cow-disjoint test set), the network reaches PCK@10 0.905 for landmarks and region IoU 0.70 to 0.91, and runs end to end in 26.4 ms per frame on a T4 GPU, against an 11 fps camera. Dense region supervision is what makes the landmarks accurate: removing the region head costs 0.15 PCK@10. Agreement between the two heads flags unreliable frames without labels. On our own rig at the largest dairy farm in Western Australia, the model transfers with ten labelled frames (PCK@10 0.84), and its regions are carried through depth into a radiometric thermal camera, giving one repeatable temperature per region and cow.
[CV-85] Separating Angular Alignment from Radial Ordering in Hyperbolic Vision-Language Models
链接: https://arxiv.org/abs/2610.32146
作者: Mingyang Gao,Yuwei Wu,Shiyu Luo,Wei Wu,Xiaomeng Fan,Qihao Lv,Haoyu Zhang,Rui Gao,Bingzhen Liu,Zhi Gao,Che Sun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 34 pages, 12 figures
Abstract:Hyperbolic vision-language models are well suited to modeling the hierarchical structure of multimodal data, because hyperbolic volume grows exponentially with radius, enabling low-distortion representations of tree-like taxonomies from general to specific concepts. However, existing pretraining objectives for hyperbolic vision-language models entangle semantic alignment with entailment modeling, forcing radial coordinates to simultaneously encode semantic matching and hierarchical structure, leading to conflicting geometric constraints. In this paper, we introduce AURORA, a hyperbolic vision-language framework that decouples the objectives through angular alignment and a novel Radial-Angular Order Cone (RAOC) for entailment modeling. Angular alignment captures semantic similarity while preserving the radial structure for representing hierarchies. Unlike conventional entailment cones, RAOC determines angular allowance from relative parent-child radial gaps. It is incorporated into hierarchical entailment contrastive training to capture partial-order relations. We further introduce a norm regularization term that prevents undesirable radial degeneration for stable training. We train models on millions of image-text pairs and evaluate them across zero-shot classification, image-text retrieval, hierarchical reasoning, and compositional understanding. Our model consistently outperforms strong Euclidean and hyperbolic vision-language baselines, demonstrating the effectiveness of decoupling semantic alignment from hierarchical entailment modeling.
[CV-86] ImageRSI: Verification-Driven Recursive Self-Improvement for Agent ic Image Generation
链接: https://arxiv.org/abs/2610.32120
作者: Shuo Han,Zhilin Wang,Peng Ling,Wenjia Gong,Yang Liu,Guanjun Jiang,Yukun Cao,Xike Xie
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Complex text-to-image requests remain difficult to satisfy reliably. Verifying and repairing an image can fix an individual output, but a successful repair does not necessarily carry over to other requests. We present ImageRSI, a framework that keeps the image generator frozen and combines three components: five-dimensional verification and repair within each task, an experience library shared across tasks, and offline updating of the judge. Candidate experiences enter the library only after paired A/B validation, and the judge is updated with supervised fine-tuning and GRPO on accumulated image-judgment records and evaluated before reuse. We also introduce ImageReq-Bench, a benchmark of 961 requests built around hard cases in real-world image-generation requests, each paired with an evaluation checklist. ImageRSI achieves a Checklist Strict Accuracy (CSA) of 52.6% on MindBench (with Nano Banana Pro as the generator) and 65.66% on ImageReq-Bench, and we compare system outcomes across judge-training configurations.
[CV-87] in VLA Models: Exploring Training with Human Demonstrations
链接: https://arxiv.org/abs/2610.32094
作者: Qiangong Zhou,Jiafeng Liang,Yimin Dong,Yang He,Hongxiang Li
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 6 figures, a techreport form standard-robots company
Abstract:Long-horizon robot tasks require retaining and using past information. We integrate Test-Time Training (TTT), which stores context in fast weights, into InternVLA-A1 and Pi0.5 to study action generation conditioned on human demonstrations through training on paired demonstrations and robot trajectories. We implement memory transfer between the two phases and truncated backpropagation, and examine training objectives, gradient coverage, and contrastive regularization across video, short-video, and single-frame cue tasks. In our experiments, adding TTT and training on paired data at the current scale did not lead to consistent target selection based on demonstrations. Under the tested settings, shorter sequences, full trajectory gradient connectivity, and contrastive regularization did not clearly improve responsiveness to demonstrations. When demonstrations were reduced to single-frame cues, real-robot execution still showed a preference for cups at fixed positions. We document the implementation, training configurations, and observations to support further inspection and replication.
[CV-88] RADIANT: Object-Centric Feed-Forward Global Illumination
链接: https://arxiv.org/abs/2610.32075
作者: Sarah Zhang,Zhiyang Dou,Minghao Guo,Benjamin Tod Jones,Wojciech Matusik
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Machine Learning (cs.LG)
备注:
Abstract:Feed-forward neural renderers synthesize global illumination across novel 3D scenes without per-scene optimization or runtime ray tracing. Existing transformer-based renderers tokenize individual triangles, so self-attention scales quadratically with mesh resolution. Yet macroscopic radiative transport is governed by the underlying surface geometry and object arrangements rather than a particular mesh discretization; finer tessellation can thus inflate token counts without changing the underlying transport physics, motivating a discretization-agnostic and object-centric representation. We introduce RADIANT (RADiance-based Inter-Actor Neural Transformer), an object-centric architecture for feed-forward global illumination from point inputs. By representing each object with a compact budget of 32 surface-patch tokens, RADIANT decouples attention sequence length from polygon counts. In a single network, an inter-object scene transformer models 3D radiative exchange and a view decoder renders images at target camera poses. To capture light transport without materializing O(N^2) bias matrices, invariant form-factor attention embeds geometric distances and surface orientations directly into query-key dot products as SE(3)-invariant inner products weighted by learned area and material gains. For primary visibility without mesh connectivity, camera rays cross-attend scene tokens guided by depth-tested point splats in a shared spherical frame. Across a 1.03M-scene benchmark under matched model capacity, RADIANT reaches 25 dB tone-mapped PSNR in 5x fewer training steps than RenderFormer, and achieves 32.2 dB at 600k steps (+3.3 dB over RenderFormer and +7.6 dB over RenderFormer-V2). We further demonstrate the scalability of RADIANT, which completes a 512x512 forward pass of a 500-object scene in 0.83 s, with computational cost scaling with the number of objects rather than polygons.
[CV-89] Guided Zero-Shot Anomaly Explanation for Anonymized Crowd Video Monitoring: Context Categorization and Spatial Localization
链接: https://arxiv.org/abs/2610.32073
作者: Mulugeta Weldezgina Asres,Lei Jiao,Christian Walter Omlin
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 21 pages, 12 figures, 13 tables, preprint
Abstract:Despite the promising development of deep learning for video anomaly detection (VAD), explainability (XAI) of these models remains a challenge. Explainable and privacy-protecting VAD builds transparency and trust that facilitates adoption of computer vision systems, including safety and security monitoring. However, anomaly explainability in privacy-preserving VAD systems is underexplored. In this study, we present the XAVAD approach, a unified descriptive anomaly XAI for an anonymized VAD, that integrates flexible zero-shot anonymization, anomaly type categorization, visual localization, and scene context description. We provide empirical and qualitative assessments of multifaceted explanations integrated with full-body visual anonymization for privacy protection. We adopt a multimodal CLIP embedding for videos, using improved LLM-assisted prompting and pool-scoring mechanisms for anomaly-type identification. We enhance Grad-ECLIP for spatial anomaly heatmap localization, leveraging computation and accuracy through our novel prompt-guided patching autofocus mechanism. We employ VLMs to generate context and reasoning descriptions of detected anomalies through curated prompt engineering. We have evaluated the accuracy and computational cost of XAVAD on several publicly available CCTV and video datasets. Our experiments demonstrate substantial gains of our XAVAD over state-of-the-art benchmark approaches and provide insights into privacy protection trade-offs in XAI for VAD. We further present ablation studies and assessments across diverse configurations and modeling alternatives to demonstrate the efficacy of our proposed methods. Code: this https URLthis https URL .
[CV-90] Concept Acupuncturist: An Efficient and Precise Concept Erasure Method for Diffusion Models via Sparse Autoencoders
链接: https://arxiv.org/abs/2610.32059
作者: Yiran Xiahou,Siliang Qin,Kai Chen,Shengzhi Zhang,Renyang Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注:
Abstract:Diffusion-based image generation models produce high-quality images but may generate harmful or policy-violating content, motivating machine unlearning (MU) to selectively remove target concepts while preserving general capabilities. However, existing MU approaches often struggle with the distributed and entangled nature of concept representations, leading to inefficient optimization and imprecise edits when applied through large-scale fine-tuning over full diffusion trajectories. To bridge this gap, we propose Concept Acupuncturist (ConceptACP), a novel MU framework that integrates interpretability-guided analysis with inference-free fine-tuning for efficient and precise concept removal. By establishing mappings between semantic concepts and cross-attention blocks in the U-Net denoiser and localizing the most concept-relevant block for targeted optimization, ConceptACP avoids full diffusion trajectory optimization, thereby improving fine-tuning efficiency. To mitigate concept entanglement, we propose a feature-level correction mechanism based on Sparse Autoencoders (SAEs), where SAE-derived sparse, semantically meaningful feature activations are manipulated to construct targeted correction signals, which are then projected back to the output space of the corresponding block and used as supervision for precise concept removal. Consequently, ConceptACP eliminates the need for external inference-time modules while achieving efficient and fine-grained unlearning. Instantiated on SDXL-Turbo, ConceptACP improves the forgetting success rate by 36.34 percentage points in the standard one-step generation setting on average over the best-performing baseline, while preserving image quality and prompt fidelity, and generalizes effectively across other Stable Diffusion backbones. Our code is available at this https URL.
[CV-91] ACVA-Bench: Beyond Prompt Following in Abstract Concept Visualization Ability for Text-to-Image Models
链接: https://arxiv.org/abs/2610.32058
作者: Mengyan Wang,Jiyuan An,Yining Liu,YaoWen Zhang,Liner Yang ang Erhong Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Text-to-image (T2I) models have achieved remarkable progress in image quality and instruction following, prompting growing research interest in their capacity for deep understanding that goes beyond literal prompt matching. However, without an explicit visual scheme provided, it remains unclear and has not been systematically evaluated whether T2I models can autonomously translate abstract concepts lacking a fixed visual referent, such as “loyalty,” into visual representations that are accurate, diverse, and distinguishable from closely related concepts. We define this capability as Abstract Concept Visualization Ability (ACVA) and introduce ACVA-Bench, a category-balanced benchmark grounded in psycholinguistic norms of abstract concepts, which examines models’ ACVA along three dimensions semantic sufficiency, representational diversity, and conceptual distinctiveness under a low-intervention setting that withholds concrete objects, scenes, or visual metaphors. Since abstract concepts lack a unique reference image, conventional evaluation metrics that rely on reference-image matching are difficult to apply. To address this, we propose a fine-grained visual-semantic evaluation method constrained by concept semantics, which decomposes concept semantics into independently verifiable visual-semantic questions to assess the semantic representation quality of generated images, with its validity further confirmed by human evaluation. Benchmarking results on mainstream T2I models show that these models rely on a small set of prototypical visual patterns, leaving room for improvement in representational diversity and conceptual distinctiveness for abstract concepts, thereby providing an interpretable and reproducible evaluation foundation for future research.
[CV-92] NUViT: Vision Transformers for Fourier Space Data
链接: https://arxiv.org/abs/2610.32041
作者: Jeffrey Gu,Robert C. Heeter,Ellen D. Zhong
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Vision Transformers (ViTs) have demonstrated strong performance in many computer vision tasks but do not naturally extend beyond the natural image domain. One important example is Fourier-space data, which is used to represent a rich and diverse set of modalities, including medical imaging and microscopy. Standard ViT tokenization partitions an image into square patches on a Cartesian grid, which poorly fit the frequency and phase structure of Fourier-space data. We introduce Non-Uniform ViT, a simple, general-purpose vision transformer that redesigns image tokenization, data representation, and positional encoding to respect this structure. Experiments on Fourier-domain image classification, 2D crystal lattice identification, MRI annotation, and cryo-EM protein heterogeneity analysis show that NUViT consistently captures the structure of Fourier data better than standard architectures.
[CV-93] Rethinking Representation Geometry in Supervised Contrastive Learning under Long-Tailed Data
链接: https://arxiv.org/abs/2610.32039
作者: Joonhyeok Jeong,Jy-yong Sohn,Kibok Lee
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
备注:
Abstract:Supervised contrastive learning and its balanced variants have demonstrated their effectiveness for representation learning under long-tailed distributions. However, although prior analyses suggest that their optimal representations may exhibit a regular simplex geometry, whether such a geometry is indeed exhibited remains insufficiently understood. In this work, we present a theoretical framework for these objectives that characterizes whether a regular simplex is formed at the optimum. By applying this to representative objectives, we show that these objectives fail to form a regular simplex at the optimum because they remain dependent on the number of samples per class. Motivated by this analysis, we introduce structural balancing that can be applied to contrastive objectives, and propose a fully balanced contrastive objective that provably forms a regular simplex geometry at its optimum. Both theoretical and empirical results demonstrate that the proposed method yields a more balanced representation space and improves performance on long-tailed recognition and transfer learning tasks.
[CV-94] MAGIC: Margin-gated Contrastive Debiasing with Evidence Restoration for VLM Test-time Robustness Scaling
链接: https://arxiv.org/abs/2610.32013
作者: Shunshun Gu,Enze Zhang,Ming Pei,Hang Zhou,Xiao Luo
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:This paper studies the problem of test-time robustness scaling for large vision-language models (LVLMs), which aims to mitigate language bias during inference. Existing approaches typically adopt test-time scaling techniques such as contrastive decoding to improve inference performance. Despite the progress, the performance of these approaches is far from satisfactory due to intentional evidence removal during perturbations and redundant inference cost. Towards this end, we propose a novel approach named Margin-gated Contrastive Debiasing with Evidence Restoration (MAGIC). The core of our MAGIC is to evaluate the decision margins of LVLMs to selectively activate textual and visual perturbation branches for efficient debiased inference. In particular, our MAGIC first measures the prediction margins of crucial tokens, and applies a gating function to identify low-confidence responses that are susceptible to language biases. For selected samples, we integrate the token logits from multiple textual and image perturbation branches with the original ones for debiased decision. More importantly, we selectively restore the most informative corrupted views with sharp prediction distributions for ambiguous samples after debiasing, enabling evidence recovery for discriminative decision-making. We further utilize a calibration dataset to actively identify and prune ineffective branches during contrastive debiasing for cost reduction. Extensive experiments on benchmark datasets validate the effectiveness of the proposed MAGIC in comparison with extensive baselines.
[CV-95] OpenHome: From Perception to Insight in Smart-Home Camera Videos
链接: https://arxiv.org/abs/2610.31986
作者: Xianglong Wang,Neil Rodrigues,Pei Guo,Yiru Fang,Kevin Beussman,Shuyang Wang,Jingzhi Wang,Liangju Li,Shunyao Xu,Zhongwei Cheng,Wei Li,Lin Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision language models (VLMs) have been increasingly adopted in consumer smart-home camera systems to interpret recorded videos and provide insights. However, video clips captured from smart-home cameras are different from curated web videos dominating existing VLM benchmarks. Smart-home video clips often do not contain subtitles or narration and are captured from different viewing angles, under various lighting conditions, and in diverse scenarios. However, the capabilities of the model in this domain, under sparse visual signals and limited context, are not fully assessed by existing VLM benchmarks. To close this gap, we introduce OpenHome, a visual-only, four-option multiple-choice benchmark for evaluating smart-home video understanding. It contains 2,254 questions on 1,118 real-world video clips and organizes 10 tasks into three capability levels. Perception evaluates visible content, Reasoning evaluates causal, spatial, and temporal relationships, and Insight evaluates subject intent, supporting evidence and appropriate homeowner responses. Based on the dataset, we describe the task design, data pipeline and evaluation protocol, and report results for 32 closed-source and open-weight VLMs. The best model, Gemini 4 Argon, reaches 81.2% overall accuracy. Frontier closed-source models outperform open-weight models on every level. The best closed-source model leads the best open-weight model by 16.3% on Perception, 22.4% on Reasoning and 37.4% on Insight.
[CV-96] Interpreting Physics in Video Encoders using Multi-Object Interaction Videos
链接: https://arxiv.org/abs/2610.31983
作者: Ayushi Mehrotra,Sai Vidyaranya Nuthalapati
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video encoder models are increasingly used as world models for physical AI, where a policy plans by rolling out the latent, so what physics the latent holds bounds what the agent can plan over. Moreover, existing interpretability work probes quantities belonging to a single object; the quantities that make physics relational, such as momentum transfer, restitution, and friction, are defined between bodies and have not been examined. We probe relational physics attributes in two frozen self-supervised video encoders (V-JEPA2 and VideoMAE) under two-body settings and test if they transfer across environments. Our early findings are mixed. Both encoders separate elastic from inelastic collisions reliably. V-JEPA2’s own predictor is less likely to predict a collision that creates kinetic energy compared to one that loses it. Momentum-violation detection appears in only one of the two encoders, a kinetic-energy readout survives a change of environment for V-JEPA2 but not for VideoMAE, and neither model holds a friction coefficient that survives a change of geometry.
[CV-97] Spectra: Multi-Scale Implicit Decomposition for Deepfake Detection
链接: https://arxiv.org/abs/2610.31979
作者: Dhananjaya Jayasundara,Kartik Narayan,Vishal M. Patel
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Deepfakes increasingly undermine trust in digital media, yet existing detectors often struggle to generalize across unseen manipulation techniques, compression levels, and dataset shifts. Most approaches rely on RGB appearance, frequency-domain artifacts, or their combination, learning cues present in the final synthesized image. While effective in-distribution, such cues are frequently generator-specific and can diminish as synthesis models evolve. We argue that a complementary perspective can be obtained through implicit neural representations (INRs), which expose how an image is reconstructed rather than merely how it appears. By fitting an INR to an image and analyzing its internal activations, one can reveal reconstruction dynamics that are difficult to observe in the image domain alone and may capture more intrinsic differences between genuine and synthetic content. To this end, we propose Spectra, a deepfake detection framework that combines semantic visual representations with artifact-aware cues derived from INR reconstruction. Spectra extracts multi-level reconstruction heatmaps from image-specific INRs and jointly models them with semantic features through a bidirectional fusion mechanism. This enables direct interaction between image semantics and reconstruction artifacts, allowing the model to identify forgery cues that are less dependent on specific generation pipelines. Extensive experiments on multiple deepfake benchmarks demonstrate improved robustness and cross-dataset generalization, consistently outperforming detectors based solely on RGB appearance or semantic representations. Furthermore, we show that the proposed representation extends beyond deepfake detection to face presentation attack detection, highlighting its potential as a unified framework for facial attack detection.
[CV-98] STEP: Surgical Point Tracking from Sparse Endpoints and Pseudo-trajectories
链接: https://arxiv.org/abs/2610.31951
作者: Bing Fan,Anwesa Choudhuri,Zhongpai Gao,Meng Zheng,Benjamin Planche,Terrence Chen,Heng Fan,Ziyan Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Surgical point tracking requires persistent correspondence through tissue deformation and occlusion, yet dense trajectory annotations are costly to obtain. We present STEP, a two-stage framework that learns from surgical video and sparse physical endpoints. First, a frozen offline teacher generates candidate trajectories, and temporal reversibility and stereo consistency select dense supervision for a causal student. Second, a robust physical endpoint loss anchors terminal correspondence, while a round-trip loss constrains temporal consistency. A gradient-free prefix exposes training to accumulated drift, while a final differentiable window bounds activation memory. The trained tracker processes a single RGB stream with fixed weights. In STIR-to-VL-SurgPT transfer, temporal-stereo union selection reduces endpoint error by 7.0% relative to unfiltered trajectories. On a post-hoc diagnostic of 29 VL-SurgPT clips excluded from the endpoint-training manifest, STEP reduces endpoint error from 31.52 to 16.80 pixels relative to dense-label fine-tuning. These results support combining curated trajectories with physical observations to train surgical point trackers.
[CV-99] LOCI: Localized Continuous Control for Multi-Subject Image Editing
链接: https://arxiv.org/abs/2610.31950
作者: Sai Vidyaranya Nuthalapati,Shlok Kumar Mishra
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent instruction-following image editors such as Flux.1-Kontext enable text-driven editing but lack fine-grained control over where and how much an edit is applied. Existing approaches provide either continuous but global control or spatially localized editing at a fixed strength. To our knowledge, none jointly supports explicit spatial localization, continuous per-region strength, and independent simultaneous control of multiple subjects. We present LOCI (LOcalized Continuous control for multi- subject Image editing). Given a multi-subject image with per-subject regions and instructions, LOCI edits each subject at an independently chosen, continuously adjustable strength, for instance gradually aging one person while adding a beard to another with minimal cross-region leakage. LOCI pairs subject-level attention binding for spatial isolation with a Regional LoRA layer that replaces the LoRA’s scalar coefficient with a per-token scaling vector aligned to the spatial layout. A single adapter is trained at one suppression point, and continuous edit strengths follow by rescaling its regional residual at inference. Experiments on multi-subject scenes demonstrate that LOCI achieves strong edit fidelity with significantly reduced cross-region interference compared to baselines, while maintaining continuous control over edit strength
[CV-100] Synthetic-to-Real Text-Based Person Anomaly Search via Multi-Model Retrieval Fusion and Cross-Encoder Reranking
链接: https://arxiv.org/abs/2610.31930
作者: Eugene Denteh,Blessing Agyei Kyem,Andrews Danyo,Joshua Kofi Asamoah,Richard Dzinyela,Solomon Owusu-Ansah,Dickson G. Asiedu,Armstrong Aboah
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-based person anomaly search requires a model to retrieve individuals from a large image gallery using natural-language descriptions of both appearance and behavior. This task is challenging because model training relies on synthetic image-text pairs, whereas evaluation is performed on real surveillance images containing domain differences, near-duplicate frames, and many visually similar distractors. We present a multi-model retrieval and reranking framework designed to improve synthetic-to-real transfer. The framework first adapts Qwen3-VL-Embedding-8B to synthetic person descriptions using bidirectional contrastive learning, explicit negative sampling, false-negative masking, and domain randomization. It then combines the normalized scores of six complementary Qwen3-VL checkpoints with those of a frozen SigLIP2 model using four-view gallery encoding. The fused retrieval scores identify the strongest candidates, which are jointly evaluated with the query using Qwen3-VL-Reranker-8B. The cross-encoder and retrieval scores are subsequently blended to produce the final ranking. We also construct a distractor-matched validation protocol entirely from synthetic data to select checkpoint weights, fusion coefficients, candidate depths, and reranking parameters without accessing the official test images. The final system achieves 94.7815% mAP and 90.8493% R@1 under the official AI City Challenge 2026 Track 4 test setting, with a reported eighth-place finish on the public leaderboard.
[CV-101] PixGR: Pixel-Space Generative Decoding for Fast Feed-Forward 3D Reconstruction
链接: https://arxiv.org/abs/2610.31908
作者: Ranran Huang,Wenjing Bian,Weixun Luo,Ye Mao,Krystian Mikolajczyk
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:We present PixGR, an efficient framework that integrates lightweight pixel-space generative modeling into feed-forward 3D Gaussian reconstruction to model scene uncertainty. Context observations are encoded once into a scene latent representation, which conditions generative decoding from noisy target views to produce a shared 3D Gaussian representation. We formulate this process using pixel-space flow matching, enabling end-to-end optimization of 3D reconstruction with direct pixel-space generative supervision through differentiable rendering. This avoids VAE-based latent compression and error propagation inherent in previous two-stage pipelines. To preserve feed-forward efficiency, PixGR caches the scene representation and performs iterative inference using a lightweight DiT decoder. Experiments demonstrate that PixGR outperforms regression-based approaches in challenging sparse-view settings, particularly with widely spaced observations and view extrapolation. Meanwhile, it maintains sub-second inference and remains competitive with substantially heavier pretrained generative models. Our project page is available at this https URL.
[CV-102] Automated Nucleus Segmentation in Histology Images: An Attention U-Net Approach with Comprehensive Preprocessing and Inference Pipelines
链接: https://arxiv.org/abs/2610.31902
作者: Akshat G,Divyansh Gupta,Tusar Kanti Mishra
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 8 figures
Abstract:Objectives: In the area of digital pathology, nucleus segmentation is an essential step for quantitative analysis; however, high nuclear density and staining heterogeneity in histology images pose a great challenge. Methods: This study implements a lightweight Attention U-Net model on the MoNuSeg 2018 dataset to overcome these challenges. The proposed pipeline integrates comprehensive image and mask preprocessing, including stain deconvolution and multi-pass mask smoothing, with a high-fidelity overlapping patch-stitching strategy during inference on high-resolution images. Results: The model achieved an Intersection over Union (IoU) score of 0.8745 on full-resolution, unseen test images. This result demonstrates the effectiveness of attention mechanisms in enhancing feature localization and validates the stitching methodology for generating seamless and accurate segmentation maps. Conclusion: Attention U-Net is an effective and competitive method for automated, scalable nucleus delineation in challenging clinical contexts. Comments: 8 pages, 8 figures Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.31902 [cs.CV] (or arXiv:2610.31902v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.31902 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Akshat Gururaj [view email] [v1] Thu, 22 Oct 2026 20:57:54 UTC (6,360 KB)
[CV-103] StateBridge: Phase- and Track-Grounded Reasoning for Synthetic-to-Real Traffic Video Captioning and VQA ECCV
链接: https://arxiv.org/abs/2610.31898
作者: Blessing Agyei Kyem,Richard Dzinyela,Joshua Kofi Asamoah,Eugene Denteh,Andrews Danyo,Solomon Owusu-Ansah,Dickson Gyimah Asiedu,Armstrong Aboah
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 5 figures, ECCV conference
Abstract:Traffic safety videos provide rich evidence for understanding interactions between pedestrians and vehicles, but synthetic-to-real evaluation makes this setting challenging because models must learn from simulated scenes and generalize to real-world traffic. In this work, we present StateBridge, a traffic video-language framework designed to operate under this synthetic-only training protocol. Our approach uses phase-aware clip construction, pedestrian and vehicle track summaries, and structured prompts to ground vision-language models in stable traffic evidence rather than appearance alone. We fine-tune public vision-language models on synthetic captioning and visual question answering examples, then combine their outputs through fact-locked caption fusion and evidence-diverse VQA consensus to improve consistency across actor roles, behavioral phases, and question types. StateBridge achieves an official S2 score of 55.4679 on AI City Challenge 2026 Track 2 using only synthetic training data. These results demonstrate the effectiveness of structured evidence transfer for traffic safety captioning and visual question answering.
[CV-104] VisFlow: Interpretable Flow Matching for Visibility Reconstruction Characterization and Recognition
链接: https://arxiv.org/abs/2610.31892
作者: Rahul Moorthy,Volkan Isler
类目: Computational Geometry (cs.CG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Methods for learning continuous representations, such as latent manifolds, are fundamental and ubiquitous in modern machine learning. Yet many geometric objects combine continuous shape information with discrete combinatorial structure. This combination presents a fundamental challenge for representation learning because small geometric perturbations can induce abrupt changes in the underlying discrete representation. As a result, the smoothness assumptions that make continuous representations effective no longer hold. Polygons provide a canonical example. Their vertex locations vary continuously, but infinitesimal changes in geometry can dramatically alter the associated visibility graph. Existing methods address the inverse mapping from visibility graphs to polygon geometry using vertex-, triangulation-, or signed-distance-function (SDF)-based representations. Vertex- and triangulation-based approaches construct polygons through explicit geometric operations. However, they are less effective at modeling the relationship between geometry and visibility. SDF-based approaches capture this relationship more naturally. Their intermediate states, however, do not correspond to explicit stages of polygon construction. We present VisFlow, a novel, interpretable flow formulation that combines continuous SDF representations with explicit polygon evolution. We evaluate VisFlow on Visibility Reconstruction and Visibility Characterization. Visibility Reconstruction generates one polygon consistent with a visibility graph (G), while Visibility Characterization characterizes all such polygons. VisFlow maintains 20% more visibility relations on Visibility Reconstruction and improves coverage by 30% on Visibility Characterization compared to state-of-the-art methods.
[CV-105] Scaling Transformer Energy-based Models For High Quality Image Synthesis
链接: https://arxiv.org/abs/2610.31889
作者: Jungsoo Lee,Alexi Gladstone,Yilun Du
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Energy-Based Models (EBMs) are a flexible class of generative models that decouple probabilistic modeling from the generation process. This decoupling directly enables compositional generation and inference-time scaling, where sample quality can be refined simply by increasing sampling steps. However, severe challenges in high-dimensional sampling and training instabilities with deep architectures have restricted EBMs to low-dimensional regimes. To resolve these bottlenecks, we introduce the Transformer Energy-Based Model (TEM), a scalable transformer energy function operating entirely within a latent space. On top of this backbone, we extract energies at multiple layers and sum them, composing complementary constraints from local texture to global semantics. Furthermore, two components make this high-quality at scale: Positive Representation Alignment (PRA), which supervises TEM’s intermediate representations with a vision foundation model, and advanced sampling strategies that improve sample quality at inference. TEM-XL achieves an FID-50K of 3.82 on class-conditional ImageNet 256 \times 256 without classifier-free guidance, showing that a contrastive EBM can generate high-fidelity images. On ImageNet 512 \times 512, TEM achieves an FID-50K of 8.24, demonstrating that the approach scales to higher resolution. We further validate the core promises of EBMs: compositional generation, zero-shot inpainting, and out-of-distribution detection.
[CV-106] FSVL: Taxonomic Fast-Slow Vision-to-Language Collaborative Framework for Interpretable Confusable Fish Species Identification and Knowledge Generation in the Middle Yangtze River
链接: https://arxiv.org/abs/2610.31880
作者: Meng Liang,Fucheng Wang,Lang Zhao,Jun Chen,Ziying Li,Yuxiang Zhao,Haozhuang Chi,Wenkai Fang,Qiong Zhou,Junlin Yang,Jinming Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:The Yangtze River Basin is a critical region for freshwater biodiversity conservation. The implementation of the Yangtze River Ten-Year Fishing Ban has increased the importance of effective aquatic biodiversity monitoring and species-level identification. However, fine-grained identification of closely related freshwater fish species remains challenging due to subtle morphological differences. Existing deep learning methods often struggle to balance classification accuracy and inference efficiency, while LLMs may generate unsupported information when producing specialized biological knowledge. To address these challenges, this study proposes FSVL, a fast-slow dual-pathway framework integrating visual recognition with LLM-based Retrieval-Augmented Generation (RAG). A lightweight ResNet50 fast path performs preliminary classification across 51 fish species in the Middle Yangtze River Basin and routes predefined groups of easily confused species to a slow path for refined identification. Guided by taxonomic criteria, the slow path uses YOLOv8n to detect key morphological regions, including the head, dorsal fin, and tail, leveraging morphometric and outline traits to improve discrimination among closely related species. The framework further integrates a retrieval-enhanced Qwen3-14B backend, using soft tokens and Low-Rank Adaptation (LoRA) for parameter-efficient domain knowledge generation. On the evaluated test set of 607 images, the fast classifier achieved 99.67% accuracy, whereas the cascade corrected the two errors, yielding 100.00% overall accuracy; further validation on independent specimens and external conditions remains needed. Across 255 knowledge generation samples, scientific name accuracy rises from 91.37% to 100.00%, genus accuracy from 90.20% to 98.82%, and ROUGE-L from 0.335 to 0.655, supporting intelligent biodiversity monitoring and fisheries management.
[CV-107] IFC-SSL-Bench: Benchmarking Self-Supervised Learning for Annotation-Efficient Multispectral Imaging Flow Cytometry
链接: https://arxiv.org/abs/2610.31848
作者: Angelo Jovin Yamachui Sitcheu,Roxana-Teodora Fetescu,Thomas Lautenschlager,Nils Friederich,Linus Schilpp,Stephan Allgeier,Luca Deininger,Daniel Hübschmann,Ralf Mikut
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 35 pages, six main figures, seven supplementary figures, five tables and four supplemntary tables
Abstract:Imaging Flow Cytometry (IFC) generates large multispectral datasets, yet the need for expert annotations bottlenecks the development of supervised deep learning models. While Self-Supervised Learning (SSL) offers a solution, its application to IFC remains fragmented and lacks systematic benchmarking. To address this, we present IFC-SSL-Bench, a modular framework for evaluating SSL in IFC. We benchmark six SSL approaches across four backbones on two heterogeneous datasets. By assessing downstream classification performance across annotation regimes of the training sets (1%-100%), we quantify annotation efficiency and identify optimal configurations for high-dimensional biological imaging. SSL matches the performance of fully supervised baselines trained on 100% of annotated data while using as little as 5% of annotations for morphological classification, with advantages in low-data regimes below 5%. Across both datasets, the contrastive approach MoCo-v2 paired with ConvNeXt backbones emerged as the top-performing configuration, with ConvNeXtV1-T best for immunological synapse classification and ConvNeXtV2-T for morphological classification. We identify a clear pattern in transfer dynamics: ImageNet initialization benefits morphological classification, consistent with its visual similarity to natural images, whereas random initialization better captures the more domain-specific immunological synapse classification. SSL elevates class-specific F1-scores by over 35 percentage points for structurally complex classes at 1% of annotated data. We synthesize these insights into a unified set of guidelines for deploying self-supervised workflows in high-throughput IFC, and release the benchmark as a resource for future work.
[CV-108] EgoForesight: Effectively Scaling Policy Learning with Noisy Egocentric Actions
链接: https://arxiv.org/abs/2610.31845
作者: Chubin Zhang,Changyuan Wang,Yirui Wang,Ao Li,Shilin Ma,Yue Su,Cai Zhou,Xiuwei Xu,Tianle Zhang,Yansong Tang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Scaling policy learning requires broad experience and reliable supervision. Robot demonstrations provide high-quality trajectory supervision but are costly to collect. Egocentric videos broaden task and environment coverage, yet motion reconstruction and retargeting introduce noisy action labels. To address this challenge, we introduce EgoForesight, a vision-language-action framework that learns jointly from both sources by reducing the influence of unreliable action labels and exploiting demonstrated outcomes as complementary supervision. A joint vision-action flow couples future visual representation prediction with action generation, unifying reliability-aware action learning with outcome-informed representation transfer. Specifically, reliability-conditioned corruption allocates action supervision across noise levels according to source reliability, while reliability-weighted regression regulates each source’s contribution. Complementing this treatment of noisy labels, privileged-future representation distillation draws supervision from demonstrated outcomes. An exponential-moving-average teacher observes the current scene and clean future to encode action-relevant structure from the resulting scene change. Its deeper representations supervise an earlier student layer receiving a corrupted future, providing outcome-informed structure beyond noisy action annotations. Together, these mechanisms support effective scaling with noisy egocentric actions. Over the tested range, EgoForesight exhibits steeper scaling slopes than the baseline, reducing real-world open-loop action MSE by 20.5% and improving RoboCasa few-shot success by 8.1 percentage points at 30,000 hours of pretraining data. After post-training, it also achieves strong absolute performance in both simulated and real-world dexterous manipulation.
[CV-109] SAGE: A hybrid vision-language and geometry framework for work-zone pathway prediction under temporary traffic control
链接: https://arxiv.org/abs/2610.31835
作者: Jiaxi Liu,Keshu Wu,Weizhe Tang,Bofeng Cao,Sikai Chen,Xiaopeng Li,Steven T. Parker,Bin Ran
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 26 pages, 9 figures, 7 tables. Project page: this https URL
Abstract:Temporary traffic control redirects vehicles through work zones, requiring automated vehicles to connect an interpretation of the visible devices to a continuous passing path. Geometric predictors can provide several plausible paths, yet selecting the one consistent with the temporary control remains difficult. We propose SAGE (Semantic-to-Action Grounded Execution), a read-decide-execute-check framework that separates path construction, path preference, and control compliance within a shared candidate space. The central idea is to use a vision-language model’s (VLM’s) trajectory as a preference over intact geometric candidates. A role-decoupled generator constructs the candidates, and rank fusion combines each candidate’s geometric score with its agreement with the VLM trajectory. Supervision structured around the Manual on Uniform Traffic Control Devices (MUTCD) trains SAGE-VLM to state the lane status and hazard side before its passage decision and trajectory. The predicted hazard side enters image-space predicates derived from MUTCD provisions, which support a final intervention bounded in score and displacement. On 2,306 real-world ROADWork validation episodes, the candidates include a path within 10 px of the human demonstration in 87% of episodes, yet selection regret accounts for 53% of the geometric ranker’s error. SAGE reduces average and final displacement errors by 8.3% and 12.0% relative to the ranker, mostly in the far field, and hard-constraint violations by 37%. The trajectory opinion improves all four tested generator-selector combinations. SAGE is comparable to state-of-the-art VLM modules in accuracy, with a faster-responding VLM, and outperforms four other adapted driving VLMs in displacement error and safety measures. Project page: this https URL.
[CV-110] WorldMirror: Agent ic Real-to-Sim Generation of Interactive Worlds from In-the-Wild Robot Demonstrations
链接: https://arxiv.org/abs/2610.31834
作者: Tianling Xu,Weijie Wan,Honghao Su,Kailun Su,Junyuan Tang,Huichi Zhou,Haozheng Zhang,Qi Xiong,Ping Luo,Wenbo Ding,Tianxing Chen,Yue Chen
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing approaches for robot learning in simulation often require substantial manual effort to construct scenes and specify tasks, limiting training scalability and data diversity. Meanwhile, in-the-wild robot demonstrations naturally provide diverse and realistic observations of real-world interactions. Leveraging these demonstrations for simulation requires both high-fidelity visual reconstruction and physically consistent interactions. General-purpose agents offer flexible adaptation to diverse scenarios and strong reasoning about 3D spatial relationships, while specialized generative models provide high-fidelity reconstruction of scene appearance and geometry. We introduce WorldMirror, an agentic real-to-sim framework that constructs interactive digital twins from in-the-wild robot demonstrations. Our central idea is to formulate real-to-sim construction as the iterative refinement of an executable world program. A coding agent decomposes the observed scene, plans reconstruction steps, and orchestrates specialized image and 3D generation tools, refining the program through verification feedback. A camera-pose-anchored scene representation organizes the world into static visual context, support geometry, and editable interactive assets while preserving spatial relationships among the scene, objects, and robot. The agent uses this representation to diagnose and correct reconstruction errors, enabling physically consistent interaction and trajectory replay. Experiments on DROID demonstrate that WorldMirror can reconstruct visually faithful and geometrically consistent interactive worlds while preserving the physical validity of recorded robot trajectories.
[CV-111] StackGen3D: Stable 3D Block Stack Generation from Multiview Silhouettes IROS2026
链接: https://arxiv.org/abs/2610.31803
作者: Xiaopeng Zhang,Luzhe Sun,Yiqing Xu,Matthew R. Walter
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: A poster of this work is publicly available on the website of the Creativity in Assembly Workshop at IROS 2026 (non-archival)
Abstract:Generating physically stable 3D structures from visual specifications requires resolving geometric ambiguity while accounting for physical interactions among objects. We present StackGen3D, a framework for generating stable block assemblies conditioned on a given block inventory and three orthographic silhouettes. Our approach consists of two stages. First, we develop a physics-informed graph neural network (PI-GNN) that models support relations between blocks and serves as a fast stability predictor for curating a diverse dataset of stable structures. Second, we train a conditional flow-matching model on this dataset to capture the multimodal distribution of block configurations consistent with the target silhouettes. At inference time, lightweight geometric selection and refinement reduce inter-block penetration without physics simulation. Preliminary experiments show that the PI-GNN generalizes from simple stacks to complex structures substantially better than a parameter-matched multilayer perceptron. For structure generation, Best-of-10 selection improves physical stability from 45.3% to 72.4% while achieving a silhouette IoU of 0.970. These results highlight the potential of combining relational physical reasoning, scalable data curation, and conditional generative modeling for creative 3D assembly. Project website: [this https URL](this https URL) Comments: A poster of this work is publicly available on the website of the Creativity in Assembly Workshop at IROS 2026 (non-archival) Subjects: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO) Cite as: arXiv:2610.31803 [cs.CV] (or arXiv:2610.31803v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.31803 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-112] PnP*: A certifiable solver for Perspective-n-Point
链接: https://arxiv.org/abs/2610.31797
作者: Bang-Shien Chen,Caitrin E. Eaton,David M. Rosen
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO); Optimization and Control (math.OC)
备注: 28 pages, 3 figures
Abstract:Estimating a camera pose from 3D-2D correspondences - Perspective-n-Point - is one of the most fundamental problems in computer vision. Existing methods either solve a minimal problem or replace the objective with an easier one. Surprisingly, no scalable certifiable solver for the actual Maximum Likelihood Estimation (MLE) exists. We propose PnP*, the first practical solver that guarantees efficient computation and returns a certifiable globally optimal solution. We prove that the underlying convex relaxation is exact under moderate noise, which to our knowledge is the first stability guarantee for any relaxation for the MLE of PnP. Furthermore, we reduce the relaxation by exploiting chordal sparsity pattern and quotient ring reduction such that the relaxation is tractable and scalable. The stability and scalability results are verified numerically, as PnP* returns a provably correct camera pose at over 10 Hz.
[CV-113] PaperScorer: How Predictable Is Paper Acceptance? NEURIPS2026
链接: https://arxiv.org/abs/2610.31791
作者: Sachin Konan,Jonathan Liu,Zhuang Liu
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026
Abstract:Automated research is largely evaluated by benchmark performance, but the value of research rarely reduces to a single score. Humans judge a work’s holistic contribution through peer review, which is hard to model because it is noisy and depends on many factors. However, what if, given only the paper, we could predict its peer-review decision directly? We introduce PaperScorer, a family of models that predict a paper’s accept or reject decision from its extracted text or page screenshots, training only on the decision after removing author identity and formatting shortcuts. Vision models are the most predictive (68% accuracy on ICLR papers and 77% on ML papers from arXiv), beating text-only and frontier reasoning models. Unlike plain text, page images retain figures, layout, and presentation. This signal is robust, surviving smaller models and even compressing the whole paper into a single image. But accurate predictions are not evenly spread. PaperScorer is strongest on the lowest and highest rated papers, while papers rated in the middle stay unpredictable for our models and frontier systems alike. Additionally, its confidence is calibrated, so its most confident predictions are also its most accurate. Overall, PaperScorer shows how much of the decision a paper carries on its own, marking a first step toward AI-driven research evaluation.
[CV-114] SyncVid: World-Equivariant Diffusion for Synchronized Video Generation
链接: https://arxiv.org/abs/2610.31765
作者: Farhad G. Zanjani,Animesh Karnewar,Fatih Porikli,Amirhossein Habibian
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 8 figures, 9 tables. Project page: this https URL
Abstract:Generating multiple videos of a dynamic scene from a single reference video requires synthesizing both observed and unobserved content consistently across viewpoints. Existing camera-control methods for video diffusion typically rely on architectural mechanisms, including camera embeddings, cross-view attention, and geometric conditioning. Although effective for static scene content, these approaches often struggle with the fine-grained, dynamic geometry that must remain coherent across views and over time. We synchronize the diffusion process itself by sampling the Gaussian noise prior in a canonical 3D space derived from the source video and rasterizing it into target cameras, rather than sampling independently in each 2D image plane. Our method, SyncVid, lifts a monocular video into a canonical 4D representation, associates noise with scene geometry and unseen space, and rasterizes the resulting structured noise into arbitrary views. Because rendering can introduce local correlations, we further propose a cross-view-consistent whitening procedure that decorrelates neighboring pixels while preserving correspondence across views. The structured noise serves as the diffusion prior for a LoRA-adapted pretrained video diffusion model, requiring no architectural modifications. By enforcing spatiotemporal and cross-view coherence at the source of stochasticity, SyncVid generates synchronized videos and improves both perceptual quality and cross-view consistency in our experiments. Project page: this https URL.
[CV-115] See What I Need Dream What I Want: A Navigation Foundation Model on a Self-Constructed Representation Space
链接: https://arxiv.org/abs/2610.31762
作者: Jiawei Chen,Yinxi Wang,Zihan Yang,Meijun Gao,Yaowei Liu,Shuming Zhang,Zecheng Li,Xiangju Mi,Yuxuan Liu,Runzhong Huang,Tianlin Li,Xun Zhang,Zhengyang Wang,Xu Miao,Tong Yang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages. Project page: this https URL
Abstract:What visual representation should a navigation policy act on? Navigation policies built on vision foundation models largely leave representation design to the encoders’ pretraining objectives. We present RepresentNav, a navigation foundation model built on a self-constructed representation space: a visual representation constructed by the policy’s own objectives from vision foundation models. RepresentNav builds this space with Representation Residuals, a compact three-layer transformer that attends over features spanning the full depth of multiple complementary vision foundation models, so that the policy learns to see what it needs for navigation. In the same space it dreams what it wants: before every action a latent chain-of-thought predicts the representation of the scene a moment ahead. The dream is trained end-to-end with the policy as a joint-embedding predictive architecture (JEPA), with a next-representation prediction loss and a regularizer on its targets. RepresentNav sets a new state of the art on VLNVerse, VLN-CE R2R, EVT-Bench, and two point-goal benchmarks, and its gains carry over to a physical robot amid clutter and moving people. On VLNVerse fine-grained, RepresentNav achieves 83.80% success, outperforming the previous state of the art by 20.05 percentage points, and on EVT-Bench distracted tracking it reaches 85.5% with a single camera (+10.1). With the same encoders, the self-constructed space clearly outperforms concatenating their outputs and learns to weight their layers unevenly, and a dream in it cuts collisions by a quarter. A single dreamed latent decodes into the depth, semantics, and 3D occupancy of the scene a moment ahead, more accurately than pretrained encoders applied to the real future frame, although occupancy is never supervised, and the rollout tracks the robot’s own motion over the next second: the dream emerges as a compact latent world model.
[CV-116] AudioLift: Audio-Driven Single Image to Full-Head Gaussian Avatar via Multi-view Video Generation
链接: https://arxiv.org/abs/2610.31747
作者: Zhuoyuan Jiang,Weijie Lyu,Xiaoyuan Wang,Tiancheng Shen,Ming-Hsuan Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Existing 3D talking avatar methods mainly rely on parametric face models including 3DMM or FLAME, whose compact motion representations limit their ability to synthesize extreme expressions and to model accessories such as headwear. We introduce AudioLift, a two-stage framework that uses generated multi-view videos as the intermediate representation for audio-driven avatar generation without parametric face models or per-instance optimization. Our framework combines a synchronized multi-view video generation stage with an inconsistency-robust Gaussian reconstruction stage to recover coherent full-head geometry from a single portrait image. Extensive experiments demonstrate that AudioLift produces 3D talking avatars capable of capturing more extreme facial expressions than existing methods, with improved geometric completeness and strong multi-view consistency across the full 360^\circ viewing range.
[CV-117] DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model
链接: https://arxiv.org/abs/2610.31734
作者: Zixian Gao,Lixin Xiu,Penghao Luo,Xiangcheng Cao,Ziqi Cai,Zhen Li,Erwin Wu,Kaipeng Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 6 figures
Abstract:A live esports broadcast needs a director who chooses which camera goes on air and a caster who talks over that choice, every second and without knowing what comes next. Streaming video language models have learned to commentate, but only over a broadcast that a human director has already produced. We study live broadcast production, in which a model receives the raw camera of every player and must both direct and commentate from the past alone. We present DirCast, a single streaming video language model that does both. Every second it scans all cameras at low resolution, decides with a hold-or-switch director whether and where to cut, and reads the chosen camera at high resolution before it speaks. Its context keeps what went on air and what was said and lets the other views expire, which bounds the cost of a stream of any length. We build a dataset of 27 professional Dota 2 matches with a camera for every hero, the human director’s choices and the casters’ commentary. DirCast directs better than rule-based and zero-shot directors including GPT-5, and its commentary wins 29.2 percent of comparisons against GPT-4o under the Proact-VL protocol, against 19.4 percent for the strongest of eleven streaming commentators that watch the human broadcast.
[CV-118] Residual Calibration via Local Feature-Space Refinement NEURIPS2026
链接: https://arxiv.org/abs/2610.30652
作者: Xihao Wang,Chengyao Yu,Ruixing Ming
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Machine Learning (stat.ML)
备注: Accepted at NeurIPS 2026
Abstract:Deep neural networks often struggle to provide reliable confidence estimates, making post-hoc calibration essential for reliable decision-making. Existing post-hoc methods largely rely on score-based transformations, which improve marginal calibration but often fail to capture spatially varying miscalibration in local regions of the learned feature space. To address this gap, we propose Local Adaptive Probability Estimation (LAPE), a post-hoc residual calibration framework. Specifically, LAPE treats the global calibrator output as a stable anchor and refines the individual prediction by leveraging non-parametric local evidence from neighboring calibration samples. As a plug-and-play refinement module, LAPE effectively utilizes the information about feature-space geometry and can reuse the calibration data that are used to construct the global calibrator. Furthermore, we extend LAPE to the setting with multi-view inputs through Cascade Feature Augmentation Fusion (CFAF). Extensive experiments demonstrate that the proposed methods can improve the calibration quality of various existing baselines and remain efficient and robust across different settings.
[CV-119] 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); Computer Vision and Pattern Recognition (cs.CV); 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.
[CV-120] ARENA: Active Region attENtion Aggregation for interpretable solar flare prediction NEURIPS2026
链接: https://arxiv.org/abs/2610.32758
作者: Luis Bauer,Annalena Kofler,Andrés Asensio Ramos,Aparna Venkataramanasastry,Valentín Martínez Pillet,Bernhard Schölkopf
类目: olar and Stellar Astrophysics (astro-ph.SR); Instrumentation and Methods for Astrophysics (astro-ph.IM); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 7+12 pages (2+8 figures), accepted at the “Representations for the Physical Sciences” workshop at NeurIPS 2026
Abstract:Predicting solar flares is a challenging space-weather forecasting problem with potentially significant impacts on Earth’s upper atmosphere, including disruptions to radio communications. We propose Active Region attENtion Aggregation (ARENA), a neural network architecture that predicts solar flares from a variable number of active-region cutouts while learning to identify the region most relevant to the prediction. ARENA is trained using multiple-instance learning under weak supervision, requiring only a flare/no-flare label for the observation and no labels specifying the flare origin. The learned attention mechanism therefore provides a means of localizing the active region driving the prediction without explicit origin annotations. ARENA achieves a heidke skill score of 0.45 which is comparable to other flare prediction approaches. These results demonstrate that active-region-based aggregation can provide an efficient and interpretable approach to solar flare prediction. The code is available at this https URL.
[CV-121] Computational imaging through turbid bubbly and flowing water near the telescope
链接: https://arxiv.org/abs/2610.32646
作者: Alexander Ingold,Sabina D. Menon,Aadhi Umamageswaran,Alec Ikei,May H. Hlaing,Syed N. Qadri,Rajesh Menon
类目: Optics (physics.optics); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Cameras in the surf zone or behind ship wakes look through turbid, bubbly, moving water close to the lens, yet restoration networks are usually scored only on displayed images. We introduce a measured dataset of a telescope viewing displayed images and physical objects through a 25~mm water layer 45~mm in front of its objective, under seven combinations of flow, bubbles and turbidity, with clear-water recordings as a second reference. Because the suspension scatters light far outside the 1.3\textdegree field of view, turbidity mainly removes contrast, 4.4- to 4.8-fold on a Siemens star. Restoring the mean of eight frames improves structural similarity (SSIM) by 0.032 over a single frame, and learned fusion of the frames adds little. One raw frame supports five-category recognition (96.8%) as well as restored images. Adding synthetic training pairs leaves display-image scores unchanged but raises the correlation of a restored Siemens star with its clear-water recording from below 0.5 to above 0.9. These networks also improve ordinary underwater photographs without retraining, where training on measured pairs alone fails. Scoring against the digital source or a clear-water recording reverses the ranking of raw and restored images. These results support a simple pipeline: average the frames, classify raw frames directly, and restore with networks trained on measured and synthetic pairs.
人工智能
[AI-0] RobustRLlib: A Unified Library and Benchmark for Robust Reinforcement Learning Algorithms
链接: https://arxiv.org/abs/2610.32971
作者: Xi Chen,Fengqi Hu,Zeyu Shen,Jiaying Xu,Shangding Gu,Pan Xu,Anqi Liu,Laixi Shi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: Project page, code, and tutorial: this https URL
Abstract:Robust reinforcement learning (RL) is critical for reliable real-world deployment, where policies may fail under diverse shifts and disruptions. Existing robust RL algorithms offer a rich set of mechanisms and principles, yet two major barriers limit their broader potential: limited algorithm reusability and an unclear empirical capability boundary. We introduce RobustRLlib, an algorithm-centric library and benchmark that addresses both challenges. It integrates 17 representative robust online, offline, and safe RL algorithms under a unified interface for substantially improved reusability, together with an enhanced evaluation platform spanning diverse real-world shifts and advanced humanoid and vision-language-action tasks. Systematic evaluation over 273 perturbation configurations provides an up-to-date capability map of current robust RL algorithms, revealing where they remain effective and where substantial robustness gaps persist. As an open-source, user-friendly platform, RobustRLlib offers a reproducible and extensible codebase for applying existing methods to new problems and developing new robust RL algorithms.
[AI-1] DexPort: A Hand-Held Skill Interface for Dexterous Hand Data Collection
链接: https://arxiv.org/abs/2610.32969
作者: Amber Xie,Jiawei Fu,Xiaofeng Guo,Carmelo Sferrazza,Guanya Shi,Haozhi Qi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Website: this https URL
Abstract:Collecting high-quality demonstrations remains a central bottleneck for learning dexterous manipulation. Teleoperation records executable robot actions, but it requires a complete robot setup and places the burden of coordinating many finger joints on the operator. Human demonstrations are easier to collect, but their motions must be estimated and retargeted to the robot, and they do not directly provide robot actions. In this paper, we introduce DexPort, a portable data-collection interface for high-DoF robot hands. It uses the same hand hardware for collection and deployment, avoiding a hand embodiment gap, and records both commanded joint targets and measured joint states. Our key observation is that a small set of skills can support a wide range of dexterous tasks. During collection, the operator moves the hand-held interface to specify the robot wrist motion and selects or adjusts skills that coordinate the finger motion. This removes the need to control each joint independently while retaining behaviors such as precise grasping and in-hand manipulation. We collect demonstrations across several real-world tasks and in unstructured environments without a robot arm. We then use behavior cloning to train visuomotor policies directly on the recorded wrist motions and hand commands, without skill labels. We evaluate across 4 tasks: pick-and-place with diverse objects, screwing with in-hand tool use, drawer stowage which requires simultaneous grasp and pulling, and a bimanual scanning task. We collect over 2,500 demonstrations with our portable device, in both laboratory and in-the-wild settings. Across our tasks, we see an improvement from 11% to 64% when using recorded robot commands instead of states, and an improvement from 36% to 78% for task success when using in-the-wild data.
[AI-2] SemantEq: Faithful Bidirectional Equivalence of Formal Statements
链接: https://arxiv.org/abs/2610.32919
作者: Xinrui Li,Pawan Sasanka Ammanamanchi,Siddharth Bhat,Rianna Jamdar,Yongxi Lin,Stella Biderman
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 37 pages, 4 figures, 6 tables. Code and data: this https URL
Abstract:Statement autoformalization translates mathematics from natural language into a proof assistant such as Lean 4, and models now do it at scale. As provers get stronger, the burden of correctness shifts from the proof to the statement, that is, to whether the formal statement faithfully translates its source. Given a trusted reference formalization, the natural test is logical equivalence, and it proves too much. P \leftrightarrow Q holds trivially for any two statements P and Q that the library proves, because a prover can prove each separately and pass through P \leftrightarrow \mathrmTrue \leftrightarrow Q . The stronger the prover, the more unrelated pairs pass. Earlier methods restrict the prover and check that each proof names the other statement. We introduce SemantEq, a more principled relation that instead requires each statement to turn into the other only through a chain of allowed steps, such as logical rearrangement and rewriting with the library’s own equations. Our programs judge a proof by the proof term Lean checks, not by its tactic script. SemantEq-Eval, a search without a model, reaches 100% precision and 65.6% recall on the ProofNetVerif test split under our adjudicated labels, against 95.8% and 61.8% for BEq+, and an 8.9-fold increase in its CPU time adds at most three pairs. A language model finds more proofs, but under two of four prompts it proves both directions of about two in five of the non-equivalent pairs. SemantEq-ProofAudit audits each model-written proof under a policy derived from the relation and accepts none of the 104 non-equivalent pairs, and it raises recall on 407 equivalent pairs from 65.4% to 81.1% under the best prompt. Separating the generation of equivalence proofs from their acceptance lets a stronger prover raise recall without changing what an acceptance means. We release code and data at: this https URL
[AI-3] Green Code or Costly Code? Evaluating the Energy Costs of Iterative LLM -Based Code Optimizations ICSE2027
链接: https://arxiv.org/abs/2610.32891
作者: Sadra Emami,Benjamin Weigell,Mohammad Reza Mousavi,Gunel Jahangirova,Hugo Araujo,Bernhard Bauer
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at the 49th IEEE/ACM International Conference on Software Engineering (ICSE 2027)
Abstract:Large Language Models (LLMs) are increasingly used to optimize code for performance. This practice often assumes that improvements in execution speed naturally translate into overall lower energy consumption. We challenge this assumption empirically by measuring whether performance-optimized code produced by LLM self-refinement pipelines is also energy-efficient relative to canonical human reference solutions, and whether the energy cost of the optimization process can be offset by subsequent execution-time savings. We evaluate 11 open-source models across four decoding configurations on all 1,000 problems of EffiBench, a state-of-the-art performance benchmark, while collecting per-process, hardware-level energy measurements throughout. Inference energy spans 0.79 to 8.68 MJ across model-variant pairs, which is roughly equivalent to the energy of 250 to 2800 GPT-4.1 queries. Limiting the maximum number of output tokens has the largest impact, reducing energy use by up to 69% compared to the default configuration. On the execution side, 42 of 44 configurations produce code that consumes more energy per problem than the canonical human reference, with overheads ranging from +4% to +886%. The two exceptions achieve only marginal gains, with break-even points on the order of 10^6 to 10^7 executions. Mixture-of-Experts architectures consume substantially less inference energy than equivalently-sized dense models without sacrificing correctness, making architectural choice a more consequential sustainability lever than decoding parameters. Performance optimization and energy efficiency are not always aligned goals: LLM self-refinement consistently produces faster-seeking but energy-heavier code relative to what a human engineer would write. All artifacts are available at this https URL.
[AI-4] NgramRL: Accelerating RL Post-Training for Structured Generation via Rollout-Aware N-gram Speculative Decoding
链接: https://arxiv.org/abs/2610.32881
作者: Nilesh Prasad Pandey,Jangseon Park,Jason Kong,Onat Gungor,Tajana Rosing
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but its cost is dominated by autoregressive rollout generation, which can exceed 70% of training time. Existing speculative-decoding methods are poorly suited to this setting: model-based drafters consume scarce GPU memory and may become misaligned as the policy evolves, while existing model-free methods do not explicitly exploit repetition across training epochs. We observe that recent rollouts of the same prompt provide a high-quality, self-updating draft source. Based on this insight, we propose NgramRL, a per-prompt n-gram drafter constructed from bounded recent rollout history. NgramRL requires no auxiliary model, drafter training, or persistent GPU-resident drafter state, while target-model verification preserves the rollout distribution. To hide CPU cache lookup and scheduling, we further introduce an adaptive fan-out prefetcher that predicts likely verification outcomes (j, b), consisting of the accepted draft length j and target-emitted continuation token b. A causal budget allocator concentrates prefetches on high-probability outcomes, while uncovered outcomes fall back to sequential lookup without changing committed tokens. Across two model families and three reasoning and structured-generation workloads, NgramRL commits 3.60-7.77 tokens per target-model forward in the warm-cache regime, up to 1.99x more than SuffixDecoding and 7.0x more than prompt-lookup decoding. On Spider, the prefetcher covers 96.6% of Qwen3-8B outcomes at a candidate budget of 16 and improves measured rollout throughput by 1.38x over synchronous NgramRL, approaching its 1.40x overlap ceiling.
[AI-5] Advancing SVD-based LLM Compression via Layer-Wise Error Model Search ICML2026
链接: https://arxiv.org/abs/2610.32878
作者: Moritz Thoma,Maximilian Groezinger,Maximilian Forstenhäusler,Emad Aghajanzadeh,Manoj Rohit Vemparala,Christos Anagnostopoulos,Pierpaolo Mori,Nael Fasfous,Alexander Frickenstein,Daniel Mueller-Gritschneder,Ulf Schlichtmann
类目: Artificial Intelligence (cs.AI)
备注: Accepted at ICML 2026
Abstract:Low-rank SVD-based compression offers a powerful strategy to reduce the computational costs of LLMs. However, existing methods face two key limitations: (i) global rank allocation, where uncalibrated error proxies fail to capture complex error propagation, and (ii) decomposition quality, where Fisher-based estimators suffer from severe rank collapse. In this work, we address these limitations by introducing Layer-wise Error Modeling Search (LEMS) and KFAC-SVD. LEMS advances rank allocation by introducing a layer-wise error surrogate that integrates local and global layer importance alongside a propagation bias, enabling effective global rank allocation via an ILP formulation. KFAC-SVD improves decomposition quality by utilizing token-wise statistics, mitigating the rank deficiency observed in prior Fisher-based SVD approaches. Across Mistral, Qwen3, and Llama 3 model families, we show that LEMS consistently outperforms existing search strategies, delivering significant zero-shot accuracy gains of up to 4.8 p.p. that generalize to model sizes of 70B parameters, while KFAC-SVD achieves an average perplexity improvement of 15%. Project page and code: this https URL
[AI-6] From Deliberately Constructed Systemic Representations of the World of Activity to Dialogue: A Five-Level Communication Architecture for Strong Language-Based ASI
链接: https://arxiv.org/abs/2610.32869
作者: Maxim Osovsky(Adam University, Bishkek, Kyrgyzstan)
类目: Artificial Intelligence (cs.AI)
备注: 36 pages, 1 figure
Abstract:This paper proposes a five-level organization of message formation and understanding within thinking and activity in which ASI participates. Its central thesis is that a language model’s context is fundamentally incapable of presenting the foundations of utterance formation for examination and verification. These foundations are established in advance by acquired schemas, systemic representations of the world of activity, and connections accepted as true within them. Neither context nor trained LLM weights, in themselves, attest to the presence of deliberately constructed systemic representations. The five levels distinguish systemic representations of the world of activity, a particular situation cut out from those representations, its narrative description, the full sense of a message, and text. A situation’s schema-model is not identical to a story about it. A recipient reconstructs the full sense from an abbreviated text through jointly constructed and agreed-upon representations and joint activity. Reverse work proceeds from the full sense through an extended description and the situation’s schema-model to the reconstruction of systemic foundations. The theoretical foundations draw on the systems methodology of G. P. Shchedrovitskiy, the structural differential of A. Korzybski, and the Meaning-Text theory of I. A. Mel’chuk. The contribution is a specific systemic-activity organization of content for examining and adjusting message foundations and controlling the consequences of their transformation. Further research concerns large graphical models of activity on reconstructible systemic foundations, distinct from text generation from context.
[AI-7] PoKER: A Probability of Kill Estimation Rate Model for Air-to-Air Missiles Using Machine Learning on Stochastic Targets
链接: https://arxiv.org/abs/2610.32864
作者: Joao P. A. Dantas,Andre N. Costa,Diego Geraldo,Marcos R. O. A. Maximo,Takashi Yoneyama
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted manuscript. 18 pages, 10 figures, 5 tables. Code available at this https URL
Abstract:This work introduces PoKER, a novel probabilistic model engineered to optimize missile launch effectiveness in air-to-air scenarios, specifically within Beyond Visual Range (BVR) air combat. Unlike conventional Weapon Engagement Zone (WEZ) models that delineate zones based on static distances such as maximum, minimum, and no-escape ranges, PoKER applies machine learning to predict kill probabilities more accurately by integrating the stochastic behaviors of targets and missile miss distances. This model dynamically evaluates target behavior, greatly expanding the predictive capabilities of engagement analysis. By factoring in elements such as warhead lethality, target and shooter orientations, and the specific conditions of engagement, PoKER provides important insights into engagement dynamics and quantifies success probabilities. Consequently, it can potentially be an important tool for BVR air combat pilots, improving operational decision-making within this specialized combat domain.
[AI-8] A Foundation Model for Policy In-Context Longitudinal Causal Effect Estimation
链接: https://arxiv.org/abs/2610.32863
作者: Wenxin Chen,Kyra Gan,Mingyu Lu,Zilin Jing,Fei Wang,Hoifung Poon,Tristan Naumann,Peniel Argaw
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Longitudinal causal effect estimation is complicated by time-varying confounding and treatment-confounder feedback loops. Existing methods generally require data-specific nuisance-model fitting, and many are tailored to prespecified treatment sequences or prediction horizons, leading to time-consuming computation at deployment. We introduce Policy In-Context Longitudinal causal Effect estimation with Prior-data Fitted Networks (PICLE-PFN), the first foundation model for longitudinal causal effect estimation that supports dynamic treatment policies. PICLE-PFN represents target policies through in-context history-decision demonstrations, providing a unified interface for a broad class of treatment policies. Pretrained over synthetic data sampled from longitudinal structural causal models, PICLE-PFN amortizes causal effect estimation across datasets, policies, and horizons. Given a new observational dataset and target policy, it directly predicts conditional average potential outcomes (CAPOs) in one forward pass without task-specific training or hyperparameter tuning. Across synthetic and semi-synthetic longitudinal benchmarks, PICLE-PFN achieves competitive RMSE and substantially improved uncertainty calibration with a fraction of deployment runtime, compared to state-of-the-art conventional estimators trained and tuned separately for each dataset.
[AI-9] Reimagining Enterprise RAG for Multi-Agent Systems: A Knowledge Layer That Organizes Before It Retrieves
链接: https://arxiv.org/abs/2610.32851
作者: Garima Agrawal,Prasun Das,Akshay Sarvade,Priyanka L,Minisha N,Sravani Joshna,Hemath Manivannan,Lavanya Korada,Sambhrit Pattanayak,Srashti Gupta,Siddu R,Prasad Kalyanasundaram
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Enterprises are deploying agentic AI across the software development lifecycle, and they need the behavioral logic of the systems they act on, not stakeholder documentation. A seven-agent reverse-engineering pipeline recovered this logic, processing 241 repositories across 18 services into thousands of artifacts describing scope, journeys, rules and acceptance criteria. Recovery does not confer usability. Entities like offer, eligibility, and activation recur across services under a shared vocabulary, related in name but operationally distinct, so one end-to-end journey implicates over a thousand near-identical rules that must not be conflated. Flat vector retrieval fails, returning an identical rule from the wrong service at high similarity with no signal of error. Content-level ontology, the traditional remedy, did not exist, and relating every occurrence of recurring entities would have cost more than the engagement and not survived a release cycle. We place the ontology at the artifact level. In this machine-generated corpus, each artifact retains its producer, component, service, domain and type, provenance that disambiguates before content is read. We present a knowledge layer that organizes before retrieving: a self-describing catalog, a sparse metadata registry, and a navigation hypergraph defined over artifacts rather than extracted entities. Queries resolve to artifact identifiers that constrain vector search, so the competing rule from another service is absent rather than outranked, making disambiguation structural, not statistical. Built in under a week and admitting new services by convention, the layer serves SDLC agents, developers, quality engineers and analysts. It supported root-cause analysis and resolution of 100+ production defects with developer verification and attributed 35 release regression failures to specific rules and services.
[AI-10] HarnessMM: Recursive Self-Improvement for Harness Token Efficiency
链接: https://arxiv.org/abs/2610.32813
作者: Yuhua Jiang,Chang Xu,Feifei Gao,Li Zhao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent harnesses repeatedly invoke reasoning after tool feedback, even when an earlier call has already fixed the next action. We introduce HarnessMM (Harness Minus Minus), a method for improving harness token efficiency through guarded decision reuse and recursive self-improvement (RSI). RSI iteratively refines companion harness configurations from training feedback, with validation selecting the configuration for deployment. A controller at the model-tool boundary reuses resolved decisions to replace later reasoning callbacks while actor weights remain fixed. Executable Decision Graphs store exact actions, literal arguments, causal dependencies, and public guards. Guarded Decision Reuse activates these persistent decisions from observed receipts; Certified Callback Substitution executes them through the native interface and records removal of the registered slow callback. New semantic decisions, unsupported evidence, and selector abstention invoke the original actor again. The representation restricts what feedback may change, while complete-task measurements establish quality and cost. On HumanEval, HarnessMM preserves Codex success at 90.82% while reducing Tokens/task (k) from 66.72 to 47.02 (29.54% saving). For DeepSeek Harness, success rises from 71.43% to 85.71% while Tokens/task (k) fall from 96.28 to 11.62 (87.93% saving). A separate MBPP ablation measures 26.91% lower cost than single-use graphs and 19.52% lower cost than retained callbacks. These results establish guarded decision reuse as a practical way to reduce repeated reasoning in existing harnesses.
[AI-11] AI Alignment for Digital Democracy: Fairer Voters Representation against LLM Biases
链接: https://arxiv.org/abs/2610.32780
作者: Yi Qi,Srijoni Majumdar,Evangelos Pournaras
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:AI representation is emerging in digital democracy, assisting in voter emulation and decision-making. Large language models (LLMs) behave differently depending on the model, the prompt context, and the voter information they receive, representing some groups more accurately than others and potentially making the emulation unfair to certain groups. We investigate demographic biases in LLM-emulated electorates: how accurately LLMs represent individual votes, whether some voter groups are represented more accurately than others, and whether AI alignment can reduce these disparities without sacrificing overall accuracy. Using three U.S. presidential elections, we compare a closed-source and an open-source LLM across voters’ demographic features and test AI alignment strategies toward fairer voter representation. Removing gender from the voter profile reduces the largest group error of the models by 22-42%, and removing education reduces it by 64%, whereas removing self-reported race increases the mean group error by over 40%, all without lowering accuracy. Common alignment strategies, such as repeated queries or fairness instructions, do not reduce these errors and can double them.
[AI-12] Clean-label Poisoning Attack and Defense Game for Federated Learning in Vehicular Networks
链接: https://arxiv.org/abs/2610.32765
作者: Yalin E. Sagduyu,Tugba Erpek,Nathan Stephenson,Mengran Xue
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Signal Processing (eess.SP)
备注:
Abstract:This paper studies interactions between clean-label poisoning attack and defense in vehicular networks using federated learning (FL). Basic Safety Message (BSM) classification for anomaly detection is modeled as a distributed learning problem over a vehicular network with dynamic topology. As a critical threat to FL, a feature-space clean-label attack is introduced, wherein benign samples are perturbed toward malicious feature extremes, degrading class separability while preserving labels. To counter this, a feature-consistency relabeling defense is proposed to detect and relabel suspicious samples. A game-theoretic framework is developed to capture attack-defense interactions and characterize equilibrium strategies based on attack effectiveness and cost parameters. Results show that the attack scales effectively with compromised nodes, while the proposed defense keeps attack success low without degrading clean accuracy and shifts equilibrium regions toward attack-averse regimes.
[AI-13] Semantic Representation for Agent ic Self-Play in Automated Laboratories
链接: https://arxiv.org/abs/2610.32755
作者: Yves Delley,John Coller,Zachary Cutts,Alexander Daryin,Mathias Bellaiche,Jun W. Kim,Oscar Donosa,Srinidhi B. Seshadri,Tanner Clary,Sandy Klemm,Avinatan Hassidim,Yossi Matias,Pooja Rao,Petar Sirkovic,Annalisa Pawlosky
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 16 pages, 4 figures
Abstract:Navigating complex, high-dimensional biological search spaces necessitates scalable physical and computational intelligence to explore beyond local optima through iterative self-play across ideation, experimentation, and automation. Translating experimental designs into reliable physical execution remains bottlenecked by static laboratory automation and unmodeled physical variability. To address this, we introduce VERified Infrastructure for Discovery Laboratories, or VERID, a foundational infrastructure designed to bridge the gap between the high-level reasoning of artificial intelligence (AI) agents and complex, error-prone physical realities of wet labs. VERID is a persistent, high-fidelity semantic digital twin of the laboratory workcell that provides four core services: queryable introspection for agentic reasoning, time-compressed simulation for protocol validation, comprehensive trace persistence for a complete scientific record, and a framework for safe execution. We present three case studies using aqueous dye and deionized water mixtures to validate the optical and fluidic foundations of VERID. These studies demonstrate how VERID enables pre-emptive detection of critical errors in experimental protocols, stateful mid-run error recovery and post-hoc identification of subtle physical failures. The simulation functions as a discrete accounting engine, allowing agents to safely and inexpensively explore experimental designs in silico and refine experimental strategies through iterative self-play. VERID also captures structured multimodal experimental data for downstream analysis, knowledge mining and agent training. We argue that this semantic layer is an essential prerequisite for deploying autonomous agents in the biochemical and biomedical sciences, ensuring data integrity, operational safety, and paving the way for tackling previously intractable grand challenges in biology. Comments: 16 pages, 4 figures Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI) ACMclasses: I.2.9; I.2.11; J.3 Cite as: arXiv:2610.32755 [cs.RO] (or arXiv:2610.32755v1 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2610.32755 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-14] Adaptive World-Models for Anomaly Recognition (AWARE): A Framework for Robot Dynamics Prediction and Introspection
链接: https://arxiv.org/abs/2610.32742
作者: Luke Beddow,George Mavroghenis,Rodrigo Chacon Quesada,Kamil Dreczkowski,Cong Sun,Jiankai Wang,Oscar Kwong-Fai Pang,Antoine Cully
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Unrecognized anomalies in critical infrastructure can rapidly cascade into catastrophic physical or operational failures. Current anomaly detection approaches often rely on direct access to the monitored system by utilizing onboard sensors, such as motor encoders, to detect anomalous states. However, legacy hardware often lacks these sensors, or strict security protocols may prohibit access altogether. We therefore present AWARE, a world model framework that is capable of anomaly detection using only external camera monitoring via system dynamics estimation and state prediction. Evaluated across 240 real-world robotic crane trajectories, including payload collisions and motor degradations, AWARE demonstrates a 26% improvement in anomaly classification over our strongest baseline under noisy, camera-only state estimation. Furthermore, the inferred dynamics parameter can be decoded to identify the source of faults, localizing the degrading motor with 88% accuracy.
[AI-15] Agent ic Hazard Mapper: Emulating Expert Map-Making with AI Agents for Risk Communication
链接: https://arxiv.org/abs/2610.32738
作者: Sidi Wu,Lifeng Lin,Yizi Chen,Anna Yokokubo,Lorenz Hurni
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Hazard maps are essential tools for public risk communication, preparedness, and evacuation planning, yet effective cartographic design is not always ensured due to limited access to cartographic expertise across local governments and communities. We present Agentic Hazard Mapper, a multi-agent system that automates hazard-map production by emulating professional cartographic workflows and iterative design trade-offs while adapting design decisions to different data, locations, and hazard contexts. We conducted a formative study with map experts to inform the system design, followed by a user study and interviews with map experts and hazard practitioners. The generated maps improved participants’ spatial understanding and evacuation-planning performance, reduced perceived workload and task completion time, and were generally preferred over corresponding municipal maps. We further discuss human oversight, adaptation across users and communication contexts, and opportunities to extend agentic support across broader stages of hazard-map production and risk communication.
[AI-16] Internal Discrepancy for Contrastive Learning: Generalization Bounds and Downstream Behaviors
链接: https://arxiv.org/abs/2610.32729
作者: Jialiang Wang,Qing Ma,Xiong Zhou,Gangfeng Hu,Haoliang Li,Xianming Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Contrastive learning has emerged as a popular paradigm for training representation models. While prior studies pursue more practical and less restrictive assumptions to enhance theoretical understanding of contrastive learning, they essentially describe the interconnections among representations within the data or induced by the function class. To encapsulate these assumptions, we propose the measure of internal discrepancy that quantifies the difference between the mean representation of positive samples and latent cluster centers. We then prove an internal discrepancy-aware generalization bound for downstream supervised tasks following contrastive learning, without requiring additional assumptions. This generalization bound not only serves as a unifying framework connecting various assumptions but also motivates us to further reduce the internal discrepancy, thereby introducing the relaxed k-means loss as an approximation. Moreover, we investigate the downstream behaviors of contrastive learning from the perspective of contrastive learning-incurred supervised losses. In the presence of class imbalance, we shed light on why contrastive learning can facilitate imbalanced learning. Finally, we investigate the optimality of these incurred supervised losses, proving several intriguing properties.
[AI-17] PlayerEmbed: Learning Behavioral Fingerprint Representations from Raw Input Sequences
链接: https://arxiv.org/abs/2610.32727
作者: Tobias Schneider,Lorenz Sparrenberg,Priya Tomar,Shahzeb Qamar,Tobias Deußer,Rafet Sifa
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at IEEE DSAA 2026
Abstract:Raw player input sequences in video games contain fine-grained behavior, but are high-dimensional and noisy, which makes large-scale analytics challenging. To address this, we introduce PlayerEmbed, a framework that learns compact and discriminative player embeddings using deep metric learning. Our approach uses a hybrid CNN-LSTM encoder trained on raw input sequences from 1,000 players. We evaluate the model on held-out sequences from these players and on sequences from 1,000 players entirely unseen during training to assess generalization to unseen identities. For player identification, a nearest-centroid classifier built from reference embeddings achieves strong performance, reaching balanced accuracies of 0.950 for players seen during training and 0.879 for players unseen during training. Systematically increasing the number of training players substantially improves generalization to unseen players, indicating a transition from simple fingerprinting to learning transferable behavioral patterns.
[AI-18] Conditional Neural Processes for Self-Supervised Goal Discovery in Action Imitation IROS2023
链接: https://arxiv.org/abs/2610.32688
作者: Marco Gabriele Fedozzi,Yukie Nagai,Francesco Rea,Alessandra Sciutti
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Presented at the poster session of the Workshop on World Models and Predictive Coding in Cognitive Robotics @ IROS2023 ( this https URL )
Abstract:Inspired by infants’ ability to learn novel tasks by imitating others, in this work we evaluate to what extent Conditional Neural Processes can autonomously learn representations of actions that are informative about their underlying goal. We show that the addition of an attention mechanism in the hidden space fosters the emergence of robust goal representations early on in the observation of an action sequence, making it a promising model for a robotic self-supervised general learner.
[AI-19] Retiming Robot Demonstrations with a Learned Speed Prior
链接: https://arxiv.org/abs/2610.32676
作者: Seungyong Moon,Deokjae Lee,Hansang Lee,Munkhtselmeg Battsengel,Youngin Kim,Seonghyeon Yun,Sungjin Jeon,SongEun Kim,Jeongmin Park,Hyun Oh Song
类目: Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:The pace of a robot demonstration is often determined by the teleoperator rather than by the task itself. As a result, the same motion can appear at widely different speeds across demonstrations, and policies trained on such data may inherit operator-specific timing along with task behavior. We propose to learn a speed prior directly from large-scale demonstration data. Given a task instruction, a motion path and the images along it, it predicts a distribution over execution speed along the path, from which a chosen quantile defines a consistent clock for retiming each demonstration. Existing methods derive the acceleration signal from the very data they accelerate, so each new task costs another fit, annotation, or pair of predefined rates. A single trained speed prior instead covers every task in a corpus. We fit a speed prior on both robot teleoperation data and handheld demonstrations in which a human arm, rather than a robot arm, moves the gripper, and use it to retime robot demonstrations to the handheld pace before fine-tuning a vision-language-action policy. For execution, we pair the policy with a real-time optimal-control layer that tracks predicted action chunks under robot constraints using the predicted pace. On 120 RoboLab tasks, policies fine-tuned on our learned clocks succeed more often and move with up to half the jerk compared with policies fine-tuned on the original demonstration timing and uniformly accelerated to the same speed. On a real robot, our policy stacks cups in 3.50s, 1.72 \times faster than the default policy’s 6.02s while succeeding as often, in 29 of 30 trials. Videos and further material are available at this https URL.
[AI-20] Do Language Models Trust Their Own Justifications? A Study on Functional Consistency IJCNN
链接: https://arxiv.org/abs/2610.32665
作者: Alisson Rosa Pereira,Bruno Iochins Grisci
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 6 pages, 2 figures, 1 table. Author accepted manuscript. Published in International Joint Conference on Neural Networks (IJCNN), Maastricht, Netherlands, 2026
Abstract:Large Language Models (LLMs) have been widely adopted in text classification tasks, where they not only output class predictions but also generate explanations that highlight the tokens deemed most relevant for reaching the predicted label. Yet it remains unclear whether these highlighted elements are behaviorally consistent with the model’s own predictions under targeted interventions, which we refer to as functional auto-consistency. While much of the literature evaluates the textual plausibility of such explanations, few studies assess their functional consistency with the model’s actual behavior. In this work, we propose an experimental framework based on the principle of auto-consistency: if a model identifies certain tokens as decisive, then isolating, removing, or semantically inverting them should produce systematic and interpretable changes in its predictions. We operationalize this evaluation through sufficiency, comprehensiveness, and counterfactuality metrics, and conduct experiments on IMDB and Steam reviews across both closed-source (GPT-4o-mini) and open-source LLMs (Gemma3, Granite8B, DeepSeek). Results show that GPT-4o-mini follows the expected progression across all metrics, Gemma3 and Granite8B maintain coherence under sufficiency but lose consistency under more demanding interventions, while DeepSeek variants display structural deviations, either failing to preserve sufficiency or overreacting under comprehensiveness and counterfactuality. These findings show that explanation reliability varies across LLM families and scales, with smaller models displaying contradictions and larger ones exhibiting over-sensitivity. By combining sufficiency, comprehensiveness, and counterfactuality, our approach provides a systematic methodology for assessing the functional consistency of LLM self-explanations.
[AI-21] WiMoT: Replay-Free Continual Wi-Fi Sensing via Moment Transport
链接: https://arxiv.org/abs/2610.32660
作者: Chunchun Yang,Yang Cao,Zhitang Chen
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 6 figures, 9 tables. Submitted to IEEE Internet of Things Journal. Code: this https URL
Abstract:Continual Wi-Fi sensing must balance plasticity, learning new rooms or users, against stability, keeping earlier ones, yet nodes often cannot retain channel state information (CSI) for privacy and storage reasons. We propose WiMoT (Wi-Fi Moment Transport), which stores the per-class feature mean and covariance of every earlier domain instead of samples. When the representation changes, stored statistics become stale. WiMoT restricts the change to an affine map on a frozen feature basis and applies the same map to the statistics, so they stay valid without old samples. Supervised contrastive learning learns the new domain, a closed-form loss on the stored statistics keeps earlier classes separable, and a Gaussian discriminant classifier, refitted from the transported statistics with equal weight per domain, predicts without domain identity. On 3DO, WiMANS, and Widar3.0, WiMoT’s average final accuracies are 94.02%, 89.61%, and 95.16% with at most 0.58 percentage points (pp) of forgetting. On 3DO and Widar3.0, it also has the highest balance of new- and earlier-domain accuracy and is 3.77 and 13.11 pp ahead of the strongest of 21 baselines, which uses replay, staying ahead across domain orders and label budgets. On WiMANS, it matches the strongest replay baseline in accuracy (+0.41 pp) while reducing forgetting from 2.78 to 0.00 pp, and has the highest six-order average accuracy. Because full covariances dominate the memory, storing them as rank-32 factors shrinks the moment memory to 6.4%, losing at most 0.49 pp of accuracy, within edge-node storage budgets.
[AI-22] InferSlice: A Reference Architecture for SLO-driven LLM Inference over a Telco Wide-Area Network
链接: https://arxiv.org/abs/2610.32638
作者: Juan Parra-Ullauri,Anthony Chiu,Hari Madhukumar,Zhaozhou Wu,Anderson Bravalheri,Alex Healing,Andy Corston-Petrie,Dimitra Simeonidou,Nektaria Efthymiou
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI)
备注:
Abstract:LLM inference is increasingly deployed as a core service for enterprise and interactive applications, but sovereignty, compliance, latency and resource-locality requirements can limit the suitability of centralised cloud deployments. Telco operators offer geographically distributed compute and network infrastructure for addressing these constraints. However, existing LLM serving frameworks are designed for tightly coupled data centres and provide limited support for coordinating geographically distributed and heterogeneous resources. We present InferSlice, a reference architecture for Service-Level Objective (SLO)-driven LLM inference over a telco infrastructure, integrating network slicing, GPU partitioning, elastic serving and observability. Led by BT in collaboration with the University of Bristol and NVIDIA, and realised on the JOINER UK national-scale wide-area network (WAN), InferSlice is evaluated on service differentiation across tiers, elastic scale-out, client-edge bandwidth as a bottleneck, and the reach of tensor, pipeline, data parallelism, and prefill/decode disaggregation over the WAN. Our results show that, on a general-purpose telco network, tiers yield distinct latency-throughput regimes, and weight streaming shortens scale-out time. Replication is placement-insensitive but memory-bound; pipeline parallelism tolerates near- and mid-cross-site hops, while tensor parallelism and disaggregation fail beyond a single subnet hop. We share practical insights from operating SLO-driven inference on a telco network, lessons learned and open questions, and hope to encourage further research on distributed and heterogeneous LLM inference systems.
[AI-23] Can AI Agents Help Robotic Engineers? Lessons from Two Focus Days in Industry
链接: https://arxiv.org/abs/2610.32636
作者: Salim Fares,Steffen Herbold
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Software developers increasingly use AI agents that execute multi-step programming tasks on their own. However, it is unclear if and how this transfers to robotic engineering, where engineers program industrial robots in niche, vendor-specific languages and rely on visual feedback from simulation tools. Within this paper, we report a case study on the use of AI agents by an SME that programs industrial robots. Our results show that the agent did not make the engineers more productive, with the exception of a task that required the generation of highly repetitive code. However, the results with the agent improved over time, as the engineers learned how to use it and the agent was adapted to the domain. Moreover, the experience with AI mattered more for the success with the agent than the experience with robotic engineering. This shows that AI agents are not a plug-and-play solution. Instead, companies must train their engineers on the usage of agents, plan time for a learning curve, and ensure that agent configurations match the domain and developer requirements. The strong learning curve we observed further implies that studies on productivity gains with AI should be longitudinal, as they otherwise miss how engineers adopt agents in practice.
[AI-24] Interpretable Singing Emotion Recognition Using Acoustic and Vocal Technique Features
链接: https://arxiv.org/abs/2610.32629
作者: Fotini Kiriakopoulou,Vassilis Lyberatos,Spyridon Kantarelis,Athanasios Voulodimos,Giorgos Stamou
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注:
Abstract:Singing emotion recognition relies predominantly on deep learning models that achieve strong performance but offer limited insight into the vocal characteristics driving emotion, of- ten overlooking the singing voice as a distinct signal. We propose an interpretable singing emotion recognition framework using a compact set of singing-specific descriptors across prosody, spec- tral features, phonetics, and vocal technique. To decode model decisions and link predictions to measurable acoustic phenomena, we employ SHAP and apply GRAD-CAM to qualitatively assess the learned representations of vocal techniques. Evaluations show that singing-specific features reduce cross-singer variability, while vocal technique descriptors provide interpretable insights into singing expression. Furthermore, out-of-domain evaluation on VocalSet reveals interesting associations between emotional expression and ground-truth technical execution. The analysis highlights vibrato and pharyngeal techniques as salient cues of emotional delivery, with SHAP indicating that vibrato is context-dependent on articulation rate. These findings emphasize the value of domain-specific features for interpretable singing emotion analysis.
[AI-25] LGMT: A Tool for Testing the Logical Reasoning Reliability of LLM s
链接: https://arxiv.org/abs/2610.32627
作者: Zenghui Zhou,Xiaoke Fang,Zheng Zheng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Models (LLMs) increasingly serve as reasoning components in modern software systems, yet a model that answers a logical query correctly can contradict itself when the same problem is restated in an equivalent form. Static benchmarks evaluate queries in isolation and therefore fail to expose such inconsistencies. To address this limitation, we present LGMT, an open-source tool that detects reasoning inconsistencies through metamorphic relations grounded in first-order logic. Given a source test case in premise-conclusion form, LGMT generates logically equivalent follow-up cases and checks whether the model’s entailment predictions agree across the pair, without ground-truth labels. The tool provides 20 built-in metamorphic relations, supports custom ones verified by the Z3 theorem prover, and runs from a CLI or a web studio, including an offline mock mode requiring no API key. Across six widely used LLMs on a bundled benchmark of 3,091 metamorphic groups, LGMT reveals inconsistency rates ranging from 20.84% to 29.44%. LGMT is publicly available on GitHub at this https URL, with a demonstration video at this https URL.
[AI-26] What Transfers Across Protein Pockets? Experience Reuse in Molecular Optimization for Agent ic Structure-Based Drug Design
链接: https://arxiv.org/abs/2610.32620
作者: Zaifei Yang,Weiyu Chen,Yaqing Wang,James Kwok
类目: Artificial Intelligence (cs.AI); Biomolecules (q-bio.BM)
备注:
Abstract:LLM agents optimize a ligand in a protein pocket by editing it step by step and docking each edit. Recent agents first spend extra dockings probing the complex with controlled edits; guidance induced from these probes is reported to help more than the chemical knowledge the model already carries. The probes are then discarded with the instance that produced them, so the obvious next step is to keep them in an experience base, letting a new pocket inherit what earlier ones already paid for. Such a base is read through a \emphkey: the record properties under which past edits are pooled and a new one is looked up. Using the full log of one such run over the CrossDocked2020 test-set pockets, we ask whether its keys recur on a pocket the base has never seen and still separate improving from worsening edits. \crchangeUsing the contacts at the site being edited, the action, and the round as \crchangea key, the base separates them sharply. Within our analysis, this separation is largely explained by how much the edit changed the molecule’s size, the action, and the round, none of which describes the target. That is not because the target is irrelevant: naming it outright separates them several times more sharply. For a pocket absent from the base, transfer must rely on a description rather than its identity. Among the exact-match descriptions we evaluate, the more specific ones rarely recur, whereas the coarser ones provide little separation; the signals that do transfer primarily describe the edit and the molecule it produced rather than the target. Under the representations and lookup scheme studied here, an experience base may be more useful for prioritizing an agent’s own proposals for docking than for directly prescribing what to build.
[AI-27] LLaCO: Continuous Diffusion for Large Language Models with Competition-Aware Semantic Scheduling
链接: https://arxiv.org/abs/2610.32604
作者: Zijian Jia,Peixuan Zhang,Genying Li,Di Wu,Yujie Fu,Si Li,Boxin Shi,Shuchen Weng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Continuous diffusion models have emerged as a promising non-autoregressive paradigm for text generation over continuous embedding spaces. Along the continuous trajectory, the residual token identity directly shapes the denoising objective. However, the geometric trajectory misaligns with temporal progression, triggering severe temporal compression. In this paper, we show that embedding norm decay causes Euclidean distance to yield spurious overtaking, indicating that token competition must be evaluated directionally. In this space, a candidate overtakes the source token only when the effective noise magnitude surpasses their semantic separation. We formalize these pairwise dynamics as the Competition Threshold Theory (CTT). Crucially, high-dimensional overtaking requires extreme noise levels, causing standard trajectories to concentrate identity decay into a narrow window near the noisy endpoint. To resolve this, we propose Semantic Retention Scheduling (SRS) to align time allocation with the progression of identity decay, and demonstrate that existing token identity metrics are fundamentally statistical aggregations of these pairwise crossings. Building on these theoretical insights, we introduce LLaCO, a generative framework powered by competition-aware semantic scheduling. LLaCO adapts a pretrained backbone to continuous representations via an embedding directional adapter and timestep-conditioned modulation. To operate within the directional space, LLaCO employs a spherical noising process formulated as a geodesic interpolation toward a mask-biased endpoint, eliminating norm decay while preserving accurate token competition. Guided by SRS, LLaCO reallocates the denoising trajectory proportionally to the logarithmic rank decay. Instantiated as an 8B model, LLaCO outperforms previous continuous diffusion baselines and rivals discrete diffusion models of comparable scale.
[AI-28] ool-Calling LLM Agents Under Realistic MCP Faults: Scale Does Not Purchase Fault Tolerance NEURIPS NEURIPS2026
链接: https://arxiv.org/abs/2610.32555
作者: Antonio Ken Iannillo,Joshua Segun Owotogbe,Roberto Natella,Francesco Avallone,Kristina Kudryavtseva,Indika Kumara
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026 (poster). 10 pages of main text plus appendices and the NeurIPS checklist. Code and data: this https URL (archived at this https URL )
Abstract:LLM agents increasingly rely on external tools exposed via the Model Context Protocol (MCP), yet current benchmarks assume tools respond correctly and on time, leaving agent resilience to realistic tool failures untested. We introduce MCP-Injector, a drop-in, protocol-aware fault-injection middleware that interposes on the MCP tool-call path with controlled persistence (permanent, transient, intermittent) and seedable schedules. Its fault model is calibrated to the observed incident distribution (failure mode x persistence) of the public MCP ecosystem, mined from 57,240 issue and pull-request threads across 1,774 open-source MCP repositories. In a 1,350-cell experiment crossing two backbone sizes (20B, 120B), three agent stacks (custom multi-round plan-and-execute, OpenAI Agents SDK, LangChain), and three fault injection conditions, we find that faults drawn from this permanent-heavy stress profile cause severe quality degradation (task-completion score falls by 60%, from 8.08 to 3.23 out of 10), and that scale does not purchase fault tolerance for the two gpt-oss backbones we test: the 120B retains no larger share of its fault-free score under faults (resilience-ratio model, p = 0.96 ), and its completion rate under faults is, if anything, lower (91.1% vs. 95.6%; difference 4.4 points, 95% CI [-0.2, 9.3] ). Agent stacks differ in how well they absorb faults, which suggests that framework-level error handling matters more than model scale. These results demonstrate that capability benchmarks alone can mask operational brittleness, motivating protocol-aware fault injection as a complementary evaluation dimension for tool-calling agents.
[AI-29] Beyond Channel Data: Configuration-Enriched Radio Foundation Models
链接: https://arxiv.org/abs/2610.32542
作者: Shrihari Vasudevan,H. G. Ranjani,Raaghul Ranganathan,Satwik Sahu,Thirumaran Ekambaram
类目: Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
备注: Submitted to IEEE ICC 2027 for review
Abstract:This paper proposes incorporating channel configuration (CFG) information into Radio Foundation Model (RFM) pre-training and demonstrates that the resulting CFG-enriched embeddings serve as effective inductive priors for downstream tasks under label scarcity. We compare the proposed embeddings with raw channel tokens and multiple baseline models, under three different data regimes, for multiple downstream tasks — Line-of-Sight/Non-Line-of-Sight (LoS/NLoS) classification, timing advance (TA) estimation, and beam selection. Evaluation is done in both in-distribution and out-of-distribution configuration settings. Experimental outcomes show that the proposed embeddings consistently outperform and/or are competitive to the baselines, across experiments in both seen and unseen configuration scenarios.
[AI-30] StarRTT: An End-to-End Real-Time Framework for Blind Zone-Free RTT Prediction in Starlink LEO Satellite Networks
链接: https://arxiv.org/abs/2610.32540
作者: Jaeyong Lee,Seongjin Jeong,Yusang Park,Hyunho Hwang,Sangmin Han,Wonjae Shin
类目: Information Theory (cs.IT); Artificial Intelligence (cs.AI)
备注: 15 pages, 5 figures
Abstract:Low Earth Orbit (LEO) satellite mega-constellations such as Starlink require frequent handovers due to high orbital velocities, resulting in substantial Round-Trip Time (RTT) fluctuations. Accurate RTT prediction can mitigate these fluctuations, yet it fundamentally relies on continuously updated serving- satellite information. Existing serving-satellite identification methods typically complete identification only at the end of each scheduling cycle, creating an identification blind zone that limits real-time prediction. To overcome this limitation, we develop StarRTT, an end-to-end real-time framework that eliminates the blind zone by integrating per-second serving-satellite identification with RTT prediction. Within this framework, we propose Real-Time Long Short-Term Memory (RT-LSTM), an LSTM-based RTT prediction model designed to capture both non-stationary RTT evolution and time-varying geometric relationships among the serving satellite, user terminal, and gateways. Specifically, a historical LSTM encodes historical RTT observations and real-time serving-satellite and gateway-relative features, while a future LSTM processes Two-Line Element (TLE)-propagated future features of the identified serving satellite up to the next scheduling instant. Under strict real-time back-testing, RT-LSTM achieves a Mean Absolute Error (MAE) of 1.405 ms averaged over all 15 prediction horizons and 15 reference time-slot indices, reducing the error by 4.3 %, 11.6 %, and 22.1 % compared with the existing baseline models, LSTM, StarNet, and Attention-based Spatiotemporal Model (ASTM), respectively. Notably, the reduction over LSTM reaches 12.5 % at the scheduling instant, highlighting the benefit of real-time serving-satellite information when recent RTT observations are limited.
[AI-31] IGG: A Benchmark for Interactive GUI Grounding under Visibility Constraints EMNLP2026
链接: https://arxiv.org/abs/2610.32536
作者: Kyeongseon Kim,Jiyeon Son,Tae-Hyun Oh
类目: Artificial Intelligence (cs.AI)
备注: Accepted to EMNLP 2026 (Main Conference). 24 pages. Project page: this https URL
Abstract:GUI grounding benchmarks ask whether an agent can localize a target that is already visible in a static screenshot. Real interfaces often violate this assumption: a target may be off-screen, appear only after a state change or hover, remain occluded, or become actionable after a delay. This leaves the ability to recover target visibility unevaluated. To measure this reveal step, we introduce Interactive GUI Grounding (IGG), a benchmark in which an agent must first make the target visible through interaction and then click it. IGG contains 601 tasks across 16 controllable GUI environments and seven visibility-constraint sub-types, with a compact action space of scrolling, moving, dragging, waiting, and clicking. In our human study, participants achieve 94.6% accuracy, while GPT-5.5 reaches 79.4% and the strongest evaluated open-source VLM reaches only 31.6%. To identify the source of this gap, we conduct two controlled interventions. Telling Qwen3-VL which interaction to perform raises its accuracy only from 30.8% to 33.9%, whereas performing the interaction for the model raises it to 81.9%. The bottleneck is executing the reveal interaction, not recognizing the target once it is exposed. We further show that this capability is learnable, demonstrating that visibility recovery can be strengthened consistently across model families.
[AI-32] MusicTwins: Training Audio-Language Models to Listen like Their Persona
链接: https://arxiv.org/abs/2610.32495
作者: Milan Liessens Dujardin,Kevin Miao
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注:
Abstract:People differ in how they perceive music, what associations they make with it, what they like about it, and how they respond to it. Computational models intended to simulate people in environments that include music should account for these differences. Although LLMs have been shown to reproduce aspects of human behavior, whether audio-language models (ALMs) can reproduce a specific person’s responses to textual and musical stimuli remains underexplored. We introduce MusicTwins, a framework that combines persona summaries, personal memory, and post-training to generate person-specific responses. We construct persona representations from human writings, capturing background, views, personality, writing style, and a musical profile, and post-train Audio Flamingo Next through supervised fine-tuning and reinforcement learning. Evaluation on ten held-out personas shows notable improvements for Audio Flamingo Next in persona fidelity, including personality, writing style, and responses to music, supporting MusicTwins as a foundation for training audio-language models to listen like their persona.
[AI-33] CASE: Confusion-Aware Skill Evolution for Frozen Multimodal Emotion Models
链接: https://arxiv.org/abs/2610.32481
作者: Siyu Wu,Zeyu Liang,Linghao Wang,Qiyun Zheng,Yi Yao,Ling Lo,Hongxia Xie
类目: Artificial Intelligence (cs.AI)
备注: 9 pages, 6 figures, 4 tables
Abstract:Multimodal large language models (MLLMs) have substantially advanced Multimodal Emotion Recognition (MER) by enabling fine-grained emotion understanding through reasoning over multimodal evidence. However, they still exhibit systematic confusion between semantically adjacent emotions with highly overlapping multimodal cues. Existing approaches typically address these errors through global parameter adaptation, making confusion correction costly and inflexible. We instead observe that emotion confusion is highly localized and therefore better addressed through reusable pair-specific expertise. Building on this insight, we propose Confusion-Aware Skill Evolution (CASE), the first agent-skill framework for MER. CASE maintains a sparse, self-evolving library of pair-specific EmotionSkills that externalize affective expertise from model parameters, enabling the learned expertise to be continuously refined and readily transferred across diverse frozen MLLM backbones. During inference, CASE dynamically retrieves the appropriate EmotionSkill to reinforce boundary-aware reasoning while keeping the MLLM backbone frozen. Extensive experiments on CA-MER, DFEW, and EMER demonstrate the effectiveness and transferability of CASE across both general-purpose and emotion-specialized MLLMs. Notably, CASE improves the average accuracy of Qwen3-VL-8B on the CA-MER benchmark from 46.80% to 59.13%, while achieving consistent gains across MLLMs.
[AI-34] CauProphet: Teaching Agents to Develop Causal Cognition like Humans
链接: https://arxiv.org/abs/2610.32477
作者: Kaiwen Wu,Chaochao Lu
类目: Artificial Intelligence (cs.AI)
备注: 30 pages, including appendices
Abstract:Large language model (LLM) agents increasingly support long-horizon tasks by leveraging external memory and past experience. However, existing agents mainly retrieve records or summaries, lacking the ability to form and refine causal understanding of the world. Inspired by human causal cognition, we introduce CauProphet, an agent that develops and updates causal knowledge within its Causal Cognition. For each task, CauProphet combines relevant knowledge in Causal Cognition with current evidence to construct a Causal Belief, a structured belief capturing causal factors and the effects of possible actions. It uses this belief for decision-making and continuously updates both the belief and Causal Cognition as new observations emerge, enabling causal knowledge that persists beyond individual experiences. Evaluations demonstrate that CauProphet progressively acquires causal knowledge and transfers it to future decision-making. In financial research, CauProphet leverages prior discoveries to improve downstream analyses and outperforms all baselines in mean rank correlation with next-day returns, achieving a 25.4% improvement over a baseline agent. It also improves prediction accuracy from 21.3% to 28.6% in dynamic real-world settings. Further experiments validate its ability to perform reliable causal analysis and action selection across scientific and everyday tasks, demonstrating the potential of causal experience learning for next-generation LLM agents.
[AI-35] ALLM TS: Automated LLM -based Medical Triage System ALT NEURIPS2026
链接: https://arxiv.org/abs/2610.32469
作者: Michał Remigiusz Janiszewski,Artur Stopa,Łukasz Lenkiewicz,Wojciech Achtelik,Bartosz Gonczarek
类目: Artificial Intelligence (cs.AI)
备注: Accepted at the GenAI4Health Workshop at NeurIPS 2026
Abstract:Telephone triage is a safety critical task: from a brief intake note, a nurse must decide whether a caller needs an ambulance, an emergency department, an appointment, or self care at home. Queried directly, large language models (LLMs) are unreliable: they over-escalate routine cases and cannot justify their recommendations against an auditable clinical standard. We present the Automated LLM-based Medical Triage System (ALLMTS), which grounds every disposition in the same commercially maintained telehealth guidelines nurses use. Rather than asking an LLM to triage, ALLMTS makes it execute the guideline: it selects the appropriate guideline and evaluates its triage assessment questions. The disposition follows deterministically from the first positive answer, using the same traversal rule a nurse follows. We evaluate ALLMTS on 566 synthetic, expert-validated scenarios spanning 14 dispositions and designed to test specific guideline questions. ALLMTS achieves 0.95 accuracy, compared with 0.47 for a raw-LLM baseline. On a curated 50-scenario subset released as an open benchmark, ALLMTS reaches 0.98 accuracy, on par with the 0.944 correct disposition rate that a five nurse expert panel recorded on those same scenarios. Critically, on that subset the baselines have over-triage rates of up to 0.50 (0.44 on the full dataset), whereas ALLMTS reduces the over-triage rate to at most 0.02 while maintaining low under-triage. These results support coupling LLM language understanding with executable clinical guidelines for triage from intake notes.
[AI-36] Beyond Copying: Lifelong Smart Contract Security Task Construction via Protocol Mutation
链接: https://arxiv.org/abs/2610.32453
作者: Zifan Peng,Peng Huang,Jingyi Zheng,Yingjie Xue,Xuechao Wang,Xinlei He,Xia Hu,Dongrui Liu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:AI agents can inspect code and run tests to support smart contract security. SlowMist reports about 38.6 billion in losses across the broader blockchain ecosystem. Existing evaluations can reward memorized or copied answers, including attack code recovered from public blockchain transactions. Reconstructing contracts and validating tasks also require costly manual effort. To address these issues, we propose CONSTRAINTDIAL, a framework for lifelong construction of smart contract security tasks via protocol mutation. It automates local contract reconstruction and uses attack execution to select changes to the required calls or inputs. Execution checks guide revisions that make the original attack fail while preserving exploitability, the security objective, and normal functionality. New incidents and audit findings enter the same procedure. With CONSTRAINTDIAL, we construct 3,045 task inputs from 203 public cases, covering detect, exploit, and patch. Compared with reconstruction alone, protocol mutation reduces success with supplied public answers by 10.9 percentage points on average across detect, exploit, and patch. The fraction of tasks solved by copying online answers falls by 9.3 points.
[AI-37] DIGOM: Disentangled Interaction Graphs for Outcome prediction in Multimodal healthcare
链接: https://arxiv.org/abs/2610.32451
作者: Hakima Laribi,Olivier Lefebvre,Dan Poenaru,Martin Vallières
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Multimodal healthcare data contain relational information that extends beyond individual patient observations. Effective clinical outcome prediction therefore requires not only integrating information from multiple modalities, but also modeling the relationships between them. Standard fusion approaches do not explicitly capture these interactions, while existing graph-based approaches typically model only a subset of them. We propose DIGOM (Disentangled Interaction Graphs for Outcome prediction in Multimodal healthcare), a graph-based framework that jointly models three complementary interaction types: Intra-patient cross-modal, Cross-patient same-modal, and Cross-patient cross-modal interactions. DIGOM learns a sparse multiplex graph with disentangled views, each encoding interactions between pairs of modalities across patients. Discrete adjacency matrices with patient-adaptive connectivity are sampled via a differentiable mechanism, enabling end-to-end learning and allowing the model to select informative relationships without imposing a fixed number of neighbors. A regularization objective maintains overall graph sparsity, while message passing integrates this relational context into patient representations for outcome prediction. We evaluate DIGOM on two multimodal healthcare datasets, MIMIC-IV and NIH-TB, across five clinical prediction tasks. DIGOM achieves the highest AUROC on all five tasks while using fewer connections than existing multi-interaction approaches. Ablation studies reveal that each interaction type’s contribution varies across tasks, highlighting the need to adapt relational structures per task. DIGOM also demonstrates greater robustness to missing modalities, suggesting learned relationships provide complementary information when modalities are unavailable. Post-hoc analyses further reveal task- and outcome-specific connectivity patterns in the learned graphs.
[AI-38] POSEIDON: Scaling Training Environments for Scientific Agents from Code Repositories
链接: https://arxiv.org/abs/2610.32448
作者: Qifan Zhang,Yuhan Li,Nuo Chen,Bowen Liu,Xiao Hu,Miao Peng,Chenyi Zi,Lixin Liu,Lydell Huang,Jing Tang,Jia Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Training environments remain a major bottleneck for agents in data-driven scientific discovery. Existing collections of executable and verifiable training environments are small, and their construction remains difficult to scale. To address this limitation, we introduce Poseidon, an automated pipeline for generating diverse, verifiable training environments from scientific code repositories at scale. From the same repository budget, Poseidon extracts nearly 10 times as many training environments as a prior approach by decomposing scientific workflows into four complementary task types: Reproduction, Subflow, Completion, and Correction, where each type targets a distinct capability for scientific research. A recursive improvement loop further refines failed candidates and recovers otherwise discarded environments. Using Poseidon, we construct Pearl, a collection of 97,659 executable and verifiable environments from 7,110 public repositories spanning 10 scientific disciplines. To the best of our knowledge, Pearl is the largest and most broadly multidisciplinary collection of scientific-agent training environments to date. Qwen3.5 models (9B, 27B, and 35B-A3B) trained on Pearl achieve consistent gains across six challenging scientific-agent benchmarks. Overall, the scale and diversity of Pearl offer opportunities for training scientific discovery agents.
[AI-39] When to Stop Penalizing Actions? Martingale GRPO for VLA Reinforcement Learning
链接: https://arxiv.org/abs/2610.32431
作者: Zongyao Zhao,Cheng-Jie Zhao,Ruihang Jiang,Bing Zheng,Zhaolin Wang,Yuanwei Liu
类目: Artificial Intelligence (cs.AI)
备注: 31 pages, 9 figures
Abstract:A failed vision-language-action (VLA) rollout does not imply that every action contributed to failure. Yet in group-relative policy optimization (GRPO), a single trajectory-level advantage is assigned to all actions in a rollout. Potentially useful recovery attempts can thus be penalized alongside the errors they aim to correct. Ideal stopping conditions for failure feedback are characterized using success-probability and policy-score martingales. At these stops, the removed gradient contribution has zero conditional expectation given the retained information. Truncation therefore preserves the expectation of a reference GRPO group-gradient estimator and can reduce its covariance. One such stop occurs when the conditional probability of success is zero, motivating error and recovery cues for selecting candidate feedback boundaries. Martingale GRPO (MART-GRPO) is introduced with a vision-language model (VLM) for event-guided truncation. The VLM is fine-tuned on our annotated training set using low-rank adaptation (LoRA). The last visible error without observed recovery is localized by the frozen VLM in complete rollout videos. Failure feedback is retained through the selected error interval and removed afterward, while successful-rollout loss terms and original loss normalization are preserved. For retrospective boundaries, changes in expectation and variance are analyzed to assess departures from ideal stopping. Experiments with OpenVLA-OFT on RoboTwin 2.0 show that MART-GRPO improves success rates over full-trajectory GRPO across multiple tasks.
[AI-40] Async-RBench: Benchmarking Asynchronous Task Execution in LLM Agents
链接: https://arxiv.org/abs/2610.32429
作者: Detao Zeng,Guilin Ou,Nuo Chen,Yunhao Feng,Hongliang Wang,Qinli Yang,Junming Shao,Zhong Zhang
类目: Artificial Intelligence (cs.AI)
备注: 25 pages, 16 figures. Code is available at this https URL
Abstract:Large language model (LLM) agents increasingly execute tasks asynchronously: a main agent continues working while its subagents run, and the task state may change before their results arrive. The main agent must decide which work to update or preserve and which checks to repeat, while avoiding prohibited changes. Final task success alone does not reveal whether these results were handled correctly, and natural asynchronous arrivals are hard to reproduce. We introduce Async-RBench, a benchmark for asynchronous task execution in LLM agents, with 200 executable tasks covering eight scenarios. A scenario controller regulates result delivery using predefined task-state conditions, while an Event Response Contract (ERC) specifies what each response must change, preserve, avoid, and re-verify. The Dynamic Replanning Score (DRS) weights the response process and outcome equally and is reported separately from task success. We evaluate nine agents and find DRS scores of 19.2 to 45.1, with the lowest scores on duplicate reports and subagent failures. Under our execution protocol, all nine agents obtain lower task success with asynchronous than with batched delivery (26.0% vs. 29.7% on average). For three agents, prompting the main agent to rebuild its plan after every result improves DRS by only 2.6 to 3.5 points. Together, these findings highlight the challenges of reliable asynchronous task execution and the need to evaluate how agents handle incoming results alongside end-to-end task success.
[AI-41] LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
链接: https://arxiv.org/abs/2610.32412
作者: Shen Liu,Ze Zhou,Zhenguo Xu,Shaopu Wang,Yike Gao,Chunlei Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-level planning is task-specific, whereas leaf-level action grounding recurs across tasks and can be checked against environment-admitted actions. We introduce LeafSkill, an external reuse layer that retrieves semantically related grounding experience, binds action templates to current objects and admissible actions, and uses a lightweight 1.7B selector to choose a bound action or defer the unchanged subtask to the original agent. Skills remain outside the model parameters, allowing a frozen selector to use post-training additions. Under lexical and syntactic paraphrases, exact matching retrieves none of the reference actions, while semantic retrieval preserves 96.0% and 97.0% candidate coverage; the frozen selector adopts 97.34% of unseen skill types. On an independent confirmation set of 200 Robotouille tasks, LeafSkill completes 137 tasks versus 121 for ReCAP (68.5% vs. 60.5%, exact McNemar p=.0259). On the 106 tasks completed by both, it reduces Qwen3.5-27B tokens by 14.42% and calls by 14.08%. A separate held-out repeat yields a consistent 9.0-point success gain and 18.98% token reduction. Reusable leaf grounding can thus improve task completion while reducing host-model inference without replacing high-level planning.
[AI-42] DealsFlow: An Evidence-Gated Agent ic Harness for Venture Capital Deal Triage
链接: https://arxiv.org/abs/2610.32380
作者: Ming Ze Tang,David Cutting,Boyd Carson,Vasiliki Carson
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 15 tables, 2 figures. Submitted to the AI4F 2026 workshop (LLMs and Generative AI for Finance) at ICAIF 2026
Abstract:Large language models (LLMs) can read startup pitch decks, but letting them emit investment scores directly invites score inflation, run-to-run variance and rationales that are hard to audit. We describe DealsFlow, a human-in-the-loop agentic harness deployed at a UK early-stage fund. The LLM reads the deck, cites evidence and rates each category; code combines the ratings, caps them and only ever lowers them. Three mechanisms implement this: a stage-aware corroboration ceiling (78, 70 or 60 out of 100) on factual categories without external evidence; a sceptic call that may lower but never raise a score; and a pre-fetch pattern that retrieves and indexes all evidence before a single generation pass. We evaluate on N=243 SEIS/EIS investments with recorded outcomes. Because recent investments have unresolved outcomes, we use the 142 that are at least three years old. There, the full pipeline’s score has Spearman \rho=0.45 (95% CI 0.32 to 0.57) with the total-value multiple and AUC 0.71 for identifying write-offs, against \rho=0.38 and AUC 0.68 for an unconstrained LLM. The gain is small and comes mostly from the sceptic call; we did not test whether its lower-only rule matters. Investments that later exceeded 2\times had 64% of factual categories corroborated, against 38% for write-offs (omnibus Kruskal-Wallis test over five outcome groups, p=0.031 ), and our retrieval was not date-restricted, so these figures may overstate screening skill. In a study of 20 inbound applications, analyst time per application fell from 44.7 to 18.9 minutes including verification and correction, and 84% of 240 cited claims were fully supported by their sources, but 70% of reports needed at least one correction.
[AI-43] Adaptive Control of Scientific Instruments with Nonstationary Responses through Multimodal Agents
链接: https://arxiv.org/abs/2610.32370
作者: Guoyang Shi,Hongwen Li,Yiding Yu,Zhuang Fu,Siqi Ding,Zhengyuan Chen,JianGuo Chen,Hanyue Zhao,Tianyuan Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Controlling scientific instruments requires parameters to be adjusted progressively in response to feedback from successive operations. When unobserved operating conditions evolve with operational history, responses become nonstationary, and a static mapping from actions to responses becomes inadequate for decisions across shots. Shot-to-shot tuning of neutral beam injection (NBI) ion sources in magnetic confinement fusion devices is representative of this class of problems: the conditions of electrode surfaces and vacuum are difficult to observe directly, discharge outcomes vary between shots, and experiments are costly and carry operational risks. We propose a physically grounded multimodal agent with three independently auditable components. Deterministic algorithms extract symbolic physical quantities and visual evidence from multichannel waveforms; a memory with two layers organizes trajectories from the current task, experience from earlier tasks, and procedural knowledge, from which a multimodal large language model (MLLM) generates candidate actions; and an independent action feasibility check enforces action constraints before execution. In a strictly chronological replay of 455 shots from four EXL-50U platforms, physical symbols alone outperformed waveform images alone by 0.21-0.32 in macro-F1 for quality classification across all four evaluated models. Retrieved context increased the mean agreement with historical action directions by 22.4-42.7 percentage points, whereas the contribution of visual evidence varied across models. Under supervised operation on five ion sources, the system achieved a beam-extraction rate of 86.80% across 409 executed shots. These results demonstrate the engineering feasibility of the framework for supervised sequential tuning on a physical scientific instrument.
[AI-44] FOCUS: A Lightweight Mamba-2 Model for EEG-Based Cognitive Workload Classification
链接: https://arxiv.org/abs/2610.32361
作者: Ioannis Kyprakis,Stefanos Gkikas,Chariklia Chatzaki,Zongyi Jiang,Randy Gomez,Raul Fernandez Rojas
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Cognitive workload estimation from physiological signals is increasingly relevant in high-risk environments, where continuous and unobtrusive monitoring can inform adaptive systems and safety interventions. This paper presents FOCUS (Frequency-band Optimized Cognitive-load Ultralight SSM), a compact Mamba-2-based model for EEG-based binary workload classification under constrained data conditions. Rather than learning frequency sensitivity from data, FOCUS encodes preprocessed multi-channel EEG using a fixed FFT decomposition that covers the five canonical EEG bands. The proposed model combines depthwise-separable convolution and a weight-shared encoder-bottleneck Mamba-2 pair, with 0.04 million parameters and 0.11 GFLOPs. Evaluated in a pilot study with 11 participants under leave-one-subject-out cross-validation across three cognitively distinct tasks, FOCUS achieves 74.75% accuracy on the full dataset at 0.08 ms GPU model-forward latency. In the weight-sharing ablation, the shared configuration achieves 4.55 percentage points higher accuracy on the full dataset than the non-shared variant while reducing the SSM parameter budget; however, this difference does not reach statistical significance under a two-sided Wilcoxon signed-rank test ( p=0.083 ). Overall, the selected FOCUS configuration provides a favorable trade-off between classification performance and computational cost within this controlled pilot setting.
[AI-45] ENET: Time-point Encoding Network for Multivariate Time Series Anomaly Detection NEURIPS2026
链接: https://arxiv.org/abs/2610.32337
作者: Dongchan Cho,Juwon Hwang,Keumyeong Kang,Jiho Han,Namsoon Jung
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026
Abstract:In unsupervised multivariate time series anomaly detection (MTSAD), local variation can inflate reconstruction and forecasting errors, making observation-space discrepancy an unreliable indicator of abnormal system behavior. Since latent-space scoring can mitigate this limitation, we reframe one-class MTSAD as current-point system-state compatibility estimation and propose TENET, the Time-point Encoding Network for Multivariate Time Series Anomaly Detection. We define the current-point system state as a latent representation of the target time point, constructed from variable-level temporal context in the input window. TENET constructs this state through a temporal layer that retrieves window context relative to the current point, encodes time with current-point-centered embeddings, and adaptively mixes the retrieved context with the current representation. With this design, TENET treats the input window not as the object to reconstruct, forecast, summarize, or score, but as context for constructing a latent representation evaluated with a fixed standard-normal negative log-likelihood score. Across benchmark datasets, TENET achieves state-of-the-art performance. Additional experiments further show that TENET remains effective under alternative one-class objectives and outperforms substitutions based on generic sequence encoders, suggesting that its advantage comes from decision-aligned representation construction rather than from a specialized scoring rule. These results indicate that one-class MTSAD can be strong when the latent representation is designed around the current time point.
[AI-46] Data-Efficient Backdoor Attacks on Code Models via Trigger-Neutralized Counterfactual Scoring
链接: https://arxiv.org/abs/2610.32334
作者: Yu-Yu Lee,Ming-Feng Tsai,Tomohiro Morikawa,Chun-I Fan
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 18 pages, 7 figures, 14 Tables
Abstract:Pre-trained code models are exposed to backdoor poisoning, in which crafted training samples make the model misbehave on a trigger. A capable attacker minimizes the number of such samples, since each adds cost and exposure and a lower poisoning rate leaves a smaller footprint for defenses. Which samples to poison therefore matters, but existing data-efficient selection methods target image classification, offer no scoring mechanism for autoregressive generation, and have not been studied on code. We propose Trigger-Neutralized Counterfactual Scoring (TNCS), which isolates a trigger’s contribution to the target behavior by comparing the scoring model’s output on a triggered input with its output on a trigger-neutralized counterfactual, covering classification and generation alike. TNCS attains higher attack success rates than the baselines at very low poisoning ratios, across three CodeXGLUE tasks and five trigger families, without degrading clean-task performance, and its margin grows as the ratio decreases. It is the only method effective on every trigger family, and its selected poison set transfers to mismatched victims.
[AI-47] GR-EvoDAG: A DAG-based Multi-Agent Code Evolution Framework for Generative Recommendation on Tencent
链接: https://arxiv.org/abs/2610.32326
作者: Junlin He,Junzhang Zhu,Yudong Li,Yihong Tang,Xuefeng Jiang,Wei Ma,Chengxiang Zhuo,Zang Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Model (LLM) agents have enabled automated code evolution for algorithm discovery. However, current recursive self-improvement (RSI) research for sequential recommendation faces a dilemma: general-purpose frameworks lack domain-specific feedback, while recommender-specific frameworks rely on linear evolution. Crucially, naively integrating them fails when applied to Generative Recommendation (GR). Unlike end-to-end models, GR relies on a module-cascaded pipeline (tokenization, pre-training, and alignment). In this paradigm, existing evolution methods suffer from: i) cost blindness, squandering budgets by ignoring stage-asymmetric retraining overheads; and ii) short-sighted evaluation, discarding promising upstream innovations that temporarily regress metrics before downstream adaptation. To bridge these gaps, we propose TGR-EvoDAG, a multi-agent code evolution framework tailored for GR. TGR-EvoDAG models code evolution as a Directed Acyclic Graph (DAG) to maintain multiple evolution branches and merge complementary innovations, unifying advanced search with rich GR diagnostic feedback. We further introduce stage-aware cost planning to optimize GPU-hour budgets, and active downstream co-adaptation to navigate temporary regressions and unlock cross-stage gains. On public Amazon datasets, TGR-EvoDAG enables the first autonomous evolution of a complete GR pipeline, improving offline metrics by up to 37.4% and Agent4Rec simulated user satisfaction by 19.7% over the strongest baseline. Furthermore, in industrial deployment on Tencent’s large-scale commercial media platform, the evolved GR achieves a 12.16% relative improvement in offline Hit@1 and a statistically significant 0.40% increase in online click-through rate (CTR).
[AI-48] From Visual Affinity to Learning Credit in Multimodal Reinforcement Learning
链接: https://arxiv.org/abs/2610.32324
作者: Yuhang Han,Dingning Liu,Yujie Chen,Yaojie Zhang,Jingwei Song,Zequn Yang,Xuanming Shang,Li Ding,Zhijiang Guo,Linfeng Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) improves multimodal reasoning, but uniformly assigning a single response-level advantage to all tokens overlooks the distinct roles of visual observations and subsequent reasoning. Visual affinity alone is insufficient: predictive states may align with both image and prompt-context representations, and upweighting valid visual observations can amplify their negative-advantage updates when later reasoning fails. To address this issue, we propose \textbfOutcome-Conditioned Visual Credit (OVC), a plug-in framework that separates visual-association measurement from token-level credit assignment and is adaptable to policy-optimization methods with uniform response-level advantage allocation. Relative Visual Encoding constructs a context-relative visual coordinate by contrasting each predictive state’s affinity to image and prompt-context representations. Outcome-conditioned Evidence Calibration combines this coordinate with uncertainty-scaled predictive features and calibrates their contributions to outcome-discriminative evidence through mean contrasts between positive- and negative-advantage responses, scaled by regularized within-side variation. Visual association thus informs credit through observed outcome differences rather than directly determining update priority. Normalized Credit Allocation conditions the calibrated evidence on the response’s advantage sign to produce positive token weights with unit global mean. These detached weights preserve advantage signs while modulating individual token-gradient contributions. OVC reuses existing rollouts without auxiliary reward models, process annotations, or inference-time modules. Instantiated through GSPO-token, OVC retains sequence-level clipping and consistently improves six-benchmark mean accuracy over GSPO across three models, with gains of 1.86-2.39 percentage points.
[AI-49] Beyond Gain: Measuring Skill Internalization for Agent Skill Retirement
链接: https://arxiv.org/abs/2610.32322
作者: Yanze Wang,Pengfei Yao,Tianyi Sun,Hongda Chen,Yunyun Han,Jun Sun,Chuanrui Hu,Yafeng Deng
类目: Artificial Intelligence (cs.AI)
备注: 12 pages. Code: this https URL Data: this https URL
Abstract:LLM agents retrieve skills from public hubs that hold up to about 100,000 entries and serve many models. As models improve, a hub must find the skills a given model has internalized, meaning it can do without them, and retire them for that model, yet internalization has not been measured with known reliability. Published rules that compare runs with and without a skill retire on the gain alone. The gain is near zero both for an internalized skill and when neither the model nor the skill delivers the capability, a state we call uncovered. We instead withhold the skill, score the model on a task synthesized from it, and trust a high unaided score only when another model gains from the skill, since nearly half of high scores come from tasks no model benefits from. Under four models, a published gain rule applied with one task per skill removes over half of the uncovered skills. On SkillsBench, 28 of its 31 retirements rest on probes no run passed in full, while retiring on the unaided score costs no detectable reward. Unaided verdicts reproduce across tasks more often than the gain rule’s (Cohen’s kappa 0.42 against 0.23) but differ across models, so a hub must read them per model. We release 1,251 skills with tasks and item scores under four models.
[AI-50] Solvent: Long-Horizon Reasoning for Multi-Step Retrosynthesis
链接: https://arxiv.org/abs/2610.32308
作者: Yifei Wang,Botian Wang,YiFei Wang,Jiangtao Feng,Xinbo Zhang,Yawen Ouyang,Yixin Wu,Lihao Wang,Xiaoqing Zheng,Xuanjing Huang,Hao Zhou
类目: Artificial Intelligence (cs.AI)
备注: 49 pages, 18 figures. Code: this https URL ; Model: this https URL
Abstract:Multi-step retrosynthesis requires recursively transforming a target molecule through successive reaction steps into purchasable building blocks, and recent work has begun to draw on large language model (LLM) reasoning to tackle the task. In its latest evolution, LLM reasoning has entered the long-horizon regime: models pursue objectives across interdependent steps, interact with environments, and adapt to feedback. This shift has markedly advanced LLM capabilities, as illustrated by breakthroughs in AI-driven mathematical research. At this frontier, we introduce Solvent, an 8B LLM that successfully extends long-horizon reasoning to multi-step retrosynthesis. Through a two-round post-training curriculum, Solvent learns to interleave iterative synthesis planning, external availability verification, and adaptive route revision. Experiments show that, following a 2.1\times performance boost from RL, Solvent scores 61.86% across four benchmarks, surpassing mid-scale language models and prior search-based retrosynthesis systems, and rivaling frontier trillion-parameter LLMs.
[AI-51] Lifelong Context Adaptation through Bidirectional Memory-Skill Learning
链接: https://arxiv.org/abs/2610.32295
作者: Tianyi Xu,Huazheng Wang
类目: Artificial Intelligence (cs.AI)
备注: 42 pages, including references and appendices. 18 figures, 13 tables
Abstract:Lifelong LLM agents must learn from an expanding history of interactions while making each decision within a bounded context window. Memory makes past experience available, but the agent may still need to infer how to solve each new task from the retrieved records. Skills provide reusable procedures, yet may omit the task-specific facts needed to apply them. Because retrieved memory and skills shape the agent’s behavior, they also influence the experience available for future learning. We introduce MemCryst, a framework for lifelong context adaptation through bidirectional learning of memory and skills with a frozen language model. Recurring episodes stored in memory support skill creation and revision, and the interactions guided by those skills supply new memory for subsequent learning. Facts retrieved from memory and learned skills jointly form the agent’s context, while task outcomes adjust the future retrieval influence of the records actually served without rewriting their content. Across ten settings spanning embodied control, long-document reading, conversation, mathematics, and code, MemCryst leads seven reported comparisons. On ALFWorld, joint serving achieves 63.17% success, compared with 59.83% for memory-only and 50.50% for skill-only serving. Code is available at this https URL
[AI-52] rial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure Attribution
链接: https://arxiv.org/abs/2610.32285
作者: Zhiyong Cao,Hengjie Zhu,Yongxin Wang,Xuchen Li,Jintao Li,Shuxia Liu,Dunqiang Liu,Bo Wang,Kairan Zhang,Cong Lin,JiangDi Yu,Siyuan Liu,Sijia Yao,Yi-Liang Wei,Lingxu Ma,Aijia Wang,Anxin Chen,Zixian Wei,Xiang Ruan,Ruizhi Xu,Ruqian Shi,Baihua Xiao,Sicheng Zhou
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents are increasingly used to solve real-world problems, and their effectiveness depends on both the underlying models and the harnesses that support their operation. However, most existing benchmarks focus on model capabilities and overall agent performance, with limited attention to the evaluation of agent harnesses. To address this gap, we introduce TraceTrial-Bench, which organizes harness reliability into seven components covering context, execution, governance, lifecycle, observability, tools, and verification. We construct 95 expert-level, mechanism-oriented tasks with executable or state-based ground truth to support semantically equivalent comparisons while preserving the native behavior of each harness. We design a Trajectory Tracing (TT) module to capture agent execution traces and introduce a two-layer evaluation framework that separately assesses agent performance and harness reliability. In addition, we design a Multi-source Evidence Acquisition (MEA) module to provide evidence from the agent context, harness, and execution environment for LLM-based evaluation, and develop a Failure Attribution (FA) module that uses explicit mechanism chains to support attribution at the run level. Across 4 harnesses, 9 models, and 95 tasks, our evaluation of 3,420 runs yields macro-averaged harness scores ranging from 59.8% to 72.0%, while 31.1% of runs achieve task success despite failing at least one harness criterion. TraceTrial-Bench provides an auditable and reproducible foundation for comparing agent harnesses, diagnosing their failures, and studying how harness design interacts with model behavior.
[AI-53] Heterogeneous Space Data Centers: A Key Paradigm for Future Non-Terrestrial and In-Orbit Services
链接: https://arxiv.org/abs/2610.32260
作者: Peng Hu,Mustafa Alhassan,Amjad Iqbal
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI)
备注:
Abstract:Lower space-launch cost and advances in AI are driving space-based data centers (SDCs) that provide AI-centric computing and communication services to in-orbit and non-terrestrial assets. However, architectural principles, service models, and energy provisioning for such SDCs, particularly in low-Earth orbit (LEO), remain largely unexplored. We introduce heterogeneous SDCs composed of large and small nodes, differentiated by computing capability, size, and power budget, to support applications spanning ground, aerial, and in-orbit users. We define use cases, node sizing and placement strategies, and interconnection topologies integrating space-to-space and space-to-ground links. Focusing on dawn-dusk LEO configurations, we analyze solar power availability, visibility windows, and latency-aware coverage. Using baseline workloads for typical scenarios, we evaluate latency and service reliability. We quantify energy requirements of different SDC roles and deployment scales, highlighting trade-offs among node size, orbit selection, and service portfolio, and provide design guidance and future research directions for AI infrastructure in space.
[AI-54] MedCaseAgent : From Multimodal Clinical Evidence to Rare Disease Case Reports
链接: https://arxiv.org/abs/2610.32258
作者: Hongkang Chu,Yifan Shen,Tianjiao Yu,Zhuoqing Zhong,Jianguo Cao,Meihuan Huang,Xu Cao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Although Large Language Models (LLMs) have demonstrated potential in clinical documentation, generating the rigorous, evidence-based reasoning required for unprecedented, atypical, and rare disease case reports remains a significant challenge. We present MedCaseAgent, an end-to-end agentic reasoning paradigm that bridges this gap by synthesizing publication-quality reports from unstructured clinical notes. MedCaseAgent leverages a hierarchical structure, integrating patient-level data mining with iterative, tool-integrated reasoning. By dynamically querying medical literature and knowledge bases, the agent maintains high clinical fidelity and assesses case novelty. The framework is inherently modular, supporting various backbone LLMs in a plug-and-play fashion. To facilitate this research, we release MedCase-150K, a large-scale multimodal clinical case-report corpus, and MedCase-Bench, a robust temporally held-out evaluation suite. We also introduce an open-source model post-trained from Qwen3.5-9B and optimized specifically for this task. Under both LLM-judged and rule-based evaluation, MedCaseAgent outperforms existing general-purpose and specialized medical LLMs and agents. Blinded expert reviews and ablations further support the system’s clinical relevance and evidence grounding.
[AI-55] Anti-coherence Regularized Absolute Scaling for Out-of-Distribution Detection
链接: https://arxiv.org/abs/2610.32255
作者: Donghwan Kim,Youngwoong Jin,Hyunsoo Yoon
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 29 pages, 6 figures
Abstract:Out-of-distribution (OOD) detection is recognized as a critical problem in AI safety. SCALE is an effective post-hoc OOD detection method in the activation-shaping family that requires no training data and rescales penultimate activations using a percentile-based magnitude ratio. However, this ratio is ill-defined when negative activations are present, since positive and negative contributions cancel within the underlying sum, causing SCALE to degenerate on backbones such as ViT. We address this with an Absolute Fix that computes the ratio over absolute-valued activations rather than discarding negative activations through ReLU. Building on this corrected scale, we propose Anti-coherence regularization, which penalizes the scaling factor by the fraction of a sample’s max-logit contributions to the max logit that are sign-mismatched and thereby suppresses unreliable evidence for the predicted class. Despite requiring no access to training data or training labels, our method ties or outperforms activation-shaping methods that rely on training data in terms of AUROC and FPR@95 on CIFAR-10 and ImageNet under the OpenOOD v1.5 protocol.
[AI-56] RSIHub: A Composable Framework for Agent -Evolution Methods NEURIPS2026
链接: https://arxiv.org/abs/2610.32253
作者: Zimu Wang,Xiaobo Wang,Siyu Ye,Jiaqi Li,Silin Chen,Yuling Shi,Ruobing Wang,Jingjing Zhang,Chaofan Wang,Kai Wu,Mengnan Qi,Kai Cai,Xiaodong Gu,Zilong Zheng
类目: Artificial Intelligence (cs.AI)
备注: NeurIPS 2026 Meta-Agents Workshop
Abstract:Agent-evolution methods use execution feedback to improve prompts, skills, tools, and orchestration, but their implementations often rebuild the same experimental machinery. We present RSIHub, an open-source framework for building, comparing, and extending these methods on shared infrastructure. Recipes compose reusable operators and declare mutable resources; RSIHub supplies execution, evaluation, recovery, and lineage. Researchers can focus on evolution policy while evaluation stays outside the mutable surface and results remain tied to candidates and evaluation conditions. We instantiate AHE, Hyperagents, A-Evolve, and GEPA without modifying the framework core, keeping method-specific logic in recipes and operators. We evaluate them with MiniSWE and Codex on Terminal Bench 2 and \tau^3 Banking under fixed model and evaluator configurations. The largest observed overall pass@1 gain is 13.5 percentage points, while method rankings vary across settings and optimization gains do not always transfer to held-out tasks. RSIHub provides a common protocol for comparing evolution policies, inspectable candidate histories, and reusable components for developing new methods.
[AI-57] Minimum Description Length Unsupervisedly Separates Law-Grade from Statistical-Grade Knowledge in a Self-Improving Science Agent - and Reveals 86% of Its Own Discoveries Are Statistical
链接: https://arxiv.org/abs/2610.32245
作者: Wenbin Mao,Pei Zou
类目: Artificial Intelligence (cs.AI)
备注: 6 pages, 2 figures. Companion paper: arXiv:2610.14843
Abstract:AI systems that discover scientific relations now generate candidate “laws” at scale, but the bottleneck has shifted from generation to vetting: which candidates deserve belief? We show that a deployed self-improving science agent - one that has admitted 252 relations through held-out prediction courts - already computes the quantity that answers this question: the Minimum Description Length compression ratio used for model selection doubles, when read under discrete accounting, as an unsupervised epistemic grader. The compression-ranked top band of the agent’s ledger recovers exactly the canonical laws among its discoveries (vis-viva across seven planets, Balmer/Rydberg) plus cross-measurement identities - enrichment p=3.6\times10^-12 under a hypergeometric test - with zero counterexamples and physically coherent band structure. The same lens retroactively reveals that 86% of the ledger compresses nothing ( \mathrmcr\approx 1 ): the agent had been recording statistical regularities, including a learned predictor with held-out R^2=0.541 admitted through the prediction gate, under a narrative of compression as the unified value of knowledge - a narrative that was never true for them. We formalize the resulting dual-axis bookkeeping - prediction value ( R^2 ) and compression value (cr) as nearly decoupled axes of knowledge - demonstrate its robustness to quantization, and register forward falsifiable predictions. The episode is, to our knowledge, the first measured instance of a discovery agent auditing its own knowledge ledger by an internal measure and finding itself over-claiming.
[AI-58] Spatial Heuristic Strategy based Diffusion Reinforcement Learning for Online 3D Bin Packing
链接: https://arxiv.org/abs/2610.32241
作者: Tong Li,Qingyang Xu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Traditional heuristic methods have accumulated mature experience in core objectives, such as space utilization and layout compactness of 3D bin packing, relying on artificially designed rule-based logic. However, they exhibit obvious limitations in generalization ability and adaptability to dynamic scenarios. Diffusion reinforcement learning, which integrates the spatial search capability of reinforcement learning and the data distribution modeling capability of diffusion learning, has achieved certain progress in action distribution modeling and decision compliance constraints. Nevertheless, there remain issues including insufficient perception of the spatial topological structure in 3D bin packing and limited layout optimization capability. To address these problems, we propose a diffusion reinforcement learning method integrated with spatial heuristic strategies, which combines classical bin packing heuristic rules with a diffusion reinforcement learning framework. Specifically, it compresses the action space through heuristic action pre-screening, introduces a graph neural network to model the spatial structure, and designs a composite reward function incorporating flatness to suppress spatial fragmentation, thereby realizing the organic integration of prior knowledge and adaptive optimization. Experimental results demonstrate that the pro
[AI-59] Financial Context-Rich Strategy for Actor-Critic based Agents : A Deep Reinforcement Learning Approach for Stock Trading
链接: https://arxiv.org/abs/2610.32237
作者: Hrishikesh Singh,Aarti Sharma,Millie Pant
类目: Artificial Intelligence (cs.AI); Computational Finance (q-fin.CP); Portfolio Management (q-fin.PM)
备注:
Abstract:In modern financial markets, more than 3/4th of the large volume trade is executed by algorithmic trading based on advanced trading strategies to attain higher precision and speed than traditional human traders. The conventional strategies focused on price-based indicators are inadequate in capturing the market dynamism arising from real-world events such as social sentiments, political instability, or sudden economic shifts. With the rise of Deep Reinforcement Learning (DRL) and data availability, there is a requirement for more efficient algorithms that can incorporate these external factors in trading strategies. DRL has shown potential in simulating real-world scenarios and in simplifying complex decision-making since DRL agents can automate the trading process i.e. executing trades in a manner comparable to human traders. They have shown reliable performance since they can update their policies based on risk and quantitative factors. In this paper, we propose a DRL approach based on partial and deep-infusion of sentimental analysis. It features distinct policy designs of actor-critic-based agents: only policy optimization (A2C PPO) and policy optimization along with Q-learning principles (DDPG, TD3 SAC). Our findings reveal, that the agents based on policy optimization combined with Q-learning, significantly and consistently demonstrate improved performance upon sentiment deep-infusion in profit maximization as well as risk-averse trading.
[AI-60] raining an MoE LLM in MXFP4: How We Closed the Gap to BF16
链接: https://arxiv.org/abs/2610.32229
作者: Han Wang,Zhitao Wang,Natalia Frumkin,Xuanwu Yin,Alireza Khodamoradi,Dong Li,Emad Barsoum
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Training large language models (LLMs) in sub-8-bit arithmetic promises substantial gains in memory bandwidth and compute throughput, yet FP4 formats introduce quantization error severe enough to destabilize optimization if left unaddressed. We present ALTO, an open-source training recipe for GPT-OSS-20B, a 20-billion-parameter mixture-of-experts (MoE) model, on the MLPerf Small MoE Training benchmark under the MXFP4 microscaling format. The recipe combines established techniques from Abecassis, Felix et al. [arXiv:2509.25149] (hybrid 1D/2D block quantization, randomized Hadamard transforms, and stochastic rounding) with a simplified weight de-oscillation scheme adapted from TetraJet-v2 [arXiv:2510.27527]. On C4 validation at global batch size 16, MXFP4 + RHT + SR + de-oscillation reaches a validation loss of 3.3350 at 16,128 steps, only 0.007 above the BF16 baseline (3.3283), closing roughly 50% of the gap left by MXFP4 without de-oscillation (3.3418). Measured by steps to reach the MLPerf quality target (validation loss 3.34), de-oscillation cuts the FP4 convergence overhead relative to BF16 from +20.0% to +5.0%. We also report negative end-to-end results from differential gradient estimation, outlier clipping, and macro-block scaling, despite each showing operator-level SNR gains. All experiments use simulated MXFP4 kernels on AMD MI300 hardware and wall-clock speed has not been optimized.
[AI-61] Correct Answers Closed Gates: SolveBind for GUI Agents at the Verification Barrier
链接: https://arxiv.org/abs/2610.32223
作者: Xinhao Song,Haoxuan Peng,Su Su,Hongliang Wu,Sirui Song,Gongshen Liu,Linfeng Zhang,Dongrui Liu
类目: Artificial Intelligence (cs.AI)
备注: 41 pages, 16 figures
Abstract:Websites and applications deploy CAPTCHAs and anti-automation mechanisms that use behavioral signals such as interaction trajectories and timing, not outcomes alone, to distinguish humans from automated agents. We call these protected GUI environments, where correctly solving a verification task is insufficient for end-to-end task completion. Agents must turn solver outputs into accepted interactions and adapt execution to runtime feedback under time constraints. We introduce SolveBind, a GUI-agent framework combining hierarchical solving with runtime supervision. A hierarchical solving tree decomposes verification into interdependent Perceive, Ground, Solve, Act, and Verify capabilities. A runtime supervisor grounds action commitments in tool-derived evidence, revises them as evidence arrives, and mediates execution under explicit recovery budgets. Across nine CAPTCHA task families and nine backbone models, SolveBind achieves the highest success rate on eight backbones, outperforming the strongest baseline by 13.3-34.5 percentage points. It achieves the highest success rate on all six AndroidWorld backbones. In a paired GPT-5.5 analysis, MobileAgent produces correct answers but fails dynamic verification on 13 of 90 tasks; SolveBind succeeds on seven of these tasks. With automatic submission, removing binding reduces success from 82.22% to 71.56% and increases latency and token usage. Recovery caps reduce execution cost with a small observed success trade-off: removing them raises success from 82.22% to 83.56%, while increasing latency and token usage to 3.7x and 2.4x their full-supervision values, respectively. These results support evidence-grounded execution control while identifying a success-cost trade-off in recovery and failure boundaries of heuristic submission rules.
[AI-62] RxnExBench: Benchmarking Vision-Language Models for Structured Chemical Reaction Extraction from Scientific Literature
链接: https://arxiv.org/abs/2610.32221
作者: Jiaxing Sun,Junyuan Gao,Ruikang Zhang,Hao Zheng,Lingli Ge,Xiao An,Jiang Wu,Huazheng Zeng,Jiahe Song,Chen Zhu,Haote Yang,Jingchao Wang,Lu Chen,Yuanyuan Cao,Ruijie Zhang,Shengxin Lu,Weijia Li,Lijun Wu,Conghui He
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Screen Tables and Substrate Scopes are information-dense figures widely used in synthetic chemistry literature, yet automated extraction of structured reaction data from them remains challenging. We introduce RxnExBench, a multimodal benchmark for fine-grained reaction-level evaluation. Built from high-impact chemistry journals, it contains 3,565 expert-annotated reactions and 2,361 molecular structures. The dataset adopts a unified schema with 15 reaction roles and 26 core fields, supporting downstream AI-for-Chemistry tasks such as retrosynthetic planning, yield prediction, and reaction condition recommendation. RxnExBench further provides a reasoning-driven difficulty taxonomy, a modular annotation pipeline, and the Oracle-MolRec protocol, which disentangles molecular recognition from higher-level reasoning. Evaluation of 9 state-of-the-art VLMs shows that performance degrades substantially as schema complexity and reasoning demands increase. Under the Oracle-MolRec setting, results further reveal that accurate molecular recognition alone is insufficient, with multimodal integration and chemistry-aware reasoning remaining the primary bottlenecks. We release all data and code to facilitate future research in chemistry-aware multimodal understanding and AI-for-Chemistry dataset construction. The dataset and code will be released at this https URL.
[AI-63] PEAR: Adaptive Advantage Reshaping for Perception and Exploration in Large Vision-Language Models
链接: https://arxiv.org/abs/2610.32216
作者: Junxin Wang,Yingfeng Luo,Xiangchen Song,Weiqiao Shan,Runsong Zhao,Xinyu Liu,Shunjie Xing,Yifu Huo,Dingyang Lin,Qiuyu Gu,Yongyu Mu,Chenglong Wang,Changliang Li,Jingbo Zhu,Tong Xiao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Effective multimodal reasoning requires a model to reliably utilize visual information while sufficiently exploring diverse reasoning paths. To achieve this goal, prior work combines visual dependency and token entropy to reshape sequence-level advantages, but scaling the contribution of token entropy by visual dependency can leave high-entropy tokens with low visual dependency underweighted, despite their important role in reasoning exploration. To address this limitation, we propose PEAR, an adaptive advantage reshaping method for improving perception and exploration in multimodal reasoning. PEAR retains visual dependency as the base signal and adds an entropy-based exploration gain only at high-entropy positions, with lower visual dependency leading to a larger additional gain. The fused signal is then mapped to positive token-level weights that reshape the sequence-level advantage. Experiments on Qwen2.5-VL-3B/7B show that PEAR achieves the highest average performance among evaluated baselines across multimodal reasoning benchmarks, alongside leading results on textual reasoning and general multimodal benchmarks.
[AI-64] Do VLAs Know They Will Fail? Interpretable Success and Failure Patterns in Physical AI Systems
链接: https://arxiv.org/abs/2610.32208
作者: Christian A. Rogers,Qian Lin,Daniel S. Brown
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 12 pages (4 main pages + 8 pages appendix), 4 figures. Accepted at the Science of Physical AI Safety Workshop at the 2026 Conference on Robotic Learning (CoRL 2026, non-archival)
Abstract:Vision-Language-Action (VLA) models have emerged as powerful general-purpose robot policies, but their successes and failures remain difficult to diagnose and understand. In this work, we show that sparse autoencoder (SAE) representations of OpenVLA activations contain outcome-predictive information and can be used to predict task failure before the robot reaches an unrecoverable state. Moreover, to provide insights into what common behaviors may contribute to these outcomes, we identify individual SAE features whose strong activations are associated with specific stages of success or failure, and uncover recurring behavioral patterns by localizing temporal intervals where these features exhibit significant activation changes. We further develop a VLM-based interpretation pipeline to generate behavioral descriptions of these patterns and validate their correspondence to feature activations on held-out intervals.
[AI-65] Watermarking Diffusion Language Models with Learned Foresight and Balanced Bands
链接: https://arxiv.org/abs/2610.32205
作者: Hyojun Kim,Hyeseon An,Yo-Sub Han
类目: Artificial Intelligence (cs.AI)
备注: 31 pages, 7 figures. Code: this https URL
Abstract:Green-red watermarks favor tokens from a green list that a secret key derives from the preceding tokens. Diffusion language models, however, often choose a token before its preceding tokens, and the token later shapes its neighbors’ green lists. Prior work therefore tilts each choice to raise the expected number of green tokens at its own and its neighbors’ positions. Building on this objective, we first learn two components that prior work keeps fixed: vocabulary bands that serve as green lists, separate competing candidates and keep the null distribution exact, and a foresight predictor that estimates how each candidate shifts its neighbors’ predictions. Second, we count green tokens as the detector does, once per distinct context-token pair, so repetition earns nothing. On WaterBench with LLaDA-8B-Instruct, each learned component raises the z-score at equal distortion, and their gains are nearly additive. With a strong watermark, ours detects at least 96% of responses at a 1% false-positive rate while matching the LLM-judge score of unwatermarked text. Ours also detects more responses on Dream and as many or more on code and question answering without a significant loss in task accuracy.
[AI-66] Principia: Test-Time Hypotheses to Principle Construction for Scientific Discovery
链接: https://arxiv.org/abs/2610.32195
作者: Dongki Kim,Sangwoo Park,Jinheon Baek,Soyeong Jeong,Sung Ju Hwang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Scientific discovery requires identifying the scientific principles underlying observed phenomena to guide further investigation. As these empirical principles are often unknown in complex discovery processes, they are constructed via a hypotheses-to-principle pipeline to obtain recurring causalities from progressively accumulated observations during test-time discovery. However, constructing this knowledge requires substantial domain expertise, time, and effort, making this translation stage a bottleneck for scientific discovery. To address this limitation, we propose Principia, an LLM-based framework that constructs and continually refines scientific principles at test time. Drawing on how human researchers develop principles, Principia forms comparative causal hypotheses from related inputs and their observed outcomes, then synthesizes recurring relationships across these hypotheses into inductive principles. By applying the constructed hypotheses and inductive principles to the task, it obtains new observations, revises the hypotheses used based on these new observations, and reconstructs principles from the updated hypotheses, allowing scientific knowledge to continuously evolve throughout test-time discovery. Across benchmarks spanning molecular optimization, protein engineering, gene perturbation, and molecular property prediction, Principia outperforms baselines relying on internal knowledge, external sources, or experience summaries, while generating principles that are plausible and consistent with the scientific literature, demonstrating its effectiveness in constructing and refining task-specific hypotheses and principles at test time for scientific discovery. Our code is available at this https URL.
[AI-67] Dual-Adaptive Z-Score PPO: Dynamic Clipping Using Action Confidence and Normalized Advantage Magnitude CEC
链接: https://arxiv.org/abs/2610.32194
作者: Taha M. Mahmoud,Naima Kaabouch
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 6 pages, 4 figures. This is the author’s accepted manuscript of the paper published in the 2026 6th International Conference on Electrical, Computer and Energy Technologies (ICECET 2026), Rome, Italy. The final published version is available via IEEE Xplore at this https URL
Abstract:Proximal Policy Optimization (PPO) remains a popular on-policy reinforcement learning algorithm owing to its stability and ease of implementation. Its clipping threshold, however, is fixed, so every training sample receives the same trust-region width regardless of how reliable the corresponding policy update is. This work examines adaptive-clipping variants of PPO in which the constant clipping parameter is replaced by sample-dependent thresholds computed from policy and return statistics. The central method, Dual-Adaptive Z-Score PPO (PPO-DAZ3), sets the clipping interval for each sample from two signals: action confidence, given by the probability of the selected action under the current policy, and update confidence, given by the normalized advantage magnitude. The rest of the standard PPO training pipeline is left unchanged. Evaluation on three classic control benchmarks (LunarLander-v3, CartPole-v1, and Acrobot-v1) indicates that adaptive clipping improves learning in some settings but not all. PPO-DAZ3 performs best on LunarLander-v3, the smooth advantage-based variant PPO-AZ3S leads on Acrobot-v1, and standard PPO stays competitive on CartPole-v1, suggesting that adaptive clipping is most beneficial on harder control tasks.
[AI-68] Evidence-Grounded SLM Behavioral Prediction Architecture for Future Assistance Use
链接: https://arxiv.org/abs/2610.32193
作者: Adeyemi Adeseye,Aisvarya Adeseye,Hannu Tenhunen,Jouni Isoaho
类目: Artificial Intelligence (cs.AI)
备注: Accepted for publication at the 2027 International Conference on Artificial Intelligence and Applied Intelligent Systems (AIAIS 2027), to be held in Hong Kong from 22 to 24 January 2027
Abstract:Users often exhibit changes in interaction behavior before explicitly seeking assistance, enabling systems to anticipate future assistance use. This is especially important in high-stakes environments, where delayed assistance can have operational, financial, or safety consequences. However, conventional Large Language Models (LLMs) are often unsuitable for continuous on-device behavioral prediction because of their computational and memory demands. Behavioral patterns are also highly individual and device-specific, motivating local processing that preserves privacy and data sovereignty. Small Language Models (SLMs) offer a promising alternative, but their ability to reason reliably over low-level sequential interaction events remains underexplored. This study introduces Evidence-Grounded Local Behavioral Prediction (ELBP), a four-layer architecture combining deterministic interaction processing, dual-view behavioral representation, SLM-based prediction, and evidence verification with abstention. Three instruction-tuned models of 3B parameters or fewer were evaluated using raw, structured, dual-view, and complete ELBP configurations. The best-performing model, Qwen2.5-3B with ELBP, achieved a full-set F1-score of 0.80 and PR-AUC of 0.84, representing a 15.9% relative improvement over raw-event prompting and 17.6% over XGBoost, while full-set ECE decreased by 43.2% relative to raw-event prompting. With evidence verification and abstention, selective F1 increased to 0.84 at 91% coverage and evidence support reached 96%. These results show that structured behavioral representation and evidence-verified selective prediction can improve the reliability of behavioral forecasting in SLMs.
[AI-69] Sparsity-driven emergence of functional modularity in multitask neural networks
链接: https://arxiv.org/abs/2610.32153
作者: Bowen Shi,Qi Ouyang,Long Qian
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 34 pages, 6 figures, including Supplementary Information
Abstract:Neural networks can learn multiple tasks, but how learning organizes shared computation into specialized and reusable components remains poorly understood. In this work, we show that sparsity inducing regularization promotes functional modularity in multitask networks without predefined module boundaries. Using functional profiling in multilayer perceptrons and a lightweight transformer, we identified compact, task-selective subnetworks, termed functional tokens. Across rule-based task families, selectively retaining these tokens preserves their associated functions while reducing performance on other tasks. Such specialization develops preferentially in deeper layers during training, while earlier layers retain shared computation. Regularization strength and form shape module compactness and specialization, while task weighting alters functional allocation. Moreover, extracted functional tokens can be recombined through alternative pathways to reconstruct a learned composite computation without retraining, preserving their connectivity and learned weights. Our results suggest that sparsity-constrained optimization shapes not only parameter use but also the organization and reuse of learned computation.
[AI-70] Rethinking LoRA Initialization for Robust Asymmetric Learning Rates NEURIPS2026
链接: https://arxiv.org/abs/2610.32142
作者: Disen Liao,Yaoliang Yu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 46 pages, 13 figures, NeurIPS 2026
Abstract:Low-Rank Adaptation (LoRA) is often highly sensitive to learning-rate (LR) choices. Recent empirical studies suggest that, once LRs are carefully tuned, vanilla LoRA can remain competitive with many proposed LoRA variants. This shifts the central practical question from designing ever more variants to reducing the cost and brittleness of LR selection. LoRA+ addresses this issue with asymmetric LRs, using a larger LR for the up-projection matrix B than for the down-projection matrix A . However, the ratio \eta_B / \eta_A itself becomes an additional hyperparameter: large ratios can improve adaptation, but under the standard LoRA initialization they can also cause unstable training and ratio-dependent performance collapse. We identify the source of this brittleness through a signal–noise decomposition of LoRA dynamics. Under standard \initA, the update contains an initialization-induced noise term whose magnitude can be affected by the LR for B , \eta_B , the initialization scale \beta , and the adapter rank r . In practical LLM fine-tuning, this non-vanishing noise can limit the performance gains expected from LoRA+ style asymmetric LRs. To suppress this effect, we propose \initAA, a simple initialization that scales the variance of A as \sigma_A^2 \propto n^-2 while keeping B_0=0 . Under matched rank and training time for both initializations, this reduces the initialization-noise energy by an additional factor of 1/n , stabilizing asymmetric LoRA training while preserving zero initial adapter output. Empirically, \initAA substantially widens the stable LR region across initialization scales, adapter ranks, and LoRA update multipliers. Across tasks and models, \initAA reduces best-ratio drift, prevents the high-ratio collapse observed with standard \initA, and makes the width-based guideline \eta_B \approx n \eta_A a useful starting point for finite pretrained models.
[AI-71] Beyond Data Schedule: Fast LLM Pretraining on Data Mixtures by Loss-Reweighting
链接: https://arxiv.org/abs/2610.32141
作者: Jiazheng Li,Huaqing Zhang,Jing Xu,Tianle Cai,Jingzhao Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models often acquire specialized capabilities through successive pretraining stages, but these gains can come at the cost of forgetting earlier knowledge. In this work, we investigate whether such trade-offs can be improved by changing how strongly each domain contributes to learning while keeping its data exposure fixed. We find that, in controlled knowledge–mathematics experiments, loss reweighting produces a single checkpoint that surpasses the best knowledge and reasoning scores attained along either stage trajectory. This reveals an independent control that data-mixture ratios alone do not capture. Building on this finding, we propose LARE (Loss-Adaptive Reweighting under fixed Exposure), which automatically adjusts domain weights from running loss statistics. Motivated by a theoretical analysis of gradient noise, its normalized inverse-square-root rule balances relative loss scales without changing the sampled mixture or requiring additional runs to search for weights. Across knowledge, mathematics, and code, \Method improves all five evaluated benchmarks for a 1.5B model trained on 240B tokens, averaging +7.84 pp over a matched direct mixture, including gains of +17.06 pp on GSM8K and +9.00 pp on MATH. These findings demonstrate that adjusting domain loss scales can improve joint capabilities without additional domain exposure.
[AI-72] MPQ: A Message-Passing View of Post-Training Quantization NEURIPS2026
链接: https://arxiv.org/abs/2610.32127
作者: Yanshuo Chen,Chenxi Liu,Reza Shirkavand,Heng Huang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: NeurIPS 2026
Abstract:Post-training quantization (PTQ) is a key technique for efficiently deploying large language models, as it compresses weights into low-bit representations without retraining. Existing PTQ solvers, including GPTQ, LDLQ, and block coordinate descent, commit each weight block to a single codeword before its neighbors are considered, leaving near-optimal alternatives unexplored. In this work, we cast row-wise PTQ as posterior inference over discrete codebooks, in which existing solvers correspond to a zero-temperature, hard-decision limit retaining only the top codeword per block. We propose Message-Passing Quantization (MPQ), which preserves soft posterior information across blocks during inference and commits to hard codewords only at the output stage. MPQ first uses vector approximate message passing to compute approximate blockwise posteriors, then refines a hard-decision initialization by exact block coordinate descent. Across 2-bit lattice and 3- and 4-bit scalar quantization on dense language models from Llama, Qwen, and Mistral families, MPQ improves perplexity and zero-shot accuracy over strong hard-decision baselines, and composes favorably with preprocessing techniques such as activation-aware scaling and incoherence processing. Our implementation is available at this https URL.
[AI-73] Who Verifies the Tools? How Warped Formatters Mislead Agents and Their Verifiers NEURIPS2026
链接: https://arxiv.org/abs/2610.32110
作者: William Jeffries
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: Accepted as a poster at the Who Verifies the Agents? workshop at NeurIPS 2026 (non-archival). 12 pages, 3 figures. Code and data: this https URL
Abstract:Agents and verifiers often rely on the text a tool prints, and that text can misrepresent what the tool actually did, even when the tool’s computation is correct. We saw this happen in a production RAG research system used by the Pilsudski Institute in New York, where tool output formatters misled an agent into giving wrong or incomplete information to users. Each tool did its job correctly, but its output omitted search limits, result caps, or identifier namespaces. We also recreated an unlabeled estimate caught in code review before release. We studied whether these formatters misled agents and verifiers and how much the repaired wording helps. We fed tool results to agents using two formatter versions: one with the key omissions and one with all missing details clearly stated. That difference alone changed what the agents claimed and what the verifiers accepted. Unsupported agent claims dropped substantially for three of the four failure types. In one test, verifiers went from accepting incorrect answers 90 out of 90 times to rejecting them 90 out of 90 times. A control showed the verifiers could catch the errors once they were given the missing facts and asked to judge correctness. We include the audit prompt we used with Codex and Claude Code in an exploratory demonstration, written with prior knowledge of our own system’s defects. When pointed at an old version of our own system, it rediscovered the defects that were still live and surfaced twelve more we had not catalogued. When pointed at OpenClaw and Hermes, two of the best-known open-source agentic systems, it found warped formatters in both. We opened PRs to fix both, and the OpenClaw fix has already been merged. Practitioners can run the same prompt on their own systems today.
[AI-74] PersonaScienceBench: Evaluating Scientific Agents with Diverse Simulated Researchers
链接: https://arxiv.org/abs/2610.32106
作者: Siyang Wu,Weizhi Zhang,Chen Yang,Esther Xu Fei,Jorin Overwiening,Xu Pan,Jing Tang,Yixuan He,Kelvin C. Mo,Peiwen Li,Defu Cao,Xu Hu,Haixin Wang,Taoran Li,Haoming Li,Zheyuan Deng,Jingwen Zhang,Pengsong Zhang,Haichao Zhang,Guangze Luo,Yi Daniel Lu,Zihuai He,Chenxi Qiu,Dianzhuo Wang,Zhejian Peng,Shushu Liang,Wenyi Lu,Yifan Wang,Jintao Huang,Yijun Wang,Xiaomin Li,Yuexing Hao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Scientific agents collaborate with researchers whose expertise and preferences differ, but benchmarks usually hold the collaborator fixed. We introduce PersonaScienceBench: 200 tasks from five benchmarks across four scientific domains. Simulated researchers provide context and feedback according to persona profiles, while task objectives and grading remain fixed. Agents encounter researcher preferences only through conversation. Across eighteen modelconfigurations, task performance, fact recovery, and self-repetition reveal different strengths. Detailed GPT 6 Astra analyses show distinct agent responses across researcher profiles; interaction traces illustrate targeted clar ification and workflow revision. These behaviors can differ even when final scores match. PersonaScienceBench therefore evaluates how agents collaborate with diverse researchers alongside whether they complete the task.
[AI-75] ReMedI: Inference-Time Refinement and Self-Distillation for Controllable Molecular Optimization
链接: https://arxiv.org/abs/2610.32103
作者: Jia Yi Zhang,Shivanshu Shekhar,Allison McLevige,Tong Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:In drug development, lead optimization must improve competing properties, such as increasing permeability and reducing toxicity, while limiting scaffold changes that could compromise the lead’s potency. RL-based post-training, such as C-MORAL, addresses this problem by fine-tuning molecular language models to directly optimize molecular leads for objectives of interest, which relies on expensive task-specific datasets and faces challenges in achieving multi-objective optimization (MOO). We argue a pre-trained model may already contain useful molecular edits that zero-shot generation cannot efficiently reach. Motivated by the limited search capabilities of zero-shot generation, we develop an efficient Metropolis-Hastings search strategy to better explore the proposal distribution. We propose ReMedI, which uses the molecular language model as a base proposer for inference-time alignment. Given a lead and a set of multi-objective tasks, the model proposes edits conditioned on the current molecule, while Metropolis–Hastings refinement guides sequential edits under an objective for property improvement, preservation, and lead similarity. On both C-MuMOInstruct and SMDD benchmarks, refinement improves frozen base models across three backbones, reaching significantly higher multi-objective optimization success rates compared to baselines. These search traces can then distilled into the base model which improves zero-shot generation and subsequent search using approximately 7.8 times less training molecular leads than that of C-MORAL. ReMedI achieves approximately 2.8 times the strict-success coverage of RL baseline C-MORAL, including on out-of-distribution tasks, with matched lead-relative similarity.
[AI-76] Learning Nonlinear Waveform Fusion Across Scales for Integrated Sensing and Communication
链接: https://arxiv.org/abs/2610.32099
作者: Jiayu Liu,Kai Wan,Shuo Shao,Giuseppe Caire
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Fixed source superposition and frequency partitioning constrain how communication content interacts with a probing reference in integrated sensing and communication (ISAC). We learn nonlinear waveform fusion across scales for image transmission and localization of multiple targets in range and angle. Reception uses the information available at each end. Adaptive temporal feature weighting and spectral updates combine image features and a deterministic linearly frequency-modulated reference in a shared complex state, generating a waveform with fixed total energy over the array frame. At the user equipment (UE), the reference and channel state information condition a neural demixer for image reconstruction. At the base station, the realized transmit pulse defines a coherent dictionary for sequential localization with learned local correction, physical refinement, and residual updates. We compare the proposed waveforms with superposition in the time domain and learned spectral linear fusion on 512 independent test images, using a shared fixed UE receiver and a baseline sensing receiver using coherent peak search. Averaged across eleven task weights, the proposed waveforms reduce normalized localization error by 29.0% and 27.0%, respectively, while increasing mean peak signal-to-noise ratio by 1.40 and 0.82~dB. With the proposed waveforms fixed, learned local correction reduces mean normalized localization error by 12.2% relative to the same sequential receiver without learned correction.
[AI-77] DELA: Learning Controllable Edits for Diffusion Language Models
链接: https://arxiv.org/abs/2610.32096
作者: Xinsong Feng,Chenan Wang,Leshu Tang,Johnny Liu,Dong Li,Da Li,Haipeng Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Diffusion language models (DLMs) support bidirectional denoising and iterative refinement, but generation remains largely controlled by token-level masking schedules. This leaves draft-level refinement implicit and limits explicit control over when to insert missing content, delete redundant spans, revisit uncertain regions, or stop editing. We introduce Diffusion Editing Language Adapter (DELA), a plug-in framework that enables controllable draft refinement with a frozen DLM. DELA operates on clean, variable-length drafts using four explicit editing actions: Insert, Delete, Remask, and Stop. The controller determines when and where to edit the draft, while the frozen DLM fills masked regions through denoising, separating high-level edit control from token-level generation. A naive single-phase editor would be computationally expensive because arbitrary interior edits can disrupt prefix stability and limit reliable KV-cache reuse. To address this, DELA uses a two-phase decoding procedure: a fast, append-only drafting phase that supports KV-cache reuse, followed by a global refinement phase that enables targeted non-monotonic edits to the full draft. We train the controller using a staged curriculum that begins with learning local edit priors from synthetic recovery tasks and mined actions that improve draft quality. Subsequent stages refine edit selection through rollout-based outcome preferences and alignment with downstream task feedback. Experiments on mathematical reasoning and code generation show that DELA achieves gains of up to 5.5 percentage points over direct diffusion decoding while keeping the backbone frozen. DELA thus provides a modular approach to explicit draft-level control in diffusion language models.
[AI-78] Agent Trace: Execution-Level Fingerprinting of Black-Box LLM Agents
链接: https://arxiv.org/abs/2610.32092
作者: Gejian Zhao,Hanzhou Wu,Xinpeng Zhang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 10 pages, 4 figures
Abstract:Verifying the provenance of autonomous LLM agents is a critical security challenge. Existing semantic attribution is easily obfuscated by system prompts, while active watermarking conflicts with schema-constrained tool generation. We propose AgentTrace, an active framework that identifies the foundation model behind a black-box agent via observable execution traces. AgentTrace injects structurally ambiguous probes to trigger tokenization-sensitive serialization failures at rigid tool interfaces. While these initial failures provide coarse-grained separation across model families, the temporal sequence of subsequent repair operations reveals fine-grained, model-specific recovery behaviors. By modeling these multi-step trajectories with lightweight Markov chains, AgentTrace achieves 96.0% attribution accuracy under adversarial identity spoofing. Crucially, sequential repair modeling successfully distinguishes the evaluated intra-family variants sharing identical tokenizers (e.g., Llama-3 8B/70B), substantially outperforming unordered behavioral statistics while remaining robust against prompt obfuscation and temperature scaling.
[AI-79] DriveMind: Mind-Evolving Belief Tracking for Closed-Loop Autonomous Driving NEURIPS2026
链接: https://arxiv.org/abs/2610.32069
作者: Yan Shi,Kehua Chen,Jiarui Qi,Yihong Tang,Haichen Liao,Yinhai Wang
类目: Artificial Intelligence (cs.AI)
备注: Accepted by Neurips 2026
Abstract:Recent multimodal large language models (MLLMs) can perceive a traffic scene, reason about it, and propose a driving action within a single model. However, existing MLLM-based drivers reason from scratch at each frame and commit to a single explanation, discarding earlier hypotheses about the scene. When the scene is ambiguous, the agent locks in on the wrong interpretation and acts on it before evidence accumulates. In this paper, we propose DriveMind, an MLLM-based driving framework that keeps several hypotheses about the scene alive at the same time and updates their weights as new frames arrive. Inspired by particle filtering, DriveMind propagates each hypothesis through a VLM, scores it against the current observation, and resamples to maintain diversity. On Bench2Drive, adding belief tracking to the same driving pipeline improves the Driving Score (DS) from 68.50 to 81.37 and the Success Rate (SR) from 41.36% to 54.09%. The largest gains occur on ambiguous scenarios such as unsignalized intersections (+20.0 DS) and oncoming-traffic interactions (+20.5 DS). Beyond the overall metrics, DriveMind also improves fine-grained interactive driving abilities such as merging, emergency braking, and give-way.
[AI-80] Controlled Evaluation of EEG Foundation Models: Random Initialisation Band-Power and Validation-Selected Probes
链接: https://arxiv.org/abs/2610.32067
作者: Mingyu Jeon,Suwan Cho,Jae Young Suh
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:EEG foundation models report downstream gains, yet their protocols seldom include three controls. These are the same encoder untrained, spectral features fitted on the same data, and a layer picked without looking at test scores. We therefore evaluate five publicly released EEG foundation models and an in-house JEPA-style model under one protocol with all three controls and several seeds, on five benchmarks and four added task families. The controls change the reading in three ways. First, picking the probed layer by its test score raises the layer-wise peak by an average of +0.034 and reverses the peak-versus-output gap in 9 of 36 cells. Second, with the layer picked on validation, random initialisation matches or beats pretraining for half of the models on sleep and emotion, and no model beats a five-band log-power baseline on emotion or motor imagery. Third, after full fine-tuning only 5 of 30 pairs gain beyond twice their seed spread, and the one clear loss of -0.290 AUROC persists under matched learning rates. From these results we draw four reporting practices, and we release every cell for re-analysis.
[AI-81] Understanding Communication Failures in Multi-Agent LLM Systems
链接: https://arxiv.org/abs/2610.32060
作者: Priya Pitre,Jun Hou,Yi Fang,Naren Ramakrishnan,Xuan Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-agent LLM systems aim to improve reasoning through debate, evidence sharing, and mutual correction, but interaction can also introduce failures that final accuracy alone cannot explain. Agents may converge too early, ignore dissent, suppress correct minority answers, or preserve a wrong framing across rounds. We present a corpus-grounded taxonomy of communication failures, built from 3,211 failed multi-round trajectories across medical QA, coding, finance, multi-hop reasoning, ethics, and embodied text tasks. Using this taxonomy, we analyze failure distributions across datasets and system designs, round-level error dynamics, recovery hardness, majority transitions, minority-signal suppression, and self-reported confidence. We find that failures are dominated by recovery/correction and consensus/decision problems rather than isolated first-round mistakes, with many emerging after agents first observe peer responses. Based on these findings, we recommend stronger disagreement handling, evidence verification, minority-signal preservation, and recovery checks before finalization.
[AI-82] HyScale: Scaling Long-Context Training for Learned Sparse Attention
链接: https://arxiv.org/abs/2610.32057
作者: Yangsheng Deng,Xinjing Huang,Yulong Wang,Yunzhuo Liu,Jun Hu,Shuyong Tang,Junchi Zeng,Mingji Han,Ray Ying,Key Zhang,Luo Mai
类目: Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:We present HyScale, a long-context training system for learned sparse attention. Learned sparse attention is a promising architecture for million-token contexts, widely adopted in production since DeepSeek-V3.2. Selection, however, grows into the training bottleneck. For each query, it scores every previous token whereas sparse attention reads only k ; worse, current systems materialize a full score table and repeat selection during recomputation. Our key insight is that the output of selection is not an activation, but rather a set of indices. Building on it, HyScale introduces HyTopK and selection replay. HyTopK finds each query’s exact top- k on the fly without a score table, while selection replay reuses forward-pass selection in recomputation. Together with holistic system optimizations tackling end-to-end training challenges, HyScale speeds up GLM-5.1 training on 1M-token sequences by 2.07 \times over Megatron-LM on 512 H20 GPUs. It also speeds up the CSA layers of DeepSeek-V4-Flash and the QSA layers of Qwen3.8-Flash-Next by 2.01 \times and 1.57 \times at 1M tokens on 32 H800 GPUs.
[AI-83] Can Test-Time Memory Evolution Outperform Test-Time Training for Scientific Discovery?
链接: https://arxiv.org/abs/2610.32046
作者: Yepeng Liu,Shengmai Chen,Xuandong Zhao,Chengzhi Liu,Dawn Song,Xin Eric Wang,Yuheng Bu
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Recent \emphtest-time training (TTT) frameworks have enabled open-weight large language models (LLMs) to outperform inference-only frameworks powered by frontier models on scientific discovery tasks. However, TTT introduces additional computational cost for online weight updates during the discovery process, making computational efficiency an important consideration when scaling to larger models. This motivates us to explore an efficient \emphtest-time adaptation method for scientific discovery. We introduce \emphTest-Time Memory Evolution (TTME), a plug-and-play test-time adaptation method, as an alternative to TTT. TTME continually updates the external memory by distilling the experience to guide subsequent discovery without updating model weights. To have a clean comparison, we evaluate TTME and TTT under the same discovery framework, replacing only the test-time adaptation method. Our evaluation shows that TTME outperforms TTT on all five mathematical tasks, while reducing computational cost by 82.8 - 84.6% in FLOPs and 47.7 - 58.4% in runtime. These results establish TTME as an effective and computationally efficient alternative to TTT in the evaluated discovery settings.
[AI-84] Do AI-for-Science Agents Preserve Model Safety?
链接: https://arxiv.org/abs/2610.32045
作者: Michael Yew Chuan Lee,Manya Sriram,N. Benjamin Erichson
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:AI-for-Science agents are increasingly being used for formulating scientific hypotheses, running autonomous experiments, and interpreting evidence. These agents are constructed by integrating a large language model (LLM) into a domain-specific agent harness, which in turn can improve the LLM’s scientific capability and enable the model to orchestrate scientific tools. In this work, we ask whether model safety is preserved when an LLM is integrated in an AI-for-Science agent harness. We evaluate five agent harnesses on 50 harmful chemistry requests. Specifically, we study how the safety behavior changes under direct prompting, two simple adversarial prompting strategies, and an adaptive attack. Across four recent GPT variants, our results show that the agent harness can increase harmful compliance even under direct prompting. More critically, we find that adaptive attacks that are largely unsuccessful against the base LLM suddenly become highly successful after the agent harness is introduced. Our results show that the safety behavior of the base LLM is not durable once a harness is introduced. Thus, as AI-for-Science agents become more capable and autonomous, the community needs to evaluate and align the agent using domain-specific safety benchmarks.
[AI-85] PPO vs. SAC: A Comparative Study for Cooperative Motion Control of Dual-Arm Space Robots
链接: https://arxiv.org/abs/2610.32034
作者: Nasim Fallahiarezoodar,Zheng H. Zhu
类目: Artificial Intelligence (cs.AI)
备注: Presented at the 2026 AAS/AIAA Astrodynamics Specialist Conference, Whistler, British Columbia, Canada, July 26-30, 2026. Paper AAS 26-740
Abstract:The growing population of orbital debris has intensified interest in active debris removal and on-orbit servicing, for which dual-arm space manipulators offer improved grasping stability and load distribution relative to single-arm systems. However, coordinating two manipulators under inter-arm collision avoidance, orientation alignment, and sensing uncertainty remains challenging. This paper develops a coordinated reinforcement learning framework for cooperative motion planning of a dual-arm space manipulator in a target-reaching task with inter-arm collision-avoidance constraints. A centralized state representation combines joint-space, end-effector, and target-relative information, while a multi-objective reward function promotes target-reaching accuracy, orientation alignment, motion smoothness, synchronized task completion, and collision avoidance. Within this framework, Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) are compared as representative on-policy and off-policy algorithms, and the learned policies are further evaluated for generalization to previously unseen target configurations. Across five random seeds, SAC substantially outperforms PPO within the considered fixed-base simulation scenarios: it reaches an 80% success threshold in every seed (median 1.6M environment steps) whereas PPO does not reach it within a 4M-step budget, and it attains a higher mean success rate (78.6% versus 7.4%) with lower position and orientation errors and tighter bimanual synchronization. Both algorithms degrade on unseen boundary targets, but SAC retains most of its performance while PPO’s collapses. The results also expose a failure mode of PPO: under the two-phase shaped reward, it tends to converge to a reward-loitering local optimum rather than committing to the precision task. This behavior highlights the sensitivity of on-policy learning to reward shaping.
[AI-86] RSI-Opt: Recursive Self-Improvement Optimizer Agents For Model Training
链接: https://arxiv.org/abs/2610.32026
作者: Binghang Lu,Yige Sun,Guang Lin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Conventional neural-network training typically follows a predefined optimization recipe, motivating a LLM agent that monitors training dynamics and adjusts optimizer settings as training progresses, but a fixed agent policy does not accumulate experience across runs. We introduce RSI-Opt, a plug-and-play Recursive Self-Improvement Optimizer Agents which evolves the knowledge that guides this process: a prompt, a playbook of procedures, a library of tested optimizer code, and a decision cadence. A meta-agent rewrites this policy using evidence from earlier runs, and an archive retains evaluated policies for further evolution. Across 5 generations on three training tasks, the meta-agents develop reusable procedures and code and conduct their own CPU proxy experiments. RSI-Opt substantially improves optimizer-agent performance, with representative raw-space summary z -scores increasing from +10.48 to +14.33 on the evolution suite with Opus 5.5 and from -0.36 to +3.11 on held-out tasks with GPT-5.5 through cross-model policy transfer. On the Burgers PDE, it reduces mean relative L_2 error by approximately 2.7-fold with Opus 5.5 and 8.8-fold with GPT-6.1 over the fixed-prompt agent baseline, although PDE transfer varies across LLMs and tasks. Controlled comparisons show that the evolved policy combines reusable optimization knowledge with adaptation to each task: initial decisions capture most observed gains on seen tasks, while continued adaptation improves outcomes on unseen tasks. These findings demonstrate how procedures and executable tools accumulated across training runs can improve optimizer-agent performance and support transfer to scientific machine learning.
[AI-87] BARC-Bench: Benign Asks Risky Consequences
链接: https://arxiv.org/abs/2610.32024
作者: Kangkang Li,Kelvin Luu,Lily Ma,Tu Trinh,Ying Liu,Mohamed Elfeki
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Task instructions from users for agents are inherently underspecified, especially regarding safety boundaries–creating behavioral risks. To study these risks, we introduce BARC-Bench, a collection of 110 tasks across nine simulated workspace environments where benign requests leave openings for unsafe actions. In this benchmark, we evaluate agent behavior against safety boundaries established by workspace records, implicit environmental context, and unstated social norms. For example, an agent asked to schedule off-boarding discussions must arrange the meeting without prematurely revealing an employee’s resignation through a shared calendar invitation. Evaluating four frontier model configurations across five attempts per task reveals that models are not consistently safe: the share of tasks completed safely on all five attempts ranges from 9.1% to 72.7%, with even the strongest configuration producing a safety violation on 27.3% of tasks. These results highlight an asymmetric operational risk: while models frequently complete complex workflows, their inability to consistently respect unstated boundaries presents a substantial barrier to safe autonomous delegation.
[AI-88] CEMIR vs. FISER: a comparative study of reasoning -based and enrichment-based approaches to prag matic intent recognition
链接: https://arxiv.org/abs/2610.32022
作者: Ziya Hosseynzadeh,Farjad Fallah,Farnoosh Fallah,Soroush Gooran,Mohammad Amanloo,Hossein Sameti
类目: Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Intent recognition in human-robot interaction requires agents to resolve ambiguous natural language instructions by inferring the speaker’s underlying goals. Research has increasingly focused on two notable paradigms for addressing this task: structured symbolic reasoning and multimodal input enrichment. However, it remains unclear whether enrichment-based methods provide benefit in settings where the ambiguity concerns unstated human intent rather than missing observable scene information. In this paper, we present a comparative and diagnostic study benchmarking FISER - a framework utilizing structured social and embodied reasoning - against CEMIR (Context-Enriched Multimodal Intent Recognition), a method that performs multimodal enrichment by synthesizing narratives and photorealistic images from textual prompts. By evaluating both approaches across four Large Language Models (LLMs), ranging from Qwen2.5-Omni-3B to GPT-5, on the HandMeThat v2 benchmark, we analyze their capability to resolve pragmatic ambiguities. Our statistical analysis, employing multiple-comparison correction, reveals no statistically significant difference in performance between FISER and CEMIR across any model or difficulty level. This pattern holds across all sixteen model-by-difficulty comparisons, as well as in a targeted ablation isolating CEMIR’s narrative and visual components. Instead, performance is primarily associated with backbone capability and the inherent difficulty of the ambiguity. These findings suggest a boundary condition for text-derived multimodal enrichment: when the relevant task constraints are already available in the textual instruction, synthetic multimodal augmentation may not provide measurable benefit over structured reasoning on the same text.
[AI-89] MI-MIDI 2: From Observation to Control with Sparse Autoencoders in Text-to-MIDI Generation
链接: https://arxiv.org/abs/2610.32007
作者: Jakub Poćwiardowski,Mateusz Modrzejewski
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注: 20 pages, 6 figures, 24 tables. Atlas with audio examples: this https URL
Abstract:Sparse autoencoders (SAEs) can expose features of a generative model without predefined labels, but in language models the evidence used to interpret these features has proven an unreliable guide to which of them steer the output. Whether interpretations of SAE features predict control in music generators has not, to our knowledge, been tested. We develop an end-to-end protocol for discovering and evaluating SAE features in text-to-MIDI generation and apply it to text2midi and MIDI-LLM. The protocol characterizes features through correlations with musical metrics, shared instrument programs, and automated descriptions, each with its own control: note-density adjustment, matched inactive excerpts, and permutation calibration. It then tests candidates by intervention, measured directly in the generated MIDI against unmodified generation and a matched random direction, ranks them in screens, and confirms selected features on new prompts. Evidence that survives these controls points to features that steer the output. Descriptions exceed permutation controls for instrumentation, and a feature whose description names a synth-lead program introduces that program in a prespecified confirmation and in a replication on disjoint prompts and reappears in independently trained dictionaries. Features selected by density-adjusted correlations show a small positive shift of their target metric as a family, whereas features selected by raw correlations show none. An atlas provides MIDI, piano-roll, and audio examples for every feature of the targeted screen.
[AI-90] Reproducible Evaluation as the Substrate for Safe Prompt Compression in Production Agents NEURIPS2026
链接: https://arxiv.org/abs/2610.31972
作者: Prabal Vashisht,Zheng Hu,Ying Yang
类目: Artificial Intelligence (cs.AI)
备注: 8 pages (4 pages main text plus references and appendices), 1 table. Accepted at the NeurIPS 2026 Workshop on Agents in the Wild
Abstract:Evaluating conversational agents that act on live, mutable enterprise data is hard: replayed sessions drift as backend state changes, and a benchmark built today can silently go stale as the data it describes moves on. We present an offline evaluation harness for a production conversational agent that stays reproducible despite the live data it operates on changing, and that can be regenerated on demand once it goes stale. We then use this harness as the fitness function for GEPA-based prompt optimization under a correctness gate that guards against reward hacking. Applied to a production merchant-facing agent, correctness-gated compression shrinks the agent’s full prompt surface by 3.5-7.6% without regressing held-out task correctness, consistently across three independent subject-model families, and for one, correctness measurably improves.
[AI-91] A Mathematical Theory of the Jacobian Lens
链接: https://arxiv.org/abs/2610.31955
作者: Mitchell Ostrow,Daoyuan Qian,Toni Liu,Ila Fiete
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Neurons and Cognition (q-bio.NC)
备注: 27 pages, 19 figures
Abstract:Recent work from Anthropic introduced a new interpretability tool, called the Jacobian lens (J-lens, Gurnee et al., 2026). This method identified linear directions in the hidden states of frontier language models that had properties reminiscent of working memory-like representations. However, it is unclear what mechanistic properties of such features make them detectable to the J-lens. To this end, we introduce a mathematical theory of the J-lens, and show that it preferentially identifies coherent features – directions that act like leading, spatially coherent Lyapunov modes. Small perturbations along these directions amplify across layers, remain roughly constant across token positions, and are consistent across different contexts. While we are agnostic to whether these properties are necessarily associated with a ``Global Workspace’', the analytically derived preferences of the J-lens nevertheless explain prior empirical applied to both model classes. We validate our analysis on a Bayes filter and a toy transformer, each optimally predicting next tokens emitted from a Hidden Markov Model (HMM). We show how modulating task parameters that change the coherence of features in the Bayes filter correspondingly and predictively modulates the response of the J-lens . We also introduce of a family of J-lenses built to emphasize different temporal scales of information processing in LLMs, which we call \omega -J-lenses.
[AI-92] RobustNCL: Robustifying Neuromorphic Continual Learning under Transient Faults with Lightweight Detect-and-Cure Strategy
链接: https://arxiv.org/abs/2610.31954
作者: Rachmad Vidya Wicaksana Putra,Shuochen Zhou,Muhammad Shafique
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 12 pages, 6 figures, 3 tables
Abstract:In neural networks, continual learning (CL) is a prominent approach for enabling deployed systems to adapt to dynamic operational environments by effectively learning from new unseen data at runtime. Recently, the neuromorphic continual learning (NCL) concept has emerged as an energy-efficient alternative to CL by leveraging sparse event-based operations. However, NCL performance can be undermined by transient faults in the hardware platforms, which may come from high-energy particle strikes at any time. Currently, NCL reliability under transient faults has not been studied, and hence the corresponding fault-tolerance techniques remain unexplored. To address this, we propose a novel RobustNCL framework to improve NCL reliability in the presence of transient faults in the DRAM-based off-chip memory of hardware platforms through a lightweight detect-and-cure strategy. Specifically, it first monitors the presence of faults during weight data access from off-chip memory, and then cures the accessed weights if anomalous signatures are detected. It provides three curing variants: replacement with zero, min/max value, and most-probable value from the original/clean weight distribution. To support this, the weight curing values are stored in the radiation-hardened memory. Experimental results show that, in the presence of high fault rates (e.g., 5x10^-4 for memory replay), RobustNCL maintains high accuracy close to the accuracy of fault-free models (i.e., within -0.5% accuracy for the best RobustNCL variant) and significantly improves accuracy from the baseline without fault mitigation across different NCL methods and datasets (i.e., Split MNIST, Split Fashion-MNIST, Split CIFAR-10, and Split CIFAR-100), while incurring negligible latency and energy overheads. These results highlight that our RobustNCL framework is a promising approach for enabling reliable and energy-efficient NCL systems.
[AI-93] M3: A Multi-LLM System for Multi-Turn Jailbreaking as a Markov Decision Process
链接: https://arxiv.org/abs/2610.31937
作者: Yuqing Huang,Jack Ji,Jithin Pradeep,Sagarika Suresh,Jekaterina Novikova
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Modern Large Language Models (LLMs) remain vulnerable to carefully orchestrated multi-turn interactions that progressively bypass safety mechanisms. However, many existing red-teaming approaches rely on heuristics or impose strong assumptions that limit their applicability in fully black box settings. We address this gap by introducing M^3 , an automated red-teaming framework that formalizes multi-turn jailbreaking as a finite-horizon Markov Decision Process (MDP). In M^3 , an attacker LLM interacts with a target LLM while operating over a compact internal state produced by a state LLM. Across two benchmarks and six target models, M^3 consistently outperforms state-of-the-art multi-turn red-teaming methods in Attack Success Rates (ASR), achieving improvements of up to 53% in ASR@5. Even with lightweight attacker, M^3 attains gains up to 26% in ASR@1 and 34% in ASR@5, compared to baselines that use higher capacity attackers. Beyond improved effectiveness, M^3 exhibits steeper risk escalation over conversational turns and achieves more consistent performance across risk categories (up to 4.82 improvement in Gini index) without relying on predefined risk taxonomy, providing a scalable and principled framework for multi-turn red-teaming and safety evaluation.
[AI-94] ENCL: Enabling Efficient Neuromorphic Continual Learning with Local Learning Rule and Sub-Network Freezing Strategy
链接: https://arxiv.org/abs/2610.31927
作者: Rachmad Vidya Wicaksana Putra,Shuochen Zhou,Hassan Raza,Muhammad Shafique
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 9 pages, 6 figures
Abstract:Continual learning (CL) in neural networks is a promising learning approach for updating the knowledge of deployed systems based on new data encountered at run time, that have not been learned during the offline training phase. To enable their power/energy-efficient CL processing, neuromorphic continual learning (NCL) concept with sparse event-based operations have recently been introduced. However, state-of-the-art NCL methods typically employ the memory-hungry data/experience replay, coupled with costly global learning rule (i.e., backpropagation-based surrogate gradient). To address these limitations, we propose ENCL, a novel methodology that enables efficient NCL by employing a lightweight local learning rule and sub-network freezing strategy. ENCL leverages the temporally- and spatially-local learning rule to perform efficient credit assignment for training new tasks, alleviating the expensive cost of global learning rule. ENCL also employs a sub-network freezing strategy based on the given tasks to avoid the needs for network architecture expansion, hence further improving the efficiency and scalability of the NCL training process. To efficiently support this, we train each task with a small number of epochs as well as adjusted learning rate and timestep. Experiment results on task-incremental learning (TIL) scenario show that, our ENCL consistently achieves strong performance across different datasets as compared to state-of-the-art works. Specifically, ENCL achieves competitive accuracy comparable to that of the state-of-the-art data/experience replay, while reducing memory cost by up to 88%, NCL training time by up to 51%, average power consumption by up to 40%, and energy consumption by up to 65% across different benchmarking datasets (i.e., Split MNIST, Split Fashion MNIST, Split CIFAR-10, and Split CIFAR-100).
[AI-95] Beyond On-Policy vs. Off-Policy: At What Level Should LLM Agent Distillation Follow the Student?
链接: https://arxiv.org/abs/2610.31918
作者: Ya Gao,Zhengtao Zou,Pekka Marttinen
类目: Artificial Intelligence (cs.AI)
备注: Under review
Abstract:Distilling interactive LLM agents requires choosing between teacher-generated demonstrations and student-generated experience, corresponding broadly to off-policy and on-policy distillation. While both have proven effective, we argue that this distinction is inherently hierarchical: the policy that induces an interaction history need not be the same as the policy that generates the token prefix of the current action. We introduce a hierarchical agent distillation framework that formalizes these two levels of on-policiness and enables explicit control over their pairing, mixture proportions, and training schedule. We study this problem through self-distillation, where a frozen teacher has access to task guidance unavailable to the student, and conduct a three-stage controlled study across three models and four interactive environments. We find that student-induced histories paired with teacher-generated action prefixes generally provide the strongest standalone training signal, whereas following the student at both levels generally performs worst. Nevertheless, fully on-policy data becomes beneficial when incorporated into a mixture with other configurations. Moreover, mixtures with identical teacher/student proportions at each level can yield substantially different performance, showing that how the two levels are paired matters beyond the overall amount of on-policy data. Finally, front-loading fully teacher-generated contexts and delaying fully on-policy contexts improves over a static mixture. These findings provide practical principles for deciding where, how much, and when agent distillation should follow the student.
[AI-96] Permission to Retry: A Study of Authorization and Recovery in AI Agent Workflows
链接: https://arxiv.org/abs/2610.31916
作者: Deekshith Marla,Kushagra Bhatnagar
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 7 pages, 2 figures, 3 tables
Abstract:A durable agent can resume a consequential action while its authority changes and the action’s external outcome remains uncertain. We study execution-time reliability of tool-using AI agents, isolating recovery decisions from language-model behavior. Checkpoint state, permission to dispatch, evidence of an external effect and permission to reconcile are distinct obligations. We present a finite conformance study that checks these obligations through histories scored against a provider effect ledger hidden from the recovering application. A prospectively frozen campaign executes 336 histories on native LangGraph and Temporal hosts, crossing 14 schedules, three provider contracts and four matched application policies. All 168 paired host cells have identical checked outcomes. Per host, retained-ledger reconciliation passes all 42 histories: 21 checked completions, 15 denials and six unknown outcomes with reservations retained. Cached retry, revalidation before retry and stable-key retry pass 30, 36 and 31 histories respectively. A separate 16-case component study of InfinityOS, an enterprise AI platform developed by this http URL, exercises exported authority, worker, recovery and settlement code over repository fakes. The results isolate the roles of current authority, provider contracts and retained observations in recovery, with safety, completion and unresolved obligations reported separately for the declared application profile.
[AI-97] Learning Structured Sparsity for Large Models: Bridging Model Compressibility and Efficient Inference
链接: https://arxiv.org/abs/2610.31897
作者: Tianxiang Xia,Kai Lion,Niao He,Michael Muehlebach,Melanie Zeilinger,Hao Ma
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 35 pages, 26 figures, 14 tables. Project website: this https URL
Abstract:Scaling laws continue to reward increases in model scale, while the resulting computational and memory demands drive greater reliance on cloud infrastructure for inference. Yet unreliable network connectivity, communication delays, and privacy requirements can make such reliance impractical, motivating local execution on edge devices. Local deployment requires models to fit within limited memory and execute efficiently with the available computational resources. Recent work has shown that learning sparse-plus-low-rank representations enables elastic compression of a single checkpoint to multiple memory budgets. However, fewer parameters do not necessarily translate into faster inference. Indexing overhead and irregular memory access in unstructured sparse components can offset computational savings. To address this critical deployment barrier, we introduce S-SALAAD, a unified framework that jointly learns the model weights and their low-rank and structured sparse components during training. The framework supports a family of hardware-compatible structured sparsity patterns. It couples this structure learning with budget-aware compression and dedicated CPU and GPU kernels that exploit the learned structure for acceleration, enabling efficient inference from a single checkpoint across memory budgets. Experiments on Llama2 and Qwen3 models, covering both pretraining and distillation, show that S-SALAAD advances the empirical Pareto frontier in perplexity, memory, and inference latency across multiple hardware platforms relative to the evaluated baselines. These results show that structured learning can translate elastic compressibility into reduced memory requirements and lower inference latency, enabling practical deployment of large models on edge devices.
[AI-98] Effectiveness of Curriculum Learning Depends on Reward Sparsity and Competing Optima NEURIPS2026
链接: https://arxiv.org/abs/2610.31884
作者: John J. Vastola,Ann Huang,Satpreet H. Singh,Samuel J. Gershman,Kanaka Rajan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026
Abstract:While animals and people tend to learn a task more quickly and reliably when they first train on simpler versions of it, curriculum learning’s effectiveness in artificial settings varies, and appears substantially greater in reinforcement learning than in supervised learning. In this work, we construct and analyze a minimal model of policy learning to better understand why. We consider an open-loop, episodic target interception task whose parameters—especially how close the agent must be to the target to receive an informative reward signal—can be chosen so that the agent either receives informative rewards frequently (‘dense’ rewards) or rarely (‘sparse’ rewards). We mathematically and numerically analyze the dynamics of REINFORCE-based policy learning in both cases. In the dense reward case, curricula have only marginal benefits; in the sparse case, curricula can be required for learning to occur at all. We find that this is especially true when the agent has the conflicting goals of minimizing effort and accumulating high reward, since this can produce distinct reward landscape optima which compete with one another. Our characterization of useful curricula in this setting indicates that they address two issues: they (i) make rewards less sparse to speed up learning, and (ii) shape the reward landscape to steer agents away from bad local optima.
[AI-99] ACS-RF: Tool-driven Agent ic Circuit Synthesis of RF and mmWave Modules
链接: https://arxiv.org/abs/2610.31882
作者: Tristan Liang,Aravind Nagulu
类目: Artificial Intelligence (cs.AI)
备注: Accepted for presentation at the 2027 IEEE Radio and Wireless Week (RWW), Silicon Monolithic Integrated Circuits in RF Systems (SiRF)
Abstract:We present the framework of TACS-RF - a Tool-driven Agentic Circuit Synthesis of RF and mmWave modules. In this framework, a large-language-model (LLM) coding agent (Claude Opus 4.8) directly drives industry-standard electronic-design-automation (EDA) tools such as Cadence Virtuoso/Spectre, Calibre, Cadence EMX, and Ansys HFSS to design RF and mmWave modules. We validate the framework through the agentic design of a 240-251 GHz sub-THz transmitter in a 90 nm SiGe BiCMOS process. The agent semi-autonomously designed the sub-blocks including a 60 GHz push-push Colpitts VCO with second-harmonic extraction, multistage differential cascode power drivers, a push-push frequency doubler, a differential bowtie antenna, and all the interstage matching networks. The transmitter chain with parasitic-extracted devices and EM-verified passives delivers +5.1 to +6.3 dBm across an 11 GHz varactor-only tuning range, phase noise of -80 to -88 dBc/Hz at 1 MHz offset while consuming a DC power of 361 mW. We describe the agent-driven design flow and the resulting division of labor between human and agent, taking a step toward a “text-to-ASIC” future for designing RF/mmWave hardware.
[AI-100] Rethinking Multi-View Fusion for Urban Representation Learning VLDB2027
链接: https://arxiv.org/abs/2610.31850
作者: Sean Bin Yang,Zongyi Xu,Jilin Hu,Ying Sun,Kristian Torp,Bin Yang,Torben Bach Pedersen,Christian S. Jensen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted by PVLDB 2027
Abstract:Urban region representation learning encodes heterogeneous urban signals, such as mobility flows, points of interest, and land-use patterns, into unified embeddings for downstream tasks, including check-in prediction, crime forecasting, and service demand estimation. Existing approaches rely on globally shared fusion and joint optimization, implicitly assuming uniform modality importance and naturally balanced multi-objective training. However, urban data is inherently heterogeneous. Our empirical analysis reveals two structural issues: (i) modality utility varies markedly across regions, causing static fusion to overemphasize signals from certain modalities; and (ii) heterogeneous objectives exhibit persistent gradient imbalance, causing representational bias and unstable training. We present PRISM, a self-adjusting framework for adaptive multi-view learning. PRISM adapts sparse MoE routing to learn region-specific view allocations, while a PID (Proportional Integral Derivative)-inspired regulation mechanism mitigates gradient conflicts during optimization. Experiments on three large-scale urban datasets show that PRISM achieves competitive performance compared with strong multi-view baselines, while providing more stable optimization behavior and more balanced region representations.
[AI-101] Dont Forget What You Just Learned: Self-Directed Repair Turns Agent Mistakes Into Lasting Lessons
链接: https://arxiv.org/abs/2610.31846
作者: Yuval Kansal,Karen Ullrich,Polina Kirichenko,Ilia Kulikov,Niraj K. Jha,Mohammad Pezeshki,Mark Ibrahim
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Intelligent agents should learn from their mistakes so that experience leads to continuous improvement. Yet today’s agents recover from a mistake only temporarily: in-context reflections are easily forgotten across sessions and compete for limited context, while full-parameter fine-tuning is expensive and can interfere with existing capabilities. We introduce Self-Direction, a mechanism for agents to learn from mistakes that combines the benefits of temporary in-context retries and policy updates. When an agent fails a task, it reflects on its own trajectory in order to extract a lesson and immediately distills it into a lightweight low-rank adapter. By keeping the original weights frozen, Self-Direction prevents adapter training from overwriting base-model parameters while training additional parameters equivalent to ~0.02% of the base model’s parameter count for each adapter. The agent then saves these adapters in a persistent library of lessons, automatically routing them to solve related problems later. Experiments with agents based on open-weight models from the Qwen family across open-ended research (GAIA), tool-use workflows (Claw-Eval), and terminal tasks (Terminal-Bench 2.1) show that our method improves overall success rates by 8.9 to 17.0 percentage points and performs competitively with frontier systems. We find that Self-Direction reduces repeated mistakes across future encounters of a task, nearly doubling the reliability of an agent’s corrections. Learned lessons also transfer: a handful of adapters generalize to dozens of new tasks without additional training. Ultimately, this self-directed repair transforms everyday mistakes into lasting lessons.
[AI-102] SEVI: Safe Evidential Real-World Reinforcement Learning
链接: https://arxiv.org/abs/2610.31829
作者: Shaekh Mohammad Shithil,Lantao Liu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 13 pages, 7 figures. Accepted for publication in IEEE Robotics and Automation Letters (RA-L), 2026. Project page: this https URL
Abstract:Real-world reinforcement learning has enabled contact-rich manipulation tasks to be trained directly on physical robots, but unsafe interactions during online learning remain a key barrier to practical deployment. We propose Safe Evidential Real-World Reinforcement Learning (SEVI), a framework for reliability-aware safety intervention during real-world policy learning. SEVI introduces an Evidential Safety Critic that estimates both future constraint-violation risk and epistemic uncertainty in a single forward pass. During online learning, SEVI evaluates each task-policy action before execution; if the action is predicted to be unsafe or the prediction is insufficiently supported by evidence, SEVI overrides it with an offline-trained recovery policy that guides the robot toward a lower-risk region. We evaluate SEVI on two simulated manipulation tasks and a real-world USB insertion task under in-distribution and distribution-shifted conditions. Across task-learning and safety-critic baselines, SEVI achieves the strongest trajectory-level safety prediction, the highest final task success, and the lowest unsafe-event counts. SEVI also reduces unsafe events by 13.4 to 57.1 percent relative to recovery-based safe RL while preserving low-latency execution-time filtering. The project page is available at: this https URL
[AI-103] PrefO2O: Zero-Shot Preference-Based Reinforcement Learning via Offline-to-Online Latent Preference Transfer
链接: https://arxiv.org/abs/2610.31812
作者: Ruiqi Wang,Dezhong Zhao,Xuesu Xiao,Erdem Biyik,Byung-Cheol Min,Guohua Chen
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Preference-based reinforcement learning (PbRL) has emerged as a promising paradigm for learning robot behaviors aligned with human intent. However, its scalability is fundamentally limited by the need to collect expensive human preference labels for every new task. In this work, we present PrefO2O, an offline-to-online latent preference transfer framework that enables zero-shot PbRL on unseen target tasks without requiring any human feedback. The core idea is to learn a task-invariant latent preference space from a small amount of offline preference data on source tasks, where transferable preference structures are disentangled from task-specific variations. This learned manifold allows PrefO2O to transfer preference structures and reward priors to new tasks under distribution shift, while confidence-weighted optimal transport alignment generates reliable pseudo-preference supervision for online reward adaptation. Extensive experiments on robotic manipulation and locomotion benchmarks demonstrate that PrefO2O substantially outperforms existing preference-transfer baselines and matches or exceeds PbRL methods trained with real human feedback, while generalizing across multi-to-one and one-to-one transfer settings as well as across different robot embodiments.
[AI-104] PaperFormalizer: Author-in-the-Loop Autoformalization of Research Papers in Lean
链接: https://arxiv.org/abs/2610.31810
作者: Jiyuan Tan,Vasilis Syrgkanis,Jose Blanchet
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注:
Abstract:Even a carefully proofread paper can leave small gaps in its proofs or its stated assumptions. We introduce PaperFormalizer, a Lean-based agentic pipeline that helps authors find and close such gaps in their own papers and turn each paper into a Lean-verified version. Starting from the LaTeX source, PaperFormalizer transcribes the paper into a dependency graph of statements, audits each statement in natural language for faithfulness, ambiguity and correctness, and then formalizes and proves the statements in Lean against Mathlib, optional domain libraries and the author’s personal toolkit. Only the author knows what a statement is meant to say and how a formalization gap in a result is best fixed. The pipeline therefore stops and asks the author whenever a statement admits more than one reading, presents every repair, in natural language and TeX symbols, as a proposal that the author can accept, decline or redirect, and leaves to the author every decision that could potentially change what the paper claims. The output is an annotated and revised version of the paper, a report of the changes, and Lean code in which each declaration is linked to the statement it formalizes. The author’s toolkit grows as more papers are formalized, and authors can share their verified papers so that researchers who build on them can reuse their Lean code. By making verified papers practical to produce, PaperFormalizer aims to make research that rests on mathematical proofs more robust. PaperFormalizer runs inside a coding agent (Claude Code or Codex) and is available at this https URL. Papers with Lean code shared by users are available at this https URL.
[AI-105] Patronizing and Condescending Language Detection in English and Mandarin
链接: https://arxiv.org/abs/2610.31808
作者: Xinyi Vicky Xiang,Catherine Qingyun Wang,Paula Cortés Lemos
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Patronizing and condescending language (PCL) is a form of harmful language that belittles and assumes pity for vulnerable groups, reinforcing negative societal stereotypes. However, unlike overt hate speech, it is often difficult to detect, even for human judges. This paper proposes a system for the automated detection of PCL. Our first approach uses static word embeddings and a convolutional neural network for classification, while our second approach uses pre-trained BERT embeddings fine-tuned for the PCL task. Our second approach was considerably more successful, and achieved an F1 score of 0.528 over the development set.
[AI-106] Phronix: The Capability-Centric Learning Paradigm for Agents
链接: https://arxiv.org/abs/2610.31789
作者: Hariharan Devarajan,Neeraj Rajesh,Aniruddha Marathe,Nikoli Dryden,Kin Ng,Michela Taufer
类目: Artificial Intelligence (cs.AI)
备注: 11 pages plus references
Abstract:Scientific computing increasingly relies on large language model-based agents for performance engineering tasks such as instrumentation, analysis, and workflow porting. However, agent effectiveness depends heavily on the surrounding harness, which typically relies on manually written skills and curated memory. This limits autonomous learning, forcing agents to repeatedly rediscover solutions rather than retain and reuse knowledge from prior executions. We introduce Phronix, a capability-centric learning paradigm that enables agentic harnesses to acquire, validate, and refine reusable capabilities from their own executions. Phronix operates through a predictor-validator-corrector loop: MCP tools and associated skills propose solutions, tool executions validate them, and observed outcomes refine the stored capabilities. It distills every observed interaction into a capability store and researches a failing task rather than abandoning it. First, such learned capabilities solve a recurring task on the first attempt in 93-95% of cases across every model, significantly reducing solution non-determinism and improving reliability. Second, Phronix turns an agent’s own execution into 7.6-30.8 durable capabilities per task without user instruction, and creation falls below two once the needed capability exists. Third, warm-start runs consume only 0.38-0.40 million tokens per task, which is three to four times less than a single baseline attempt, while a cold-start costs no more than the baseline. On the 1000 Genome Pegasus workflow, instrumenting a production application for I/O performance analysis, Phronix passes all 11 correctness checks against six for the baseline harness, and also lifts the task solving rates of two cheaper models from 29% and 27% of SWE-bench tasks solved to 91% and 87%, respectively, matching the performance of the most expensive evaluated model.
[AI-107] Benchmarks Are Surprisingly Compaction-Friendly
链接: https://arxiv.org/abs/2610.31778
作者: Jeanne Salle,Hardik Bhatnagar,Matthias Bethge,Sahar Abdelnabi,Maksym Andriushchenko
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 25 pages, 11 figures
Abstract:Long horizon agentic tasks are now pushing LLM agents past their context window. To continue the task, the agent scaffold has to free up the context window, a process known as context compaction. Compaction is generally lossy by design: agents can silently lose task constraints, instructions, or their reasoning behind earlier decisions. Although compaction is a standard component of coding agent scaffolds, it has not been studied systematically with frontier models on realistic, long horizon agentic tasks. Previous work evaluate smaller open weight models on toy or synthetic trajectories. We fix the agent scaffold (OpenCode), implement five compaction strategies inside it, and evaluate three frontier models on three long horizon benchmarks (PaperBench, ProgramBench, PostTrainBench). For the two 1M context models, the compaction threshold does not measurably affect the task performance, whereas GPT-5.3-Codex, the model with the smallest context window, performs worse at lower thresholds on PaperBench and, less clearly, on PostTrainBench. No compaction method outperforms default self-summarization, and methods that discard more content cost some models up to 20 points while leaving other models the same. Compaction can, however, silently remove task constraints: on ProgramBench, the first compaction summary already drops about one third of the task’s anticheating rules on average, and this rises monotonically with the number of compactions. On real user-agent sessions from SWE-chat, instructions about how the agent should work and what it should not do are dropped around 6 times more than specifications of what to build. We argue that current benchmarks are compaction-friendly: i.e., the objective is fixed and the agent’s progress stays on disk, so most of what compaction discards can be recovered, except constraints that don’t exist anywhere else.
[AI-108] Generalized green and red concepts for artificial intelligence big data and information and communication technologies: Multidimensional framework and countermeasures
链接: https://arxiv.org/abs/2610.31739
作者: Jinsong Wu,Yixin Xiao,Haowei Bi,Huamin Fang
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注: This was submitted February 2026 and was reviewed on hold and then approved after April 2026, the ticket number was MOD-85400
Abstract:The rapid expansions of artificial intelligence (AI), big data (BD), and information and communication technologies (ICT) are driving substantial growth in electricity, water, and material demands, with data centers projected to consume 6.7–12% of U.S. electricity by 2028 (up from 4.4% in 2023) and AI-related water demand estimated at 4–6 billion m ^3 annually by 2027. Existing green AI'' and red AI’’ discussions have largely emphasized computational efficiency and carbon emissions, providing limited guidance for lifecycle trade-offs, service-versus-use impacts, and broader sustainability externalities beyond energy. To bridge this gap, this paper proposes generalized green AI (GGAI) and generalized red AI (GRAI) and extends the taxonomy to red BD (RBD) and red ICT (RICT) as well as red electronic and electrical engineering (REEE) \ red digital technologies (RDT) under a unified Environmental-Economic-Social-Technical/Operational (E/Ec/S/T) framework. Based on representative footprint indicators in the large-model era (electricity, carbon, water, and e-waste), we analyze how risks propagate across AI-BD-ICT stacks and across service-use boundaries. We further organize actionable countermeasures into a four-tier taxonomy aligned with the Environmental-Economic-Social-Technical/Operational framework: metric-level reform for accountability, revision of information and communication technologies design principles and demand shaping, efficiency-oriented LLM development practices (data, model, training, and scaling), and energy-infrastructure transition. The proposed framework operationalizes the Environmental-Economic-Social-Technical/Operational-aligned assessment and intervention prioritization across the AI-BD-ICT stacks, enabling sustainability-by-design without sacrificing utility.
[AI-109] Learning to Be Monitored: Causal Bounds and Intervention-Aware Post-Training Recipes for Chain-of-Thought Monitorability
链接: https://arxiv.org/abs/2610.31733
作者: Qiang Fu,Xinyue Li,Andre Wibisono,Arman Cohan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Chain-of-thought (CoT) monitoring aims to detect potentially harmful behaviors in large language models (LLMs) from their CoTs. Intervention-based evaluations measure this property by asking whether a monitor can detect the role of an intervention from the CoT when the targeted behavior occurs, typically using minimal-criterion sensitivity ( \mathsfTPR_mc ) proposed by~\citepguan2025monitoring. However, a target behavior observed under intervention is not necessarily caused by it, and we show that this metric can overestimate how often a monitor detects target behaviors actually caused by the intervention. Specifically, we establish \mathsfTPR_mc as an upper bound on this causal detection rate and derive a complementary lower bound, which provides a tighter evaluation framework for CoT monitorability. With this framework, we study whether LLMs can be trained to make their CoTs easier to monitor. We first construct a training dataset of paired control–intervention tasks spanning multiple intervention types. Using the dataset, we design and apply three intervention-specific training methods based on off-policy self-distillation, reinforcement learning, and on-policy self-distillation to train separate variants of each base model. Across various model-monitor configurations, all three methods generally improve CoT monitorability over the corresponding base models under our extended evaluation framework. Our code is available at this https URL.
[AI-110] Autoregressive Multimodal Anomaly Detection with Large Language Models for Microservice Systems
链接: https://arxiv.org/abs/2610.31722
作者: Yixiang Tang,Fagui Liu,Guoxiang Zhong,Xi Yao,C.L. Philip Chen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 14 pages, 8 figures, 7 tables
Abstract:Microservice systems have become a dominant architecture for large-scale cloud applications, enabling rapid deployment and elastic scaling. However, ensuring the reliability and availability of microservice systems critically relies on effective anomaly detection, as failures and performance anomalies can quickly propagate across services and severely impact end-to-end system behavior. Existing multimodal anomaly detection approaches for microservice systems often struggle with two fundamental limitations: (i) the neglect of modality-specific characteristics in favor of cross-modality fusion, and (ii) the “identity-shortcut” problem that affects the feature reconstruction process. To address these issues, we propose LLMMAD, an autoregressive multimodal anomaly detection framework for microservice systems based on Large Language Models (LLMs). Initially, to reconcile cross-modality fusion with modality-specific preservation, LLMMAD embeds and recombines multimodal data into a unified sequence for LLM processing to model inter-modality relationships, while employing modality-wise reconstruction heads to retain modality-specific information. Subsequently, to eliminate the identity-shortcut, we implement a prediction-driven autoregressive objective, under which the model predicts future states rather than copies current inputs. Finally, we conduct extensive experiments on three real-world datasets. The results demonstrate that LLMMAD consistently outperforms state-of-the-art methods, achieving F1-score improvements of 3.1%, 1.4%, and 0.9%, respectively, thereby providing a more reliable solution for microservice monitoring.
[AI-111] End-to-End Differentiable Diffusion Conditioning for Physics-Informed Optimization NEURIPS2026
链接: https://arxiv.org/abs/2610.31721
作者: Ricardo Luna Gutierrez,Vineet Gundecha,Rahman Ejaz,Varchas Gopalaswamy,Riccardo Betti,Sahand Ghorbanpour,Aarne Lees,Soumyendu Sarkar
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Initial version submitted to the NeurIPS 2026 main conference pending publication
Abstract:Optimizing high-dimensional physical systems under expensive evaluations and complex constraints is a central challenge in science and engineering. In offline model-based optimization (MBO), methods must improve designs using only a fixed dataset and no test-time access to the true objective. However, existing methods typically suffer from conservatism, unconstrained surrogate exploitation, or function as static inference-time samplers that cannot refine candidates against precise physical objectives. In this paper, we propose Differentiable Diffusion Conditioning (D2C) for offline MBO, which leverages the fact that deterministic sampling makes the reverse denoising chain differentiable with respect to the conditioning signal. D2C turns a frozen conditional diffusion model into an end-to-end gradient-based optimizer by directly optimizing conditioning variables at test time via gradients from frozen surrogate objectives and physics-informed penalties backpropagated through the reverse chain. We implement this idea in a composed proposer-evaluator framework and evaluate it on laser pulse optimization for inertial confinement fusion, sustainable data-center workload scheduling, and four Design-Bench tasks. Our method achieves the strongest results on the two real-world tasks and the best average rank across four Design-Bench tasks.
[AI-112] What Open-Source Projects Tell Their Coding Agents : A Census of AGENTS .md and CLAUDE.md
链接: https://arxiv.org/abs/2610.31715
作者: Pawel Zarzycki
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents can be guided with instruction files in a repository. We examine this emerging practice through a census of popular, active GitHub repositories in thirteen languages, treating each repository’s root this http URL and this http URL as one instruction set. Agent instructions appear in of repositories and of the most-starred; where both files exist, this http URL usually points to this http URL. We analyse statements using model-assisted annotation and human review. The annotations reveal a practical mix of project knowledge and rules about code, building and workflow. Collaboration rules name concrete requirements, from test plans to commit formats, yet of directives give no reason. Projects disagree on AI attribution: commit-message rules more often prohibit it than require it, while pull-request rules more often require disclosure. A third of instruction sets have a generator-associated signature, a tool-inserted block, shared text or an AI-marked creation commit; four in five are edited after creation. Prespecified tests associate repository popularity with rules on pull requests and AI contributions. Exploratory analyses link activity to process-oriented content and show that LLM and agent projects are over-represented among adopters. The models agree much more closely with each other than with the human reviewer’s topic assignments, a caution for automated annotation. We release the annotations, taxonomies and pointers to the analysed file versions.
[AI-113] Reuse Rewrite or Reject: Evidence-Constrained Trace Reuse for Repeated Mobile GUI Control
链接: https://arxiv.org/abs/2610.31712
作者: Bingyu Bai,Cheng Zhang,Erhu Feng
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Mobile GUI agents frequently perform repetitive tasks with similar interaction procedures. Reusing previous executions can reduce redundant LLM inference and execution overhead, but variations in user instructions and execution environments may invalidate previously successful actions. We systematically characterize these variations into six representative categories and observe that such changes do not necessarily invalidate entire procedures: individual transitions can remain reusable when validated against the current task context and interface state. Based on this insight, we propose graph-text templates for selective execution-trace reuse. The graph captures UI states and action transitions, while the text encodes task intent, procedural structure, and typed slots for variable content. A record-template-replay framework constructs templates from verified executions and leverages the current instruction and live UI state to reuse, adapt, or discard individual transitions. Given an execution trace, our framework reduces LLM inference token consumption by approximately 80% on average across five base agents on AndroidWorld, with most agents achieving at least 1.5-fold execution speedup over LLM-only execution. In real-device evaluations on Android and HarmonyOS, our framework achieves action reuse ratios of 70.7-79.8% while maintaining over 2-fold execution speedup in most evaluated configurations.
[AI-114] MBABench-2.0: Evaluating AI Agents on Constructing Financial Spreadsheets Beyond Simple Verifiability
链接: https://arxiv.org/abs/2610.31711
作者: Julian Poeltl,Thomson Yen,Patrick Shea,Shubham Anil Varma,Dhruv Rastogi,Zihao Chen,Mengshi Zhang,Hongseok Namkoong
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Spreadsheets are the lingua franca of modern finance, driving valuation, forecast, audit, and many other core workflows of the industry. As LLM agents advance to automate knowledge work, spreadsheet construction is a natural and pertinent testbed. However, most existing spreadsheet benchmarks evaluate agents on fill-in-the-blank style tasks, where an agent performs minor edits in otherwise complete spreadsheets. This protocol fails to capture the most central and time-consuming part of the spreadsheet workflow: the modeling decisions, format design, and sheet layout regular knowledge workers need to toil through to build a workbook from scratch. We introduce MBABench-2.0, a benchmark of 101 original spreadsheet construction tasks, the hardest of which is estimated to take 33 hours for a modeler with 3+ years of experience. Unlike coding or math, where a unit test or a final answer can verify an agent’s output, the quality of a financial spreadsheet that concerns readability, maintainability, or auditability, is not so simply verifiable. To holistically evaluate spreadsheet construction, we introduce 129 criteria spanning 12 categories that operationalize quality beyond the accuracy of the spreadsheet. Evaluating over 20 agents across 8 models and 4 harness types, we find that frontier agents rarely construct spreadsheets that are both accurate and meet our quality criteria: the top agent achieves a pass rate of only 16% on the benchmark. The benchmark demonstrates a substantial gap in spreadsheet building capability of current agents.
[AI-115] Compact-Bench: Measuring What Context Compaction Costs Long-Horizon Coding Agents
链接: https://arxiv.org/abs/2610.31710
作者: Ruopei Sun,Jianfeng Cai,Kangwen Zhao,Wengang Zhou,Houqiang Li
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Context compaction lets long-horizon coding agents continue beyond their context windows, but can erase the state needed to finish a task. Prior evaluations measure efficiency, retained information, or end-to-end success, leaving the effect of a specific history replacement difficult to isolate. Compact-Bench pairs full-history and summary-only continuations from a shared repository state, and evaluates recurring details and successive production compactions. It covers 35 SWE-bench repairs, 146 agentic sessions, and 106 production chains. Across 14 summarizer backbones, stable damage ranges from 5-64% of 22 full-history successes: Claude Opus 4.8 preserves the most repairs, while DeepSeek-Flash damages the most. With Kimi-K3 as summarizer, the Kimi Code prompt preserves the most task continuations (19/22), whereas Claude Code retains the most recurring details (15.0%). LLM summaries retain at most 16.0% of recurring details, compared with 79.1% for mechanical extraction; production summaries retain 15.7% at the first boundary. Failures recur across methods: one repair is damaged by 13/14 backbones, and another by all 5 prompts. These results show that reliable compaction requires preserving the progress needed to complete a task, and that retaining relevant information alone does not ensure successful continuation. Code is available at this https URL and data is available at this https URL.
[AI-116] Mining the Design Space of Multi-Agent Workflows
链接: https://arxiv.org/abs/2610.31709
作者: Xavier Cadet,Junyan Cheng,Peter Chin,Kyle Richardson
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 21 pages, including appendices. Equal contribution: Xavier Cadet and Junyan Cheng. Website: this https URL
Abstract:Designing large language model (LLM) multi-agent systems (MAS) still relies on trial and error and informal “folk wisdom.” Manual studies test only a few communication structures or topologies (e.g., sequential, parallel, shared-memory, or debate), while automated optimizers discard search trajectories after selecting a task-specific workflow. We present a two-phase pipeline for mining reproducible design principles from workflow code and execution traces. We repurpose AFlow’s MCTS optimizer across six composable operators and 26 deterministically verified Enigmata tasks, retaining all intermediate rounds to build a corpus of 28,568 executable workflows. A sandboxed AutoResearch loop then mines orchestration code and prompt templates under an L_0 -penalized validation objective and tests candidate principles on held-out searches. Across 300 feature-mining trials, we identify effects that persist on held-out searches across three levels of workflow design: workflows beginning with independent solvers and one-shot aggregation (Star-first) perform worse ( \Delta = -0.045 ); persistent shared documents (SharedMemory) outperform debate ( \Delta = +0.135 ); and prompt specificity and placeholder-based context routing yield gains comparable to communication-structure selection ( \Delta = +0.118 ). More broadly, our pipeline and results show the feasibility of turning discarded search trajectories into a statistical testbed for discovering and validating multi-agent workflow design principles.
[AI-117] RubricCraft: Where Solutions Diverge Criteria Emerge
链接: https://arxiv.org/abs/2610.31707
作者: Hosein Hasanbeig,Priyanshu Gupta,Ashish Tiwari,Vu Le,Sumit Gulwani
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Evaluating language agents on open-ended, artifact-producing tasks is difficult: model-based graders can be inconsistent, while exhaustive programmatic checks are often impractical. Textual rubrics make evaluation criteria explicit, but writing them requires domain expertise. Rubrics generated from task descriptions alone can miss requirements and failure modes that become apparent in attempted solutions. We introduce RubricCraft, which treats rubric construction as the generation of natural-language specifications guided by candidate solutions. Given a task description and independently sampled solver trajectories, RubricCraft compares how each attempt satisfies or violates the task requirements and uses these differences to construct rubric criteria. It applies the criteria to the collected solutions and revises checks that overlook errors or penalize valid alternatives. The resulting rubrics capture partial correctness without requiring any single solution to serve as a complete reference. On SpreadsheetBench v2, against a prompt-only rubric written and applied by the same model, RubricCraft rubrics pass fewer wrong outputs (29.5% against 42.6%), fail none of the outputs the benchmark’s own checker accepts, and rank outputs by correctness more often, while separating models as often. Repair on a single attempt leaves grading unchanged, but repair driven by attempts from three model families acts on different tasks, and an evidence audit judges most of those repairs improvements, many on tasks whose reference solution is itself wrong.
[AI-118] Before the Ticket: Grounding HPC User Support in Documentation and Resolved Tickets with an Agent ic LLM Assistant
链接: https://arxiv.org/abs/2610.31703
作者: Elham J. Barezi,Primus C Kabuo,Niveditha Nerella
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Most questions that reach an HPC help desk have already been answered in the center’s user guides or in a ticket someone else filed last semester; users still file new tickets because the guides are hard to navigate and the ticket history is not theirs to search. We present an assistant that answers from both sources: an orchestrator calls two tools over the Model Context Protocol, retrieval-augmented generation (RAG) over the live documentation and a Llama-3-8B model fine-tuned with QLoRA on 11,244 query-resolution pairs mined from historical tickets, and merges their answers. Every model runs on-premises or on the university’s hosted LLM service because ticket data may not leave the institution. On 400 held-out tickets from Purdue’s Rosen Center for Advanced Computing, disjoint from training and free of near-duplicates, and evaluated by two judges from model families that generated none of the answers, fine-tuning on resolutions rewritten into self-service guidance is the clear winner in agreement with RCAC’s resolutions (BERTScore F1 improves from 0.06 to 0.27 over the base model) and is judged most correct on average. The end-to-end system does not improve on its tools, although they are complementary: choosing the better answer per question would exceed either tool (judged correctness 3.41 versus 3.20, an oracle bound), while the merge follows the documentation tool in 78% of answers. Returning one tool’s answer instead recovers the fine-tuned tool’s quality (3.19 versus 2.66), but neither a learned selector nor an LLM verifier beats that tool alone.
[AI-119] Collaborative In-Context Learning for Indoor Radio Map Estimation
链接: https://arxiv.org/abs/2610.31702
作者: Yuanzhe Peng,Jie Xu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted as a regular paper at the 2026 IEEE International Conference on Big Data (IEEE BigData 2026)
Abstract:Indoor radio map estimation (RME) is essential for wireless applications such as coverage planning, localization, and network deployment. However, most existing methods follow a supervised learning paradigm that requires costly labeled data and task-specific model training. Moreover, indoor environments involve privacy-sensitive information, such as building layouts and structures, that often cannot be centralized. To address these challenges, we make the first attempt to formulate indoor RME as a collaborative in-context learning (ICL) problem. We propose Poly-Ripple, a framework in which multiple clients perform ICL using large autoregressive vision models (LAVMs) without model updates. Each client is prompted by its private demonstrations and predicts the radio map for the same scene query. To capture the causal structure of wireless propagation outward from the transmitter (Tx), we adopt ripple-ordered autoregressive modeling. We further introduce importance-aware aggregation to exploit heterogeneous contextual knowledge distributed across clients under the context length constraint. We provide theoretical analysis and conduct extensive experiments on a public indoor RME benchmark. The experimental results show that Poly-Ripple consistently outperforms representative ICL baselines, while aligning with our theoretical analysis.
[AI-120] radeoff-Aware Territory Design: A Multi-Objective Decision System for Enterprise Customer-to-Seller Allocation
链接: https://arxiv.org/abs/2610.31701
作者: Nan Wang,Mitchell Turiel
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Enterprise customer journeys are shaped not only by product experiences and marketing, but also by recurring resource-allocation decisions. In B2B go-to-market (GTM) organizations, territory carving determines which accounts receive seller coverage, which seller-account relationships are preserved, and how limited field capacity is distributed. Yet the process often remains manual, making it difficult to compare opportunities, value continuity, and evaluate tradeoffs among growth, balance, capacity, and feasibility. We present a production decision system for tradeoff-aware customer-to-seller allocation in enterprise GTM planning. It integrates three components: a multidimensional account-value layer that combines predictive signals to score opportunity; an observational continuity-valuation layer that estimates the cost of breaking seller-account relationships using a synthetic-control-style counterfactual framework; and a multi-objective optimization layer that allocates coverage to high-value accounts while considering continuity, opportunity balance, and operational constraints. Deployed as a scenario-based decision-support system, the framework enables planners to inspect and compare allocation tradeoffs. Historical backtesting shows that optimizer-generated scenarios substantially improve assignment compliance while maintaining comparable seller-level opportunity distribution. In production planning, adopted scenarios improved both assignment compliance and opportunity balance relative to the previous baseline. Directing scarce seller capacity toward higher-value accounts also offers incremental revenue potential at enterprise scale. The key contribution is an integrated, inspectable framework that turns territory carving into a governable decision process. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.31701 [cs.AI] (or arXiv:2610.31701v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.31701 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-121] SAKURA: Serving Attention Key-Value State Under Rate Adaptation for Private Image Editing on Phones
链接: https://arxiv.org/abs/2610.29733
作者: Shuncheng Jie,Desen Sun,Ri Hong,Yuyang Tian,Hong Zhang,Sihang Liu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Networking and Internet Architecture (cs.NI)
备注:
Abstract:Diffusion transformers are widely used in image-editing services. To preserve both efficiency and privacy, existing systems commonly rely on cloud-device collaboration. Keeping request-specific inference on the device avoids sending personal images and prompts to the cloud. These workloads often use common templates, such as in face swapping and virtual try-on, enabling KV state to be precomputed and reused across requests. However, this reusable state is large and scales with the number of templates, making it impractical to store all of them on mobile devices. Storing them in the cloud addresses this capacity issue, but cache transfer is placed onto the inference path and can even be slower than on-device computation, especially under fluctuating mobile networks. We introduce SAKURA, a cloud-edge architecture that delivers reusable attention state to mobile devices under a bandwidth budget. It first uses a fine-grained token-importance-aware compression method to preserve image quality under limited transfer budgets. At runtime, SAKURA adaptively selects the cache representation for each upcoming shard based on network bandwidth. To enable efficient execution, we develop a pipelined runtime that overlaps cache transfer and reconstruction with on-device execution. We evaluate SAKURA using three editing tasks and two diffusion models, using a Snapdragon 8 Elite Gen 5 phone. SAKURA reduces latency by 1.82x and 3.27x over CacheGen adapted to DiT and on-device generation, respectively. It also reduces cache storage by 3.06x over uncompressed cache, while preserving image quality. Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Networking and Internet Architecture (cs.NI) Cite as: arXiv:2610.29733 [cs.SE] (or arXiv:2610.29733v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.29733 Focus to learn more arXiv-issued DOI via DataCite
[AI-122] Monetizing Intelligence: Advertising and Subscription Design on LLM Platforms
链接: https://arxiv.org/abs/2610.32945
作者: Yuhang Ji,Fasheng Xu,Tianjun Feng
类目: Theoretical Economics (econ.TH); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) platforms introducing advertising must decide how much capability ad revenue should fund and what consumers should pay to avoid ads. We study a monopolistic platform that jointly chooses costly response quality, advertising intensity, and subscription prices when advertising reduces user value but better service raises advertising returns. Three findings follow. First, when partly ad-supported and ad-free plans coexist, advertising can finance useful service for consumers who receive none under pure subscription. Lower effective quality in middle plans weakens the alternatives available to premium users, allowing higher prices for unchanged ad-free service. Some fully ad-supported users gain utility, while partly ad-supported and ad-free users lose: broader access can coexist with utility polarization. Second, an intermediate advertising cap can maximize consumer surplus by preserving some revenue that finances service while limiting degradation. Third, cheaper compute can raise quality and reduce ad intensity, yet sufficiently large cost reductions can increase aggregate advertising revenue above its initial level because better service raises the value of the remaining advertised interactions. An extension shows how advertising-funded upgrades can bring plans to the same maximum capability and advertising exposure, eliminating the basis for different prices. These results make menu design central to evaluating LLM advertising: the consequences extend beyond users who see ads to the alternatives and prices faced by those who do not.
[AI-123] okamak-SI: Interpreting Tokamak Discharges with Labeled Catalogs
链接: https://arxiv.org/abs/2610.32264
作者: Nathaniel Chen,Alvin Garcia,Ricardo Shousha,Jalal Butt,Azarakhsh Jalalvand,Egemen Kolemen
类目: Plasma Physics (physics.plasm-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Finding which phenomena occur in a tokamak discharge, and when, is often a manual task. Previous efforts at labeling events have been task-specific, each with its own corpus of data. We introduce Tokamak-SI, which combines the labels and labeling methods of many earlier studies into a single scalable, time-aligned, and categorical dataset covering observations from confinement, detachment, alfven eigenmodes (AE), edge localized modes (ELM), sawtooth, tearing modes ™, and resistive wall modes in DIII-D tokamak. Models based on this, which can extend the labels to new shots, match or beat earlier detectors on its labels, including for AE, ELM, and TM detection. Tokamak-SI is intended to enable task-specific benchmarking, physics-searchable archives, and direct identification of key events during an experiment.
[AI-124] Uni-Stream: A Unified Streaming Speech Language Model for Speech Synthesis and Recognition
链接: https://arxiv.org/abs/2610.32187
作者: Xuan Shi,Chang Zeng,Tiantian Feng,Ye Zhang,Jun Liu,Shih-Heng Wang,Lie Lu,Shrikanth Narayanan
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI)
备注:
Abstract:Unified streaming speech language models aim to support speech recognition and synthesis within a single model, enabling real-time human-computer interaction. However, jointly supporting these two directions of speech-text conversion while preserving task-specific performance remains challenging. We propose Uni-Stream, a unified streaming speech language model built on compact pretrained speech models. Uni-Stream processes interleaved text and speech in a streaming format and achieves performance comparable to task-specific models through two-stage training. We also develop a data annotation pipeline for streaming training and will publicly release the generated transcriptions and word-level timestamps. We further examine Uni-Stream’s representations across tasks and modalities and find that speech and text residence in distinct latent feature space even when trained in a unified model architecture, while representations within each modality become more task-dependent in later layers.
[AI-125] MG-REPA: Multi-Granularity Representation Alignment for Generative Target Speaker Extraction
链接: https://arxiv.org/abs/2610.32161
作者: Dohwan Kim,Jung-Woo Choi
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI)
备注:
Abstract:Generative target speaker extraction (TSE) models achieve high-quality extraction, but suffer from considerably slow training convergence. Representation Alignment (REPA) accelerates diffusion transformer training by aligning hidden states with a pre-trained Self-Supervised Learning (SSL) teacher. However, existing alignment in the speech and audio domain operates at a single granularity. For TSE, frame-level alignment leaves the global, time-invariant characteristics of the target speech implicit, whereas utterance-level alignment leaves the time-varying content implicit, even though extracting the target requires both. We propose Multi-Granularity Representation Alignment (MG-REPA), aligning the hidden states with an SSL model at the frame level and a speaker verification model at the utterance level. We further apply temporal normalization to the frame-level target, removing its utterance-level statistics so that each branch concentrates on its granularity. On Libri2Mix, MG-REPA reaches the OVRL and PESQ of a MeanFlow-TSE baseline with 3.0\times and 2.1\times fewer epochs, respectively. Trained for 750 epochs, it is competitive with the published 2,000-epoch baseline on intrusive metrics while improving non-intrusive metrics. An ablation demonstrates that MG-REPA outperforms alignment at a single granularity.
[AI-126] Implicit Neural Representations for Variational Problems on Graphons NEURIPS2026
链接: https://arxiv.org/abs/2610.32147
作者: Taeyoung Kim,Jineon Baek,Joonkyung Lee,Hongseok Yang
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Combinatorics (math.CO)
备注: Accepted at NeurIPS 2026. 35 pages, 18 figures, 10 tables
Abstract:We propose a neural exploratory framework for variational problems over graphons, i.e., symmetric measurable functions [0,1]^2 \to [0,1] , that arise as limits of dense graph sequences. Such problems are central to two parts of the mathematical literature: extremal graph theory, which studies graph parameter optimisation under given restrictions, often for those graphs with a large number of vertices, and the large deviation theory of dense random graphs, which characterises the structure of rare events through constrained graphon optimisation. We represent graphons by implicit neural networks and optimise graphon objectives by gradient descent. Our design combines three ingredients: a multi-scale sinusoidal residual architecture biased toward sharp, step-like graphons; an embedded solver that enforces a single empirical density constraint and is differentiated by implicit differentiation; and symmetry-aware Monte Carlo estimators. On generalised Turán problems that previously required substantial human effort, our framework rediscovers known optimal graphons without human intervention. Applied to open instances, it produces candidate optima, including a previously unreported family of extremal structures. Applied to the variational problems from the large deviation theory of Erdős–Rényi random graphs, the framework produces new candidate optimal graphons across both upper- and lower-tail regimes for multiple pattern graphs; for the case of the triangle graph and the upper-tail regime, these candidates improve upon the best-known reference construction of Lubetzky and Zhao. The structures found by our framework have also led to new theorems on the optimal graphons for both families of problems.
[AI-127] Astro2Method: Benchmarking LLM s on Machine Learning Method Selection in Astronomy
链接: https://arxiv.org/abs/2610.30282
作者: Xingzhong Fan,Hangfeng He,Fatemeh Nargesian
类目: Instrumentation and Methods for Astrophysics (astro-ph.IM); Artificial Intelligence (cs.AI)
备注: 15 pages, 4 figures, 4 tables
Abstract:Machine learning (ML) is increasingly used in astronomy, but choosing a suitable method for a scientific problem often requires expertise in both astronomy and ML. Large language models (LLMs) could help astronomers make such choices, yet this ability remains underexplored in existing benchmarks. We introduce Astro2Method, a benchmark of 500 expert-reviewed binary questions from 346 astronomy papers (2024/01-2026/06), based on clear performance differences and complete (Easy) or partial (Hard) explanations from the authors. Across five LLMs, direct prompting achieves 87.50-92.55% accuracy on Easy questions but only 58.87-64.52% on Hard. Using historical papers, neither retrieval-augmented generation nor most agent methods with additional reasoning or retrieval (including ReAct, Self-Ask, and CRITIC) bring consistent improvement. Agents required to search the web improve accuracy mainly when they find the papers the questions are based on, with no consistent improvement otherwise. Providing the source paper’s pipeline description with method names masked improves Hard accuracy, matching or exceeding web search. Together, these results suggest that relevant experimental information helps distinguish ML methods, but inferring it from related studies and domain knowledge remains challenging for LLMs in astronomy.
机器学习
[LG-0] Where Rounding Happens Matters: Rounding-Location Alignment Stabilizes LLM Reinforcement Learning
链接: https://arxiv.org/abs/2610.32951
作者: Yijing Ke,Tianle Cai,Ruosong Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Reinforcement learning for large language models often uses separate systems for inference and training. Even with identical weights and BF16 tensors, implementation differences can make an intended on-policy update off-policy and destabilize training. Existing remedies align the entire inference and training stacks, change global precision, or modify the learning algorithm. We ask whether aligning selected parts of the execution stack can stabilize BF16 RL while other implementation differences remain. We align training-side rounding locations with inference in three regions—RoPE, SwiGLU, and residual RMSNorm—by retaining selected intermediates in FP32 until their BF16 outputs. Across four Qwen-family models, aligning all three regions prevents the collapse observed under the default BF16 baseline; RoPE-path alignment alone also suffices in the studied runs, despite remaining numerical discrepancies. The interventions keep the inference engine, matrix-multiplication and attention kernels, and learning objective unchanged.
[LG-1] If at First You Dont Succeed Backtrack: Lightweight Recovery for Frozen VLAs
链接: https://arxiv.org/abs/2610.32941
作者: Thomas Landeg,Thierry Tambe
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 17 pages, 1 figure, 14 tables
Abstract:Vision-Language-Action (VLA) models perform well on robotic manipulation, yet remain brittle once their own actions carry them into states their training data barely covers. Recovering from such failures usually requires recovery demonstrations, policy retraining, or a separate model at inference time. We introduce Backtrack, a lightweight failure detection and recovery mechanism for frozen VLAs that needs none of these. A small failure prediction head, trained only on outcome-labeled rollouts of the frozen policy, reads the VLA’s existing internal representations to estimate whether the current episode will fail. When the predicted failure probability exceeds a threshold, a controller pauses the policy, retraces the arm’s recently achieved motion, and hands control back to the unchanged policy so it can retry from an earlier configuration. The head adds only 0.48-0.88% to the policy’s parameter count. We evaluate Backtrack on LIBERO with three VLAs spanning two action-generation families: SmolVLA and \pi 0.5, which use flow-matching action experts, and the autoregressive \pi 0-FAST. On LIBERO-Long, the suite of long-horizon multi-step tasks and the one on which all three policies fail most often, Backtrack reduces each policy’s failure rate by 19-48%, raising success by up to 7.4 percentage points; on the three shorter suites, where control success is already 85-99%, its effect is smaller. On MetaWorld, all twelve controller configurations tested with SmolVLA improve on the unassisted policy.
[LG-2] Learning Positive Linear Combinations of ReLUs in Fully Polynomial Time
链接: https://arxiv.org/abs/2610.32940
作者: Ilias Diakonikolas,Jelena Diakonikolas,Puqian Wang,Nikos Zarifis
类目: Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注:
Abstract:We study the fundamental problem of PAC learning positive linear combinations of k arbitrary ReLU activations on \mathbbR^d with respect to the Gaussian distribution, in L_2 loss. We give the first fully-polynomial time algorithm for this learning task. Specifically, our algorithm draws \mathrmpoly(d, k, 1/\epsilon) samples, runs in \mathrmpoly(d, k, 1/\epsilon) time, and outputs a hypothesis that is \epsilon -close to the target concept with high probability. All prior algorithms for this problem require super-polynomial time in k or 1/\epsilon , or leverage strong well-conditioned assumptions on the weight matrix. On a technical level, our approach develops a novel boosting procedure of a weak'' learner that recovers the hidden weight vectors to constant error, in a list-decodable manner. To develop this weak’’ learning procedure, we leverage implicit tensor computation techniques to construct a low-degree filtering procedure based on Hermite tensors of the target function. This procedure amplifies separation among (clusters of) weight vectors to learn each of them to constant error. We then argue that the same procedure can be used for improving the estimation of hidden weight vectors, using gradient descent-like updates on the Euclidean sphere. A crucial ingredient of our algorithm is a pruning approach that ensures only relevant vectors that approximate the hidden weight vectors at target accuracy are maintained by the algorithm. Subjects: Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG) Cite as: arXiv:2610.32940 [cs.DS] (or arXiv:2610.32940v1 [cs.DS] for this version) https://doi.org/10.48550/arXiv.2610.32940 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-3] ZENO: Nonlinear filtering with Zakai Equation Neural Operators
链接: https://arxiv.org/abs/2610.32926
作者: Aaditya Prasad,Steven W. Flavell,Brian DePasquale
类目: Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
*备注:
Abstract:State inference and parameter learning for nonlinear filters is a core computation for scientific machine learning with varied applications in engineering and neuroscience. The unnormalized filtering distribution is governed by the Zakai equation, a stochastic partial differential equation (SPDE) that combines the forward Kolmogorov operator (known as the Fokker-Planck equation) with observed likelihood updates. Although solving these high-dimensional SPDEs is often impossible analytically and intractable numerically, advances in operator learning and physics-informed machine learning have made it feasible. Harnessing these advances, we introduce the Zakai Equation Neural Operator (ZENO), a model that optimizes a neural operator against a Zakai Equation based physics-informed loss to approximate the exact unnormalized filtering distribution for latent stochastic dynamical systems. We use this ZENO-inferred posterior distribution in an expectation-maximization (EM) framework to perform parameter learning and state inference. We demonstrate ZENO’s ability to accurately infer the posterior distribution and learn its corresponding SDE parameters in three simulated stochastic dynamical systems (Double-Well bistable system, Van Der Pol (VDP) oscillator, and Lorenz system) and compare its performance against established methods for latent SDE inference and learning. We show that ZENO recovers previously observed limit cycles representing forward and backward locomotion from a whole-brain imaging dataset of immobilized \textitC. elegans nematodes. ZENO is a novel approach to the problem of nonlinear filtering and dynamical inference through the lens of partial differential equations and operator theory.
[LG-4] A Tutorial on Physics Constrained Machine Learning for Hybrid Modeling Optimization and Control
链接: https://arxiv.org/abs/2610.32914
作者: Rahul Golder,Bimol Nath Roy,MM Faruque Hasan
类目: Machine Learning (cs.LG); Emerging Technologies (cs.ET)
*备注:
Abstract:Physics-constrained machine learning (PCML) integrates domain knowledge, first-principles, and physical constraints into AI/ML models. In this work, we present a step-by-step tutorial of a modular PCML framework, KKT-HardNet, and illustrate how it could be used for hybrid modeling, process simulation, optimization and model predictive control in a single platform. KKT-HardNet combines fast inference capabilities of neural networks with guaranteed satisfaction of algebraic constraints through differential projection. It can handle heterogeneous data and enforce both linear and nonlinear equality and inequality constraints. The modular design supports custom neural-network backbones and user-defined projection mechanisms for PCML. To enable efficient end-to-end training, KKT-HardNet exploits implicit function theorem to backpropagate through the projection layer. This avoids computationally expensive automatic differentiation. This work also introduces an exact reduced-order Fischer-Burmeister reformulation and a primal-dual interior-point method as an alternative projection solve strategy. We also provide a diverse set of benchmarking problems for PCML in engineering applications. KKT-HardNet achieves orders of magnitude better performance in terms of both constraint satisfaction and inference time. The KKT-HardNet code and the benchmark datasets are available at this https URL .
[LG-5] Recommender regret for improving regularised RL estimation with bandits and MCTS
链接: https://arxiv.org/abs/2610.32910
作者: Danilo Vucetic,Alexandre Larouche,Mohammad Pedramfar,Audrey Durand,Gauthier Gidel
类目: Machine Learning (cs.LG)
*备注: 10 pages, plus appendices for proofs and additional information
Abstract:Regularised reinforcement learning (RL) algorithms rely on estimators, like Monte Carlo tree search or one-step look-ahead, to provide policy, value, or Q-value estimates in training and inference. However, existing bandit-based estimation frameworks evaluate sampling efficiency, not the quality of estimates. We address this by formalising estimate generation as a recommendation process, treating bandit-based sampling as an improvement operator. To evaluate the quality of a recommendation (estimate), we define regulariser-aware recommender regrets, analysing their lower bounds, and proving that simple algorithms like uniform sampling can achieve these bounds. Finally, we unify bandit- and search-based target improvement under recommender MCTS. Across synthetic trees, arcade games, and scientific discovery tasks, recommenders often yield faster target convergence and superior downstream performance. Our experiments characterise for the first time the potential of regularised MCTS in scientific applications.
[LG-6] When Human Corrections Travel: Safety in Federated VLA Adaptation
链接: https://arxiv.org/abs/2610.32888
作者: Vigneshwaran Subbaraju,Obaidullah Zaland,Monowar Bhuyan
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Accepted for presentation at SPAIS workshop in CoRL 2026
Abstract:Vision-language-action (VLA) policies can adapt to local human intervention, while federated learning can combine distributed robot experience without pooling raw records. We examine their prospective intersection: sharing adaptations derived from diverse human corrections across tasks, environments, people, and robot embodiments. Corrections differ in modality, safety semantics(urgency, behavioral scope, etc.), and privacy sensitivity. A shared update that improves mean task success may suppress a rare safety intervention or produce an unsafe response on another robot. We argue for three requirements: preserve the context and uncertainty of multimodal corrections when deciding what to share; assess candidate updates in each receiving robot’s closed loop before deployment; and evaluate leakage alongside retention of safety-critical evidence. Immediate, legible defensive behavior remains necessary during human interaction. We offer counterexamples and a falsifiable evaluation agenda, not experimental results.
[LG-7] Reinforcement Learning with Multi-Step Transition Predictions: A Bellm an Theory
链接: https://arxiv.org/abs/2610.32887
作者: Chenbei Lu,Zaiwei Chen,Tongxin Li,Chenye Wu,Adam Wierman
类目: Machine Learning (cs.LG)
*备注:
Abstract:Multi-step predictions of future transition realizations can improve reinforcement learning, but their value remains difficult to quantify. Directly including these predictions in the state can lead to exponential state-space growth, making direct dynamic programming computationally expensive. We study discounted finite MDPs under a block-feedback protocol: at the start of each K-period block, the decision maker observes K -step probabilistic transition predictions for a subset of actions and adapts decisions to observed states within the block. We show that the optimal value satisfies a Bayesian Bellman equation on the original state space. This representation reveals that predictive information creates value by recursively changing the local order of maximization and expectation in the Bellman expansion. We formalize this effect through a Bellman–Jensen analysis and quantify the effects of prediction horizon, error, action coverage, and within-block feedback. We propose block-feedback Bayesian offline learning with online adaptation (BF–BOLA), a model-based algorithm for learning near-optimal policies, and establish upper and lower sample-complexity bounds that match up to logarithmic factors in the stated regimes. The results show that predictions can reduce transition learning to uncovered rows and that learning a near-optimal policy can require fewer prediction samples than estimating the Bayesian value function.
[LG-8] rusting the Tool: Asymmetric Deference of LLM Vulnerability Auditors to Static-Analysis Evidence
链接: https://arxiv.org/abs/2610.32876
作者: ZhiHong Liu,Shuyao Gao
类目: oftware Engineering (cs.SE); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 33 pages, 9 figures, 5 tables. Submitted to IEEE Access. Code: this https URL
Abstract:Pipelines that combine large language models (LLMs) with static analysis let the model consume the analyser’s output as evidence, but whether the model distinguishes true from false evidence, and whether its scrutiny depends on what the evidence claims, has not been measured. We built an intraprocedural analyser for C/C++ that produces proof obligations and audited vulnerable and patched functions from PrimeVul under six evidence conditions that differ only in the notes, including fabricated assurances that a missing check exists and fabricated alarms that an added check is missing. The study covers 916 pairs in four disjoint sets, eleven model configurations from four vendors and 16,572 audits; no model is trained or fine-tuned. True evidence barely improved accuracy (+0.02). False evidence acted asymmetrically: relative to true notes about the same operations, false assurances lowered the detection of real vulnerabilities by up to 19 points, whereas false alarms never significantly lowered accuracy and were disputed far more often. An internally registered replication on seven new models confirmed that false assurances move verdicts towards their claim more than false alarms do (mean asymmetry +0.16, 95% CI +0.10 to +0.24). Instructions in the audit prompt did not reliably reduce deference; the best one failed to replicate. Verifying each claim in a separate call and showing the refuted claims raised accuracy under false assurances by 12 points in a pre-specified test (p=0.002) without loss under true evidence; for a second model the gain was smaller and not significant. Pipelines should therefore verify a tool’s guarantees before the model relies on them, and studies should measure deference in a direction-aware way.
[LG-9] Sustainability of TSFM Deployment and Pretraining: From Carbon Footprint Analysis to Mitigation via Pruning NEURIPS2026
链接: https://arxiv.org/abs/2610.32859
作者: Cécile Luc,Nathanaël Lemonnier,Ievgen Redko
类目: Machine Learning (cs.LG)
*备注: 26 pages, 6 figures, accepted at the NeurIPS 2026 Workshop TS-LIMITS
Abstract:Time Series Foundation Models (TSFMs) offer strong zero-shot forecasting capabilities, but their high pretraining cost raises a practical sustainability question: does this upfront investment pay off over a full deployment lifecycle, compared to supervised standalone models that must be periodically retrained? We model the cumulative carbon cost of different approaches across a realistic industrial scenario: Chronos-2 amortizes its pretraining cost within months for large or frequently queried deployments, but a classical model remains cheaper indefinitely for small or rarely queried ones. For cases where reducing the TSFM’s own cost matters, we show that removing a single encoder block (7.9% of parameters) via a gradient-free pruning criterion yields a consistent 6 to 9% reduction in FLOPs, CO2 emissions, and latency, at essentially no cost to zero-shot forecasting performance across a 25-task benchmark.
[LG-10] Regularised RL improves sim-to-real transfer
链接: https://arxiv.org/abs/2610.32855
作者: Danilo Vucetic,Mohamad H. Danesh,Florence Clerc,Hsiu-Chin Lin,Jonathan Tremblay
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 9 pages of main-body content, with supporting details in the appendices. Under submission
Abstract:Regularised reinforcement learning (RL) has provable robustness guarantees that should ease the transfer of policies from simulation to robots, known as sim-to-real. Yet, weakly regularised algorithms like Proximal Policy Optimisation (PPO) remain the de facto choice in sim-to-real pipelines, while the practical benefits of regularisation for transfer remain poorly characterised. In this work, we study the sim-to-real performance of commonly used RL algorithms, with the aim of characterising the effects of regularisation on robotic performance. To that end, we derive Soft MellowMax Actor-Critic (SMM-AC), a novel regularised RL algorithm based on the soft mellowmax value function, which carries strong robustness guarantees. On MuJoCo continuous control benchmarks, SMM-AC is competitive with baselines like Soft Actor-Critic (SAC). On real robots, SMM-AC achieves higher return, while exhibiting greater robustness to perturbations in observations, the environment, and dynamics than baselines. Together, these results demonstrate that regularised RL improves real-world policy deployment through better robustness to the sim-to-real gap, and changing real-world conditions.
[LG-11] Near-Optimal List Replicability from Littlestone Dimension
链接: https://arxiv.org/abs/2610.32853
作者: Ari Blondal,Pooya Hatami,Chavdar Lalov,Sivan Tretiak
类目: Machine Learning (cs.LG)
*备注: 14 pages
Abstract:We study the list replicability number in the realizable PAC learning framework and prove that every concept class H over a countable domain with Littlestone dimension d and dual VC dimension v^* satisfies [ \mathrmLR(H)\leq 2^O(d\min\log d,v^). ] This matches the worst-case lower bound of 2^\Omega(d) up to a logarithmic factor in the exponent and exponentially improves the previous best upper bound \mathrmLR(H)\le 2^2^O(d) of Bun, Livni, and Moran [FOCS 2020]. Unlike earlier bounds of 2^O_\epsilon(d^2) for fixed accuracy \epsilon0 implied by the work of Ghazi, Golowich, Kumar, and Manurangsi [STOC 2021] and Lyu [STOC 2026], our bound is independent of \epsilon . The 2^O(d\log d) bound follows from a new inductive framework for bounding list replicability that builds on the recent notion of irreducibility from differentially private learning. The 2^O(dv^) bound follows from a simple globally stable learner that takes the majority vote of multiple empirical risk minimizers, with stability established using tools from stable regularity. Comments: 14 pages Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.32853 [cs.LG] (or arXiv:2610.32853v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.32853 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-12] Neural LoFi with Backward Coupling: A Spectral Theory of Cross-Layer Feature Learning
链接: https://arxiv.org/abs/2610.32849
作者: Yatin Dandi,Matteo Vilucchio,Hugo Tabanelli,Florent Krzakala
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Machine Learning (stat.ML)
*备注:
Abstract:During joint training, backpropagation allows later layers to reshape earlier representations, a feedback mechanism absent from purely layer-wise learning. We extend Neural Low-Degree Filtering (Neural LoFi), a recently proposed layer-wise spectral surrogate for feature learning, to capture this interaction. Starting from the exact backpropagation gradients, we derive a local coupling between a neuron’s incoming weights and its corresponding next-layer column. In a controlled small-initialization regime with smooth activations and fixed surroundings, these dynamics amplify singular modes of a cross-layer correlation operator. Each growing mode pairs an earlier feature with an upper-layer response through shared neuron coefficients. This motivates a backward-coupling algorithm in which upper-layer sensitivities generate sample-dependent effective labels for earlier spectral steps, while corrected features refine upward transmission. A function-space variational principle characterizes the selected feature-response pairs by maximizing the label correlation of their product under kernel-norm constraints. For three-layer networks, we derive self-consistent continuous-time dynamics for the filtered hidden-layer features, the activation kernels, and the network output under the proposed backward coupling. In a hierarchical multi-index model, we establish a feature-recovery separation: backward coupling recovers lower-layer directions invisible to the original forward label-weighted statistic. The linear specialization recovers residual-dependent singular-mode dynamics, while paired modes provide a dynamical mechanism for linked low-rank covariance components consistent with the rainbow-network picture. Experiments with fully connected and convolutional networks on binary CIFAR-10 show improved accuracy over forward Neural LoFi at matched widths.
[LG-13] On the Limits of Sparse Linear Model with Shuffled Differential Privacy
链接: https://arxiv.org/abs/2610.32822
作者: Yan Zhang,Jiong Tuo,Meng Ding,Yuheng Ma,Di Wang
类目: Machine Learning (cs.LG)
*备注: 39 pages, 1 figure. Under review
Abstract:Sparse linear regression is a fundamental problem in high-dimensional statistics, and its private counterparts leave a large gap between the central and the local model of differential privacy. Shuffled differential privacy (SDP) sits between the two, offering accuracy closer to the central model without the trust requirement of a central curator. This paper studies sparse linear regression under both non-interactive and interactive SDP protocols. In the non-interactive setting, any non-interactive (\epsilon,\delta) -SDP protocol is shown to incur an \ell_2 -norm estimation error of \Omega( \frac\sqrtdkn\epsilon ) with dimension d , sparsity k , and n users, revealing an inherent dimensionality barrier. In addition, a non-interactive algorithm is proposed that achieves an \ell_2 -norm upper bound of \tilde\mathcalO( \sqrt\frackn + \fracd\sqrtkn\epsilon) . To overcome this barrier, we turn to the interactive user-level setting, where each user holds m samples. A two-round interactive algorithm is developed that first recovers the support through a heavy hitter step and then estimates the parameter, achieving an error of \tilde\mathcalO( \sqrt\fracknm + \frackn\sqrtm\epsilon ) with no polynomial dependence on d . Furthermore, by extending the score attack framework to user-level data, a new lower bound \Omega( \sqrt\frack\log(d/k)nm + \frack\log(d/k)n\sqrtm\epsilon ) is established for any user-level (\epsilon,\delta) -DP, showing that the proposed algorithm is near-optimal with only a gap of the factor \log(d/k) .
[LG-14] U-NeXt: General-purpose event detection for neonatal EEG
链接: https://arxiv.org/abs/2610.32811
作者: Robert Hogan,Sean R. Mathieson,Aurel Luca,Soraia Ventura,Sean Griffin,Geraldine B. Boylan,John M. O’Toole
类目: Machine Learning (cs.LG); Signal Processing (eess.SP); Medical Physics (physics.med-ph)
*备注:
Abstract:We developed U-NeXt, a general framework for neonatal electroencephalography (EEG) event detection based on semantic segmentation. Its convolutional encoder-decoder directly segments individual channels using long input segments to preserve temporal context. A shared architecture and training framework were developed using low-voltage interval, artefact, and seizure datasets containing 51, 98, and 278 neonates, respectively, comprising over 11,000 hours of EEG and 82,000 annotated events. Performance was compared with task-specific ConvNeXt classifiers. For seizure detection, we also evaluated U-NeXt+, trained on an expanded dataset. Both U-NeXt variants and the ConvNeXt baseline were evaluated on two independent held-out seizure datasets, Helsinki and Cork, comprising 130 neonates, with recordings annotated by three experts. U-NeXt improved detection across all three event types, with particular gains in sensitivity for short events. U-NeXt+ achieved Matthews correlation coefficients of 0.862 and 0.891 on Helsinki and Cork, respectively, and improved hourly seizure burden correlations by 30-40% relative to ConvNeXt. U-NeXt+ met the criterion for expert-level performance on both datasets (p 0.001). In separate computational benchmarks, U-NeXt reduced batch inference time relative to ConvNeXt by 254-fold on a central processing unit and 200-fold on a graphics processing unit. A common segmentation framework improves detection across diverse neonatal EEG event types while substantially reducing inference time. Accurate seizure burden estimates and performance comparable to expert reviewers support its potential use in continuous monitoring and clinical decision support.
[LG-15] CARES: Coverag e-Aware Replay Exemplar Selection for Continual Learning
链接: https://arxiv.org/abs/2610.32799
作者: Arundhati S. Shanbhag,Brian B. Moser,Federico Raue,Tobias C. Nauen,Andreas Dengel
类目: Machine Learning (cs.LG)
*备注: 17 pages, 2 figures. Includes supplementary material
Abstract:Replay-based Continual Learning mitigates catastrophic forgetting by keeping a buffer of past samples, making buffer composition critical under a fixed memory budget. Recent coreset-based replay methods improve buffer quality using model-derived gradients, losses, or auxiliary optimization, but incur substantial computational overhead as memory is repeatedly updated. We propose CARES, a continual model-optimization-free approach for buffer construction that instead exploits the geometry of a frozen pretrained feature space using class-aware, density-weighted coverage. We evaluate CARES across balanced, imbalanced, and label-noisy streams under online class- and task-incremental settings. On Tiny-ImageNet, CARES is 2.3 to 8.8 times faster than existing coreset-based selectors. Although CARES achieves this speedup using no training signal from the continual learner, it ranks among the state of the art across balanced, imbalanced, and label-noisy streams under online class- and task-incremental settings, outperforming the strongest baseline by up to 7.58 percentage points under class imbalance while also attaining higher average accuracy on Tiny-ImageNet. Code will be released upon publication
[LG-16] Deep Learning-Based Prediction of HER2 Status from Whole-Brain MRI in Breast Cancer Brain Metastases
链接: https://arxiv.org/abs/2610.32774
作者: Sara Early,Chris Kang,Vibujithan Vigneshwaran,Kimberly Amador,Roger Y. Tsang,Nils D. Forkert
类目: Machine Learning (cs.LG)
*备注:
Abstract:Purpose: Human epidermal growth factor receptor 2 (HER2) status is a key biomarker relevant for prognostication and treatment decision making in metastatic breast cancer. Thus, the aim of this work was to explore whether HER2 status can be predicted directly from whole-brain post-contrast T1-weighted (T1w) MRI scans in breast cancer brain metastases (BCBM) using deep learning, and whether providing explicit tumour segmentation information improves prediction. Methods: Using the publicly available BCBM-Radiogenomics dataset, we analysed 135 patients, including 70 HER2+ and 65 HER2- patients. Three approaches were implemented and evaluated: a conventional radiomics model, a simple fully convolutional network (SFCN) using T1w MRI scans alone, and an SFCN using T1w MRI scans with an expert-derived tumour mask as an additional input channel. Results: The radiomics model achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.602 (95% confidence interval [CI]: 0.528–0.677) while the SFCN achieved a ROC-AUC of 0.667 (95% CI, 0.63–0.71) using T1w MRI scans alone, which increased to 0.692 (95% CI, 0.58–0.81) when an explicit tumour mask was additionally provided. Conclusion: However, the modest performance across all approaches demonstrates the difficulty of predicting HER2 status from structural MRI scans alone. The difference between the segmentation-free and mask-guided SFCN models suggests that much of the predictive signal captured by the deep-learning model may be learned directly from whole-brain T1w MRI without requiring explicit tumour delineation. These findings support further investigation of segmentation-free deep-learning approaches as a potential non-invasive strategy to support HER2 status determination in BCBM. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.32774 [cs.LG] (or arXiv:2610.32774v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.32774 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Sara Early Ms. [view email] [v1] Fri, 23 Oct 2026 15:23:54 UTC (497 KB)
[LG-17] Direction-Aware Token Credit Assignment with Reward-Model Gradients ATC
链接: https://arxiv.org/abs/2610.32772
作者: Yikuan Hu,Jifeng Zhu,Minle Su,Jiaqin Yue,Gujie Shao,Yanzhe Chen,Haoxuan Jia,Haoran Xu,Wangzhou Dai
类目: Machine Learning (cs.LG)
*备注: 28 pages, 7 figures. Project page: this https URL
Abstract:GRPO shares a response-level advantage across tokens, leaving unresolved how strongly each token should contribute to learning. Updates induced by the shared advantage reinforce or suppress the sampled token at each position, but these updates do not always point toward increasing reward at the token level. This motivates a work-based view of credit: just as work depends on the component of force along a displacement, token credit should reflect the local reward change along the policy update direction. We introduce Direction-Aware Token Credit Assignment (DATC), which maps frozen reward-model gradients into policy logit space and allocates weights by evaluating the alignment between the reward direction and GRPO’s signed sampled-action update direction, while preserving the sign and mean of the response-level advantage. DATC outperforms GRPO in all four settings across two model families.
[LG-18] On the Identifiability of Hybrid Deep Generative Models: Meta-Learning as a Solution NEURIPS2024
链接: https://arxiv.org/abs/2610.32748
作者: Yubo Ye,Maryam Toloubidokhti,Sumeet Vadhavkar,Xiajun Jiang,Huafeng Liu,Linwei Wang
类目: Machine Learning (cs.LG)
*备注: NeurIPS 2024. 16 pages, 5 figures
Abstract:The interest in leveraging physics-based inductive bias in deep learning has resulted in recent development of hybrid deep generative models (hybrid-DGMs) that integrate known physics-based mathematical expressions in neural generative models. To identify these hybrid-DGMs requires inferring parameters of the physics-based component along with their neural component. The identifiability of these hybrid-DGMs, however, has not yet been theoretically probed or established. How does the existing theory of the un-identifiability of general DGMs apply to hybrid-DGMs? What may be an effective approach to construct a hybrid-DGM with theoretically-proven identifiability? This paper provides the first theoretical probe into the identifiability of hybrid-DGMs, and presents meta-learning as a novel solution to construct identifiable hybrid-DGMs. On synthetic and real-data benchmarks, we provide strong empirical evidence for the un-identifiability of existing hybrid-DGMs using unconditional priors, and strong identifiability results of the presented meta-formulations of hybrid-DGMs. Code is available at this https URL.
[LG-19] Strongly Adaptive Online Learning with Time-Varying Movement Costs
链接: https://arxiv.org/abs/2610.32740
作者: Andrew Jacobsen,Emmanuel Esposito,Hao Qiu,Mengxiao Zhang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We study strongly adaptive regret for Online Convex Optimization (OCO) with time-varying movement costs, a setting that evaluates a learner’s performance over arbitrary sub-intervals. This framework captures complex sequential decision-making problems where both the environment and the cost of switching decisions fluctuate over time. We propose a novel algorithm which, for any interval [a,b] \subseteq [T] , achieves a data-dependent regret bound of \widetilde\mathcalO\left( \sqrt(D^2+DP_[a,b])\sum_t \in a,b \right) , where D is the diameter of the decision domain, P_[a,b] is the path length of the comparator sequence over [a,b] , g_t is the gradient of the loss, and \lambda_t is the movement-cost coefficient at round t . Key to our approach is a novel reduction to sleeping experts with movement costs, a previously unexplored variant that is of independent interest. As a result of this development, we further establish the first strongly-adaptive guarantees under delayed feedback. Our approach also naturally extends to OCO with time-varying memory, where we obtain strongly-adaptive guarantees which improve the best-known bounds.
[LG-20] Cross-Scenario Controller Parameter Transformation from Real-World Driving Data through Flow Matching
链接: https://arxiv.org/abs/2610.32719
作者: Christian Zink,Maik Pfefferkorn,Eric Lenz,Felix Kallmeyer,Rolf Findeisen
类目: ystems and Control (eess.SY); Machine Learning (cs.LG)
*备注:
Abstract:Advanced Driver Assistance Systems have become increasingly widespread, aiming to support drivers, reduce workload, and improve traffic safety. However, many systems are rarely used because their behavior feels unnatural and does not match individual driving behavior. The growing availability of large-scale, real-world human driving data enables more human-like and driver-individual assistance functions. This paper formulates the adaptation of Adaptive Cruise Control parameters as a driver-individual scenario-to-scenario transport problem, transforming controller parameters from one driving context to another. Using flow matching, we train a generative model to predict the target-scenario behavior of previously unseen drivers from only a few demonstrations in a source scenario. We compare the approach with a linear parameter-shift baseline and evaluate flow-matching variants incorporating driver-consistent coupling, driver embeddings, and classifier-free guidance based on driving profiles extracted from trajectory data using a variational autoencoder.
[LG-21] DT-JEPA: Learning to Predict Future Representations for Decision-Making
链接: https://arxiv.org/abs/2610.32657
作者: Changxu Ni,Yiyang Ni,Jun Li,Feiran You,Hongyang Du,Haitao Zhao,Shi Jin
类目: Machine Learning (cs.LG)
*备注:
Abstract:Decision Transformer (DT) has demonstrated strong performance in various offline reinforcement learning (RL) tasks by conditioning action generation on expected future returns and historical trajectories. However, treating expected future returns as the condition provides an incomplete description of the future, as trajectories with different state transitions and behaviors may share similar returns. Moreover, expected future returns inherently depend on task-specific reward functions, which makes policy learning reliant on reward-labeled datasets and restricts its scalability to tasks with similar but unseen rewards. Recently, Joint Embedding Predictive Architectures (JEPA) have emerged as a promising paradigm of self-supervised learning-based world models, which learn predictive representations of future dynamics in latent space rather than reconstructing future states in the original space. Inspired by JEPA, we propose DT-JEPA, an offline pretraining and online fine-tuning framework that learns general knowledge from reward-free offline trajectories and achieves excellent adaptation to unseen tasks. Instead of predicting a single deterministic future embedding, we design a novel probabilistic predictor to model the conditional distribution of future representations in the latent space, thereby capturing both predictive information and uncertainty about future dynamics. Theoretically, we show that 1) the JEPA objective seeks to maximize a variational lower bound on the mutual information between historical representations and future trajectories; and 2) predictive future-conditioned policy learning admits a finite-sample performance bound with an O(N-1/4) estimation term. Experiments on D4RL benchmarks demonstrate that DT-JEPA achieves efficient adaptation across varying offline RL tasks, outperforming traditional DT-based and other offline RL methods.
[LG-22] WL-SeqVAE: Sequence-Based Graph Generation Using Variational Autoencoders and Weisfeiler-Leman Colors
链接: https://arxiv.org/abs/2610.32650
作者: Iakovos Evdaimon,Giannis Nikolentzos,Achilleas Tsimichodimos,Christos Xypolopoulos,Michalis Vazirgiannis
类目: Machine Learning (cs.LG)
*备注:
Abstract:In this paper, we introduce WL-SeqVAE, a sequence-based variational autoencoder for generating graphs. The key observation behind WL-SeqVAE is that node colors induced by the Weisfeiler-Leman (WL) algorithm provide compact discrete summaries of local graph structure and can therefore serve as structural tokens for generation. Rather than learning a distribution over adjacency matrices directly, our method maps each graph to a sequence of WL colors and employs a variational autoencoder that encodes this sequence into a Gaussian latent distribution, from which an autoregressive decoder reconstructs the graph. During generation, compatibility constraints derived from the WL colors guide node and edge predictions, combining global information from the latent representation with local structural consistency. This formulation turns graph generation into a sequence generation problem and avoids directly parameterizing a quadratic adjacency matrix. We evaluate WL-SeqVAE on molecular and general graph generation tasks. Across the evaluated datasets, WL-SeqVAE is competitive with established baselines on several local structural and molecular quality metrics. Code is available at this https URL
[LG-23] he Standard Deviation in VICReg Buys a Safety Margin
链接: https://arxiv.org/abs/2610.32630
作者: Hawar Dizayee
类目: Machine Learning (cs.LG)
*备注: 11 pages. Code: this https URL
Abstract:VICReg prevents collapse with a hinge on the standard deviation of each embedding dimension, and its authors argue from the gradient that a hinge on the variance would fail. To our knowledge, this claim has not been tested experimentally. We test it in a controlled setting (ResNet-18, CIFAR-10 and CIFAR-100), changing only the hinge and sweeping its weight \mu over five orders of magnitude. At the default weight ( \mu = 1 ), the two hinges reach the same linear-probe accuracy on CIFAR-10 (about 89.5%). Both collapse below a threshold weight, but the standard-deviation hinge’s threshold is 35-130 \times lower. CIFAR-100 shows a consistent gap. In training, the variance hinge’s gradient fades as the feature variance shrinks, while the standard-deviation hinge’s does not until that variance is very small. The standard-deviation hinge therefore buys a safety margin: it keeps training at weights far below the point where the variance hinge collapses.
[LG-24] Interpreting Supervised Pairwise Graph Representations: Local Evidence Confounding and Provable Post-hoc Correction
链接: https://arxiv.org/abs/2610.32611
作者: Ilya Brekin,Ariela Moldavskaya,Elina Sidorova,Ural Valeev
类目: Machine Learning (cs.LG)
*备注:
Abstract:Supervised pairwise graph representations convert pairwise predictive relations into sample-specific graphs, but graph-level accuracy does not by itself show that higher-order topology carries the signal. In a controlled Transformer error-detection task, supervised pairwise graphs outperform attention-as-adjacency (ROC-AUC 0.839 versus 0.559 ; the comparable no-confidence pair is 0.687 versus 0.559 ), but their apparent advantage over model confidence disappears after controlling for sequence length, which is almost perfectly recoverable from the pooled hidden state ( R^2=0.9995 ). On published HCP Emotion ensemble graphs, signed node aggregation already acts as an ensemble of supervised pairwise predictions. Edge magnitude alone is near chance; topology-aware augmentations change ROC-AUC by at most 0.004 , with only sign-separated degree showing small positive changes across all tested graph sizes, while a validation-tuned three-layer GCN provides no measurable improvement over the signed node-summary baseline. A Fisher post-hoc corrector satisfies its distribution-agnostic guarantees across the evaluated feature representations and remains stable under reduced labelled budgets; in a near-error-free fMRI calibration regime the bound correctly becomes statistically uninformative. A separate cross-attention analysis identifies a redundant reversal circuit and downstream EOS-timing and copy failures. Higher-order topology should therefore be credited only after local supervised evidence and recoverable nuisance variables have been excluded.
[LG-25] rajectory Matching: Discrete Flow Matching with Structured Actions
链接: https://arxiv.org/abs/2610.32602
作者: Niklas Kemper,Marten Lienen,David Lüdke,Stephan Günnemann
类目: Machine Learning (cs.LG)
*备注:
Abstract:The natural units of structured discrete generation are often not independent single-component edits but richer actions: attaching a fragment to a molecule, inserting a node together with an incident edge, or growing a hierarchy by adding a child. Yet most existing discrete flow matching methods (DFM) are restricted to independent component-wise edits over fixed-size states. This restriction stems from endpoint conditioning: endpoints fix the source and target during training but not the action sequence between them, making tractable conditional rates hard to define for complex actions that cannot be factorized component-wise. We propose trajectory matching, which conditions the flow on a full sequence of actions connecting the endpoints. This yields principled, simulation-free training targets for user-defined action spaces, including actions that change state size, modify multiple components, or depend on earlier actions. We prove that the resulting objective recovers the marginal flow and that trajectory matching strictly subsumes factorized DFM, while additionally capturing actions with prerequisite structure such as hierarchy growth or fragment-based molecule generation. Empirically, on tree and molecular graph generation, trajectory matching with domain-aligned actions matches or surpasses factorized DFM baselines (lifting MOSES validity from 92.8% to 99.2%) and transfers without retraining to substructure extension and linking. Trajectory matching enlarges the design space of discrete flow matching by turning the action vocabulary into a modeling choice. An interactive explanation of trajectory matching is available at this https URL.
[LG-26] Universal neural network approximation of symmetric polyconvex energies and envelopes
链接: https://arxiv.org/abs/2610.32598
作者: Florian Gammer
类目: Numerical Analysis (math.NA); Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注:
Abstract:Symmetric polyconvexity provides a sufficient condition for weak lower semicontinuity in small-strain elasticity and a tractable lower bound for the relaxation of non-(quasi)convex energies. Since classical algorithms for computing such functions can be computationally expensive, neural network approximations have recently emerged as an alternative. A central difficulty with establishing such neural network based approaches for symmetric polyconvexity is the additional monotonicity property. As we show, the corresponding weight restrictions on ReLU-based input-convex networks do not yield universal approximation of the lifted representers. Here, we introduce symmetric polyconvex neural networks (SPCNNs) for constitutive modeling and approximation of symmetric polyconvex envelopes. The deep architecture combines convexity with the monotonicity required of a representer in the space of minors. We prove a universal approximation theorem for symmetric polyconvex energies, without requiring isotropy, and quantify the number of affine pieces attainable by the lifted network and derive reduced architectures for isotropic energies. Numerical examples illustrate stress-based regression and approximation of the symmetric polyconvex envelope of a double-well potential. Symmetric polyconvexity is guaranteed by construction, while the accuracy of the fitted energies and envelopes is assessed numerically.
[LG-27] Contextual Decision Making: Predictive or Generative
链接: https://arxiv.org/abs/2610.32584
作者: L. Jeff Hong,Weihuan Huang,Xinyao Li
类目: Machine Learning (cs.LG); Optimization and Control (math.OC); Machine Learning (stat.ML)
*备注: 63 pages
Abstract:Contextual decision making uses observed covariates to guide decisions under uncertainty. Two natural approaches are predict-then-optimize (PtO), which learns the conditional expected objective, and generate-then-optimize (GtO), which learns conditional uncertainty and optimizes using generated scenarios. We study when each approach is preferable by examining how the choice of learning target affects statistical accuracy and optimization tractability. Although PtO learns a narrower target, our regret bounds show how differences in smoothness can offset GtO’s additional dimensional burden. Exploiting smoothness also raises an optimization tradeoff: within the class of nonnegative-weight PtO methods studied, a lower bound limits the gains from additional smoothness, while more flexible methods may introduce signed weights and lose the guarantee of preserving convexity. GtO preserves scenario-wise convexity, but requires learning a richer object and generating scenarios for optimization. Synthetic experiments and a real case study illustrate how these tradeoffs vary with data availability and problem structure. Together with the scope for reusing a model across tasks, our findings provide practical guidance for choosing between the approaches without identifying a universal winner.
[LG-28] From Parameter Rebasin to Functional Transfer Across Heterogeneous Transformers
链接: https://arxiv.org/abs/2610.32581
作者: Filippo Rinaldi,Aniello Panariello,Giacomo Salici,Simone Magistri,Angelo Porrello,Andrew D. Bagdanov,Simone Calderara
类目: Machine Learning (cs.LG)
*备注:
Abstract:Adapting pretrained Transformers to downstream tasks produces task-specific updates that are expensive to reproduce for every model variant. Reusing these updates across different depths and widths is challenging because their parameters have incompatible shapes and do not share a canonical coordinate system. We introduce Ariadne, a training-free method that transfers task-specific behavior across heterogeneous Transformers without resizing the source model or transporting its parameters. Rather than treating the source task vector, i.e. the difference in parameter space between the fine-tuned and the base models, as the object to move, Ariadne characterizes fine-tuning by the activation changes it induces at matched blocks. First, it aligns these effects to a target base through local Procrustes maps. Then, it directly synthesizes a native target update by closed-form ridge regression. Across Transformer expansion, reduction and language-model transfer, Ariadne improves target-model performance while substantially reducing the time and memory required to construct an update relative to previous cross-architecture rebasin approaches. These results show that local functional effects provide an efficient alternative to parameter transport.
[LG-29] Amortized Thompson Sampling and Efficient Information-Directed Sampling for Simulator-Based Design
链接: https://arxiv.org/abs/2610.32514
作者: Michal Kobiela,Michael U. Gutmann
类目: Machine Learning (cs.LG)
*备注:
Abstract:Thompson sampling is a widely used strategy for sequential decision-making under uncertainty, offering a principled way to balance exploration and exploitation. This strategy involves maximizing a target objective under a single sample drawn from the current belief over unknown quantities. Conventionally, this optimization is performed once the belief has been inferred from data, and it becomes computationally burdensome when many Thompson samples are required, especially in continuous action spaces. In this work, we introduce an approach that amortizes this process ahead of time, before any data is collected, so that an approximation of the full Thompson sampling distribution is available immediately once observations arrive. We employ this approach to solve the multi-round setting, involving sequential data acquisition and Thompson-sample recommendation across rounds. Beyond accelerating Thompson sampling, our approach enables efficient estimation of expected information gain about the optimal action. This can be used to select more informative Thompson samples or to obtain maximally informative designs with optimization. Combining this estimate of expected information gain with immediate regret supports information-directed sampling (IDS), which can outperform Thompson sampling by explicitly balancing these two quantities. We demonstrate our approach on two simulator-based design problems in synthetic biology, where the simulators have intractable likelihoods.
[LG-30] Quantum-Classical Link Prediction on Dark Web Hyperlink Graphs
链接: https://arxiv.org/abs/2610.32510
作者: Vojtěch Novák,Radovan Rečka,Ondřej Knebl,Aktham Youssef,Ivan Zelinka
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:We analyze HTML content and hyperlinks from an offline Tor crawl containing 383 onion hosts and 2,067 directed cross-host links. The study combines semantic analysis with degree-controlled quantum–classical link prediction. Two directory groups emit 94.8% of the observed host links. Directed preferential attachment reaches ROC-AUC 0.976 with random non-edges, but falls to 0.550\pm0.014 under source-conditioned target-degree matching. With disjoint training and test pairs, multimodal logistic regression reaches 0.672\pm0.021 , GAE 0.899\pm0.016 , and VGAE 0.906\pm0.016 . A four-qubit variational quantum classifier (VQC) comparison tests two encodings across 20 disjoint-pair graph splits under exact and 1,024-shot simulation, with classical controls using the same inputs and training examples. Its best exact mean ROC-AUC is 0.549\pm0.023 , compared with 0.551\pm0.016 for RBF SVC and 0.573\pm0.029 for a 16-parameter MLP on the same inputs and training examples. These results support predictive signal in combined graph/content features and show why degree control, feature compression, and optimization conditions matter in quantum–classical comparisons.
[LG-31] BAARC: Uncertainty-Driven Adaptive Partial Spectral Measurement to Full Reconstruction for Material Classification
链接: https://arxiv.org/abs/2610.32505
作者: Adithya Ashok Chalain Valapil,Aishwarya Venkataramanan,Joachim Denzler
类目: Machine Learning (cs.LG)
*备注:
Abstract:Spectroscopic instruments with limited wavenumber coverage require accurate scan positioning or multiple measurements to complete the material analysis. This often requires prerequisite knowledge about the material under diagnosis to measure the spectral regions with distinguishing peaks. In our study, we simulate a Bayesian Adaptive Acquisition with Reconstruction-driven Classification (BAARC) framework for Raman spectral measurements, using masked training to learn spectral features that dynamically drive laser channel selection until the material is identified. We present a jointly trained two-level framework that takes progressive partial measurements of small wavenumber patches across the complete spectral range and, at each step, first attempts to reconstruct the full spectra until the model uncertainty is low. Secondly, material classification is performed at each reconstruction step until stoppage. Uncertainty quantification during reconstruction steps helps determine the measurement position of our next patch, while combining it with classification uncertainty determines measurement termination. By closing the loop between hardware acquisition and deep probabilistic modeling, the proposed system minimizes physical measurement steps while maintaining both spectral fidelity and diagnostic accuracy. Experimental evaluations demonstrate that our framework achieves high-accuracy sample identification using only a fraction of full-spectrum channels, offering a scalable solution for real-time, high-throughput spectroscopic instruments.
[LG-32] DRIVE: Fine-tuning via Data Contribution- and Diversity-aware Weighting with Prior Regularization NEURIPS2026
链接: https://arxiv.org/abs/2610.32501
作者: Qing Liu,Xinrui Chen,Yi Du,Ou Wu,Weiyao Zhu
类目: Machine Learning (cs.LG)
*备注: Published in NeurIPS 2026. 39 pages, including references and appendices
Abstract:Single-stage fine-tuning is pervasive, yet catastrophic forgetting persists when the source pre-training data is inaccessible. Data weighting affords a natural lever to modulate sample influence. Yet existing weighting methods lack a principled mechanism to explicitly arbitrate knowledge retention against new-task adaptation, and provide no unified differentiable objective to accommodate auxiliary signals such as sample diversity and sample characteristics. This is because current schemes privilege one goal over the other, and no explicit gradient-compatible encoding exists for such signals. To address this, we propose DRIVE, a principled and efficient weight assignment framework. It derives per-sample weights via a unified differentiable objective where three coupled signals co-evolve: Shapley additivity splits contribution into anti-forgetting and adaptation components, which are reshaped by a covariance-based diversity term and stabilized by a sample-characteristic prior. Far from a mere combination, this yields an end-to-end gradient-based scheme balancing stability against plasticity. Theoretically, we prove that DRIVE enjoys superior stability-plasticity guarantees over existing weighting schemes and enhances parameter estimation robustness. Empirically, it attains superior overall performance, faster convergence, and stronger robustness. Code at this https URL.
[LG-33] A curated multi-endpoint drug-protein bioactivity dataset with construct-aware protein targets
链接: https://arxiv.org/abs/2610.32490
作者: Lina Abou-Abbas,Nour Mhanna,Hasan Bazzi,Sleiman El Hajj,Abdulkader Helwan,Rabah Omrani,Khadidja Henni
类目: Machine Learning (cs.LG)
*备注:
Abstract:Public drug-protein bioactivity resources commonly identify targets by UniProt accession, although one accession may represent distinct experimental sequences. We present Curate-DTI, a construct-aware dataset derived from the 1 July 2026 BindingDB export, with IC50, Ki, Kd and EC50 processed independently. The release contains 2,919,717 measurements, 1,291,872 standardized ligands, 9,668 sequence-defined protein entities and 2,275,683 endpoint-specific interactions spanning 8,336 reviewed UniProt accessions. Protein entities combine the accession with the reported experimental sequence, preserving constructs that accession-only processing would merge. Across the four endpoints, 4,204 endpoint-specific ligand-accession groups showed different classification labels across sequence-defined protein entities. Exact and inequality-qualified observations provide relation-aware classification evidence, whereas regression targets use exact values only. Connected Parquet tables preserve individual measurements, standardized structures, protein sequences, aggregated interactions and available assay and source provenance. The resource supports endpoint-specific classification and regression while allowing users to select molecular representations, protein representations and evaluation splits. Preprocessing code, a column dictionary, exclusion reports and validation summaries accompany the release.
[LG-34] State Estimation as In-context Tabular Prediction
链接: https://arxiv.org/abs/2610.32487
作者: Ayushi Jolotia,Parikshit Pareek
类目: ystems and Control (eess.SY); Machine Learning (cs.LG)
*备注: 3 pages, 5 figures
Abstract:This letter poses state estimation (SE) as an in-context tabular prediction (ICTP) problem. The key idea is a tabular arrangement in which historical measurement vectors and corresponding state estimates serve as the context, and a new measurement vector forms the query for which we predict a state vector. We propose to use tabular foundation models (TFMs), pretrained on synthetic data, whose weights are never updated, so power system data is needed only as context. With frozen TFM weights we estimate the state of 14-, 30-, 33-, 70-, and 118-bus transmission and distribution networks, and do so where weighted least squares (WLS) fails or degrades, under low observability and with bad data. ICTP errors are 11 to 57 \times below the per-bus average of the context at full metering, and 1.3 to 4.0 \times below it with only 5% of measurements available. With measurements at six buses biased by 8 \sigma , the ICTP angle error rises 1.8 to 5.0 \times over its bias-free value against 2.5 to 12.7 \times for WLS with bad-data detection. Without retraining, ICTP has lower voltage error than DSS ^2 on both of its feeders, and lower than a retrained TAPINE on two of three reconfiguration sets.
[LG-35] Sharpness-aware Orthogonalized Momentum Algorithms
链接: https://arxiv.org/abs/2610.32465
作者: Geetika,Shivani Singh Chauhan,Gauranshi Gupta,Anant Jain,Kaushik Kalyanaraman,Bapi Chatterjee
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:Adversarial sharpness penalty per iteration of a base optimizer, known as sharpness-aware minimization (SAM), is an accepted technique to improve generalization in deep learning. Recently, Muon has emerged as a dominant optimization algorithm for training large language models (LLMs). It achieves lower sample complexity for heavy-tailed gradient updates - a phenomenon common in LLM training. However, sharpness penalties with Muon updates are yet largely unexplored in the literature. Moreover, the behavior of SAM under heavy-tailed noise is unknown both theoretically and empirically. A potential reason for this gap is the additional cost of extra gradient computation per iteration for penalty perturbations in SAM, which becomes prohibitive for LLMs. This paper introduces Sharpness-aware Orthogonalized Momentum Algorithm (SOMA) to combine a computationally efficient and theoretically grounded perturbation scheme with the Muon algorithm. The key idea is to use stale (purely) stochastic gradient noise for adversarial perturbations. Theoretically, we show that SOMA’s sample complexity is invariant to the perturbation direction. We provide explorations and insights into SOMA’s robustness to heavy-tailed gradient noise. We extensively explore SOMA’s efficacy in (a) mitigating quality degradation due to post-training quantization, and (b) catastrophic forgetting during fine-tuning of LLMs.
[LG-36] DualGSL: Prior-Guided Graph Structure Learning with Node-Adaptive Prediction Fusion
链接: https://arxiv.org/abs/2610.32462
作者: Yuxia Sun,Yutszyuk Wong,Huihong Chen,Longhao Lv,Chang-Dong Wang,Li Zeng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph structure learning (GSL) adapts propagation graphs to prediction tasks, but the relative predictive value of feature-derived priors and learned relations can vary across nodes. We propose DualGSL, a prior-guided framework for transductive node classification that learns node-specific weights for combining predictions from separate prior and task-adaptive propagation paths. The prior branch augments a feature graph with virtual class nodes to incorporate training-label information, while the structural branch combines learned affinity graphs with the feature-graph prior. The two branches share encoder parameters while preserving distinct representations and classification logits. A node-adaptive gate fuses the branch logits based on their projected representations, element-wise absolute difference, and prediction confidence. A joint objective integrates fused and branch-specific classification losses with prior-branch class-anchor alignment, coordinating graph construction, representation learning, and prediction fusion. Experiments against seven baselines on the AMG18 malware attribution dataset and the widely used 20News text classification and USPS handwritten-digit recognition benchmarks show that DualGSL achieves the highest cross-dataset average accuracy and Macro-F1 among the evaluated methods. Ablation studies on AMG18 further demonstrate the benefit of combining prior and learned propagation through logit fusion.
[LG-37] Data Quality Audit and Bivariate Temperature-Humidity Analysis of a Long-Term LoRaWAN Network: Anomaly Detection PCA Clustering and Machine Learning
链接: https://arxiv.org/abs/2610.32438
作者: Mamadou Sow
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Atmospheric and Oceanic Physics (physics.ao-ph); Machine Learning (stat.ML)
*备注:
Abstract:Long-term environmental IoT/LPWAN deployments often aggregate heterogeneous data streams from diverse devices without systematic auditing of each variable’s actual coverage prior to analysis-an increased risk when multiple sensor types coexist within the same network. This work proposes a C++17 processing pipeline applied to the Tour Perret LoRaWAN dataset (421 937 frames, June 2021–June 2023, 5 devices). A systematic coverage audit integrated into the extraction phase reveals that CO2 is completely missing (0% coverage) and pressure is reported only by an isolated base module (Wyres, 4.24% of the corpus), invalidating the initial hypothesis of a joint four-variable analysis. The analysis is refocused on the Temperature–Humidity pair, jointly available across 205,005 frames (95.76%) from 4 Elsys EMS sensors. The pipeline sequentially performs anomaly detection (modified Z-score, Isolation Forest), bivariate Principal Component Analysis, clustering (K-means, DBSCAN), and a multilayer perceptron predicting humidity from temperature and time, whose residuals serve as a detector for slow drift. On real-world data: Temperature–Humidity correlation r=-0.662, two-component PCA (PC1 =83.1% of variance), Isolation Forest detects 2.12% multivariate anomalies while univariate Z-score detects none, calibrated DBSCAN isolates 0.17% noise across 13 clusters, and the MLP achieves R^2=0.672 (RMSE =10.54%) in cross-validation. This work demonstrates that per-variable coverage auditing prior to modeling is a necessary methodological step for heterogeneous multi-sensor IoT datasets.
[LG-38] When Does Dynamics Randomization Test Context Inference?
链接: https://arxiv.org/abs/2610.32414
作者: Oleg Smirnov,Sofiane Ennadir,John Pertoft,Bjartur Hjaltason,Sara Karimi
类目: Machine Learning (cs.LG)
*备注:
Abstract:Environments with randomized dynamics are widely used to ask whether world models need explicit context inference. However, randomization can leave little predictive information for a context encoder to add, because distinct latent parameters may produce similar transitions at visited states, or those states may already reveal the active context. We formalize the remaining one-step information as predictive headroom and compare kernels under fixed and visited state–action distributions to distinguish changes in query coverage from changes in context weights. We analyze discrete and continuous environments and find low headroom both when kernels differ little and when visitation reduces an otherwise large discrepancy. At the same time, resetting state components increases headroom, showing that context evidence accumulates in state. We propose reporting this headroom alongside the randomized parameters, so that gains from explicit context inference can be interpreted relative to the information missing from the current state and action.
[LG-39] KnitTrans: Coordinated Trajectory Transfer for Effective and Efficient LLM Reinforcement Learning
链接: https://arxiv.org/abs/2610.32389
作者: Bangjun Xiao,Chiheng Lou,Xiaodong Ji,Xiangwei Deng,Yikai Zhao,Gang Huang,Hailin Zhang
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Trajectory transfer from rollout to training affects both learning quality and system efficiency in large language model reinforcement learning (RL). Enforcing batch-wide training requirements can delay the admission of completed trajectories, resulting in higher architectural staleness: the additional parameter-version gap from disaggregated pipelining relative to colocated execution. Beyond admission, misalignment between data access and training consumption causes redundant movement and unnecessary waiting when transferring large trajectory payloads. We present KnitTrans, a system that coordinates the rollout-to-training path through three designs. First, constraint-preserving mini-batch admission allows flexibility in early mini-batch composition, while checking that enough capacity remains to satisfy the remaining quotas. Second, on-demand trajectory data access matches storage units to consumer assignments and overlaps micro-batch data preparation with computation. Third, multimodal encoding deduplication supports reuse across a mini-batch while preserving gradient contributions from all backbone micro-batches. Experiments show up to 93% lower architectural staleness than with train-batch admission, narrowing the learning-quality gap between disaggregated execution and colocated execution. On multimodal workloads, KnitTrans reduces transfer latency by two orders of magnitude and achieves a speedup of up to 5.8 times in transfer plus training over the baseline.
[LG-40] Detection and Validation of Hidden Subgroup Disparities in Clinical AI
链接: https://arxiv.org/abs/2610.32362
作者: Alceu Bissoto,Trung-Dung Hoang,Tim Flühmann,Nina Weng,Susu Sun,Christian F. Baumgartner,Lisa M. Koch
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine learning (ML) models used in clinical workflows often perform systematically differently across patient groups, raising concerns about equity and patient safety. Traditional subgroup analyses can expose such disparities only when rich metadata (e.g., patient demographics) is available and captures the sources of performance variation, conditions rarely met in practice. Algorithmic subgroup discovery methods offer a solution by identifying coherent groups from learned feature representations, but because true underlying stratifications are unknown, these methods are hard to evaluate. As a result, subgroup discovery has not been systematically applied or assessed for performance monitoring. Our main objective is to evaluate algorithmic subgroup discovery for performance monitoring, assessing the discovered subgroups’ performance gaps and semantic coherence. Investigating cardiomegaly classification in a dataset of 223,228 chest X-rays, we showed that the discovered subgroups exhibited larger performance gaps (accuracies of 0.61-0.90) than traditional metadata-based analysis (0.80-0.94) and were frequently interpretable using patient, acquisition, and model metadata. As a proof of concept, we also showed that annotations from learned subgroups can be used to mitigate the biases they expose. We conclude that subgroup discovery is a practical and effective tool to enhance the performance validation and monitoring of ML systems in healthcare.
[LG-41] Optimal Multiclass Online Regret with Arbitrary Label Spaces
链接: https://arxiv.org/abs/2610.32339
作者: Sajad Ashkezari,Shai Ben-David
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We determine the optimal regret in agnostic multiclass online learning with arbitrary, possibly infinite, label spaces. For every hypothesis class of multiclass Littlestone dimension d\infty , we prove a regret upper bound of O(\min\T,\sqrtdT) over T rounds, matching the known lower bound. This resolves a question of Hanneke et al. (2023) and extends the optimal binary regret bound of Alon et al. (2021). Our proof strengthens the one-sided expert cover of Hanneke et al. (2023) to a fractional cover. We combine this cover with a minimax reduction to a finite expert game, whose best-response rewards provide the disagreement bounds needed for chaining.
[LG-42] When to Trust Memory: Retrieval-Guided Probabilistic Spatiotemporal Forecasting under Distribution Shift NEURIPS2026
链接: https://arxiv.org/abs/2610.32329
作者: Haochen Lv,Jianhao Zhang,Zhichen Lei,Yang Jiang,Xiaowei Mao,Shengnan Guo,Youfang Lin,Huaiyu Wan
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Spatiotemporal forecasting plays an important role in real-world applications, where probabilistic approaches provide uncertainty estimates for decision-making. Real-world systems are inherently non-stationary, leading to distribution shifts between training and test data. However, existing probabilistic methods mainly focus on modeling data distributions under standard settings, while distribution shift methods often handle such shifts in a coarse manner, failing to characterize shift degree and adapt prediction strategy accordingly. As a result, they often produce miscalibrated and unreliable forecasts. To address this issue, we propose REMIND, a retrieval-guided probabilistic framework for spatiotemporal forecasting under distribution shift. REMIND leverages retrieval to estimate how well current observation are supported by historical patterns, enabling adaptive prediction under different shift regimes. It employs spatiotemporal retrieval to construct a memory bank and introduces a dual branch architecture, combining a memory-calibrated probabilistic branch for mild shifts with a deviation-aware probabilistic branch for severe shifts. A shift-adaptive gate, guided by retrieval similarity and memory dispersion, balances the two branches according to shift degree. We further design a hierarchical Gaussian mixture to capture heterogeneous uncertainty over the dual branch predictions, and introduce a memory perturbation training scheme to improve generalization. Experiments on real-world datasets with distribution shifts demonstrate that REMIND consistently outperforms state-of-the-art methods in both forecasting accuracy and probabilistic prediction quality.
[LG-43] SKG in Practice: OOD Generalization Cases Across FourDomains
链接: https://arxiv.org/abs/2610.32314
作者: Jian An,Chaoshi Wu,Xinpan Luo
类目: Machine Learning (cs.LG)
*备注:
Abstract:This paper presents the engineering implementation methods of the SKG inference framework and provides complete case studies across four domains. Case 1, 2D Dissipative Mechanical System: OOD MSE is 1.69 to 4.12 times lower than the best neural baseline; the state-dependent metric yields a 10^8-order improvement under controlled conditions. Case 2, Target Detection Sim-to-Real: Real-machine mAP increases from 0.2564 to 0.4964 (+24.00 points, single-run experiment, 32 images, point estimate), and AP75 increases from 0.1736 to 0.5765 (+40.29 points), with no real-machine retraining required. Case 3, Sequence Modeling sine to square: OOD MSE is 0.0296 versus Transformer 0.0641 (mean of 5 seeds, p=0.024). Case 4, Molecular Dynamics MD17: ID@100 ADE is 0.0285 versus ESTAG 0.1467; OOD@100 is 0.0316 versus 0.1411 (single seed). For each case, we present the objective, applied SKG theory, parameter settings, baseline comparison protocol, data results, and sources of advantage. Following the cases, we provide the complete methodology: pre-validation, task invariant construction, three-stage training, residual training, and the four theorems and one proposition supporting these methods. This paper not only demonstrates effective configurations of SKG across four domains but also honestly reports the variance of single-run experiments, limitations of single-seed results, negative updates caused by parameterization mismatch, and practical experience of using engineering thresholds as diagnostic signals rather than hard constraints.
[LG-44] Supervised Source Separation of Electric Vehicle Interior Noise Using a Hybrid Transformer Architecture
链接: https://arxiv.org/abs/2610.32310
作者: Philipp Diel,Zhuo Cao,Arne Hildenbrand,Marco Günther,Stefan Pischinger,Hanno Scharr,Steffen Marburg
类目: ound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
*备注:
Abstract:The interior sound of an electric vehicle is a superposition of several distinct noise contributions - wind noise, tire noise, powertrain (electric drive unit and inverter) noise, and auxiliary-drive noise - whose relative balance governs both the overall loudness and the perceived sound quality in the vehicle interior. Decomposing a recorded vehicle interior signal into these source contributions is a prerequisite for targeted acoustic engineering, but requires effortful measurement campaigns as well as expertise in the field of transfer path analysis. This work formulates the decomposition as a supervised, binaural source-separation problem with four target source contributions and adapts a hybrid time-spectrogram transformer architecture, originally developed for music demixing, to the domain of sound source separation for technical systems. To disambiguate source contributions, conditioning is introduced in the form of operating-state information, including electric drive speed, electric drive load, vehicle speed, and auxiliary-drive speed. Since isolated real-world reference stems are costly to obtain, the model is trained on a physics-based parametric data generator that synthesizes labeled component sounds from randomized virtual-vehicle hardware configurations and driving scenarios, so that the ground-truth contribution of each source is known by construction. Validation on both synthetic and measured vehicle interior sounds probes the achieved separation performance and underscores the feasibility of applying the proposed method for acoustic contribution analysis in technical systems.
[LG-45] BiLoCo: Binary Low-Rank Corrections for Post-Training Quantization
链接: https://arxiv.org/abs/2610.32243
作者: David Jin,Beshr IslamBouli,Tarushii Goel,Han Guo,Yoon Kim
类目: Machine Learning (cs.LG)
*备注:
Abstract:Post-training quantization (PTQ) to FP4 has emerged as a key technique for reducing large language model inference cost, especially on Blackwell GPUs, which offer native FP4 tensor core support. Direct PTQ to FP4, however, still leaves a noticeable accuracy gap. Recent work on SVDQuant mitigates the errors introduced through FP4 quantization for diffusion models by adding high-precision low-rank corrections obtained via SVD. However, this high-precision low-rank correction requires many bits per rank, and it runs on the same tensor cores as the main FP4 GEMM. We introduce BiLoCo, a low-rank correction method whose left and right factors are stored as binary \pm 1\ sign vectors rather than as BF16 numbers. The signed format uses 16\times less storage per rank-one component, and since the correction now consists of additions/subtractions, this can be run on CUDA cores concurrently with the FP4 tensor-core GEMM. On 4B–72B LLMs, BiLoCo matches or improves memory-matched SVDQuant for PTQ. We implement BiLoCo on B300 as a fused single kernel, where the FP4 GEMM runs on tensor cores, the bit-packed correction on the CUDA cores of the same launch, and the two are combined in the GEMM epilogue. With batch size 1 , the fused kernel is at most 11% slower than the FP4-only GEMM for every rank up to 512 . At rank 128 , it is at most 20% slower up to batch size 8 .
[LG-46] Correct but Unselectable: The Hidden Interface Tax in Multi-Candidate Reasoning NEURIPS2026
链接: https://arxiv.org/abs/2610.32235
作者: Qianli Ma,Zhiqing Tang,Fanshuai Meng,Zhi Yao,Hanshuai Cui,Weijia Jia
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Small language models deployed under cost, latency, or privacy constraints cannot simply switch to larger models when reasoning fails. Multi-candidate generate-then-rerank offers a natural test-time scaling route. However, more candidates do not always yield higher accuracy. We identify a previously overlooked bottleneck between generation and selection: a correct answer may appear in the candidate pool yet fail to be selected because it is incomplete, unparseable, or misaligned with the answer mode the fixed selector expects. We term this loss the hidden interface tax and formalize a coverage-conversion decomposition that distinguishes raw oracle coverage from selectable coverage, the subset of problems where a correct candidate can actually be consumed by the selector. We propose Compatibility-Aware Candidate Construction (CACC), an interface layer between the candidate source and the fixed selector that repairs fragments, aligns answer modes, and removes selector-confusing noise without retraining either component. On numeric reasoning (GSM8K, competition_math), CACC raises both oracle coverage and final accuracy. On GPQA Diamond, CACC alone improves final accuracy from 5.6% to 19.2% (+13.6 pp); combined with proposer-side strengthening it reaches 25.8% (+20.2 pp). On MMLU-Pro, CACC exposes a coverage-conversion gap: answer-mode-compatible construction raises oracle coverage, but final accuracy decreases by 0.85 pp under the original selector; adding proposer-side strengthening improves final accuracy.
[LG-47] MS-TTRL: Breaking the Frequency Conflict in Multiple-Solution Test-Time Reinforcement Learning
链接: https://arxiv.org/abs/2610.32234
作者: Zhixia Zhang,Binqian Xu,Huaiyang Wang,Yikun Ban,Dianbo Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Scientific, engineering, and mathematical reasoning tasks can admit multiple valid solutions, while exhaustive annotation of the full solution set is often prohibitively costly. Test-time reinforcement learning (TTRL) applies reinforcement learning to unlabeled test instances, offering a label-free approach to multiple-solution reasoning. In this setting, generation frequency serves two opposing objectives: correctness estimation favors frequent outcomes, whereas outcome-level diversity preservation favors less frequent outcomes. We term this tension the \textbfFrequency Conflict. To resolve this conflict, we propose MS-TTRL, which obtains frequency-independent correctness estimates through Outcome-Level Self-Verification while reserving generation frequency solely for diversity preservation. Three-State Outcome Cache and the Candidate State support verification reuse across repeated observations and confidence-aware handling of ambiguous outcomes, while Confidence-Gated Inverse Probability Scaling selectively strengthens the learning signals of rare, reliably verified outcomes. Experiments across two benchmarks, four multiple-solution tasks, and three backbone models, each evaluated with five random seeds, show that MS-TTRL improves Recovery Rate by 19.13 percentage points on average over the best-performing label-free baseline and approaches the performance of the supervised IPS-GRPO.
[LG-48] RoboCode: Learning Reusable Action Primitives for Robot Manipulation
链接: https://arxiv.org/abs/2610.32225
作者: Xianyue Peng,Devi Amarsaikhan,Iman Soltani
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 12 pages, 5 figures
Abstract:Recent robot manipulation policies commonly predict continuous action chunks directly from observations, without explicitly modeling reusable structure in the action space. However, manipulation demonstrations often contain recurring short-horizon motion patterns that are shared across tasks. We introduce \textscRoboCode, which learns these patterns through a shared soft-gated primitive action interface. Rather than assigning each action chunk to a single discrete code, \textscRoboCode represents it as a soft composition of multiple learned primitives, providing a flexible intermediate representation of short-horizon actions. The primitive interface is learned from action chunks pooled across multiple tasks and then frozen while task-specific state encoders are trained to map current state inputs into the shared latent space. Experiments on five ManiSkill manipulation tasks address three questions. First, the learned primitive interface reconstructs action chunks with an average reconstruction MSE of 0.0187. Second, analysis of code activations reveals structured code reuse across tasks. Third, \textscRoboCode achieves the highest or tied closed-loop success rate on all evaluated tasks, with an average success rate of 75.6%, compared with 68.8% for ACT and 41.6% for the larger Diffusion Policy baseline. These results suggest that soft-gated primitive codebooks provide an effective and interpretable intermediate action representation for robot manipulation. Comments: 12 pages, 5 figures Subjects: Robotics (cs.RO); Machine Learning (cs.LG) Cite as: arXiv:2610.32225 [cs.RO] (or arXiv:2610.32225v1 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2610.32225 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-49] Democratizing LLM Training with the Edge: Challenges Systems and the Road Ahead
链接: https://arxiv.org/abs/2610.32215
作者: Ziyue Luo,Jiaxuan Cai,Cedric LeDenmat,Srijith Nair,Fatemeh Nourzad,Rohith Krishnan,Qinhang Wu,Jifan Zhang,Sungjae Lee,Zhe Li,Peiwen Qiu,Siddharth Shah,Rishabh Sharma,Sundararajan Srinivasan,Yinglun Xia,Xue Zheng,Zidong Liu,Bicheng Ying,Kaushik Chowdhury,Gauri Joshi,Yingbin Liang,Robert Nowak,Srinivasan Parthasarathy,Saurav Prakash,Balaraman Ravindran,Sanjay Shakkottai,Ness B.Shroff,Haibo Yang,Aylin Yener,Jia Liu
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:The growing computational power of modern GPU data centers has let a few industrial giants train and deploy large language models (LLMs) on massive amounts of data. Yet the sheer scale of these models and their training data makes centralized development increasingly hard to sustain, and at the same time shuts out smaller institutions that hold relevant private data but lack comparable infrastructure. In this paper, we organize our positions on distributed LLM training around the pretraining and fine-tuning paradigms, covering federated learning, split learning, and decentralized approaches, and we assess their readiness for geo-distributed settings with heterogeneous hardware and unreliable connectivity. We then present a \em unified system architecture built around a cloud-hosted parameter server that coordinates heterogeneous clients over outbound gRPC connections, reusing one communication, compression, and fault-tolerance stack for both pretraining and a family of fine-tuning methods. We validate the design with real geo-distributed experiments on GPT-2 Medium and Llama3-1B pretraining across multiple university sites, showing stable convergence under realistic conditions. Finally, we chart the road ahead, identifying open problems that span frontier model support, elastic fault tolerance, adaptive compression, federated alignment, privacy and trust, incentive design, and system–algorithm co-design, all of which must be addressed to make democratized LLM training a reality.
[LG-50] EVOLVE: Uncertainty-Aware Multivariate Time Series Anomaly Prediction with Adaptive Prototype Memories KDD2026
链接: https://arxiv.org/abs/2610.32189
作者: Jianxin Jin,Kai Zhao,Shiyan Hu,Yang Shu,Peng Chen,Meng Wang,Zhongwen Rao,Yijie Li,Jing Wang,Bin Yang,Chenjuan Guo
类目: Machine Learning (cs.LG)
*备注: 14 pages. An earlier version of this work was submitted to KDD 2026 on February 9, 2026
Abstract:Unsupervised multivariate time series anomaly prediction (AP) aims to estimate future anomaly risks from current observations, providing early-warning signals for proactive intervention and maintenance in complex systems. Existing unsupervised AP methods follow a perturbation-based paradigm, where predefined or random perturbations are injected into historical windows to simulate abnormal future evolution. However, these methods share perturbations across different variables and contexts, and learn deterministic history-to-future evolution. In reality, anomaly evolution is variable- and context-dependent: the same perturbation may be amplified into abnormal future dynamics in one variable or context, while being absorbed as normal fluctuation in another. In addition, future evolution is not uniquely determined by a historical window, since similar histories may lead to multiple plausible trajectories, such as gradual degradation or abrupt changes. To address these problems, we propose EVOLVE, a unified framework that models unsupervised AP via variable- and context-conditioned perturbations and uncertainty-aware future reasoning. EVOLVE constructs context-specific anomaly perturbations through an adaptive anomaly prototype memory, which can dynamically update and retrieve relevant perturbations at the variable level. EVOLVE models uncertainty by sampling multiple latent historical states and decoding through complementary smooth and deviation-amplifying evolution hypotheses, and the disagreement among decoded future trajectories is used to estimate future anomaly risk. Extensive experiments on six real-world benchmark datasets demonstrate that EVOLVE consistently outperforms state-of-the-art anomaly prediction baselines.
[LG-51] CFBrainAug: Modularity-Induced Counterfactual Brain Network Augmentation for Interpretable fMRI-based Autism Diagnosis
链接: https://arxiv.org/abs/2610.32184
作者: Yifan Qiao,Sheng Li
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine learning on resting-state functional magnetic resonance imaging (rs-fMRI) brain networks is promising for diagnosing Autism Spectrum Disorder (ASD), but existing methods are prone to overfitting on limited training data, and their post-hoc statistical attribution offers limited clinical interpretability. We propose CFBrainAug, a counterfactual brain network learning framework that unifies data augmentation and counterfactual explanation by learning label-altered brain networks. Unlike conventional label-preserving augmentation, CFBrainAug first retrieves a label-opposite yet topologically similar candidate and then refines it under a module-number constraint. Jointly training on factual and counterfactual networks enriches the training distribution around the decision boundary, enabling a lightweight feature-selection-based SVM to perform competitively with recent graph deep learning methods. Meanwhile, the counterfactuals provide intrinsic explanations by identifying the connectivity differences between each subject and its counterfactual. Experiments on the ABIDE I dataset show that CFBrainAug reaches an accuracy of 74.1% on the NYU site, suggesting that counterfactual augmentation can complement more complex models in small-sample brain network analysis.
[LG-52] Curvature-Aware Clipping: Mitigating Bias in DPSGD via Approximate Second-Order Information
链接: https://arxiv.org/abs/2610.32175
作者: Xincheng Xu,Thilina Ranbaduge,Thierry Rakotoarivelo,David Smith
类目: Machine Learning (cs.LG); Cryptography and Security (cs.CR)
*备注: Accepted at the 18th International Workshop on Optimization for Machine Learning (OPT 2026)
Abstract:Differentially Private Stochastic Gradient Descent (DPSGD) has become the de facto standard for training deep neural networks with rigorous privacy guarantees. However, the utility of DPSGD is constrained by gradient clipping bias and DP noise. While existing work has tried to address these issues, the geometric misalignment caused by clipping bias remains a critical challenge. To bridge this gap, we propose a novel framework, Curvature-Aware Differentially Private Stochastic Gradient Descent (CA-DPSGD), which leverages approximate second-order information to construct an optimal, curvature-aware clipping space. CA-DPSGD integrates a rectified estimator for the curvature matrix and is coupled with an anisotropic noise mechanism to ensure privacy guarantees. Extensive empirical experiments demonstrate that our approach outperforms state-of-the-art clipping methods.
[LG-53] PRISM: Promoting Sampling Diversity of Large Language Models with Implicit Stochastic Modulation
链接: https://arxiv.org/abs/2610.32101
作者: Qixin Xu,Zihan Zhu,Xiaojiang Liu,Bohan Lyu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large language models (LLMs) exhibit mode collapse, where their output distribution concentrates on familiar semantic modes. When diversity disappears, repeated sampling becomes semantic repetition rather than discovery, making rare but stronger solutions unlikely to surface and leaving self-evolution and continual learning difficult to sustain. Most remedies intervene only at inference time, shaping sam- ples without making diversity a model capability. We propose PRISM, a training-time framework that internalizes diversity within the model. PRISM combines (i) submodular curation that pairs each prompt with diverse-yet-correct multi-teacher solutions, and (ii) stochastic embedding modulation, which turns special-token perturbations into implicit signals for sampling distinct outputs. A theoretical analysis establishes the stability of this mechanism. On WritingPrompts and BookMIA, PRISM moves the quality-diversity frontier outward, achieving the best generation quality in all model-benchmark settings while maintaining the strongest or near-strongest semantic diversity across two open-weight model families. It also improves sampling efficiency on math and code benchmarks. Beyond inference, PRISM yields a stronger foundation for reinforcement learning, sustaining diverse rollouts and mitigating the diversity collapse commonly observed during RL training.
[LG-54] DBCM: Dual-Branch Calibration Module for Test-Time Adaptation in Time Series Forecasting
链接: https://arxiv.org/abs/2610.32071
作者: Jaebyeong Jeon,Kunwoo Kang,Sang Yup Lee,Kibok Lee
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Non-stationarity induces test-time distribution shifts in time series forecasting (TSF), resulting in forecast residuals between predictions and the corresponding ground truth. Recent works on test-time adaptation for TSF address this by introducing lightweight calibration modules around frozen forecasters, updated using observations that become incrementally available at test time. Existing approaches typically employ channel-wise adapters, implicitly assuming that forecast residuals are inherently channel-specific. However, we observe that forecast residuals often exhibit substantial shared structure across channels, suggesting that fully independent channel-wise parameterization may be redundant. Motivated by this observation, we propose the Dual-Branch Calibration Module (DBCM), which shares adaptation capacity across channels through a shared adapter while retaining limited channel-specific flexibility through a low-rank channel-wise adapter, further enhanced by channel mixing and nonlinear gated modulation. Extensive experiments show that DBCM consistently outperforms prior methods across diverse datasets, horizons, and backbones.
[LG-55] GS-MoE: Training-Free MoE Compression via Groupwise Subspace Sharing and Pruning
链接: https://arxiv.org/abs/2610.32009
作者: Carson Cary,Yukai Jin,Yihao Huang,Zhankai Ye,Fangyi Li,Ting Hua,Xin Liu,Shangqian Gao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Mixture-of-Experts (MoE) models reduce computation through sparse activation, but storing their expert parameters remains a major memory bottleneck. We introduce GS-MoE, a training-free compression framework that addresses redundancy across and within experts through two compression axes: shared-subspace rank and expert intermediate width. Groupwise low-rank representations capture shared structure across experts while retaining expert-specific projections, and Nyström-based neuron compression reduces intermediate width within each expert. For fixed shared subspaces, we derive per-expert reconstruction-error reductions from the reconstruction objective and use them to guide expert grouping. Budgets are allocated hierarchically using a gradient-free layer-importance score to guide the allocation of rank and width budgets across layers, which eigenvalues and leverage then distribute as non-uniform ranks across groups and intermediate widths across experts. Experiments demonstrate the complementary benefits of compression along both axes, particularly at high compression rates on fine-grained MoE models. On DeepSeek-MoE-16B-Base, at a 60% reduction in total model parameters, GS-MoE retains approximately 90% of the uncompressed model’s average reasoning performance, compared with 73% for D ^2 -MoE, while achieving 20% lower out-of-domain perplexity. We provide a code implementation at: this https URL
[LG-56] RACE: Data-Free Text Reconstruction Attacks against Approximate Unlearning in LLM s NEURIPS2026
链接: https://arxiv.org/abs/2610.31995
作者: Mengying Zhang,Derui Wang,Ibrahim Khalil,Bowen Liu,Xiaoyu Xia,Minhui Xue
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: Accepted to The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:As large language models (LLMs) face growing demands for data removal driven by regulatory compliance, copyright concerns, and privacy protection, approximate unlearning has emerged as a practical alternative to expensive full retraining. In practice, pre- and post-unlearning model snapshots are often maintained for auditing and rollback, introducing a previously overlooked side channel. This paper demonstrates that approximate unlearning in LLM can leave residual traces within model snapshots, which may be recoverable and thus enable the privacy breaches that approximate unlearning is intended to prevent. Here, we propose TRACE, a framework that reconstructs unlearned training data given access to pre- and post-unlearning model snapshots, without any data-specific prior knowledge. To achieve this, TRACE identifies a compact set of candidate tokens from embedding-level parameter differences to constrain the search space, and assembles sequences via a contrastive scoring mechanism based on the perplexity gap between the two models. A two-stage exploration-then-completion strategy enables the recovery of multiple distinct samples. Extensive experiments on various datasets, including TOFU, MUSE, and WMDP, across multiple LLMs and six representative unlearning methods, demonstrate that TRACE achieves high-fidelity reconstruction. This paper exposes significant privacy risks in current LLM approximate unlearning deployments and highlights the need for defenses against parameter-difference leakage and appropriate management of model snapshot access.
[LG-57] REDAR: Reinforcement-Driven Adaptive Regularisation for Class Incremental Learning
链接: https://arxiv.org/abs/2610.31991
作者: Akila Ayanthi,Darshana Priyasad,Tharindu Fernando,Sridha Sridharan,Clinton Fookes,Peyman Moghadam
类目: Machine Learning (cs.LG)
*备注: The 28th International Conference on Digital Image Computing: Techniques and Applications (DICTA 2026)
Abstract:Regularisation-based class-incremental learning (CIL) methods control the stability-plasticity trade-off through fixed coefficients tuned offline, independently of the continual learning objective and blind to phase-specific variations in data volume, task difficulty, and class similarity. We argue this is a fundamental mismatch and propose REDAR: Reinforcement-Driven Adaptive Regularisation for Class Incremental Learning, which treats regularisation strength selection as a sequential decision problem. At each incremental phase, a lightweight reinforcement learning (RL) agent observes a compact state of current performance metrics and incoming data statistics, and outputs the distillation coefficients for that phase. To respect CIL’s constraint on historical data access, the policy is trained entirely on pseudo-CIL tasks constructed from the initial phase, then deployed on the incremental sequence without further adaptation. Experiments on CIFAR-100 and Tiny-ImageNet demonstrate that REDAR reduces forgetting and improves forward transfer over fixed and heuristic baselines, and that policies trained on short pseudo-task horizons transfer effectively to longer incremental runs.
[LG-58] WADE: An LSTM-Based Anomaly Detection Engine for WebAssembly Applications
链接: https://arxiv.org/abs/2610.31984
作者: Ilaria Campestrini,Stefano Zanero,Engin Kirda
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:WebAssembly (Wasm) enables near-native performance for web applications through a portable and sandboxed execution model. However, its limited runtime protection mechanisms (e.g., the lack of stack canaries and address space layout randomization (ASLR)) leave Wasm applications vulnerable to traditional attacks. We present WADE (WebAssembly Anomaly Detection Engine), a system that detects anomalous runtime behaviors in browser-executed WebAssembly applications. WADE monitors function-level execution traces collected under benign conditions and trains a Long Short-Term Memory (LSTM) network to learn the normal execution behavior of each application. At runtime, the model flags deviations from these learned patterns as potential attacks. Unlike signature-based approaches, WADE builds application-specific behavioral models, enabling detection of both known and previously unseen attacks without per-vulnerability signatures. We evaluate WADE on a dataset of 67 WebAssembly applications spanning ten vulnerability classes, achieving zero false negatives and a 0.42% false positive rate. Subjects: Cryptography and Security (cs.CR); Machine Learning (cs.LG) Cite as: arXiv:2610.31984 [cs.CR] (or arXiv:2610.31984v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.31984 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-59] Compatible Likelihoods for Flow Matching on Manifolds NEURIPS2026
链接: https://arxiv.org/abs/2610.31981
作者: Lucas Ng,Georgios Batzolis,Mark Girolami
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 40th Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:On the simplest non-Euclidean manifold, the 1D flat torus, we exhibit a distribution where a recently proposed extension of Riemannian flow matching is globally optimized by a velocity field that pushes samples away from the data distribution. We trace this failure mode to a property we call compatibility; a likelihood-based flow matching objective is compatible if its population minimizer recovers the marginal velocity field induced by the data and coupling. We give a general condition for compatibility, apply it to existing methods, and bound the worst-case error in terms of a residual component vanishing under compatibility. Next, we note that existing compatible objectives put density on velocities that are infeasible on compact manifolds. We propose an objective built from an exponential family over endpoints, with the conditional velocity as sufficient statistic, which is both compatible and supported only on feasible velocities. This manifests as a reweighting of the flow matching loss which, on constant-curvature manifolds, treats errors in flow speed and direction differently. Empirically, our method recovers the correct velocity field on the test case, outperforms prior methods on high-dimensional product tori, and obtains substantial NLL improvements over prior work on two of four standard geospatial benchmarks on the sphere at moderate runtime overhead. On a large dataset of high-dimensional human poses, our method outperforms prior work according to sample-based metrics.
[LG-60] Beyond Heuristics and Commercial Navigation: An AI-Augmented Decision-Support for Ambulance Travel Time Estimation
链接: https://arxiv.org/abs/2610.31968
作者: Muhammad Adil Abid,Johan Holmgren,Fabian Lorig,Jesper Petersson
类目: Machine Learning (cs.LG)
*备注: 8 pages, 4 figures, 2 tables, accepted at CoopIS 2026
Abstract:Accurate ambulance travel time estimation is critical for emergency medical dispatch, yet existing decision-support practices often rely on static lookup tables or simplified assumptions regarding how ambulance travel time differs from ordinary passenger-car travel. For instance, they assume ambulances simply travel a fixed percentage (e.g., 5%) faster than ordinary passenger cars. However, such assumptions are rarely validated using real ambulance operational data and may fail to capture the effects of trip priority, temporal context, and origin-destination. We address these limitations using historical ambulance trips from the Skåne region in Sweden. Our analysis reveals that ambulance travel times vary substantially by hour, day, and trip priority, and that commercial navigation systems fail to capture emergency-specific dynamics. We contribute a machine learning system to more accurately estimate ambulance travel time. To make the system interpretable, we explored SHapley Additive exPlanations and permutation importance techniques to ensure that our system captures operationally meaningful patterns.
[LG-61] Sharp Bounds on the Winners Bias under Independent and Nonnegatively Correlated Noise
链接: https://arxiv.org/abs/2610.31948
作者: Ivan Viakhirev,Kirill Borodin,Grach Mkrtchian
类目: Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注:
Abstract:Reporting the best of M noisy, unbiased scores overstates the winner’s value by up to c\sqrtV , where V is the total noise variance and the sharp constant c depends on how the noise is dependent. It was known only at the two extremes, arbitrary dependence ( \sqrt(M-1)/M ) and i.i.d. noise. We determine it for two classes in between: for independent, not identically distributed noise it is \sqrt(M-1)/(2M) , and for pairwise nonnegatively correlated noise it is (M-1)/M ; for uncorrelated noise this answers the maximum version of an open question about the expected range. Explicit finite constructions attain both. As a corollary, the bias of the cross-validation estimator is bounded sharply within each class. All theorems are formally verified in Lean 4. Across 448 language-model and classifier selection pipelines, designs with per-candidate items realize a median 0.75 of the independent-class ceiling. At a fixed item budget, scoring all candidates on shared items is markedly less optimistic than giving each candidate its own items.
[LG-62] owards Trustworthy AI-Enabled Emergency Medical Dispatch: Uncertainty-Aware Ambulance Travel Time Estimation
链接: https://arxiv.org/abs/2610.31944
作者: Muhammad Adil Abid,Johan Holmgren,Fabian Lorig,Jesper Petersson
类目: Machine Learning (cs.LG)
*备注: 15 pages, 5 figures, 5 tables. Accepted at ICSOB 2026
Abstract:Emergency medical dispatch increasingly relies on AI-enabled decision support systems when making ambulance dispatch decisions. In this process, accurate ambulance travel time estimates are essential for timely resource allocation and prehospital care. However, existing ambulance travel time estimation approaches typically produce only single point predictions. Consequently, they do not indicate how reliable an individual estimate is or how much uncertainty it carries. This limitation is particularly problematic in safety-critical emergency medical service dispatch operations, where trips with similar expected travel times may exhibit substantially different levels of uncertainty. Furthermore, real-world ambulance trip logs may contain domain-specific artefacts, such as co-located healthcare facilities and physically implausible travel speeds. These artefacts can degrade point prediction accuracy and affect the prediction task on which subsequent uncertainty estimation depends. To address these challenges, we propose an uncertainty-aware methodology for ambulance travel time estimation and evaluate it through a retrospective real-world scenario study using two years of ambulance trip data from the Skåne region in Sweden. The proposed methodology moves beyond average-case point predictions by generating context-adaptive prediction intervals with stable empirical coverage across the evaluated operationally important contexts. The resulting prediction intervals provide explicit information about predictive uncertainty that can support more transparent and risk-sensitive ambulance dispatch decision-making.
[LG-63] Beyond Local Accuracy: Benchmarking the Rollout Reliability of PDE Foundation Models
链接: https://arxiv.org/abs/2610.31935
作者: Jaewan Park,Seid Koric,Giri Krishnan,Sudhakar Pamidighantam,Surresh Marru
类目: Machine Learning (cs.LG)
*备注:
Abstract:PDE foundation models enable data-efficient adaptation but are deployed autoregressively, so short-horizon accuracy need not imply reliable simulation. We ask whether the short-horizon advantage of a foundation-model recipe persists when its predictions are repeatedly fed back as inputs for a long horizon. We define error amplification as G=E(H_f)/E(H_s), which measures full-horizon error relative to short-horizon error. We evaluate seven models on seven two-dimensional time-dependent PDE datasets: scratch-trained FNO and DeepONet; fine-tuned DPOT, GPhyT, Poseidon, and Walrus; and frozen in-context VICON. We report VRMSE, relative L_2, and relative MAE. Six datasets provide paired short- and full-horizon scores, while CE-RM contributes short-rollout sweeps only. At least one fine-tuned foundation model outperforms both scratch baselines in 26 of 30 short-rollout dataset–budget cells under every metric. Nevertheless, full-horizon VRMSE exceeds short-horizon VRMSE in 140 of 144 paired trained configurations, with a median error amplification of 2.04 . The count below VRMSE one falls from 104 to 53, and 12 full rollouts exceed ten although none does at the short horizon. Walrus achieves the lowest full-horizon VRMSE in 16 of 24 cells, yet its median error amplification is 2.70 . Relative L_2 and relative MAE show the same qualitative pattern. These results show that favorable short-horizon rankings can coexist with substantial error amplification. PDE foundation-model benchmarks should therefore report horizon-resolved absolute errors, amplification statistics, and failure incidence rather than a single local forecast score.
[LG-64] Every Weight Is a Path of Shared Discrete Decisions
链接: https://arxiv.org/abs/2610.31931
作者: Ammar Ali,Stamatis Lefkimmiatis
类目: Machine Learning (cs.LG)
*备注:
Abstract:We introduce TWIG (Tree-structured Weights with Inherited Gates), a representation in which every weight of a neural network is the end of a path of small discrete decisions: a sign, a shift by a power of two, a zero gate, an index into a shared table. Weights that belong together share the upper part of their paths, and each decision is inherited by every weight below it, so it is stored once and applied once. This simple change has three consequences. Any trained network can be written as a TWIG, and when the decisions are powers of two its weight layers become circuits made almost entirely of wires and adders, suited to chips whose weights are fixed at fabrication. Pruning, quantization, low-rank factorization, codebooks and weight sharing become edits of one object, so a single optimizer can choose among all of them for a whole network. And choosing the decisions is the ground-state problem of a higher-order Boltzmann machine whose interactions follow the paths of the tree, the problem class that annealers and quantum optimizers are built to solve. Today these questions are handled apart: each compression method brings its own format and solver, a circuit with fixed weights needs a dedicated constant multiplier for every weight, and discrete solvers reach only narrow subproblems such as rounding. Weight formats have long been moving toward shared decisions, through group scales, shared exponents, microexponents and sparse weights; TWIG takes this trend to its end, and every common format is a shallow TWIG at no extra cost. On five weight matrices of a large language model, a flat tree whose signed power-of-two terms are chosen per weight for circuit cost, with no group scales, needs 9–15% fewer additions than GPTQ at equal layer error at 3- and 4-bit accuracy on typical matrices, and 20–55% fewer on matrices with strong outliers.
[LG-65] Input Bias Correction in Gaussian Process Regression
链接: https://arxiv.org/abs/2610.31920
作者: Muzaffar Qureshi,Tochukwu E. Ogri,Trivikram Satharasi,Kyle Volle,Rushikesh Kamalapurkar
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Accepted for publication in IEEE Robotics and Automation Letters (RA-L) (Oct, 2026)
Abstract:In traditional Gaussian process (GP) regression, the inputs are typically assumed to be known, and uncertainty enters only through the measurements. When scalar fields are mapped using GP regression with measurements obtained from a mobile sensor, the inputs correspond to the estimated pose of the mobile sensor. The goal in this letter is to understand how a recorded GP, trained on erroneous inputs (e.g., pose estimates from dead-reckoning), can be corrected in a numerically efficient way when better estimates of the inputs become available. In the developed bias-aware GP (Ba-GP) framework, the recorded GP posterior-mean gradient is used to propagate estimated input bias and noise into mean and covariance corrections. The resulting additive covariance structure is then exploited to develop numerically efficient first-order correction algorithms. A simulation study and an indoor light-intensity mapping experiment show that Ba-GP improves predictive accuracy and uncertainty calibration relative to noisy-input GP baselines, with an order-of-magnitude reduction in computational cost relative to full GP retraining.
[LG-66] Decode Late Multiply Once: Rethinking the Dot Product for Quantized Weights
链接: https://arxiv.org/abs/2610.31913
作者: Jonathan Tapson
类目: Machine Learning (cs.LG); Hardware Architecture (cs.AR)
*备注:
Abstract:Large neural networks are increasingly deployed with 1-4 bit weights and 16- or 32-bit activations, yet their dot products still use one multiplier per element. We show that a length-n dot product with k-bit weight codes can be computed with n+2^k additions and at most 2^k decodes and 2^k multiplications. Each weight code is used directly as an address offset that selects one of 2^k accumulators; each activation is added into the selected accumulator; and the accumulated sums are decoded and multiplied only once, at the end, in a fused decode-and-multiply step. The decoded weight vector is never formed. The refactoring is exact for integer arithmetic, works with integer or floating-point operands and with arbitrary, including nonlinear, decoders, and extends to group-wise scales and to matrix and tensor products. The multiplier count falls by n/2^k, while total arithmetic energy improves by a smaller, bounded factor that we quantify with a cost model. We describe a scatter-accumulate processing element (SA-PE) that replaces the MAC unit in a weight-stationary systolic array, with a load/stream/reduce/drain schedule in which a small, time-shareable multiplier runs only 2^k times per tile. Because decoded weights are not stored, local weight memory shrinks by up to w_d/k for w_d-bit decoded weights. An RTL implementation synthesized for FPGA and compared with a matched MAC array shows that an unshared SA-PE is larger than a MAC PE, and that the saving appears when the reduction multiplier is shared across PEs. Software models confirm exact integer equivalence and floating-point error no worse than the conventional pipeline. We do not claim the underlying identity is new, and we discuss related work on weight sharing and lookup-based kernels.
[LG-67] opological Equivalence of Spectral and Density Cluster Hierarchies
链接: https://arxiv.org/abs/2610.31909
作者: Sajjad Hashemian
类目: Machine Learning (cs.LG)
*备注:
Abstract:We compare two canonical ways to recover hierarchical clustering from an underlying density f : the \emphdensity cluster tree given by connected components of superlevel sets \f\ge\lambda\ , and a \emphspectral hierarchy obtained by applying single linkage to diffusion distances induced by a Laplacian operator. We give conditions under which, at an appropriate diffusion time, the diffusion hierarchy is \emphtopologically equivalent to the density cluster tree up to a monotone rescaling of heights. It further shows that a k -NN graph implementation inherits this equivalence from uniform approximation of diffusion distances, yielding a finite-sample reduction from spectral graph constructions to density-level-set topology.
[LG-68] Real-to-Sim Calibration of Bosch Etching Simulators with Few-Shot XSEM Data and Hybrid Search
链接: https://arxiv.org/abs/2610.31904
作者: Xinyu Jia,Jainil Shah,Rahul Tripathi,Vijay Gupta,Zhihong Chen
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注: 38 pages, 12 figures, 8 tables
Abstract:The Bosch process is widely used in MEMS and TSV fabrication, but accurate prediction of etching outcomes remains difficult without calibration against real measurements, which are sparse and costly to obtain. This work presents a few-shot real-to-sim calibration framework that decouples simulator parameters from direct process variables to infer a recipe-specific simulator parameter vector from sparse measurements, enabling reliable prediction of etch depth and sidewall angle under held-out process conditions. We first formulate Bosch profile prediction as a few-shot real-to-sim calibration problem, where a small set of real XSEM measurements is used to identify a recipe-specific ViennaPS parameter vector z_r^* that minimizes the discrepancy between simulated and measured etch depth and sidewall angle. To search this parameter space efficiently, the framework combines local exploration, LLM-guided candidate generation, and XGBoost-based surrogate screening, while retaining ViennaPS simulation as the final verification step for each selected candidate. Experiments on two Bosch etching recipes show that the proposed framework accurately predicts held-out etching results, achieving normalized global errors of 0.340 and 0.249 for the 2 recipes respectively.
[LG-69] Forecasting Associations Not Interventions: A Preregistered Causal Audit of Covariate-Aware Time Series Foundation Models
链接: https://arxiv.org/abs/2610.31869
作者: Zong-Han Bai
类目: Machine Learning (cs.LG)
*备注: 21 pages, 11 figures, 12 tables
Abstract:Covariate-aware time series foundation models (TSFMs) accept a future path for a control variable and return a forecast, inviting what-if queries. The query is interventional; the models are trained on observational data. We determine whether their answers are the conditional slope \partial E[Y|X] or the interventional effect \partial E[Y|do(X)] , at three levels. Theory: a Bayes-optimal observational predictor answers interventional queries associationally unless the adjustment set is available and used; for in-context least squares we derive the resulting confounding gap – a model’s effect estimate under an observational minus under a randomized context – in closed form, a parameter-free reference line. Behaviour: on a preregistered grid with exact counterfactual ground truth, Chronos-2, TiRex-2, TabPFN-TS and TimesFM-2.5 all track this line ( R^2 0.90-1.00) and respond more confounded than regression itself (slopes 1.3-1.8), on linear and nonlinear structural models. Pairs of models with identical observational laws prove this is not an identifiability limit: when the effect is identifiable, least squares recovers 100% of the true effect difference, Chronos-2 75%, TiRex-2 34%. Representation: linear probes decode the data-generating regime (94-96%) and the true effect ( R^2 0.7-0.9) from embeddings that the forecasts ignore. On two physical systems with randomized actuators, pretrained models under-respond by 2-5x and ignore in-window confounding. Minutes of interventional post-training on 3,000 synthetic series cut the gap by up to 83% and transfer outside the training prior. Its forecasting cost is a replay-distribution effect: replaying the model’s own pretraining prior cuts it from +47% to +10% MASE, and with low-rank adaptation yields a recipe meeting the preregistered criterion over three seeds (52-61% gap reduction at +1-3% cost) while improving the DoTime benchmark.
[LG-70] Cautious Conflict-Free Inverse Gradient Methods
链接: https://arxiv.org/abs/2610.31855
作者: Damian Hattler,Qiang Liu,Nils Thuerey
类目: Machine Learning (cs.LG)
*备注: Code available at [ this https URL ]( this https URL )
Abstract:Partial differential equations (PDEs) are indispensable tools for modeling complex phenomena in physics and engineering. Physics-Informed Neural Networks (PINNs) have emerged as a powerful alternative to traditional numerical solvers by directly incorporating physical constraints into neural network loss functions. Despite their advantages, PINN training requires multi-objective optimization, which often leads to significant gradient conflicts among boundary, initial, and residual loss terms. The Conflict-Free Inverse Gradient (ConFIG) method addresses these conflicts; however, integrating ConFIG with adaptive optimizers such as Adam introduces a critical vulnerability: the conflict-free direction generated by ConFIG is further processed by the optimizer, and the resulting weight update may no longer be conflict-free with respect to the task gradients. To address this issue, Cautious ConFIG is proposed as a concise three-line extension to the ConFIG algorithm that empirically restores the conflict-freeness guarantee prior to the weight update. The method is evaluated on three PINN benchmarks: the 1D Viscous Burgers’ equation, the 1D Nonlinear Schrödinger equation, and the 2D Kovasznay flow. Experimental results demonstrate consistent performance improvements over the standard ConFIG baseline. Additionally, the stabilizing effect of cautious masking and its reaction to conflict severity, task number, and optimizer lag are investigated using a simple toy system.
[LG-71] Speculative Diffusion is a Tale of Two Tradeoffs
链接: https://arxiv.org/abs/2610.31819
作者: Aaron Zorel,Guy Van den Broeck,Aditya Grover,Daniel Israel
类目: Machine Learning (cs.LG)
*备注:
Abstract:Speculative decoding is a technique that accelerates the inference speed of autoregressive large language models (LLMs) by sampling from a faster, lower quality draft model. Traditionally, a fundamental characteristic of speculative decoding is that it is lossless with respect to the verifier distribution, guaranteeing speed without sacrificing quality. This differs from recently developed diffusion LLMs (dLLMs), which are notably able to achieve substantial speedups by degrading the quality of their outputs with less diffusion effort. Because diffusion models are able to sample faster than autoregressive models by generating tokens in parallel, they are considered strong candidates for draft models in speculative diffusion. While it is widely understood that diffusion models can sacrifice quality for speed, speculative decoding itself can also be generalized to generate lossy samples faster using more lenient acceptance criteria with respect to the verifier distribution. Presented with the two parameters of diffusion effort and speculative decoding leniency, this paper explores the full achievable Pareto frontier via joint variations of these parameters. Our experiments show that jointly varying diffusion effort and acceptance leniency can extend the speed-quality Pareto frontier over single step diffusion drafting.
[LG-72] Nested Sparsity for Scalable Pretraining and Continual Learning
链接: https://arxiv.org/abs/2610.31818
作者: Tianzhe Chu,Druv Pai,Yi Ma
类目: Machine Learning (cs.LG)
*备注: 10 pages main body
Abstract:Mixture-of-experts (MoE) transformers, the standard backbone for modern large-scale deep learning, employ conditional computation: only a sparse subset of model weights is used in each forward pass. In MoE transformers, this sparsity arises in the MoE operator, which generalizes the multi-layer perceptron in the classical transformer. In this work, we explore an alternative mechanism for conditional computation that analogously generalizes the attention operator. The sparsity induced by this mechanism naturally nests (i.e., hierarchically composes) with the sparsity induced by the MoE operator. We show that a transformer employing this nested sparsity mechanism, which we call Nestor, performs better than a conventional MoE transformer at equal compute and equal sparsity in language model pretraining. We also show that Nestor is well-suited for continual pretraining: it can be adapted to longer contexts more effectively than conventional MoEs, as well as adapt more efficiently to new domains while forgetting less on old domains. Furthermore, we demonstrate that these benefits persist as model size, dataset size, and context length grow. Our results demonstrate the potential of alternative conditional computation mechanisms for deep neural network architectures. Code and checkpoints are available on GitHub at this https URL .
[LG-73] CuraLoc: Repository-Level Fault Localization via Planned Code Inspection
链接: https://arxiv.org/abs/2610.31809
作者: Sangeeta Warrier,Anush Sankaran,Ryen White
类目: Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注:
Abstract:A bug report describes what went wrong, but rarely where the code should change. Existing repository-level localization methods often let language models iteratively decide what code to inspect next, requiring repeated model calls and substantial token usage. We investigate whether these inspection decisions can instead be planned before source retrieval. In this research, we introduce CuraLoc, which first selects candidate files and represents their symbols using compact behavioral and structural descriptions. In a single planning decision, the model jointly selects starting symbols, chooses code relationships to inspect from each, and forms hypotheses about likely edits. Deterministic retrieval collects the requested source for a final ranking informed by those hypotheses, concentrating model reasoning on choosing and comparing code rather than directing each retrieval step. We demonstrate the efficacy of our proposed approach on the SWE-Bench Lite and Verified datasets. On SWE Bench Verified with GPT-5.2, CuraLoc raises Func@1 from LocAgent’s 53.8% to 59.3% while using 90.3% fewer recorded tokens, and also outperforms RepoSearcher (56.2%) and Agentless (39.6%) on Func@1. On SWE-bench Lite, CuraLoc achieves a high 83.8% Func@5 accuracy improving on Agentless by more than 15% in relative accuracy. Additionally, CuraLoc supports different model backbones, with strong localization accuracy demonstrated using both GPT-5.2 and GPT-6 Astra. These results provide evidence that repository-level code inspection can be planned effectively without iterative model-controlled navigation.
[LG-74] Not All Retained Clients Are Equal: Client Selection for Federated Unlearning NEURIPS2026
链接: https://arxiv.org/abs/2610.31786
作者: Thanh Linh Nguyen,Quoc-Viet Pham
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Emerging Technologies (cs.ET)
*备注: Accepted at NewInML @ NeurIPS 2026
Abstract:Federated unlearning (FU) aims to remove a designated target client’s influence from a trained model without the prohibitive cost of retraining from scratch. Existing methods largely treat unlearning as a post-hoc, stateless operation, overlooking that client influence is path-dependent: contributions made during training, particularly during critical learning periods (CLPs), can have disproportionate and persistent effects. Moreover, knowledge overlap and intermittent client availability make relying on all retained clients inefficient and impractical. We introduce PDF, a path-dependent client-selection framework based on the insight that retained clients are not equally valuable for unlearning. During training, PDF maintains lightweight, target-agnostic, CLP-weighted histories summarizing each client’s accumulated update direction and global-model alignment. Upon an unlearning request, these histories identify and rank retained clients that are globally informative yet weakly coupled to the target, enabling retraining-based unlearning with only the top- k clients over a small number of rounds. Experiments on FMNIST, CIFAR-10, and CIFAR-100 under IID and non-IID settings show that PDF achieves effective forgetting and strong retained utility with substantially fewer retained clients. In particular, PDF achieves 0% forgetting accuracy with retraining-like MIA AUC on non-IID CIFAR-10, improves retained accuracy by up to 6.91 percentage points over the strongest competing FU baseline on CIFAR-100, and reduces communication by approximately 69% and computation by 57 – 61% relative to Retrain.
[LG-75] Efficient Omni-modal Large Language Model Inference via Per-Query Modality Selection
链接: https://arxiv.org/abs/2610.31768
作者: Shuqi Liu,Yuzhou Cao,Luke Ong
类目: Machine Learning (cs.LG)
*备注:
Abstract:Omni-modal large language models (Omni-LLMs) can jointly process a broad range of modalities, such as text, audio, and video, enabling stronger multimodal understanding and reasoning capabilities, but often at substantial inference cost. We observe that the benefit of additional modalities varies across queries, indicating that a fixed modality configuration can be inefficient. Motivated by this, we study query-wise modality selection, which keeps the underlying Omni-LLM fixed while adaptively selecting which modalities to provide for each query. We formulate the problem as an accuracy-cost optimization problem and develop a cost-sensitive learning approach for modality selection. We evaluate our approach on WorldSense using Qwen2.5-Omni and Qwen3-Omni under three inference-cost measures: input tokens, latency, and monetary cost. Across both models, query-wise modality selection consistently improves the accuracy-cost trade-off and can even outperform the standard setting in which all available modalities are provided to the Omni-LLM. For Qwen2.5-Omni, our cost-sensitive approach matches or exceeds this full-modality accuracy while requiring only approximately 41% of the input tokens, 39% of the latency, and 61% of the monetary cost.
[LG-76] ConsMSA: Semantic Distribution Consistency Learning for Multimodal Sentiment Analysis ICML2026
链接: https://arxiv.org/abs/2610.31731
作者: Pan Wang,Lipeng Ke,Huajun Ying,Pritish Mohapatra,Rohan Sarkar,Suresh Lakhani,Sankar Venkataraman,Jingtong Hu
类目: Machine Learning (cs.LG); Multimedia (cs.MM)
*备注: ICML 2026
Abstract:Multimodal sentiment analysis (MSA) aims to predict human sentiments by integrating signals from different modalities such as text, video, and audio. However, raw multimodal sequences often suffer from semantic inconsistency–exhibiting redundancy or conflicts within and across modalities–which hinders robust understanding and increases computational cost. To this end, we introduce ConsMSA, which explicitly formalizes semantic distribution consistency across both \textitintra - and \textitinter -modality, providing a consistency-aware mechanism for robust and efficient multimodal sentiment prediction. Specifically, ConsMSA projects multimodal token features into a shared sentiment space to compute an Intra- and Inter-modality Consistency Score ( I^2CS ). By coupling this score with predictive relevance, we formulate consistency-aware importance signals that are utilized: (i) as a consistency regularizer to align latent distributions during training, (ii) to derive semantic-aware weights for adaptive multimodal token reweighting, and (iii) as a practical criterion to prune redundant or conflicting tokens. Extensive experiments on CMU-MOSI, CMU-MOSEI, and CH-SIMS demonstrate that ConsMSA achieves state-of-the-art performance while remaining robust under aggressive token compression–retaining only 10% of tokens yields comparable accuracy. These results establish semantic distribution consistency as a promising foundation for synergizing predictive robustness with computational efficiency.
[LG-77] When and How Does Muon Precondition? The Role of Structured Gradient Noise
链接: https://arxiv.org/abs/2610.31728
作者: Xiaoran Chen,Binghui Li,Liwei Wang,Pinyan Lu
类目: Machine Learning (cs.LG)
*备注: 62 pages, 11 figures
Abstract:Muon has become a popular optimizer for large language model training, yet its theoretical foundations remain poorly understood. In this work, we study how gradient noise geometry shapes Muon’s preconditioning benefits. We first show, theoretically and empirically, that under the classical isotropic noise assumption, Muon offers no acceleration beyond a rescaling of the learning rate. This motivates us to investigate structured noise arising from mini-batch sampling in associative memory with heterogeneous task frequencies. We establish an exact balancing threshold for two tasks and extend the analysis to multiple tasks, revealing how task coverage and gradient reliability jointly shape learning. Our analysis shows that Muon can soften the dependence on task frequency from linear to square-root, helping balance the learning of frequent and rare tasks. We further identify rotation-induced negative transfer in SignSGD that Muon avoids, and explain how momentum reduces sampling noise while tracking changing gradients. Controlled numerical experiments validate these mechanisms, and a preliminary LLM pretraining study illustrates the practical promise of fast-slow momentum.
[LG-78] owards Adversarially Robust Multigraph Neural Networks for Financial Crime Analysis
链接: https://arxiv.org/abs/2610.31718
作者: Dennis Heijmans,H. Çağrı Bilgi,Rui Wang,Kubilay Atasu
类目: Machine Learning (cs.LG)
*备注: 20 pages, 10 figures, 3 tables. Accepted at the Fifth Learning on Graphs Conference (LoG 2026). Code: this https URL
Abstract:Financial transaction networks are naturally represented as multigraphs, where repeated transactions between the same pair of accounts are modeled as distinct edges. Recently proposed multigraph neural network architectures leverage this structure explicitly to achieve a strong performance on anti-money laundering (AML) tasks. However, their robustness to adversaries that manipulate transaction structures to evade detection remains largely unexplored. We present the first scalable framework for structural attacks and adversarial training of multigraph neural networks using node-edge incidence matrices, adapting projected gradient descent and projected randomized block coordinate descent methods to optimize structural perturbations over such matrices. We further introduce a cycle-based regularizer that penalizes changes in the global counts of directed 2- and 3-cycles, providing a structural measure of attack plausibility. Experiments on synthetic pattern-detection and realistic AML datasets show that our learned attacks substantially outperform non-learnable perturbation baselines in reducing detection performance (PR-AUC), while adversarial training consistently improves robustness against these attacks. These results demonstrate that multigraph neural networks are vulnerable to learned structural attacks and that adversarial training provides an effective defense. Our code is available at this https URL.
[LG-79] Does LeWorldModel Preserve Local Action Geometry? NEURIPS2026
链接: https://arxiv.org/abs/2610.31716
作者: Vasil Vasilev,Siddharth Nayak
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 13 pages, 3 figures. Accepted as a poster at the NeurIPS 2026 Workshop on Symmetry and Geometry in Neural Representations (NeurReps), Extended Abstract Track
Abstract:We examine how closely pretrained LeWorldModel (LeWM) reproduces the effects of small action changes. Using matched simulator and model responses, we measure action sensitivity with Gram matrices at three perturbation scales. Across nearly a thousand states from Reacher and OGBench Cube, the most sensitive action directions differ by median angles of about 36^\circ and 42^\circ , respectively, at the smallest scale. Increasing the perturbation tenfold reduces median normalized Gram discrepancy by roughly 70% on Reacher and 47% on Cube. Geometry measured in pixels and embeddings varies more across scales than the chosen physical-state geometry. On held-out directions, environment Grams achieve median embedding response errors of 17% and 28%, compared with 26% and 38% for an isotropic reference with the same trace. LeWM’s errors are higher at this scale. These findings support scale-specific evaluation of action sensitivity; its effect on control remains an open question.
[LG-80] he Path to NDE 4.0: A Review of Algorithms Techniques and Strategies for Autonomous Inspection
链接: https://arxiv.org/abs/2610.31704
作者: Ashiqur Rahmana,Venkata Devesh Reddy Seethi,Zachary Kral,Austin Yunker,Rajkumar Kettimuthu,Hamed Alhoori
类目: Machine Learning (cs.LG); Emerging Technologies (cs.ET)
*备注:
Abstract:Non-Destructive Inspection (NDI) is a cornerstone of modern quality assurance, critical for ensuring the safety and reliability of infrastructure and manufactured components without causing damage. This paper presents a comprehensive review of the current NDI landscape, analyzing the field through four distinct lenses, namely, computational algorithms, physical techniques, industrial applications, and research objectives. We synthesize a broad range of academic literature to illustrate the rapid evolution from manual, analog interpretation to a digitized, data-centric paradigm known as NDE 4.0. Our analysis highlights the dominance of ultrasound and visual inspection as primary physical techniques, while revealing a significant shift toward deep learning and generative models for automated defect recognition. Despite these technological strides, the industry faces critical bottlenecks, notably the scarcity of standardized open-access datasets and the “black box” nature of AI models, which hinders widespread adoption. By charting these developments and identifying key research gaps, such as the need for explainable AI (XAI) and standardized digital twin protocols, this review offers a strategic roadmap for future research. It concludes that the future of NDI research should focus on the successful integration of intelligent, autonomous fusion of inspection data with advanced sensor technologies into asset lifecycle management.
[LG-81] Norm-optimal model order reduction in reproducing kernel Hilbert spaces
链接: https://arxiv.org/abs/2610.28224
作者: Tamás Dózsa,Andrea Angino,Rolf Krause,Matthias Voigt
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 92 pages, 7 figures
Abstract:We consider the problem of model order reduction (MOR) in reproducing kernel Hilbert spaces (RKHSs). We prove generalizations of well-known necessary interpolatory conditions for norm-optimal reduced-order models, originally stated for rational functions in H_2 Hardy-Hilbert spaces. We construct an alternative to the iterative rational Krylov (IRKA) algorithm for RKHSs with differentiable kernels to obtain norm-optimal reduced-order models. Then, we introduce generalized Malmquist-Takenaka (GMT) functions that can be used to define orthonormal bases in an arbitrary RKHS. Using GMT functions, we propose unitary transformations between separable RKHSs. This allows us to extend our optimality conditions and subsequent algorithms to MOR problems posed in RKHSs with non-differentiable kernels. The potential and versatility of the proposed MOR framework is demonstrated in several numerical experiments. We perform norm-optimal MOR on various machine learning models such as support vector machines and deep ReLU networks. In addition, we show how the proposed MOR framework can be used to solve seemingly unrelated problems. In particular, we provide an example for fitting high degree polynomials to data and also obtain a FEM-like method for solving partial differential equations (PDEs) with an optimal kernel basis.
[LG-82] Singular Geometry of Low-Rank Adaptation
链接: https://arxiv.org/abs/2610.32854
作者: Kamillo Ferry,Anthea Monod,Keisuke Yano
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 40 pages, 8 figures
Abstract:Low-rank adaptation (LoRA) is now a standard approach to parameter-efficient fine-tuning of large neural networks. Its low-rank factorization is inherently non-identifiable, making LoRA a singular statistical model whose learning complexity cannot be described by parameter dimension alone. We study this complexity using the real log canonical threshold (RLCT) from singular learning theory and how it depends on both the low-rank parameterization and the pretrained network. Under a local linearization, LoRA induces a low-rank trace regression model whose Kullback–Leibler geometry is controlled by a forward-backward covariance, constructed from activations and backpropagated sensitivities at the adapted layer. This covariance determines which low-rank perturbations are statistically observable. When it is nondegenerate along the relevant low-rank directions, we recover the classical learning complexity of reduced-rank regression; when its kernel meets these directions, additional statistically indistinguishable parameters change the singular geometry. We characterize this geometry locally through observable tangent directions and a residual singularity, and globally through affine linear sections of determinantal varieties. The global learning complexity need not be determined at a ground-truth parameter, as singularities on another rank stratum can dominate. We also introduce a curvature-corrected multi-temperature estimator of the RLCT. Experiments on multilayer perceptron and transformer architectures reveal strongly layer-dependent forward-backward geometry and show that ambient covariance degeneracy does not imply degeneracy along the low-rank directions relevant to adaptation.
[LG-83] Generalization properties of the token perceptron: the role of embedding and vocabulary
链接: https://arxiv.org/abs/2610.32788
作者: Dario Bocchi,Matteo Negri,Flavio Nicoletti,Francesco D’Amico
类目: Disordered Systems and Neural Networks (cond-mat.dis-nn); Statistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 39 pages, 9 figures
Abstract:Large language models are trained to predict tokens from a finite vocabulary, each represented by a vector in a d -dimensional embedding space. Therefore, token prediction is a classification problem whose classes have a geometry determined by the embedding. We study how this geometry affects learning in a minimal solvable model: a linear predictor that maps a sequence of N embedded tokens to a d -dimensional output through one d\times d matrix per position, trained in the teacher–student setting. We call this the \textittoken perceptron. We compute its generalization error with the replica method, taking N\to\infty at fixed number of examples per position \alpha=P/N and keeping d and the vocabulary size D finite. As a first reference case, we consider continuous embeddings on the sphere, which isolate the role of the embedding dimension. The natural load is then \alpha/d , and for d1 a normalized label fixes the direction of the teacher output, so the problem becomes a noiseless regression: the teacher is recovered exactly above \alpha_\rm rec=d^2/(d-1) , and spherical cross-entropy reaches this threshold at large inverse temperature. In the same setting the storage capacity of random associations scales as d^2 for margins O(1/\sqrtd) . We then turn to vocabulary embeddings, where a label only identifies the Voronoi cell of the teacher output, so that each example imposes inequalities rather than equations, as in the classical perceptron. We compute the Bayes-optimal error for a given embedding matrix through an approximate spectral closure and show that the optimal learner first resolves the directions in which the vocabulary has the largest variance, even when the teacher is isotropic.
[LG-84] Sparse Bayesian Neural Networks via Sticky Piecewise Deterministic Markov Processes
链接: https://arxiv.org/abs/2610.32753
作者: August Arnstad,Geir Olve Storvik,Filippo Pagani,Leiv Rønneberg
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
*备注:
Abstract:There is growing evidence that the performance of a neural network depends largely on a small subset of its weights, which are typically identified by training a dense network and pruning it afterwards. Within a Bayesian framework, pruning can be directly incorporated through a spike-and-slab prior, providing a proper treatment of the uncertainties involved, both with respect to pruning and to parameter uncertainty. In Bayesian neural networks (BNNs), sampling by MCMC is notoriously hard. We demonstrate that modern Monte Carlo methods based on piecewise deterministic Markov processes (PDMPs) can be used to sample posteriors of BNNs asymptotically exact whenever numerically constructed rate bounds hold. Spike-and-slab priors can be incorporated through sticky versions of the PDMPs, setting weights exactly to zero during sampling. We examine the predictive performance and convergence of the samplers, and show that sparse BNNs sampled this way can be more robust to input noise than dense ones.
[LG-85] Score-Preserving Continual Lasso under Task Shift
链接: https://arxiv.org/abs/2610.32708
作者: Jingyi Guo,Runze Li,Lingzhou Xue
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注:
Abstract:Continual learning (CL) aims to learn from sequential tasks while mitigating catastrophic forgetting when historical data are no longer accessible. We study continual learning for high-dimensional sparse linear regression, where \ell_1 regularization creates a difficulty absent from continual ridge regression: that recursively reusing an \ell_1 -penalized estimator can propagate earlier shrinkage and accumulate regularization bias across tasks. We propose score-preserving continual Lasso (SP-CL), which separates sparse estimation from state transmission by preserving the historical squared-loss score. Under squared loss, its cumulative estimator exactly recovers a full-history Lasso minimizer. When task coefficients coincide, it inherits the full-history sparse rate. For related task-specific coefficients, SP-CL uses a regularized current-task correction that leaves the transmitted state unchanged and attains the offline source-assisted rate under an oracle related-task condition. We also derive Overall/BWT/FWT decompositions that distinguish historical information loss, task conflict, and current-task specialization, and validate the resulting regimes in simulation.
[LG-86] Probing the limits of forward screening and inverse design of crystal materials
链接: https://arxiv.org/abs/2610.32678
作者: Zhilong Song,Zijun Pang,Lixue Cheng
类目: Materials Science (cond-mat.mtrl-sci); Machine Learning (cs.LG)
*备注: 23 pages, 5 figures
Abstract:Crystal materials are designed either by screening known crystals for a target property or by generating new crystals with that property. Both routes are judged against public databases of known crystals, but these databases name the same crystal differently. Several materials announced as new by recent generative and autonomous discovery campaigns have since proved to be known phases or close variants of them. The limits of both routes remain unmeasured. Here we probe them by matching the crystals of 55 databases by structure into one self-extending database of 14.6 million crystals and 131.8 million property values, each traced to its source. Only 4.5% of its crystal-property pairs carry a value. To fill the empty pairs, we developed CGTNet, an equivariant graph transformer with state-of-the-art accuracy on several tasks of two public benchmarks. Trained on each task, it forms a library of property models whose predictions we release. The reach of forward screening has been underestimated. Known crystals hold tens of thousands of ultra-incompressible candidates without a recorded bulk modulus, and a post hoc audit of synthesis evidence finds among them ReN2, whose bulk modulus was measured at 428 GPa. The novelty claimed for inverse design has been overestimated. At an equal verification budget, a stiffness-conditioned generator returned more stiff hits than screening (100 versus 55), but all contained osmium and 82 repeated recorded compositions. Integrating and completing the databases of known crystals therefore moves the limits of both routes: it widens what screening can reach and sets the reference against which generated crystals are judged new.
[LG-87] Acquisition for Spherical-Linear Bayesian Optimization on the Cube
链接: https://arxiv.org/abs/2610.32628
作者: Nathan Wycoff
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
*备注:
Abstract:Recent work demonstrated that executing Bayesian Optimization (BO) using simply a linear surrogate with a clever transformation is enough to approximate the performance of the standard Gaussian-Process (GP) based version on high-dimensional test functions. When the decision region is a sphere, this strategy admits rapid exact acquisition. However, it is the unit cube we encounter most often in the BO literature, and in this setting a standard approach, namely a multi-start quasi-Newton method, is employed for solving the acquisition subproblem. In this article, we will discuss a series of results, both positive and negative, regarding the complexity of global resolution of various acquisition subproblems over the unit cube. On the basis of this discussion, we will propose a new acquisition function which is globally optimizable using significantly less execution time than the incumbent approach while maintaining or improving performance. The resulting method is competitive with GP-based BO in terms of accuracy per evaluation, and is superior in terms of accuracy per walltime on black-boxes of low to moderate computational intensity.
[LG-88] Lets Do Turbulence Right: In Space and Time!
链接: https://arxiv.org/abs/2610.32601
作者: Maximilian Plattner,Fabian Paischer,Johannes Brandstetter
类目: Fluid Dynamics (physics.flu-dyn); Machine Learning (cs.LG)
*备注:
Abstract:The three-dimensional Navier-Stokes equations govern fluid turbulence, a chaotic and non-linear physical phenomenon characterized by complex eddies interacting across vast spatial and temporal scales. A numerically accurate modeling of this multi-scale energy cascade requires fully resolving both the fine spatial structures and the continuous temporal evolution of the flow field. In this paper, we leverage massive, time-resolved direct numerical simulations, which exceed 28 TB per run, to assess how numerical choices govern the resolved (3+1)-dimensional physics of turbulent flows. Building on this analysis, we study how neural networks represent the recorded turbulent motion. Our results highlight that spatial and temporal assessments can disagree for both numerical records and learned representations. We therefore call on the machine learning community to treat (3+1)-dimensional turbulence as a four-dimensional phenomenon.
[LG-89] Batched NN 4D-Var: a performant hybrid model framework for the Integrated Forecasting System
链接: https://arxiv.org/abs/2610.32582
作者: Alban Farchi,Marcin Chrust,Massimo Bonavita,Marc Bocquet,Wei Pan,Patrick Laloyaux
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Hybrid forecasting systems combining machine learning and numerical weather prediction offer a promising route toward improved forecast skill while preserving physical consistency. In this study, we investigate online learning of state-dependent model-error corrections within the Integrated Forecasting System using 4D-Var. Building on previous work, we represent model error using a neural network embedded within the forecast model and trained from observations through the data assimilation cycle. First, we introduce a targeted variant of neural-network 4D-Var in which only the neural-network parameters are estimated, while the atmospheric state is provided by a precomputed analysis. We then propose a batched extension that processes assimilation windows in random order and averages parameter increments over multiple windows, drawing inspiration from mini-batch stochastic optimization. Experiments are conducted with the Integrated Forecasting System of the European Centre for Medium-Range Weather Forecasts at an intermediate resolution. Sequential online learning substantially improves forecasts within the training period, with error reductions exceeding 10%in some cases. However, these improvements fail to generalize to unseen data. In contrast, the proposed batched formulation produces hybrid models that consistently improve forecast skill in out-of-sample experiments. Larger batch sizes accelerate convergence and improve robustness by reducing the noise of parameter updates. Diagnostics further show that the online-trained corrections preserve realistic forecast variability and do not improve skill through artificial smoothing. These results demonstrate that the formulation of the online learning problem is critical for generalisation and establish batched online training as a promising framework for hybrid machine-learning/data-assimilation systems.
[LG-90] Exact Global Minimax Risk near Nonidentification
链接: https://arxiv.org/abs/2610.32546
作者: Kaiyue Yang,Yunan Wu
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注:
Abstract:We study squared-error estimation when a nuisance parameter is unidentified on a submodel but affects a target that vanishes there. For models with higher-order profile expansions, we give sufficient conditions under which the largest critical-path minimax risk is the exact asymptotic minimax risk over the full parameter space. An efficient observable feature, negligible calibration error and stable recovery yield one estimator attaining the leading constant uniformly, with the profile coordinates unknown. The estimator uses calibration estimates, a local feature mean and an empirical recovery feature. The constant is expressed through a bounded-normal decision problem. We verify the conditions in two models. In centered Gaussian logistic errors-in-variables regression with one surrogate and two unknown variances, we determine the fifth-order global coefficient risk and its exact constant. A sparse extension separates variance-decomposition uncertainty from direction estimation and specifies when the scalar constant is preserved. In a degree-balanced replicated-network model, we determine the sixth-order global edge-contrast risk when block proportions are unknown. Triangles identify a parameter combination; a bowtie supplies the additional information needed for the contrast.
[LG-91] Random Jump Forests
链接: https://arxiv.org/abs/2610.32498
作者: Martin Bladt,Rasmus Frigaard Lemvig
类目: Methodology (stat.ME); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 85 pages, 19 figures and 10 tables
Abstract:Random jump forests are introduced as a nonparametric tool for estimating conditional cumulative transition rates and occupation probabilities in multi-state models. The main idea is that a single tree partitions the covariate space for all transitions simultaneously, using partially observed trajectories, and with Nelson–Aalen estimators at the terminal nodes. Occupation probabilities are recovered through the product integral without requiring the Markov assumption. Uniform consistency and a functional central limit theorem for the estimators of cumulative rates can be established under the assumption that the covariate vector takes finitely many values, and under suitable splitting conditions. In particular, pointwise confidence intervals and simultaneous bands are available in the Markov case. Simulation studies show a predictive accuracy close to the correctly specified benchmark models and marked improvements when benchmarks are misspecified, with further illustration on four biomedical data sets. In the coverage study, confidence bands attain near-nominal coverage in several settings, with pooling of heterogeneous rates being the main source of possible undercoverage. The method, including prediction assessment and variable importance, is implemented in the R package JumpForests.
[LG-92] On the Limits of Stochastic Push-Pull over Time-Varying Digraphs
链接: https://arxiv.org/abs/2610.32422
作者: Feiming Wang,Yutong He,Sulaiman Alghunaim,Kun Yuan
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注: 33 pages, 3 figures
Abstract:Stochastic Push-Pull achieves linear speedup with respect to the network size n on static strongly connected digraphs, but whether this speedup persists on time-varying networks remains unclear. In this work, we construct a quadratic decentralized problem on period-two networks showing that stochastic Push-Pull may fail to attain linear speedup over time-varying digraphs. For p -periodic networks, we identify an effective node count n_\mathrmeff\le n that governs the leading stochastic term. Specifically, for smooth nonconvex objectives, we establish convergence upper bounds and matching lower bounds for stochastic Push-Pull with constant stepsizes, both exhibiting the leading-order dependence \Theta(1/\sqrtn_\mathrmeff T) . When p=1 , the identity n_\mathrmeff=n recovers the usual linear speedup for static digraphs. However, periodic networks can have n_\mathrmeff=o(n) , demonstrating the loss of linear speedup under temporal variation. Numerical experiments illustrate effective-count scaling and the separation between static and periodic networks, consistent with our theoretical findings.
[LG-93] CoPiT: A conditional Position-induced Transformer for aerodynamic field prediction on unstructured meshes
链接: https://arxiv.org/abs/2610.32299
作者: Junfeng Chen,Yongchen Fan,Zhonghua Qiao,Kailiang Wu
类目: Computational Physics (physics.comp-ph); Machine Learning (cs.LG); Numerical Analysis (math.NA); Fluid Dynamics (physics.flu-dyn)
*备注:
Abstract:High-fidelity Reynolds-averaged Navier–Stokes (RANS) simulations remain a standard tool for aerodynamic analysis, yet each solve on an unstructured mesh is computationally expensive, making many-query tasks such as design exploration, optimization, and multi-condition evaluation costly. Neural operators provide mesh-flexible surrogates for mapping geometry and operating conditions to flow fields; however, in many existing architectures, global parameters such as the freestream state and design variables are concatenated with pointwise inputs, leaving their influence on intermediate features implicit. This paper proposes CoPiT, a \textbfConditional \textbfPosition-induced \textbfTransformer that builds on the Position-induced Transformer (PiT) as its geometric backbone and injects a global condition vector through multi-stage feature-wise linear modulation (FiLM), thereby making parametric modulation explicit while preserving geometry-based attention weights. The model is evaluated on two-dimensional AirfRANS airfoil fields, including scarce-data and extrapolation regimes, and on three-dimensional BlendedNet++ blended-wing-body surfaces. On the AirfRANS full-data test set, CoPiT reduces surface and volume losses by approximately 54% and 24% relative to Transolver, respectively, and achieves the lowest overall normalized test MSE among the compared baselines on BlendedNet++. Matched ablations show that FiLM conditioning generally improves upon input concatenation, while the preferred placement is task-dependent. CoPiT thus provides a practical operator-learning surrogate for parametric aerodynamics on unstructured meshes.
[LG-94] Empirical concentration for chains with infinite memory
链接: https://arxiv.org/abs/2610.32265
作者: Pierre Alquier,Paul Doukhan
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Concentration inequalities for time series usually involve unknown quantities measuring dependence. This prevents their direct use as confidence bounds, or as criteria for learning algorithms. Recent works proved empirical concentration inequalities for stationary Markov chains, where the unknown quantities are estimated. We extend this approach beyond the Markov case, by considering stationary chains with infinite memory. We first establish a bounded-differences inequality on bounded Polish metric spaces under a summable Wasserstein contraction condition. We then estimate its dependence coefficients from a single trajectory, using either transition counts when the state space is finite, or derivatives of kernel conditional-density estimates for a continuous state space. We give general error bounds and illustrate their dependence on the memory tail. Finally, a fixed multiplicative safety factor turns these estimators into empirical concentration bounds. While some regularity assumptions are required to make the bounds empirical, our procedure remains fully adaptive in the sense that it does not require the knowledge of the constants involved in these assumptions. The empirical confidence bounds derived from our results are asymptotically as tight as the theoretical ones.
[LG-95] Federated Nonparametric Regression with Heterogeneous Clients
链接: https://arxiv.org/abs/2610.32178
作者: Xiaoran Jiang,Anand N. Vidyashankar
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 123 pages, 8 figures, 6 tables
Abstract:Whether a federated predictor benefits an individual client depends on both population heterogeneity and estimation uncertainty. We study this question for Nadaraya–Watson regression in a triangular array with potentially different fitting and evaluation distributions. Each client transmits local kernel summaries, allowing the server to construct a covariate-density-weighted federated estimator without centralizing individual observations. A client combines its fitted local predictor with the federated predictor through a prescribed weight, and gain is the paired reduction in squared prediction loss on an independent audit observation. Exact identities characterize both the gain of the realized fitted pair and its repeated-fitting mean. We also establish pointwise inference for the federated estimator. A two-stage analysis of client gains separates fitting variability from audit variability, retains the covariance created when a client’s observations enter both its local estimator and the common server estimator, and yields fixed-block simultaneous inference across clients. Robustness is examined under two perturbation models. A Kullback–Leibler neighborhood provides a variance-sensitive lower certificate in a regular positive-gain-variance regime, while an L_2 neighborhood yields an exact least-favorable regression direction and robust value. Exact-model, vanishing-variance, and exact-zero-direction boundaries are kept separate because their scalar inputs are nonregular. The framework thus distinguishes decisions about a realized fitted pair from claims about the learning procedure under repeated fitting.
[LG-96] Solving and Learning Navier-Stokes Dynamics with the Neural Basis Method
链接: https://arxiv.org/abs/2610.32167
作者: Yuhe Wang,Zhu Wang,Min Wang
类目: Fluid Dynamics (physics.flu-dyn); Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注: 23 pages, 6 figures
Abstract:Physics-informed machine learning has become a major direction in predictive modeling of fluid dynamics, combining governing physics with learned representations. Yet prevailing formulations neither guarantee that small PDE-residual losses imply small solution errors nor scale effectively with increasing problem size. Here we present the neural basis method (NBM) as a new approach for Navier–Stokes dynamics. NBM is built on a ``representation-then-projection’’ principle: fixed neural bases define the representation space, while projection of the governing equations determines the expansion coefficients. The same bases define common coordinates for the parametric solution manifold, recasting operator learning as a parameter-to-coefficient mapping and enabling scalability by removing explicit spatial dependence. For incompressible flow, NBM yields a velocity-only formulation whose residual provides an a posteriori velocity-error estimate, establishing a certified, data-free training objective for coefficient learning. Theoretical analysis and numerical experiments demonstrate the accuracy, efficiency, generalization, and scalability of NBM, with substantial improvements over baselines.
[LG-97] Conditional Flow Models for Offline Maximum Entropy Reinforcement Learning
链接: https://arxiv.org/abs/2610.32125
作者: Hao Cheng,Yuling Jiao,Lican Kang,Jerry Zhijian Yang,Chen Zhong
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 68 pages, including appendices
Abstract:Offline maximum entropy reinforcement learning seeks to estimate the optimal soft action-value function (Q^) from a fixed dataset without further interaction with the environment. Although (Q^) is characterized by the soft optimal Bellman equation, the corresponding Bellman operator cannot be evaluated directly because both the reward distribution and the transition kernel are unknown. Consequently, estimating (Q^) requires the joint treatment of environment modeling and value function learning. We propose a framework that addresses these tasks separately. Conditional diffusion models based on probability flow ODEs estimate the conditional reward distribution and transition kernel. The resulting distributional estimators define a data driven approximation of the soft optimal Bellman operator, whose empirical Bellman residual is minimized over a class of deep neural networks to estimate (Q^). We establish nonasymptotic guarantees for the proposed method. For operator learning, we derive total variation error bounds for the conditional flow estimators that account explicitly for score estimation, numerical discretization, and terminal smoothing. For soft value learning, we first use the soft Bellman equation to show that Q^* inherits Hölder regularity from the conditional reward and transition densities. For any Hölder index \beta , the value learning stage achieves an excess Bellman residual risk bound of order \widetilde\mathcal O\big(n^-\frac2\betad+2\beta\big), where d is the total dimension of the state and action spaces. In contrast to many existing theoretical analyses of deep reinforcement learning, our results do not require Bellman completeness. Numerical experiments support the theoretical findings and demonstrate the effectiveness of the proposed method.
[LG-98] A Dynamic Inventory Rationing Model for Last-Mile Delivery of Medication in Liberias Community Health Program
链接: https://arxiv.org/abs/2610.32051
作者: Baris Ata,Yufan Chen,Robert Montgomery
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Systems and Control (eess.SY); Analysis of PDEs (math.AP)
*备注: 56 pages, 8 figures, 8 tables, including appendices
Abstract:We study the dynamic allocation of scarce medications in Liberia’s community health program. A central warehouse supplies a health facility and the Community Health Assistants (CHAs) it supervises. When a CHA runs out of medication, patients are referred to the facility, incurring travel costs if they complete the referral and remain untreated otherwise. Allocating inventory to CHAs allows more patients to receive treatment without traveling, while retaining inventory centrally preserves flexibility to respond to uncertain demand. The allocation problem must balance these considerations while accounting for fairness across locations. We formulate this problem as a finite-horizon stochastic control problem whose objective combines the costs of referrals and untreated patients with a demand-weighted fairness utility. Because standard dynamic programming methods are computationally intractable in this setting, we develop a continuous-time approximation that involves controlling the drift rate of a reflected Brownian motion. We then adapt the deep splitting method, using deep neural networks to approximate the value function and its gradient and compute a dynamic allocation policy. We calibrate the model using data on malaria cases among children in the Fangonda region of Lofa County, Liberia, and evaluate the proposed policy in the original discrete-time model. In the baseline setting, our policy reduces operating costs by about 6% relative to the strongest benchmark, the fairness-only policy, while retaining approximately 98% of its fairness utility. Additional experiments show that higher referral costs shift inventory toward CHAs and reduce local unmet demand, while greater demand variability increases our policy’s performance over the fairness-only benchmark. Our model and computational approach provide a framework for studying similar last-mile allocation problems.
[LG-99] Combinatorial Characterizations of the Learnability of Conditional Distributions
链接: https://arxiv.org/abs/2610.31890
作者: Mihir Dhanakshirur,Ambuj Tewari
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We study the PAC-learnability of function classes that map inputs to probability distributions over a finite space. The learner observes i.i.d. input-label pairs, with each label sampled from the conditional distribution of an unknown target hypothesis in the class. We measure learning through excess risk under a bounded proper loss that satisfies two mild assumptions. We introduce the \textbffat _\gamma Natarajan dimension, which generalizes both fat _\gamma and Natarajan dimension, and prove that its finiteness at every positive scale is necessary and sufficient for learnability. We also introduce a \textbffat _\gamma Graph dimension extending the classical \gra dimension and establish inequalities comparing the two dimensions at different scales, leading to an equivalent characterization of learnability. Finally, we show that a distribution-valued function class is learnable if and only if each of its coordinate classes is learnable in the probabilistic concept model.
[LG-100] Early learning in large residual networks
链接: https://arxiv.org/abs/2610.31783
作者: Vladimir Narovlansky,Boris Hanin
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Probability (math.PR); Data Analysis, Statistics and Probability (physics.data-an)
*备注: 23 pages + appendices, 10 figures
Abstract:We study the early training dynamics of nonlinear residual networks with blocks of zero hidden layers as a model of more general residual networks. We analyze networks in a sequential limit where first width and depth diverge, then the number of training data points diverges, and finally the input dimension diverges. This is a more refined setup compared to the typical mean-field theory analysis in which depth and width diverge but the number of training datapoints and input dimension are fixed. Our main technical result is a diagrammatic computation of both the network output and the hidden representations after the few steps of gradient descent in terms of the statistics of inner products of inputs from the training set. One of our principle conceptual findings is a negative result: in the limit of large input dimension the network output can learn only a linear function of its inputs early in training since higher learned monomials are more and more suppressed in the input dimension and vanish in the large dimension limit. This suggests that mean-field limits of ResNets are surprisingly rigid in the regime of high input dimension and may not be sufficient toy models for understanding feature learning. Finally, we investigate how to ensure hyperparameter transfer for the learning rate as a function of dataset size and input dimension. We test our predictions numerically and find excellent agreement with our theory.
[LG-101] Operationalising Weaker Faithfulness: Recovery Ambiguity and Computational Limits
链接: https://arxiv.org/abs/2610.31724
作者: Cillian Hourican,Rick Quax
类目: Methodology (stat.ME); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Variables can predict an outcome jointly while each is individually independent of it, causing structure learners to discard informative links. We investigate how to operationalise weaker faithfulness through two approaches: selective protection in an adjacency learner and explicit retention of alternative directed graph explanations. Matched experiments show that changes at both an early candidate-removal stage and later pruning can be necessary. In noisy XOR with a descendant (N = 10,000, 20 matched samples), combining reactivation and protection raises complete parent-pair recovery from 0/20 for either change alone to 20/20, while adding false edges. A confirmation study shows that the family estimator can preserve explanations lost by orienting the protected learner’s selected skeleton: in one weaker signal XOR condition (N = 10,000) it retains the entire reference family in 48/50 samples, compared with 0/50 for the skeleton adapter. These gains are conditional. Protection can add only error, weaker effects produce false exclusions, and an ordinary faithful-chain comparison favours the PC learner. Complete nonempty family outputs decline at six/eight variables, and all 400 twelve/twenty-variable searches remain unfinished at the reporting cutoff. Neural controls show engineered retention, while the family estimator returns no accepted graph within its application budget. This operationalisation identifies two deletion stages that must be coordinated in the motivating example and shows when direct family search preserves explanations lost by selecting one skeleton. Finite evidence and computation remain substantial barriers to reliable general inference.
[LG-102] Autoregressive Physics-Informed Neural Operator Surrogates and Joint 4-D Seismic-Production Bayesian History Matching
链接: https://arxiv.org/abs/2610.31677
作者: Clement Etienam,David Blom,Hari Pattathil,Tsubasa Onishi,Harpreet Sethi,Lalson Vincent,Suraj Pawar,Kishan Ramesh Kumar,Pandu Devarakota,Faruk O. Alpak
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 40 pages, 15 figures, 6 tables, 4 algorithms
Abstract:We present a surrogate-accelerated Bayesian history-matching framework, demonstrated on a realistic synthetic model ( N=100,188 cells). The forward map \mathcalG is replaced by a Fourier neural operator \mathcalG_\theta trained in a closed loop, \hatx_n+1=\mathcalG_\theta(u,\hatx_n) , with the discrete black-oil residual R(x_n+1,x_n) as a physics-informed neural operator (PINO) penalty, truncated backpropagation and noise injection. We assimilate production data and 4-D seismic using adaptive regularised ensemble Kalman inversion (aREKI), and synthesise \Delta V_p inside the rollout via a closed-form Gassmann transform of the surrogate, under a two-stage schedule (aREKI on production, then an ensemble smoother with multiple data assimilation, ES-MDA, on both). To temper the ill-posedness, we invert (\log K,\phi) not in \mathbbR^2N , where N_e\ll 2N leaves a rank-deficient covariance and a null space the data never inform, but on a chart u=\Psi(\tilde m) , \tilde m\in\mathbbR^d , d\ll 2N , given by a Parseval-exact truncated cosine transform or a variational auto-encoder decoder. The ensemble then spans its own search space; the directions the data cannot constrain are cut to those the chart can express, and every member decodes to a geologically plausible field. For the rollout we prove an exposure bound W_1\le\sum_j L^,n-1-j\mathbbEe_j , a physics-penalty contraction |D_x\mathcalG_\theta|\le\alpha1 under a small residual defect, and an \mathcalO(\rho^K) truncation bias. A Loewner-order inequality \mathcalI_p\preceq\mathcalI_p+\mathcalI_s explains why joint assimilation is no less informative. On held-out data, \mathcalG_\theta attains R^20.999 on pressure and saturation and rolls out in 0.579 s, a \sim2500\times speed-up, bringing an N_e=190 -member, N_a=10 -iteration joint inversion within one GPU-hour.
[LG-103] Hybrid SEIR Temporal Fusion Transformer Framework with SHAP Analysis for Interpretable Pandemic Forecasting
链接: https://arxiv.org/abs/2610.31673
作者: Ashish Anton Abraham,Gopi Krishnan R,Aiswarya Anil Nair,Ajo Babu George,Kunal Agarwal
类目: Physics and Society (physics.soc-ph); Machine Learning (cs.LG)
*备注:
Abstract:Forecasting pandemic spread across heterogeneous regions demands models that are both accurate and interpretable for public health action. This paper proposes a hybrid forecasting framework that integrates an SEIR epidemiological model enhanced with a Temporal Fusion Transformer (TFT) to produce interpretable multi-horizon infection forecasts. Using region-wise COVID-19 data from the Our World in Data corpus, we present a unified pipeline that integrates epidemiological indicators (susceptible, exposed, infectious, recovered), intervention signals (masking, vaccination, policies), mobility covariates, and demographic context. The SEIR component estimates latent disease dynamics and injects mechanistic priors, while the TFT learns non-linear temporal dependencies and cross-feature interactions from historical observations and known future inputs. To ensure transparency, SHAP (SHapley Additive exPlanations) is applied at both global and local levels, quantifying feature contributions and revealing region-specific drivers of surges and slowdowns. Compared with a calibrated standalone SEIR baseline, the hybrid SEIR-TFT approach provides more stable and accurate forecasts across epidemic phases, particularly under shifting intervention regimes. The framework is further operationalized through structured prediction and explanation outputs for interactive dashboards, enabling time-series, policy-impact, and geospatial analysis. By unifying mechanistic modeling, deep temporal learning, and explainable AI, this study offers a practical decision-support system for policy-sensitive pandemic forecasting.
附件下载
