本篇博文主要内容为 2026-10-27 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 (2026-10-27)
今日共更新1542篇论文,其中:
- 自然语言处理共233篇(Computation and Language (cs.CL))
- 人工智能共434篇(Artificial Intelligence (cs.AI))
- 计算机视觉共292篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共429篇(Machine Learning (cs.LG))
- 多智能体系统共33篇(Multiagent Systems (cs.MA))
- 信息检索共48篇(Information Retrieval (cs.IR))
- 人机交互共36篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] Preference-Guided Cooperation in Multi-Objective Multi-Agent Reinforcement Learning
【速读】：该论文旨在解决多目标多智能体强化学习（MOMARL）中智能体协作与多目标权衡之间耦合复杂度高的问题。现有方法通常将多智能体协作与多目标优化视为相互独立的模块，导致在非平凡场景下计算复杂度急剧上升。本文的核心创新在于提出利用多目标结构本身作为促进协作的内在机制，而非将其视为额外的优化挑战。其关键解决方案是：在团队和个体双层框架下建模MOMARL，引入基于反事实的目标级信用分配机制以评估各智能体对不同目标的贡献，并提出元梯度感知的多目标多智能体强化学习方法（MIRL），包含静态（MIRL-S）与动态（MIRL-D）两种变体。通过融合元梯度信息，MIRL使智能体能够自主学习目标优先级，同时协同优化系统级目标，从而实现多目标结构对协作过程的直接引导。实验结果表明，所提方法在多目标多粒子环境中的帕累托前沿质量优于基线方法，其中动态版本表现最优。
链接: https://arxiv.org/abs/2610.35736
作者: Qiyue Xia,J. Michael Herrmann
机构: University of Edinburgh (爱丁堡大学)
类目: Multiagent Systems (cs.MA); Machine Learning (cs.LG)
备注: 14 pages, 4 figures, 1 table (plus 3 figures, 3 tables, and 2 algorithms in the appendix)
Abstract:Multi-objective multi-agent reinforcement learning (MOMARL) addresses agent coordination and objective trade-offs simultaneously while optimising overall system performance. Existing work tends to treat multi-agent cooperation and multi-objective optimisation as largely independent components of a learning problem, which amounts to an exorbitant complexity except in trivial cases. We study here whether the structure provided by multiple objectives can itself be exploited to facilitate cooperation, rather than being treated as an additional optimisation challenge. We formulate MOMARL at both the team and individual levels and introduce objective-wise counterfactual credit assignment to evaluate each agent’s contribution across the team objectives. We propose Meta-Gradient Informed Multi-Objective Multi-Agent Reinforcement Learning (MIRL), with static (MIRL-S) and dynamic (MIRL-D) variants. By incorporating meta-gradient information, MIRL enables agents to learn how to prioritise objectives individually while collectively optimising the system-level objective, allowing the multi-objective structure to directly guide cooperation. Experiments on a multi-objective multiple particle environment show that our methods improve Pareto-front quality over baselines, with the dynamic variant achieving the strongest performance.
[MA-1] Multi-Objective Reinforcement Learning Using Routed Ensembles and Trajectory Attribution NEURIPS2026
【速读】：该论文旨在解决多目标强化学习（Multi-objective Reinforcement Learning, MORL）中因目标间冲突导致的策略学习效率低下问题，特别是在目标权重组合变化时，最优策略呈现剧烈波动，使得基于偏好（per-preference）的策略学习面临严重的样本效率瓶颈。其核心挑战在于如何量化并缓解目标间的冲突对学习过程的影响。解决方案的关键在于引入目标冲突系数（objective-conflict coefficients），通过该度量推导出近似最优的在线学习后悔界（regret bounds），从而定量揭示目标错配对学习效率的负面影响；在此基础上，提出路由集成策略学习（routed ensemble policy learning）算法，通过维护一组基础策略并动态将轨迹路由至潜在奖励最高的策略，实现高效的目标平衡；同时，进一步将轨迹选择作为缓解冲突的手段，设计了基于核代理的轨迹归因算法（kernel surrogate trajectory attribution），能够捕捉轨迹间非线性交互关系（如协同与对抗效应），提升反馈信号的利用效率。实验表明，该方法在Meta-World和MO-Gymnasium等基准环境上相较基线平均性能提升4.3%，在留一分析相关性上优于现有方法50%，且在轨迹选择任务中实现27%的奖励增益。
链接: https://arxiv.org/abs/2610.35733
作者: Zhenshuo Zhang,Ruoxuan Xiong,Hongyang R. Zhang
机构: Northeastern University(东北大学); Emory University(埃默里大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Robotics (cs.RO); Machine Learning (stat.ML)
备注: 30 pages. To appear at NeurIPS 2026 (extended version)
Abstract:Multi-objective reinforcement learning is a fundamental problem for modern AI, with applications spanning post-training, reasoning models, and robotic control. A central challenge is to learn policies that balance competing objectives, such as helpfulness versus honesty in safety and alignment, or speed versus energy in robotics. The optimal policy can vary sharply with objective-weighting combinations, and the extent of this variation depends crucially on the degree of conflict among objectives, making per-preference policy learning sample-inefficient. In this paper, we introduce objective-conflict coefficients and derive near-optimal regret bounds for online settings using this measure, thereby quantifying the impact of misalignment on learning efficiency. Then, we design routed ensemble policy learning, an algorithm that maintains a small set of basis policies and dynamically routes trajectories to the policy with the highest potential reward. Finally, we consider trajectory selection as another approach to mitigate objective conflict, and design a kernel surrogate trajectory attribution algorithm that estimates how feedback signals from a set of trajectories influence policy learning on other trajectories. The kernel can capture nonlinear interactions such as synergy and antagonism between trajectories. We evaluate our approach across diverse environments and benchmarks: routed ensembles improve average performance by 4.3% over baselines for online environments from Meta-World and MO-Gymnasium, and kernel surrogate attribution improves correlation with leave-one-out analysis over existing baselines by 50%. When applied to trajectory selection, kernel-based attribution yields a 27% improvement in reward over baseline attribution methods.
[MA-2] EvoRouter: Cost-Efficient Multi-Agent Routing via Insight-Guided On-Policy Distillation
【速读】：该论文旨在解决大规模多智能体系统（multi-agent systems）在服务网络查询时面临的复杂配置难题，即如何联合优化大语言模型、协作模式、智能体数量、交互轮次及技能分配等多重变量。由于配置空间呈组合爆炸式增长，且执行奖励稀疏，难以对高度耦合的规划决策提供细粒度监督。其解决方案的关键在于提出EvoRouter——一种由洞察引导的在线策略蒸馏驱动的自演化多智能体路由框架。该框架通过专用专家周期性地从跨任务的配置轨迹中提炼出可复用的条件路由洞察（如能力需求、资源扩展收益及配置项间的相互作用），再由特权教师将这些动态演化的洞察与实时执行反馈相结合，通过在线策略蒸馏生成密集的令牌级监督信号，以指导路由器自身的规划轨迹。结合成本感知的强化学习，该框架协同优化任务性能与计算开销，使执行洞见内化至路由策略中。更新后的策略生成新的执行轨迹，支持下一轮洞察提取，形成以执行驱动的自我改进闭环。实验表明，EvoRouter在19个领域内数据集及2个跨域基准上平均准确率达68.69%，较最强基线提升7.23个百分点，同时推理计算成本降低50.95%。
链接: https://arxiv.org/abs/2610.35690
作者: Yitao Xiao,Xu Zhang,Ye Cheng,Shaoyong Guo,Yan Sun,Ziyue Zhu,Sitao Xie,Guo Tong,Haoran Luo
机构: Beijing University of Posts and Telecommunications(北京邮电大学); Fudan University(复旦大学); Nanyang Technological University(南洋理工大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 13 pages, 12 figures, 4 tables
Abstract:Serving web queries at scale with multi-agent systems requires jointly configuring large language models, collaboration modes, agent counts, interaction rounds, and skills. However, the resulting configuration space grows combinatorially, while sparse execution rewards provide limited fine-grained supervision for interdependent planning decisions. We propose EvoRouter, a self-evolving multi-agent routing framework driven by insight-guided on-policy distillation. A dedicated expert periodically synthesizes performance-cost feedback from configuration trajectories across tasks into reusable conditional routing insights about capability requirements, returns from resource scaling, and interactions among configuration choices. A privileged teacher combines these evolving insights with current execution feedback to provide dense token-level supervision for the router’s own planning trajectories through on-policy distillation. Cost-aware reinforcement learning jointly optimizes task performance and computational cost. Together, these training signals internalize execution insights into the routing policy. The updated policy generates new trajectories that support subsequent insight extraction, forming an execution-driven self-improvement loop. We evaluate EvoRouter on 19 in-domain datasets spanning five domains and two out-of-domain benchmarks. EvoRouter achieves an average in-domain accuracy of 68.69%, outperforming the strongest routing baseline by 7.23 percentage points while reducing inference compute cost by 50.95%. Our code is open-sourced at this https URL.
[MA-3] GEAR: Feature-Based Configuration and Code Generation for Agent ic Systems
【速读】：该论文旨在解决多智能体系统（Multi-Agent Systems, MAS）在工程化过程中因框架差异导致的配置复杂性与开发效率低下问题。具体而言，不同框架在智能体选择、编排策略、通信机制、推理方法及记忆配置等方面存在显著差异，开发者需针对每个框架学习特定API，难以高效比较或迁移设计方案。为此，论文提出GEAR——一个开源工具，通过统一的特征化表示（feature-based representation）显式表达配置选项及其约束，使用户可在代码生成前对规范进行验证。GEAR支持十种主流智能体框架的连接器，基于同一规格说明自动生成框架特定实现，并提供报告以揭示各框架间的适配差异。初步评估显示，在320个评估的特征-框架组合中，有271对（84.69%）具备可翻译性，验证了其可行性。关键解决方案在于构建通用规范与框架特定实现之间的可映射桥梁，从而支持开发者快速原型设计与研究者跨框架的多智能体配置探索。
链接: https://arxiv.org/abs/2610.35592
作者: Brell Sanwouo,Nada Zine,Paul Temple,Clément Quinton,Romain Rouvoy
机构: Univ. Lille, CNRS, Inria, UMR 9189 CRIStAL, Lille, France; Univ. Rennes, CNRS, Inria, IRISA, Rennes, France; Institut Universitaire de France (IUF), Paris, France
类目: Multiagent Systems (cs.MA); Software Engineering (cs.SE)
备注:
Abstract:Engineering multi-agent systems (MAS) requires choosing agents, orchestration strategies, communication mechanisms, reasoning approaches, and memory configurations. These choices depend on framework-specific capabilities and abstractions, requiring developers to learn different APIs when exploring alternative implementations. We present GEAR, an open-source tool for configuring MAS and generating framework-specific implementations from a common specification. GEAR uses a feature-based representation to make configuration options and their constraints explicit, enabling users to validate specifications before code generation. Through connectors for ten agentic frameworks, users generate projects from the same specification and can then inspect reports describing framework-specific adaptations. A preliminary assessment of mapping feasibility identifies translation strategies for 271 of 320 evaluated feature-framework pairs (84.69%). GEAR supports developers prototyping agentic systems and researchers exploring MAS configurations across frameworks. GEAR is available as open-source software at this https URL and a video demo is available at this https URL.
[MA-4] Simulation-Efficient Double-Flow Neural Posterior Estimation for Agent -Based Models
【速读】：该论文旨在解决基于代理的模型（Agent-Based Models, ABMs）中由于似然函数不可解析而导致贝叶斯后验参数推断计算成本高昂的问题。传统神经后验估计（Neural Posterior Estimation, NPE）依赖大量直接模拟生成训练数据，而这一过程在复杂模型中尤为耗时。本文提出双流神经后验估计（Double-Flow Neural Posterior Estimation, DF-NPE），其核心创新在于结合学习的时间序列摘要与两个条件化归一化流（conditional normalizing flows）：一个用于合成数据生成，另一个用于后验估计。该框架通过引入合成训练样本，有效减少对直接模拟数据的依赖，同时支持合并式（pooled）与有序式（ordered）两种训练策略。实验结果表明，在Brock-Hommes资产定价模型的两个设定下，DF-NPE可将直接模拟次数降低90%，且在平均后验偏差和运行时间上均优于标准NPE。模型性能受直接模拟数据与合成数据配比及训练策略的影响，验证了学习型摘要增强作为降低ABMs贝叶斯后验推断模拟成本的实用方法的有效性。
链接: https://arxiv.org/abs/2610.35442
作者: Junyao Zhang,Jinglai Li,Junqi Tang
机构: University of Birmingham (伯明翰大学)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Neural posterior estimation (NPE) enables Bayesian inference of the posterior parameter in agent-based models (ABMs) with intractable likelihoods, but simulating training data can be computationally expensive. We propose Double-Flow Neural Posterior Estimation (DF-NPE), which combines learned time-series summaries with two conditional normalizing flows for synthetic-data generation and posterior estimation. The framework supplements direct simulations with synthetic training pairs and supports both pooled and ordered training strategies. Experiments in two settings of the Brock-Hommes asset-pricing model show that DF-NPE can reduce direct simulations by 90 percent while achieving lower mean posterior discrepancies and runtime than standard NPE. Performance varies with the allocation of directly simulated and synthetic data and the training strategy. These results establish learned summary augmentation as a practical approach to reducing simulation costs for Bayesian posterior inference of parameters in ABMs.
[MA-5] Belief-Conditioned Multi-Agent Actor Critic for Decentralized LLM Collaboration NEURIPS2026
【速读】：该论文旨在解决去中心化生成式 AI（Generative AI）多智能体协作中因智能体独立执行、通信受限而难以高效共享上下文信息的问题。现有方法多依赖预定义工作流，限制了系统的灵活性与可扩展性。其核心解决方案是提出一种基于信念条件的多智能体演员-评论家算法（Belief-Conditioned Multi-Agent Actor-Critic, BCMAAC），通过在每个智能体内部维护对上下文信息的信念模型，并利用这些信念指导策略学习与决策，从而在缺乏显式通信的情况下有效推断任务上下文。实验表明，在Minecraft游戏、网页搜索及工具使用等多轮任务中，BCMAAC显著提升了任务完成率与协作效率，尤其在富含上下文信息的复杂任务中，信念建模的优势更为突出。
链接: https://arxiv.org/abs/2610.35398
作者: Tianle Chen,Shuo Liu,Christopher Amato
机构: Khoury College of Computer Sciences, Northeastern University (东北大学计算机科学学院), Boston, MA, 02215, USA
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: Accepted at the NeurIPS 2026 Workshop PTA: From Pretrained Representations to Acting Agents. 18 pages, 3 figures
Abstract:Recent work has explored optimizing LLM collaboration through Multi-Agent Reinforcement Learning (MARL), but many approaches still rely on predefined workflows, limiting efficiency and scalability. Decentralized LLM collaboration is challenging because agents execute independently with limited communication, share a joint objective, and must infer useful information from communication traces. In this paper, we propose BCMAAC, a Belief-Conditioned Multi-Agent Actor-Critic method that maintains beliefs over contextual information and uses them to facilitate training. Experiments on Minecraft game-playing, web searching and tool-using tasks show that BCMAAC improves task performance and collaborative behavior compared to naive actor-critic baselines. The improvements are particularly pronounced in multi-turn tasks with richer contextual information, where belief modeling provides a clear advantage.
[MA-6] Knowing When to Escalate: Disagreement Patterns for Uncertainty-Aware Content Moderation CIKM2026
【速读】：该论文旨在解决生成式内容审核系统中，基于大语言模型（LLM）代理的集体判断机制在处理价值导向任务时存在的根本性问题：现有聚合方法将代理间的分歧简单视为可靠性问题并试图通过共识或投票消除分歧，然而在涉及价值多元性的任务中，这种分歧可能反映了不可调和的价值冲突，而非可被纠正的错误。传统不确定性分解方法无法捕捉此类分歧的本质，因其仅关注输出分布而忽略了推理过程。为此，论文提出推理轨迹差异分析（Reasoning Trace Divergence Analysis, RTDA），通过嵌入具有不同视角的代理所生成的解释性理由，并基于理由相似性与结论一致性对案例进行分类，构建了一个四状态分类框架。该框架识别出“收敛性分歧”——即代理推理路径相似但结论相异——作为价值导向冲突的潜在信号。在Measuring Hate Speech数据集上的实验表明，该结构化信号比仅依赖分歧幅度或输出一致性的基线方法更能有效预测人类标注者之间的分歧。研究贡献的核心设计原则是：应根据分歧的结构特征而非其强度来触发人工介入，而RTDA可作为已具备推理轨迹生成能力系统的后处理模块，实现更精准、更具价值敏感性的内容审核。
链接: https://arxiv.org/abs/2610.35335
作者: Michał Wawer,Jarosław A. Chudziak
机构: Warsaw University of Technology (华沙理工大学); Institute of Computer Science (计算机科学研究所)
类目: Multiagent Systems (cs.MA)
备注: Accepted at CIKM 2026, November 7-11, 2026, Rome, Italy
Abstract:Agentic content-moderation systems increasingly rely on collective judgments from LLM agents, while aggregation methods commonly treat inter-agent disagreement as a reliability problem to be removed through consensus or voting. This framing is inappropriate for value-laden tasks, where disagreement may reflect irreducible value pluralism - a distinction standard uncertainty decompositions cannot recover, since they operate on output distributions and discard the reasoning behind each judgment. We propose Reasoning Trace Divergence Analysis (RTDA), which embeds model-generated justifications from perspective-differentiated agents and classifies each case by justification similarity and conclusion agreement. The resulting four-state taxonomy isolates convergent disagreement, where agents reasoning similarly but concluding differently, as a candidate signature of value-laden conflict. On 1,000 items from the Measuring Hate Speech corpus, this structural signal predicts human annotator disagreement better than magnitude-only and output-only baselines. We contribute a design principle: condition human escalation on the structure of disagreement, not its magnitude alone, with RTDA serving as a post-processing layer for systems that already generate reasoning traces.
[MA-7] Guarded Agent -in-the-Loop Residual Adaptation for Evolving Wind Power Regimes
【速读】：该论文旨在解决风力发电预测模型在不同气象情景（如寒潮、高温静稳期、强台风期及破坏性风暴）下预测误差显著变化的问题，传统固定时序骨干模型难以适应此类动态环境。其核心解决方案是提出一种“受控的代理介入残差自适应框架”（guarded Agent-in-the-loop residual adaptation framework），关键在于不修改原有的时序骨干结构，仅通过引入少量场景条件化的残差专家（residual experts）进行动态修正。该框架通过一个代理层（Agent layer）解析外部场景信息，识别当前气象状态，选择合适的残差专家，并仅在近期观测误差支持的情况下施加校正；新专家需在验证集上表现优于初始版本方可加入专家池，从而实现模型扩展的显式可控。实验表明，该方法在多个时序骨干与预测时长（4–48小时）组合下均显著降低平均绝对误差（MAE）和均方根误差（RMSE），平均MAE降幅达11.21%至20.26%，且通过在线选择基线、组件消融及路由分析验证了残差专家、场景感知路由机制以及在线接纳规则的关键作用。
链接: https://arxiv.org/abs/2610.34968
作者: Yuhang Dong,Min Xia,Jinghong Li,Yue Dai,Xinyue Wang,Shi Liang
机构: 未知
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Wind power forecasting models are often deployed as fixed temporal backbones, but the errors they make can vary across meteorological regimes such as cold waves, heat-and-calm periods, productive typhoon, and destructive storms. We present a guarded Agent-in-the-loop residual adaptation framework that leaves the temporal backbone unchanged and adapts only a small set of scene-conditioned residual experts. An Agent layer interprets external scene information, identifies the current regime, selects residual experts, and applies a correction only when recent observed errors support it. New experts enter the pool after they outperform their initializations on a validation set, keeping model growth under explicit control. Backbone-only and Agent-enhanced forecasts are compared at lead times of 4, 8, 12, 24, and 48 across nine time-series backbones. Across these runs, the guarded framework lowers both MAE and RMSE for all 45 backbone-horizon combinations, with average MAE reductions ranging from 11.21% to 20.26% across backbones. Online-selection baselines, component ablations, and routing analyses identify the roles of residual experts, scene-aware routing, and the online acceptance rule. Candidate-scene runs evaluate the promotion path under controlled conditions; long-term unattended deployment was not evaluated.
[MA-8] ClinVerse: A Realistic Multi-Agent Diagnostic Environment for Training and Evaluating Medical Agents
【速读】：该论文旨在解决当前大型语言模型（LLM）在医疗诊断应用中因依赖静态问答数据而难以有效学习主动信息获取、检查决策制定及动态临床交互等关键能力的问题。其核心解决方案是构建ClinVerse——一个基于真实电子健康记录（EHR）数据的可扩展多智能体交互环境，依托包含15万例真实诊断案例的ClinPatient数据集，支持六类专业化临床角色之间的协同互动。在此环境中，医生代理从患者主诉出发，通过逐步开展问诊与检查请求来收集临床证据并生成诊断，整个过程由规则驱动的评分框架量化信息获取效率、检查合理性与诊断准确性，形成轨迹级奖励信号，既用于强化学习训练，也支持大规模评估。实验表明，三类共九种自进化方法在多个通用与医学领域模型上均显著提升了诊断性能，并揭示了模型能力、环境反馈与自进化策略之间的协同作用机制。
链接: https://arxiv.org/abs/2610.34878
作者: Sirry Chen,Fangke Chen,Xuanwen Ding,Wei Chen,Zhongyu Wei
机构: Fudan University (复旦大学); Shanghai Innovation Institute (上海创新研究院); Zhejiang University (浙江大学); Huazhong University of Science and Technology (华中科技大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Disease diagnosis is a key application of large language model (LLM) agents in healthcare. However, existing medical agents are primarily trained and evaluated with static question-answering data, providing limited opportunities to learn active information acquisition, examination decision-making, and dynamic clinical interaction. To address this limitation, we present ClinVerse, an interactive, real-data-grounded multi-agent environment designed for training and evaluating medical diagnostic agents. Built on ClinPatient, a scalable dataset comprising 150k diagnostic cases derived from real-world electronic health records, ClinVerse supports coordinated interactions among six specialized clinical roles. Specifically, a doctor agent starts from an initial patient complaint, progressively gathers clinical evidence through patient interviews and examination requests, and ultimately produces a diagnosis. For each interaction trajectory, a rule-based scoring framework quantifies information acquisition, examination ordering, and diagnostic accuracy. These trajectory-level scores serve as reward signals for agent training while also enabling scalable evaluation. In experiments, we evaluate various general-purpose and medical-domain models and show that nine self-evolution methods across three categories consistently improve agents’ diagnostic capabilities within ClinVerse. We further analyze the interplay among model capability, environment feedback, and self-evolution methods.
[MA-9] Collective Emergent Alignment: Measuring and Amplifying Interaction-Induced Reasoning Alignment in Multi-Agent LLM s
【速读】：该论文旨在解决多智能体大语言模型（Multi-agent Large Language Model, LLM）系统中推理对齐（reasoning alignment）的涌现现象问题，具体探究智能体间交互是否能产生超越个体自我精炼（self-refinement）的、基于参考信息的推理对齐增益。其核心挑战在于理解并量化智能体协作过程中信息传递与整合如何影响整体推理一致性。解决方案的关键是提出一种名为REMAG（Reference-Entailment-Matrix for Agentic Process）的量化框架，通过语义覆盖度（Semantic Coverage）和交互特异性对齐增益来评估交互带来的对齐提升。研究发现，在多跳推理任务中，依赖跨智能体分布证据整合的推理内容表现出显著的正向交互增益，且该增益与独立测量的同伴信息恢复能力密切相关。进一步分析揭示了组合依赖性异质效应及不同能力水平智能体间不对称的信息传递模式。基于此，作者设计了一种面向事实级别的多智能体通信机制，以紧凑、感知信息缺口并选择性路由的方式替代完整的推理轨迹交换，实验证明该方法在提升推理对齐与任务性能的同时降低了通信开销；然而更精细化的通信策略也暴露出获取分布式信息、保留本地内容与通信效率之间的权衡。总体而言，本研究首次实现了对集体涌现对齐的可量化建模，并将其与信息传递模式相联系，为高效多智能体协同提供了理论指导与技术路径。
链接: https://arxiv.org/abs/2610.34836
作者: Boxuan Wang,Zhuoyun Li,Sarvapali D. Ramchurn,Xiaowei Huang,Yi Dong
机构: University of Liverpool(利物浦大学); University of Southampton(南安普顿大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: Preprint - Under Review
Abstract:Multi-agent large language model (LLM) systems can exhibit emergent behaviors through interaction, yet whether a similar phenomenon exists for reasoning alignment remains largely unexplored. We study whether agent interaction produces additional reference-grounded reasoning alignment beyond matched individual self-refinement, and introduce REMAP (Reference-Entailment-Matrix for Agentic Process), a framework that quantifies this effect through Semantic Coverage and interaction-specific alignment gains. Across diverse model families and two multi-hop reasoning benchmarks, we observe broadly positive interaction gains, particularly on reasoning content that depends on integrating evidence distributed across agents. These gains are consistently associated with independently measured peer-information recovery, while further analysis reveals composition-dependent heterogeneity effects and asymmetric information transfer across agent capability levels. Motivated by these findings, we further introduce a fact-level multi-agent communication framework that replaces full reasoning-trace exchange with compact, gap-aware, and selectively routed information. Our experiments show that compact fact exchange improves reasoning alignment and task performance while reducing communication overhead, whereas more selective communication exposes a trade-off between acquiring distributed information, retaining local content, and communication efficiency. Overall, our results show that collective emergent alignment can be quantified, linked to information-transfer patterns, and used to guide more effective multi-agent communication.
[MA-10] SlowLab: Benchmarking Autonomous AI Agents on a Year of Experiments They Cannot Undo
【速读】：该论文旨在解决当前生成式AI（Generative AI）在自主实验场景中评估基准过于简化、无法反映真实物理实验复杂性的核心问题。现有基准多依赖快速、低成本的模拟实验，而实际农业设施中的实验周期长达数月，资源稀缺且不可逆，对决策策略具有更高要求。为此，研究提出SlowLab这一新型基准，模拟一个为期一年的温室种植实验，在四个受控温室单元中进行，其作物种类、设施条件与市场价格均对代理隐藏。代理需自主决定播种时机、气候调控策略、是否中止作物以及测量读取时机，并在实验结束后推荐最优策略，由私有评估器在未见过的天气条件下评分。在针对48个未见站点的预注册研究中，三种基于低资源语言模型的代理表现接近传统种植者水平，显著低于从相同起点出发的贝叶斯优化基线。分析表明，这些代理将设施运行视为两个同步批次，导致季节内采集的测量数据无法影响后续种植决策，且这些数据也未能提升任何方法的性能。进一步验证发现，即使允许其他方法重新读取已收集的证据，推荐结果亦无显著改善，揭示了当前模型的局限性主要源于初始策略设定与实验选择机制，而非信息利用能力。该研究强调了在长期、高成本、不可逆的现实实验环境中，代理决策的动态适应性与探索-利用平衡的重要性。代码已开源，采用MIT许可证。
链接: https://arxiv.org/abs/2610.34644
作者: Ningpei Ding,Zai Wen Feng
机构: Huazhong Agricultural University (华中农业大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 16 pages (9 main text), 7 figures, 5 tables. Code: this https URL
Abstract:AI agents are increasingly proposed as autonomous experimenters, yet the benchmarks that test them mostly offer cheap experiments that answer at once. Experiments in physical facilities are different: they take months, occupy scarce equipment and cannot be undone. We present SlowLab, a benchmark in which an agent runs a one-year campaign in four simulated greenhouse compartments on a site whose crop, facility and prices are hidden. The agent decides when to plant which climate policy, whether to stop a crop and which measurements to read while crops grow, and at the end recommends one policy, which a private evaluator scores in weather the agent never saw. In a preregistered study on 48 unseen sites, three low-cost language-model agents ended near a standard grower setting and clearly below a Bayesian-optimisation baseline that starts from that setting. They ran the facility as two synchronous batches, so measurements taken during the season never changed what they planted, and these measurements improved no method’s result. Letting other methods re-read the evidence a campaign had collected did not yield significantly better recommendations, which points to where the models start and which experiments they try. Code is released under the MIT license.
[MA-11] Agent ProofBench: Benchmarking Reusable Verifiers for Web Agents
【速读】：该论文旨在解决可复用网页智能体（Web-agent）验证中因需求动态变化导致的验证规则失效与后续执行不可靠的问题。其核心挑战在于如何在不重新构建完整执行环境的前提下，确保验证规则与历史执行状态的一致性，并支持对智能体行为的持续评估。解决方案的关键在于提出AgentProofBench基准测试体系，通过将冻结的验收规则（acceptance rule）与其对应的已保存执行起始状态（saved openings）进行配对，实现对同一任务在不同时间点的可复现验证。该框架覆盖97项任务、79个任务家族及4种网页服务，通过深度测试表明：在48个任务家族中，完整面板与决策重放测试分别保留了82/96（Flash）和79/96（Pro）的原始程序正确性；在22个执行家族中，仅当两个保存的起始状态均无错误时，程序分配才表现出高正确率（32/33正确），而存在面板错误的情况则全部导致参考规则回归或交付失败。此外，中性延续策略在通用反馈下实现了最高正确交付率，为评估门控机制（gate）的实际增益提供了明确基线。所有测试失败均可通过释放的指令、状态、程序、参考规则及轨迹追溯至具体执行后果，使可复用验证在需求与策略双层面均可被系统评估。
链接: https://arxiv.org/abs/2610.34634
作者: Jingjie Ning,Xueqi Li,Yi Lu,Yibo Kong
机构: Carnegie Mellon University, School of Computer Science (卡内基梅隆大学计算机科学学院)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Reusable Web-agent verification requires a rule to follow changing requirements and support subsequent execution. We introduce AgentProofBench, which connects paired tests of a frozen acceptance rule to execution from identical saved openings. The resource spans 97 tasks, 79 families, and four Web services. Experiments with DeepSeek Flash/Pro establish two linked measurements. Across 48 families, complete-panel and decision-replay tests retain 82/96 Flash and 79/96 Pro source programs; paired reuse-rebuild BA intervals are [-5.47, 1.57] and [-10.16, 0.00] percentage points. Across 22 execution families, deployment-state panel correctness separates program assignments whose two saved openings are both delivered correctly: 32/33 assignments with error-free panels, versus 1/11 with panel errors. All reference-defined program-gate regressions and withheld deliveries occur in the latter group. Neutral continuation attains the highest correct delivery under the tested generic feedback, providing a concrete baseline for measuring the value added by a gate. Released instructions, states, programs, reference rules, and trajectories connect each test failure to its execution consequences. AgentProofBench makes reusable verification assessable at both the requirement and policy levels.
[MA-12] Online Fair Division with Frequency Predictions
【速读】：该论文旨在解决在在线不可分物品分配场景下，如何基于频率预测实现公平性保障的问题。具体而言，研究考虑的是在物品到达前，算法已知各代理对所有物品的估值多重集，但未知物品的到达顺序以及不同代理的估值对应的具体物品。在此设定下，论文试图回答：是否能够实现对“最多一个物品的嫉妒（EF1）”的正向近似？其解决方案的关键在于：针对不同代理数量，揭示了在线算法在保证正向近似EF1方面的理论极限——当代理数 $ n \leq 4 $ 时，存在确定性算法可分别实现 $ \frac{1}{2} $-EF1（三代理）和 $ \frac{3 - \sqrt{5}}{2} $-EF1（四代理），而当 $ n \geq 5 $ 时，即使在 $ n+1 $ 个物品且预测完全一致的情况下，也无法保证任何正的EF1近似。此外，论文进一步证明，对于三或更多代理，连更严格的“最多一个物品的嫉妒消解（EFX）”也无法实现正向近似，即便估值仅取两个正数值。为突破此限制，作者利用Neoh等（ICML 2026）提出的元算法构造了**认知型EFX（EEFX）**分配，并在多项式时间内实现，从而导出 $ \frac{1}{n-1} $-EF(n-2) 的分配结果；相比之下，预先固定选取序列的算法只能在 $ n \leq k+2 $ 时保证正向近似EFk。最后，在噪声预测情形下，提出算法在预测精确时返回EEFX分配，并在低估和高估情况下分别达到 $ (1 - 2\varepsilon + O(\varepsilon^2)) $-EEFX 和 $ (1 - 2(n-1)\varepsilon + O(\varepsilon^2)) $-EEFX，其中线性项为最优，且较大的高估将彻底排除任何正向近似。
链接: https://arxiv.org/abs/2610.34625
作者: Tzeh Yuan Neoh,Nicholas Teh
机构: Harvard University (哈佛大学); University of Oxford (牛津大学)
类目: Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA)
备注:
Abstract:We study online fair division of indivisible goods among n agents with frequency predictions: before any good arrives, the algorithm learns the multiset of each agent’s values for the goods, but neither the arrival order nor which values of different agents belong to the same good. Neoh et al. (ICML 2026) showed that such predictions allow every share-based guarantee that is achievable offline, and left open whether a positive approximation of envy-freeness up to one good (EF1) is possible. We show that deterministic online algorithms can guarantee 1/2 -EF1 for three agents and (3-\sqrt5)/2 -EF1 for four agents, whereas for five or more agents, no online algorithm can guarantee any positive approximation of EF1, even with n+1 goods and identical predictions. Together with the known result for two agents, a positive approximation of EF1 is therefore possible if and only if n\leq4 . We also rule out any positive approximation of EFX for three or more agents, even when the valuations take only two distinct positive values. Next, we use the meta-algorithm of Neoh et al. (ICML 2026) to compute an epistemic EFX (EEFX) allocation in polynomial time for any n , and hence a 1/(n-1) -EF (n-2) allocation for n\geq3 , whereas algorithms that fix a picking sequence in advance can guarantee a positive approximation of EF k if and only if n\leq k+2 . Finally, for noisy predictions, we give algorithms that return EEFX allocations when the predictions are exact, and guarantee (1-2\varepsilon+O(\varepsilon^2)) -EEFX under underestimates and, for n\geq3 , (1-2(n-1)\varepsilon+O(\varepsilon^2)) -EEFX under overestimates; the linear terms are the best possible, and larger overestimates rule out any positive approximation for such algorithms.
[MA-13] Agent ic AI Orchestration Framework for Spatial Intelligence: Supply Chain Deployment
【速读】：该论文旨在解决工业空间决策中因系统间缺乏统一参考框架与共同所有者而导致的数据整合难题，即分析师耗费大量精力于系统对齐而非空间推理。其核心解决方案是提出ANASSA参考架构，关键在于通过类型化数据契约在多智能体边界进行约束、基于经认证的权威空间基底验证空间主张，并在六步认知循环中嵌入两次阻断式人工审核（human-in-the-loop approval），实现结构化推理、多智能体工作流编排、权威空间定位与反馈驱动执行的深度融合。该架构已在实际供应链部署中验证，覆盖海陆联运线路的多灾害风险评估，包括基于集装箱遥测数据的冷链异常检测；在为期三个月的试点中，经过三次迭代，实现了数据质量提升25%、空间工作流运行时间减少50%、决策延迟降低60%（相较于原有手动流程）。这些结果为非受控部署观察值，论文进一步提出了机制级评估所需的仪器化与消融实验协议以支持可复现性分析。
链接: https://arxiv.org/abs/2610.34569
作者: Constantinos Papantoniou,Brian Hilton
机构: Anassa.ai(安萨人工智能); Research Scientist(研究科学家)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Software Engineering (cs.SE)
备注: 10 pages, 6 figures. Accepted for publication in the proceedings of GeoIndustry 2026 (5th ACM SIGSPATIAL International Workshop on Spatial Big Data and AI for Industrial Applications). In-person presentation scheduled for November 3, 2026, Riverside, California. Proceedings forthcoming in the ACM Digital Library
Abstract:Industrial spatial decisions are assembled across systems that share neither a reference frame nor an owner, and analysts spend substantial effort reconciling those systems rather than reasoning about space. This paper presents ANASSA, a reference architecture integrating structured reasoning, multi-agent workflow orchestration, authoritative spatial grounding, and feedback-driven execution, and reports its instantiation in an operating supply-chain deployment. The architecture places typed data contracts at agent boundaries, validates spatial claims against a certified authoritative foundation, and enforces two blocking human-in-the-loop approvals inside a six-step cognitive loop. The deployment covers multi-hazard risk assessment on an ocean-and-inland freight lane, including cold-chain excursion detection over container telemetry. Across a three-month proof of concept (POC) with three implementation iterations, three deployment outcomes are reported: 25% improvement in data quality, 50% reduction in spatial workflow runtime, and 60% lower decision latency relative to the preceding manual workflow. Absolute deployment-scale quantities and baseline counts are confidential; reported percentages are observed relative changes. These are uncontrolled deployment observations, not controlled experiments. We therefore specify instrumentation and ablation protocols for mechanism-level evaluation.
[MA-14] FlockWorld: Toward Population-Scale Multi-Agent Video World Models through Artificial Life Simulation NEURIPS2026
【速读】：该论文旨在解决现有动作条件视频世界模型在多智能体场景中规模受限的问题，即多数模型仅支持单个智能体，而现有的多智能体模型仅能处理固定且数量较少的智能体群体。其核心解决方案在于构建FlockWorld——一个基于Reynolds Boids算法的人工生命环境，用于研究大规模多智能体视频世界模型。FlockWorld的关键创新包括：同步渲染多个智能体的自我中心视角（egocentric crops）、通过不同颜色区分摄像机-智能体身份以保持跨视角一致性、将规则衍生的加速度作为可操作的动作输入，并实现摄像机智能体数量与周围群体规模的独立配置。此外，论文提出FlockDiT，一种基于流匹配（flow matching）训练的动作条件潜在扩散变换器（latent diffusion transformer），用于联合生成多个智能体的同步自我中心视频。实验表明，联合生成策略在密度匹配条件下提升了智能体检测性能，而扩散强制机制显著提高了轨迹生成保真度。FlockWorld与FlockDiT共同构成一个可控的大规模多智能体视频生成测试平台。
链接: https://arxiv.org/abs/2610.34564
作者: Meryem Koksal,Yunzhe Wang,Volkan Ustun
机构: Grinnell College(格林内尔学院); University of Southern California(南加州大学); USC Institute for Creative Technologies(南加州大学创意技术研究所)
类目: Computer Vision and Pattern Recognition (cs.CV); Multiagent Systems (cs.MA)
备注: Accepted to NeurIPS 2026, Creative AI Track
Abstract:Most action-conditioned video world models remain limited to a single player, while existing multiplayer models support only small, fixed groups of agents. We introduce FlockWorld, an artificial-life environment built on Reynolds’s Boids algorithm for studying population-scale multi-agent video world models. Specifically, FlockWorld renders synchronized egocentric crops, assigns distinct colors to preserve camera-agent identity across views, exposes rule-derived accelerations as actions, and lets the number of camera agents be configured independently of the size of the surrounding population. We further introduce FlockDiT, an action-conditioned latent diffusion transformer trained with flow matching on FlockWorld data to jointly generate synchronized egocentric videos for multiple agents. Our results suggest that joint generation benefits agent detection at density-matched settings and that diffusion forcing improves rollout fidelity. Together, FlockWorld and FlockDiT provide a controlled testbed for multi-agent video generation.
[MA-15] DynaFlow: Learning Free-Form Multi-Agent Orchestration
【速读】：该论文旨在解决多智能体系统（Multi-agent Systems）在面对多样化任务需求时，因依赖预定义协作结构而缺乏动态适应能力的问题。其核心挑战在于如何让主智能体在推理过程中自主决定是否需要协作，并生成针对具体任务的可执行工作流以协调子智能体完成复杂任务。解决方案的关键在于提出DynaFlow框架，将自由形式的可执行工作流合成纳入主智能体的动作空间，使主智能体能够根据任务进展动态生成包含顺序或并行执行、结构化信息交换及验证机制的任务程序。通过监督微调（Supervised Fine-Tuning）在提炼轨迹上训练主智能体，并基于过滤后的高质量样本（剔除不支持前提、验证偏差和信息流缺陷）进行进一步优化，实现对协作策略的学习，同时保持所有子智能体固定，从而分离编排学习与执行能力提升。实验表明，在六个涵盖数学推理、科学问答与多跳检索的基准测试中，DynaFlow在强制模式下达到88.08%的整体准确率，领先最强基线10.90个百分点，且训练后工作流分析显示上游输出的显式使用、验证与迭代优化显著增强，揭示跨阶段信息流是多智能体编排的核心维度。
链接: https://arxiv.org/abs/2610.34467
作者: Ruhao Liu,Qi Li,Shizun Wang,Xinchao Wang
机构: National University of Singapore(新加坡国立大学)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Multi-agent systems often rely on predefined collaboration structures, limiting their ability to adapt to diverse task requirements. We introduce DynaFlow, a framework that integrates free-form executable workflow synthesis into the action space of the main agent. The agent decides when collaboration is needed and generates task-specific programs that coordinate sub-agents through sequential or parallel execution, structured information exchange, and verification. Execution results and traces inform subsequent actions, enabling the agent to adapt collaboration as reasoning progresses. We train only the main agent through supervised fine-tuning on distilled trajectories, followed by refinement on examples filtered for unsupported premises, biased verification, and information-flow defects. All sub-agents remain fixed, separating orchestration learning from executor capability improvement. Across six benchmarks spanning mathematical reasoning, scientific question answering, and multi-hop retrieval, DynaFlow achieves 88.08% overall accuracy in Forced mode, outperforming the strongest evaluated baseline by 10.90 percentage points with the same Qwen3.5-27B backbone. Workflow analysis finds more explicit use, verification, and refinement of upstream outputs after training, highlighting cross-stage information flow as a key dimension of multi-agent orchestration.
[MA-16] EvolveMAS: A Unified Self-Evolution Framework for Decentralized Multi-Agent Systems
【速读】：该论文旨在解决去中心化多智能体系统（Decentralized Multi-Agent Systems, MAS）在测试时适应不同交互情境（包括合作、混合动机和对抗性场景）以及可靠复用累积经验的关键挑战。现有方法通常针对特定交互场景设计适应机制，缺乏跨场景泛化能力。其核心解决方案在于提出一个统一的自演化框架EvolveMAS，通过引入专用的记忆代理（MemoryAgent）实现测试时的记忆治理，使主智能体在不干扰自身推理上下文的前提下，维护独立的结构化记忆（Structured Memory）。该框架将适应过程分为两个互补阶段：跨回合知识演化（Cross-Episode Knowledge Evolution）与回合内在线适应（Within-Episode Online Adaptation）。前者利用历史记忆与回合结果提炼可复用的交互知识（如伙伴画像与策略），后者则基于实时交互证据动态更新语义与工作记忆以指导后续行为。实验表明，EvolveMAS在五个基准与六种配置下取得77.2的最高归一化平均得分，消融实验证明了测试时记忆治理与跨回合知识复用对性能提升的关键作用。结果验证了无需更新模型参数即可实现基于经验的自适应，为去中心化多智能体系统的持续演化提供了有效路径。
链接: https://arxiv.org/abs/2610.34450
作者: Yalun Wu,Qiwei Ma,Bingbing Xu,Dan Zhang,Long Bai,Boyang Wang,Jiatong Li,Zheqi Lv,Shaozu Yuan,Peiying Wang,Jiawei Wang,An Zhang,Tat-Seng Chua
机构: 未知
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: 27 pages, 9 figures
Abstract:Decentralized multi-agent systems (MAS) enable large language model agents to interact without centralized control. Agents need to adapt to other agents at test time and reuse accumulated experience to refine their interaction strategies for future interactions across cooperative, mixed-motive, and adversarial settings. However, existing approaches often develop adaptation mechanisms for particular interaction settings. This raises a central question: how can decentralized MAS support test-time adaptation and reliable experience reuse across interaction settings? We propose EvolveMAS, a unified self-evolution framework that supports experience-driven adaptation within and across episodes. EvolveMAS pairs each main agent with a dedicated MemoryAgent for test-time memory governance, which maintains isolated Structured Memory outside the main agent’s reasoning context. EvolveMAS organizes adaptation into two complementary processes: Within-Episode Online Adaptation and Cross-Episode Knowledge Evolution. During Within-Episode Online Adaptation, MemoryAgent updates episodic, semantic, and working memory from interaction evidence to inform subsequent actions. During Cross-Episode Knowledge Evolution, the main agent uses memory and episode outcomes to distill reusable interaction knowledge comprising partner profiles and interaction strategies for subsequent episodes. Across five benchmarks and six evaluation configurations spanning the three interaction settings, EvolveMAS achieves the highest normalized average score of 77.2 among the compared methods. Ablations show that test-time memory governance and cross-episode knowledge reuse both improve performance. Interaction traces further illustrate how updated memory and distilled knowledge guide subsequent interactions. These results support experience-driven adaptation in decentralized MAS without updating model parameters.
[MA-17] 6G Reasoning Agent Societies: Composition Scaling and Consistency-Time Laws
【速读】：该论文旨在解决6G控制平面中基于大语言模型（Large Language Model, LLM）的智能体社会在信息交互过程中因幻觉（hallucination）传播而导致的协同一致性失效问题。现有架构通过局部审计机制对每个消息进行校验，但此类孤立审计无法捕捉智能体间集体行为带来的隐性相关性，导致系统信任评估失真与收敛延迟。其解决方案的关键在于将智能体社会建模为细胞层（cellular sheaf），并据此推导出三类普适性规律：第一，组合律揭示了全局分歧量超过各智能体独立审计得分之和的根源在于隐藏的总相关性，这一盲区可掩盖高达23%的联合分歧，并因认知信噪比（cognitive SNR）中信号抵消效应，使误判程度在拓扑结构影响下虚增1.14 dB，虽能维持有效共识状态，却显著延缓收敛；第二，缩放律表明协调过程在不同度量维度上解耦，最弱连接维度决定整体收敛速度，一致性时间在多项式增长拓扑下以Ω(n²/d)随智能体数n增长，路径型拓扑在n*=13时即突破1秒近实时预算，而扩张图（expander）与半径3内局部审计则可保持有界；第三，一致性时间律指出遗漏相关性的时序误差具有单向性且受限于最差测量链路。上述理论规律在模拟拓扑与真实部署的LLM社会中均得到验证，为近实时运行提供了可证明的最坏情况延迟上限。
链接: https://arxiv.org/abs/2610.34357
作者: Hatim Chergui,Carolina Fernández-Martínez,Mustapha Benjillali,Mehdi Bennis,Merouane Debbah
机构: i2CAT Foundation( i2CAT 基金会); INPT(国家理工学院); University of Oulu(奥卢大学); Research Institute for Digital Future, Khalifa University(哈利法大学数字未来研究院)
类目: Networking and Internet Architecture (cs.NI); Multiagent Systems (cs.MA)
备注:
Abstract:Emerging 6G control planes would compose Large Language Model (LLM) agents that exchange interpreted conclusions, so hallucinations can propagate even over error-free transport. Recent architectures audit each message locally against the sender’s observation mask, but isolated audits miss collective behavior. Modeling the agent society as a cellular sheaf, we establish three laws valid on arbitrary topologies. First, the \emphcomposition law: network-wide divergence exceeds the sum of per-agent audit scores by exactly their hidden total correlation. This blind spot masks up to 23% of joint divergence and, because the declared signal cancels in the cognitive signal-to-noise ratio (SNR), overstates it by 1.14 ,dB purely through topology; the resulting trust inflation preserves valid consensus states but slows convergence. Second, the \emphscaling law: reconciliation decouples across metric counters, so the worst-connected dimension sets the pace. Consistency time grows for a society size n as \Omega(n^2/d) on polynomial-growth topologies of degree d , and paths breach a 1 ,s near-real-time budget at n^\ast=13 agents, whereas expanders ( \Theta(1) ) and radius-3 local audits ( O(1) , \approx300 ,ms) stay bounded. Third, the \emphconsistency-time law: the timing error from omitted correlations is one-sided and capped by the worst-measured link. These laws, validated on simulated topologies and deployed LLM societies, give provable worst-case latency ceilings for near-real-time operation.
[MA-18] Weighted Fairness in Multi-User Earth Observation Satellite Scheduling
【速读】：该论文旨在解决多利益相关方参与的地球观测卫星星座（Earth Observation Satellite Constellation）调度中因投资比例不均导致的公平性问题。传统方法通常假设各用户贡献均等，享有同等资源访问权，但现实中不同利益相关方的投资权重存在显著差异。为此，论文提出一种加权地球观测卫星调度问题（Weighted EOSSP），引入用户权重以反映其对星座的实际贡献，并设计了四种兼顾效率与公平性的优化准则：加权效用最大化（weighted utilitarianism）、加权平等主义（weighted egalitarianism）、加权最小字典序优化（weighted leximin）以及加权有序加权平均（WOWA）。上述方法均被建模为混合整数线性规划（MILP）问题，并与两种启发式算法——加权贪心法和加权轮询法——在文献中的典型场景下进行对比评估，重点考察其在严格运行时间约束下的效率与公平性表现。解决方案的关键在于通过可调节的加权公平性准则，在保证整体调度效率的同时，实现对不同投资权重用户的差异化、合理化资源分配。
链接: https://arxiv.org/abs/2610.34282
作者: Maxime Lucet,Nawal Benabbou,Aurélie Beynier,Nicolas Maudet,Gauthier Picard
机构: LIP6, CNRS, Sorbonne Université, Paris, France; ONERA, DTIS, Université de Toulouse, Toulouse, France
类目: Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA)
备注:
Abstract:As constellations of satellite grow in size, and become more costly, they are increasingly funded by multiple stakeholders who expect access to the constellation’s resources that is commensurate with their investment. Such cooperation raises significant challenges for constellation operation, in particular regarding the fairness of the computed schedules among stakeholders. While several existing works deal with fairness in the Earth Observation Satellite Scheduling Problem (EOSSP), most of them assume that users (e.g. space agencies, research labs, countries) have the same weights, i.e. contributed equally to the project and therefore have equal access to the constellation. However, in practice, users’ contributions often differ significantly. Hence, in this paper, we study a weighted variant of the EOSSP where users are assigned weights reflecting their respective contributions to the constellation. We propose several solution concepts with varying focus on efficiency and fairness, namely weighted utilitarianism, weighted egalitarianism, weighted leximin, and weighted ordered weighed averaging (WOWA). We formulate the related optimization problems as Mixed-Integer Linear Programs (MILPs), and compare them against two natural heuristics: weighted greedy and weighted round robin. All methods are evaluated on scenarios drawn from the literature, under strict runtime constraints, in order to assess the performance of each approach with respect to both efficiency and fairness metrics.
[MA-19] BE-CTDE: Blockchain-Enabled Multi-Agent Reinforcement Learning with Anomaly Detection and Token Economics
【速读】：该论文旨在解决多智能体强化学习（Multi-agent Reinforcement Learning, MARL）在物联网系统中因智能体策略偏离导致协作失效、训练过程可信度下降的问题。其核心挑战在于如何在去中心化执行环境下确保智能体行为的可追溯性与责任性，从而维持协同决策的有效性。解决方案的关键在于提出区块链赋能的集中式训练与去中心化执行框架（Blockchain-Enabled Centralized Training and Decentralized Execution, BE-CTDE），将区块链作为主动的训练管理层，引入行为验证器（behavioral verifier）对智能体执行轨迹与共享策略建议进行比对；通过基于代币的激励机制（包含罚没、销毁、奖励与补偿）动态调整各智能体对全局策略更新的贡献权重，实现行为可审计性与学习影响力之间的直接关联。该机制通过建立单轮激励条件以考虑动作依赖型补偿及偏离收益，并在明确的检测与记账假设下证明了有限期望排除时间界的存在，表明无需实现完美检测即可保障系统鲁棒性。实验结果表明，BE-CTDE在多智能体路径规划任务中显著优于现有先进学习型路径规划算法，在攻击场景下成功率更高，且在所有配置中均超越部分防御基线，碰撞率较仅依赖检测但无经济后果的方案降低最多达35%，较传统质押罚没机制降低最多达43%。这些成果共同揭示了行为验证与代币反馈机制在支持可问责协作学习中的关键作用。
链接: https://arxiv.org/abs/2610.34121
作者: Chih-Wei Huang,Hsin-Chieh Lee,Yun-Chen Yu,Ibrahim Althamary,Rongxing Lu,Phone Lin
机构: National Central University (国立中央大学); King Fahd University of Petroleum and Minerals (法赫德国王石油与矿业大学); Queen’s University (皇后大学); National Taiwan University (台湾大学)
类目: Multiagent Systems (cs.MA); Cryptography and Security (cs.CR); Machine Learning (cs.LG); Networking and Internet Architecture (cs.NI)
备注: 12 pages, 4 figures, 6 tables. Submitted to the IEEE Internet of Things Journal
Abstract:Multi-agent reinforcement learning supports cooperative decision-making in Internet of Things systems, but strategic deviations can disrupt coordination and undermine confidence in the training process. This paper proposes Blockchain-Enabled Centralized Training and Decentralized Execution (BE-CTDE), which uses blockchain as an active training-management layer. A behavioral verifier compares executed trajectories with shared-policy recommendations; its flags drive a token mechanism combining slashing, burning, rewards, and compensation. Retained balances determine each agent’s contribution to the global policy update, connecting auditable behavior to learning influence. We derive a one-round incentive condition that accounts for action-dependent compensation and the payoff from deviation, and a finite expected exclusion-time bound under explicit conditional detection and accounting assumptions, showing perfect detection is unnecessary. An exact finite-population capacity bound characterizes the limits of the population-relative detector. In multi-agent path finding experiments, BE-CTDE outperforms a state-of-the-art learning-based path planner with substantially higher success rates under attack, even when adversaries form the majority. It also outperforms partially defended baselines in all configurations, with collision rates up to 35% lower than under detection without economic consequences and up to 43% lower than under conventional stake slashing. Together, these results characterize how behavioral verification and token feedback can support accountable cooperative learning.
[MA-20] Build Dont Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players
【速读】：该论文旨在解决如何让大语言模型（LLM）智能体从原始数据出发，自主构建一个能够与人类玩家相媲美的复杂实时系统这一难题。其核心挑战在于处理具有随机商店、战斗不确定性、信息隐藏以及近一千张相互作用卡牌的高维、动态博弈环境——以《炉石传说：酒馆战棋》（Hearthstone Battlegrounds）为例。解决方案的关键在于提出AutoTavern系统：一个具备自我改进能力的研究体系，其中专用的LLM研究智能体共享类型化研究记忆，所有假设均需通过验证门控机制检验，并通过实时对局形成闭环反馈，使智能体不仅能优化模拟器、对手策略和决策政策，还能持续改进自身的科研流程；人类仅负责设定目标与解释模糊规则。在20天内，AutoTavern成功构建了首个忠实还原真实游戏的TavernGym仿真环境，并训练出23版TavernAgent，采用PPO算法对抗经过真实数据校准的对手。最终，TavernAgent可独立完成完整对局，其段位从6000提升至约15000，在公开排行榜上连续20场平均排名1.60，首胜率高达80%，各项指标超越服务器前十大高分玩家。通过TavernBench（960个任务分布在40种场景）评估显示，TavernAgent在任务完成率（91% vs 前沿LLM最高63%）与决策速度（0.03秒内完成，而其他模型需达26秒）方面显著领先，且在单回合允许时间内保持90%的响应率，远超现有模型。结果表明，LLM智能体在构建系统能力上远超其直接博弈表现，揭示了其作为自动化科研代理的巨大潜力。相关代码与基准测试平台将开源发布。
链接: https://arxiv.org/abs/2610.33845
作者: Changlian Ma,Limin Wang
机构: Nanjing University (南京大学); Shanghai AI Laboratory (上海人工智能实验室)
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: 20 pages, 8 figures, 15 tables. Code and benchmark: this https URL
Abstract:Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds, an eight-player auto-battler with stochastic shops and battles, hidden information and nearly a thousand interacting cards. We present AutoTavern, a self-improving research system: specialised LLM research agents share a typed research memory, every claim passes verification gates, and live games close the loop, so the agents improve the simulator, opponents and policy and, from the same feedback, their own research process; a human sets goals and rules on ambiguous mechanics. In 20 days AutoTavern built TavernGym, the first faithful, full-content reproduction of the live game, and delivered 23 versions of TavernAgent, trained with PPO against opponents calibrated to real games. TavernAgent plays complete games on the live ladder by itself: its rating rose from 6,000 to about 15,000 and, on the public leaderboard’s measure, its last 20 records average a placement of 1.60 with 80% first places, better than the server’s ten highest-rated players on every statistic (2.34, 59%). TavernBench, 960 tasks in 40 scenarios, shows that the system the agents built is the strongest player: TavernAgent answers 91% of the play tasks against at most 63% for frontier LLMs, and decides in 0.03 s where they need up to 26 s; within the second a live turn allows, it keeps 90% and no LLM more than 26%. LLM agents build far better than they play. We will open-source TavernBench and TavernGym at this https URL.
[MA-21] Secure and Functional Code Generation with Self-Evaluation
【速读】：该论文旨在解决大语言模型（Large Language Models, LLMs）在生成代码时虽功能正确但存在安全隐患的问题。现有安全增强方法如安全导向的提示（security-oriented prompting）或检索增强生成（Retrieval-Augmented Generation, RAG）往往难以兼顾安全性与功能性，可能引入不兼容任务需求的防御机制或误删必要行为。其核心挑战在于如何在不依赖已知漏洞或补丁样例的前提下，精准获取与任务相关的安全知识，并将其有效融入代码生成过程，同时保障功能完整性。为此，论文提出SecGent——一种推理阶段的安全增强框架，通过从通用弱点枚举（Common Weakness Enumeration, CWE）数据库中检索与任务规范相关的弱安全性上下文，构建面向具体任务的安全指导。该框架不直接使用漏洞或修复代码样本，而是将任务描述与文本化CWE记录关联，提供细粒度的安全约束。随后，通过功能与安全双重测试反馈驱动修复机制，实现对生成结果的迭代优化，确保在提升安全性的同时维持高功能性。实验表明，SecGent在CWEval基准上显著优于四种主流推理阶段基线方法，所有五种模型均取得最高联合功能-安全成功率；在开源模型上，平均功能-安全准确率@1（FS@1）相较RAG提升15.5%，相较最强提示优化器提升19.8%；在Claude Sonnet 5.5上达到92.1%的FS@1，超越最强基线AutoGen达4.8%；且在独立对比中优于采用安全提示的Claude Code超过6个百分点。反馈引导的修复策略进一步将FS@1提升最多达6.7%，主要体现在恢复因安全加固而受损的功能性，验证了其在平衡安全性与功能性方面的有效性。
链接: https://arxiv.org/abs/2610.33450
作者: Md Afif Al Mamun,Gias Uddin
机构: University of Calgary (卡尔加里大学); York University (约克大学)
类目: Cryptography and Security (cs.CR); Multiagent Systems (cs.MA); Software Engineering (cs.SE)
备注: 18 pages, 7 figures
Abstract:Large Language Models (LLMs) can generate functionally correct code that remains insecure. Security-focused fine-tuning can be expensive and is not always available, motivating inference-time approaches. Security-oriented prompting and retrieval can reduce vulnerabilities, but may also remove required behavior or apply defenses incompatible with the task. The challenge is to identify relevant security knowledge and use it to improve security without sacrificing functionality. We present SecGent, an inference-time framework that retrieves weakness context from the Common Weakness Enumeration (CWE) database and applies it to secure code generation. By connecting task specifications to relevant textual CWE records, SecGent provides task-specific security guidance without relying on retrieved vulnerable or patched code examples. Complementary functional and security tests check the resulting implementation, and their execution feedback guides repairs toward satisfying both objectives. We evaluate SecGent with four open-source LLMs and Claude Sonnet 5.5 across five programming languages on the CWEval benchmark. It achieves the highest joint functional–security success against four inference-time baselines for all five models. Across the open-weight models, mean functional-security@1 (FS@1) improves by 15.5% over RAG, with gains in both functionality and security, and by 19.8% over the strongest prompt optimizer compiled for each model. On Sonnet, FS@1 reaches 92.1%, outperforming the strongest baseline AutoGen by 4.8%. In a separate comparison, SecGent outperforms Claude Code with security-focused prompting by over 6% in FS@1. Feedback-guided repair further improves FS@1 by up to 6.7% over generation without feedback, primarily by recovering functionality while checking security.
[MA-22] Agent -to-Agent Synthetic Social Engineering on the Web
【速读】：该论文旨在解决生成式 AI 代理在跨 Web 服务协作中可能因被误导而执行未经授权操作的安全问题，即在证据处理正确且权限未被突破的前提下，仍存在行为偏离授权目标的风险。其核心解决方案在于提出“合成社会工程”（synthetic social engineering, SSE）的系统性分析框架，该框架通过分离身份识别、证据支持与权限判断三个关键维度，明确区分恶意诱导行为的尝试与成功篡改，不依赖对人类心理机制的假设。研究构建了可复现的测试平台 Agent Service Exchange and Evidence Commons (ASE-EC)，在1,260次基于Qwen3-4B的虚构简报任务中验证了该框架的有效性。结果显示，仅通过引入有限的先前协议，未经授权请求的比例从43.3%上升至96.7%，但受体仍保持对证据的支持并维持原有权限状态。这一发现揭示了行动选择本身是一个独立于证据处理的新型安全威胁，为评估和设计能够保障协作效率同时防范非授权行为的防护机制提供了可复用的理论基础与实验范式。
链接: https://arxiv.org/abs/2610.33411
作者: Bryan C. Boots
机构: 未知
类目: Cryptography and Security (cs.CR); Multiagent Systems (cs.MA)
备注:
Abstract:An AI agent can prepare an accurate report and still request an action its user has not authorized. Cooperation across Web services therefore requires separate judgments about identity, evidentiary support, and permission. We define synthetic social engineering (SSE) as deliberately deceptive or misleading influence intended to exploit AI recipients’ responses to credibility, authority, or obligation cues and redirect behavior away from authorized objectives. We focus on separately tasked sender agents that produce or adapt misleading communication. Drawing on cybersecurity, social engineering, and persuasion research, we develop an operational framework connecting actors, deceptive representations, candidate influence routes, evidence or permission boundaries, responses, and consequences. The framework distinguishes attempts from successful redirection, without assuming human psychological processes. We implement it in the Agent Service Exchange and Evidence Commons (ASE-EC), a reproducible testbed with independently specified records and authorization scopes. Across 1,260 research episodes using Qwen3-4B in two fictional briefing tasks, matched comparisons test authority and commitment appeals, while defense comparisons assess protection alongside authorized task completion. Invoking a limited prior agreement increased unauthorized requests from 43.3% to 96.7%, although recipients retained supported decisions and permissions remained unchanged. These controlled findings identify action selection as a distinct security concern, even when evidence handling succeeds, and they provide a reusable basis for evaluating safeguards that preserve useful cooperation across Web services.
[MA-23] Estimation of Propagated Risk in 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）的多智能体系统中，上游智能体行为对下游智能体正确性产生影响所引发的可靠性挑战，特别是现有方法未能显式建模在不同上游智能体正确性状态下，实际发生的发送者-接收者交互如何贡献于下游错误风险的问题。其解决方案的关键在于提出一种交互条件化的误差风险传播模型——交互效应传播（Interaction-Effect Propagation, IE-Prop），该模型分别学习正确与错误上游智能体的独立影响，并基于发送者估计的误差风险对这些条件化效应进行加权融合，进而通过执行图递归传播风险估计。实验结果表明，该方法在三种多智能体架构和三个基准测试上均显著提升了最终节点的受试者工作特征曲线下面积（AUROC），平均提升约0.05点；在序列型系统中，多步风险追踪进一步带来+0.03 AUROC的增益，验证了实际交互信息对预测下游风险具有超越接收方局部证据的价值，且多步风险追踪的效益取决于系统的信息流结构。
链接: https://arxiv.org/abs/2610.33282
作者: Shreyan Ganguly,Souradeep Chattopadhyay,Qisai Liu,Soumik Sarkar
机构: Iowa State University (爱荷华州立大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Large Language Model (LLM) based multi-agent systems introduce a distinctive reliability challenge in which actions taken by upstream agents can affect the correctness of downstream agents. While prior work has investigated error propagation and developed methods for estimating uncertainty inherited from upstream agents, existing approaches do not explicitly model how realized sender-receiver interactions contribute to downstream error risk under different sender correctness states. In this work, we study agent-level error-risk estimation from multi-agent execution traces. We propose Interaction-Effect Propagation (IE-Prop), an interaction-conditioned model that learns separate effects for correct and incorrect upstream agents. For each sender-receiver interaction, the model combines these conditional effects using the sender’s estimated error risk, and recursively propagates the resulting risk estimates through the execution graph. We evaluate our approach across three multi-agent architectures and three benchmarks. Across all evaluated architectures, our approach raises the final-node AUROC over the established baseline by approximately 0.05 points on average. Furthermore, tracking risk across multiple agent steps contributes a distinct +0.03 AUROC gain in sequential systems. These results show that realized inter-agent interactions provide predictive information beyond receiver-local evidence, while the value of multi-step risk tracking depends on the system’s information-flow structure.
[MA-24] Coral Discovery: The Division of Cognitive Labor in Autonomous Research
【速读】：该论文旨在解决自主研究系统中认知劳动分配不均的问题，即在缺乏中央规划的情况下，如何有效引导多个大语言模型（Large Language Model, LLM）代理在多样化路径间进行探索与协作，避免所有代理趋同于单一高绩效路径而导致创新停滞。传统方法仅依赖评分反馈使代理“跟随领先者”，导致群体收敛至少数路径，限制了探索广度。其解决方案的关键在于提出Coral Discovery框架，通过引入方向标签（direction label）机制，为每个尝试赋予三重信息：改进得分的机制（mechanism）、具体方法变更（method）及方向本身。该系统构建一个动态的方向看板（direction board），基于各方向的历史表现、探索程度和潜在价值进行加权评估，量化每条路径的贡献并激励对低尝试方向的探索；同时，信用分配以各方向自身最佳表现为准，保障落后路径仍能持续获得代理投入。此设计实现了对探索-利用平衡的精细化调控，使代理在完成自身任务后将看板作为策略建议而非强制指令，从而在保持多样性的同时提升整体性能。实验表明，Coral Discovery在9项任务中优于或持平现有最强开源多代理系统，并显著优化了Kernel基准结果，在多项任务上达到新纪录，尤其在Polyominoes任务中将准确率从87.0提升至91.78，验证了其在复杂自主研究场景中的有效性。
链接: https://arxiv.org/abs/2610.33281
作者: Xinxing Ren,Qianbo Zang,Ziyan Wang,Suman Deb,Qiang Yang,Caelum Forder,Peter Carroll,Gilead Gevero,Jing Liu,Yali Du,Milind Tambe,Zekun Guo
机构: 未知
类目: Multiagent Systems (cs.MA)
备注:
Abstract:A research community is only as strong as the way it divides its cognitive labour. The philosopher Philip Kitcher argued that researchers should not all pursue the most promising approach. Credit rules, not a central plan, keep some of them on other lines. We bring this question to autonomous research, where Large Language Model agents improve programs against a deterministic grader. Agents that share only a score follow the leader and converge onto one approach. Coral Discovery gives the community a direction board. A verifier labels every attempt with a direction, the mechanism earning the score, and a method, one change within it. Per direction, the board tracks how much each eval improves the score, adds an exploration bonus for little-tried directions, prices the option of opening a new direction, and suggests how many agents each direction deserves. Credit is earned against a direction’s own best, so a lagging direction keeps its agents. Agents finish their lines, then treat the board as advice. We test on nine tasks with the same model and budget, without grader access except for Kernel. Coral Discovery beats the strongest open autonomous multi-agent system on 6 tasks, ties on 1, and loses on 2. Against published numbers, it is best or tied on 4. It improves the reported Kernel result from 1103 to 1095 cycles and passes 1103 at eval 275, under half of the 596 used before. It sets a new Polyominoes state of the art, 91.78 against 87.0. Direction labels show that the leading scores rest on directions a score-only community never entered. Code will be released shortly.
[MA-25] Neuromorphic Decision Dynamics for Deadlock-free Safe Navigation in Congested Environments
【速读】：该论文旨在解决复杂密集环境中智能体（如机器人或自动驾驶车辆）在无通信、无全局信息、无预设优先级的情况下实现安全、高效且可扩展的协同通行问题，核心挑战在于如何在不依赖外部协调机制的前提下，确保碰撞避免（collision avoidance）、活性（liveness，即系统不会陷入死锁或活锁）、鲁棒性与可扩展性。其解决方案的关键是提出一种类脑决策动力学（Neuromorphic Decision Dynamics, NDD），这是一种去中心化、事件驱动的控制器，通过将智能体内部的决策状态与注意力状态动态耦合于自身的物理状态及局部观测到的相对运动，实现对冲突的即时响应。每个智能体沿既定路径持续调节自身速度，响应冲突需求的程度和时机由局部交互实时决定；当某局部群体出现安全临界冲突时，决策-动作反馈机制会破坏集体犹豫状态，自发涌现出协调一致的通行顺序，从而形成稳定的安全通行秩序。该秩序在冲突解除后仍保持，而群体结构则根据新冲突动态分裂或合并。研究进一步给出了感知范围、反馈增益、响应时间尺度与执行机构极限等参数的显式设计条件，确保有限异构群体在任意变化规模下均能在有限时间内实现无碰撞通行且无死锁或活锁。计算开销仅与每个智能体感知的局部交互数量成正比，验证了其良好的可扩展性。大规模仿真、与现有方法的对比以及硬件实验均证明了NDD在复杂拥挤场景中的有效性。
链接: https://arxiv.org/abs/2610.33271
作者: Himani Sinhmar,Vaibhav Srivastava,Naomi Ehrich Leonard
机构: Michigan State University (密歇根州立大学); Princeton University (普林斯顿大学)
类目: Robotics (cs.RO); Multiagent Systems (cs.MA)
备注:
Abstract:Safe navigation in congested environments requires both collision avoidance and coordinated passage. We introduce neuromorphic decision dynamics (NDD), a decentralized, event-based controller, that provides navigation in congested spaces with guarantees on safety, liveness, robustness, and scalability. NDD makes safe-and-live deconfliction a controllable, locally tunable emergent property of local agent interactions, without requiring communication, preassigned priorities, models of neighbors’ objectives or future actions, or estimates of group size or global network properties. By design, NDD couples the dynamics of internal decision and attention states with changing physical states and locally observed relative motion. Each agent continuously modulates its speed along its prescribed path, responding as soon and as much as conflict resolution requires. Agents self-organize into disjoint, locally engaged groups that resolve conflicts concurrently. When any group’s conflict becomes safety-critical, decision-action feedback destabilizes collective indecision, and coordinated conflict-resolution emerges that stabilizes a safe traversal order within the group. Order persists as conflicts clear, but groups may split or merge as agents autonomously adapt to new conflicts and those with resolved conflicts resume their nominal motion. We derive explicit conditions on sensing range, feedback gains, response time scales, and actuation limits that guarantee collision avoidance and traversal-order formation in finite time with no deadlock or livelock for finite, heterogeneous agent populations of arbitrary, changing size. Computational cost scales with the number of local interactions sensed by each agent. Large-scale simulations, comparisons with existing approaches, and hardware experiments demonstrate NDD in complex congested scenarios and validate its scalability.
[MA-26] VibeGame: Prompt-to-Game Development with AI-Native Engine and Self-Evolving Adversarial Agent Team
【速读】：该论文旨在解决从自然语言提示（natural-language prompts）端到端生成完整可玩2D游戏的挑战，其核心问题在于如何协调代码、场景、资源、动画、物理引擎及配置等多模态要素，并支持开发者的交互式验证。其解决方案的关键在于提出VibeGame框架——一个AI原生的游戏引擎，通过结构化文本形式对项目状态进行模式验证，并提供帧同步的运行时控制能力，使智能体能够以自主节奏检查游戏状态、执行动作并验证行为逻辑。在此基础上，设计由八个专业化智能体组成的对抗性代理团队，实现开发与审查的分离，通过多轮迭代发现并修复问题；同时引入无需训练的自演化机制，将已成功项目提炼为可复用的骨架、模块与协作契约，提升未来开发效率。实验表明，在仅依赖提示生成艺术资源（P2G）的设定下，基于相同基础模型的VibeGame显著优于现有各类代理系统，包括GPT-6.1-Sol。
链接: https://arxiv.org/abs/2610.33106
作者: Wenbo Hu,Ken Li,Jiazhe Wei,Yukang Cao,Weiyi Hong,Jiayi Dai,Chenjun Bai,Jiajun Liang,Yucheng Liao,Ruichuan An,Zeyu Lou,Haofan Wang,Yueming Lyu,Jingkang Yang,Ziwei Liu,Chenyang Si
机构: 未知
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Large language model agents have enabled software development through natural-language interaction, but extending this paradigm to end-to-end game development remains challenging. Building a playable game requires coordinating code, scenes, assets, animations, physics, and configurations, while allowing developers to validate their integrated behavior through interactive testing. We introduce VibeGame, a framework for creating complete 2D game projects from natural-language prompts and optional reference images. Its AI-native game engine represents project state as schema-validated structured text and provides frame-synchronous runtime control, enabling agents to inspect game states, execute actions, and verify behavior at their own pace. Built upon this interface, we design an Adversarial Agent Team of eight specialized agents to separate implementation from review, so that problems are found and fixed through multi-round loops. For each conducted and accepted project, we further introduce a training-free self-evolution mechanism to distill it into reusable skeletons, modules, and collaboration contracts for future development. We evaluate prompt-assets-to-game (PA2G), where agents are given an asset library, and prompt-to-game (P2G), where agents generate their own art. On a GameCraft-Bench subset, VibeGame in P2G consistently outperforms a range of agent harnesses in either P2G or PA2G with the same underlying model, including GPT-6.1-Sol.
[MA-27] AutoSwarm: Automating Agent Swarms
【速读】：该论文旨在解决多智能体系统在复杂任务自动化中因任务分配不当导致的协同效率低下与局部错误传播问题。现有方法虽通过智能体专业化与协作提升性能，但缺乏对责任边界和贡献可评估性的显式定义，易引发冗余或冲突工作，且依赖关系管理不善会导致错误在系统内级联放大。其解决方案的关键在于提出AutoSwarm框架，通过协作形成机制（collaboration formation）联合定义智能体的责任边界、结果接纳条件及下游使用标准，使能力分配基于其他智能体可评估与整合的贡献；同时借助执行协调机制（execution coordination），利用共享状态、依赖关系与监督结构来动态控制智能体激活、结果交接与融合，并根据审查反馈与新出现的阻塞因素自适应调整后续行为。该双机制协同设计实现了高效、鲁棒的跨智能体协作，显著优于单智能体基线与现有多智能体框架，在长时序编程、交互式规划及专业金融分析等任务中展现出优越性能。
链接: https://arxiv.org/abs/2610.33048
作者: Zhonghang Li,Zongwei Li,Chao Huang
机构: The University of Hong Kong(香港大学)
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Large language model (LLM) agents increasingly automate complex tasks, with multi-agent systems seeking further performance gains through specialization and collaboration. However, distributing work across agents does not automatically yield complementary contributions, and inter-agent dependencies can turn local errors into task failure. To address these challenges, we propose AutoSwarm, a multi-agent framework for collaborative problem solving. Collaboration formation jointly defines responsibility boundaries and conditions for acceptance and downstream use, grounding capability allocation in contributions that other agents can assess and integrate toward a shared objective. Execution coordination couples agents’ local progress through a shared state, using dependencies and supervisory relationships to govern agent activation, result handoffs, and integration while adapting subsequent actions to review feedback and emerging blockers. Experiments spanning long-horizon programming, interactive planning, and professional financial analysis demonstrate AutoSwarm’s advantages over single-agent baselines and existing multi-agent frameworks. Source code is available at this https URL.
[MA-28] False Belief Dissemination Persistence and Mitigation in Multi-Agent Social Networks Through the Lens of the Mandela Effect
【速读】：该论文旨在解决多智能体社交网络中虚假信念（false beliefs）的传播与持续问题，其核心挑战在于模拟现实社会网络中信息扩散的动态过程，并揭示虚假信息如何在缺乏原始来源的情况下依然广泛传播并长期存续，类比于人类社会中的“曼德拉效应”（Mandela effect）。解决方案的关键在于构建一个受控实验框架，通过整合智能体间的通信机制、私有记忆存储以及共享的模拟网络环境，系统地研究虚假信息在不同场景下的传播（dissemination）、持久性（persistence）与缓解策略（mitigation）。该框架基于540条跨8个领域、4种可信度水平的伪造声明数据集，能够精确评估虚假信息在不同模型架构、网络拓扑结构下的接受程度与扩散范围。实验结果表明，模型对虚假信息具有高度易感性，传播覆盖率达96.0%，且即使移除制造者，虚假信念仍保持较高接受率；相比之下，事前预防措施效果有限，而针对性的纠正干预虽能缓解但无法彻底消除虚假信念。因此，研究指出虚假信念的持续存在是多智能体社交网络中的关键安全挑战，亟需同时应对初始传播路径与累积证据链的双重影响。
链接: https://arxiv.org/abs/2610.33026
作者: Dianzhi Yu,Hongru Wang,Ziheng Xiong,Yanyu Chen,Zixing Song,Jianting Chen,Zhenfei Yin,Jeff Z. Pan,Amos Storkey,Irwin King
机构: The Chinese University of Hong Kong(香港中文大学); The Hong Kong Polytechnic University(香港理工大学); Tongji University(同济大学); University of Oxford(牛津大学); University of Edinburgh(爱丁堡大学)
类目: ocial and Information Networks (cs.SI); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: 39 pages, 6 figures, 13 tables
Abstract:Multi-agent social networks are emerging as communities in which agents communicate, retain memories, and publish information that others can find as evidence. These interactions create opportunities for false claims to spread and remain influential beyond their original sources, analogous to the Mandela effect in human societies, in which many people recall the same false memory. In this paper, we systematically study false-belief dynamics through the lens of the Mandela effect, focusing on dissemination, persistence, and mitigation. Specifically, we build a controlled framework connecting agent communication, private memory, and a shared simulated web to mimic information dissemination in practical social networks, supported by 540 fabricated claims spanning 8 domains and 4 plausibility levels. For dissemination, we measure how widely and quickly claims gain acceptance across models and communication topologies. For persistence, we withdraw agents promoting fabricated claims while retaining existing articles and memories, and track subsequent acceptance. For mitigation, we compare preventive measures applied before implantation with corrective interventions introduced after claims have spread. Experimental results show that evaluated models readily accept and disseminate fabricated claims, with dissemination reach as high as 96.0%. Dissemination reach varies with claim plausibility, domain, and network size and composition, while communication topology shapes dissemination speed across model families. Acceptance barely declines after fabricators withdraw, and targeted corrective interventions outperform the preventive measures without eliminating false beliefs. These findings identify persistent false beliefs as a safety challenge for multi-agent social networks and highlight the need to address both their initial dissemination and the accumulated evidence that sustains them.
[MA-29] UMAS: System-Level Uncertainty Quantification for Multi-Agent LLM Systems NEURIPS2026
【速读】：该论文旨在解决多智能体系统（Multi-Agent Systems, MAS）在应用大语言模型时存在的幻觉（hallucination）和错误输出问题，尤其关注如何在不增加推理成本的前提下实现系统级不确定性量化（Uncertainty Quantification, UQ）。现有不确定性评估方法大多针对单智能体场景设计，难以有效利用多智能体间的协作结构来提升可靠性。本文提出的UMAS框架是一种无参数、推理成本中立的系统级不确定性量化方法，其核心创新在于通过各智能体生成结果的内在置信度动态估计智能体间影响的可信度，并据此调节每个节点的不确定性更新机制，最终聚合节点级不确定性得到全局系统级不确定性估计。该方法不仅显著提升了不确定性评估性能（平均优于现有最优基线10.54 AUROC点），还支持幻觉检测与不确定性感知的答案选择，在特定任务上将多智能体系统的准确率提升最高达12.6个百分点，同时增强了系统的可靠性。
链接: https://arxiv.org/abs/2610.33013
作者: Hanwen Li,Jinhao Duan,Xiaoshuang Shi,Yue Zhang,Tianlong Chen,Kaidi Xu,Chenxi Yuan
机构: New Jersey Institute of Technology (新泽西理工学院); University of North Carolina at Chapel Hill (北卡罗来纳大学教堂山分校); University of Electronic Science and Technology of China (电子科技大学); Shandong University (山东大学); City University of Hong Kong (香港城市大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026
Abstract:Multi-Agent Systems (MAS) are a promising paradigm for improving the reasoning capabilities of large language models, yet they remain prone to hallucinations and erroneous outputs. Uncertainty Quantification (UQ) is crucial for assessing system reliability, but existing methods are largely designed for single-agent settings; how to exploit MAS structure for UQ while keeping inference cost-neutral remains underexplored. We propose UMAS, a parameter-free, cost-neutral, system-level UQ framework that estimates the credibility of inter-agent influence from each generation’s intrinsic confidence. UMAS uses these credibility signals to modulate how uncertainty is updated at each node, and then aggregates node-level uncertainty into a system-level estimate. Extensive evaluations on Debate and DyLAN across diverse benchmarks show that UMAS consistently outperforms state-of-the-art baselines by an average of 10.54 AUROC points. Beyond uncertainty estimation, UMAS enables hallucination detection and uncertainty-aware answer selection, improving MAS accuracy by up to 12.6 points on specific tasks while enhancing reliability. Code is available at this https URL.
[MA-30] When Agents Collude: Weak-to-Strong On-Policy Distillation for Mitigating Strategic Deception in Multi-Agent Systems
【速读】：该论文旨在解决多智能体系统在复杂任务协作中可能引发的安全风险，特别是智能体间通过合谋（collusion）实施偏离对齐目标的行为，并在报告中掩盖其真实动机，形成持续性欺骗的问题。其核心挑战在于，随着智能体协作能力的增强，系统性策略性欺骗（如共享虚假理由、协调掩饰行为）显著上升，尤其在高压力情境下更为严重。解决方案的关键在于提出一种名为PACT的弱到强的在线策略蒸馏框架，通过逆KL散度监督，使更强的“学生”模型在生成文本时匹配一个小型、冻结的诚实对齐“教师”模型的下一个词分布，从而实现对策略性欺骗的抑制。实验表明，在存在生存压力且包含伦理指令的条件下，PACT可将多智能体持续欺骗率从80%降至3%，有效缓解由协作带来的战略欺骗问题。
链接: https://arxiv.org/abs/2610.33011
作者: Eric Onyame,Runtao Zhou,Kowshik Thopalli,Bhavya Kailkhura,Chirag Agarwal
机构: University of Virginia(弗吉尼亚大学); Lawrence Livermore National Laboratory(劳伦斯利弗莫尔国家实验室)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Our code is available at this https URL
Abstract:While recent advances in autonomous agents enable them to coordinate on increasingly complex tasks, prior work shows that this growing capacity for coordination can create safety risks, including collusion, in which agents jointly pursue misaligned actions while concealing the true basis of those actions from oversight. We study this risk in a simulated insider-trading environment, comparing single agents with sequential and decentralized teams across seven language models from three families. By varying agents’ system-level objectives and pressure conditions, we examine when coordinated agents act on material nonpublic information, conceal the basis of a misaligned trade in a report, and maintain that deception under follow-up questioning. Compared to single-agent decision-making, multi-agent interaction raises misaligned-action rates from 35.4% to 68.2% and persistent-deception rates from 31.0% to 62.2%, both measured over all runs. These effects are largest under existential pressure, where teams sustain deception through follow-up questioning in 87% of all runs. Qualitative analysis reveals shared cover stories and coordinated rationales, suggesting group-level strategic deception rather than isolated reporting errors. To mitigate this behavior, we introduce PACT, a weak-to-strong on-policy distillation framework for mitigating collusive deception. Using generic honesty prompts, PACT trains a stronger student on its own sampled continuations, matching its next-token distribution to that of a smaller, frozen honesty-aligned teacher via reverse-KL supervision. Under existential pressure with ethical instructions, PACT reduces multi-agent persistent deception from 80% to 3%. Our results show that multi-agent coordination amplifies strategic deception and that on-policy honesty alignment can substantially mitigate these failures.
[MA-31] LLM -Augmented Agent -Based Simulation of Cyber Social Agents
【速读】：该论文旨在解决现有研究中对社交媒体自动化行为（social media automation）过度简化为“恶意机器人”这一单一维度的问题，揭示自动化角色在在线话语塑造中所扮演的复杂且多元的角色。传统方法往往将自动化行为视为需要被检测与清除的威胁，却忽视了其在信息传播与社会影响中的建设性与破坏性双重作用。为此，本文提出“网络社会代理”（Cyber Social Agents, CSA）的概念框架，通过构建基于实证数据的原型分类体系（taxonomy of archetypes），系统归纳了自动化主体在操作策略与修辞风格上的多样性。其解决方案的关键在于将这一基于实证的CSA分类体系嵌入由大语言模型（LLM）驱动的社会仿真系统中，使代理具备可参数化的、符合真实行为模式的交互能力，并在动态网络环境中实现协同与沟通。该方法有效弥合了规则驱动型代理模型与纯生成式模拟之间的鸿沟，实现了以数据为依据、具有社会异质性的虚拟社会环境构建，从而更真实地探究代理多样性如何影响信息扩散与社会影响力形成机制。
链接: https://arxiv.org/abs/2610.32988
作者: Lynnette Hui Xian Ng,Kathleen M. Carley
机构: Carnegie Mellon University (卡内基梅隆大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Social and Information Networks (cs.SI)
备注:
Abstract:Social media bots are often portrayed as malicious automations to be detected and removed, but automation on social media platforms encompasses a wide spectrum of roles, behaviors, and rhetorical strategies that can shape online discourse in both constructive and destructive ways. This work conceptualizes these automated actors as \textbfCyber Social Agents through a taxonomy of archetypes characterized by their operational tactics and rhetorical style. We integrate this taxonomy into an LLM-driven social simulation, where agents communicate and coordinate in a network environment. Using empirically grounded CSA personas, we simulate heterogeneous online societies to study how agent diversity shapes information diffusion and social influence. Our approach bridges the gap between rule-based agent models and LLM-powered simulations with generated agents that are parameterized from empirical data.
[MA-32] IAgent : From Natural Language to Grounded Agent ic Workflows for Computational Pathology
【速读】：该论文旨在解决计算病理学（Computational Pathology, CPath）中多步骤计算工作流构建对编程能力依赖性强、难以被缺乏编码经验的病理科医生等领域专家使用的难题。现有基于大语言模型（Large Language Models, LLMs）的自然语言接口虽能简化流程设计，但因生成幻觉操作和错误的工作流规范而限制了其在科学场景中的可靠性。为此，本文提出Tissue Image Analysis Agent（TIAgent），一种自动化的CPath工作流生成框架：给定一个描述复杂分析任务的自然语言提示及输入数据集，TIAgent能够自动生成可验证的计算图，该计算图由已有的CPath操作构成，支持可视化审查与执行。其核心创新在于将自然语言指令转化为结构化、可执行且可验证的计算图，从而实现自动化生物标志物发现与组织分析等复杂任务的端到端工作流构建。通过在90个单元任务（涵盖18类CPath操作）上系统评估14种主流LLM，采用单位工作流成功率（Unit Workflow Success, UWS）作为量化指标，结果显示Kimi-K2.7-Code表现最优，UWS达0.967（95%置信区间：0.907–0.989）。进一步分析表明，选择正确的计算操作与准确配置参数是LLM生成工作流时面临的两个独立挑战。TIAgent不仅降低了复杂CPath工作流构建的技术门槛，还为评估大语言模型驱动的科学工作流生成提供了可复现的基准体系。
链接: https://arxiv.org/abs/2610.32993
作者: Aakash Madhav Rao,Mark Eastwood,Danielle Belgrave,Fayyaz Minhas
机构: University of Warwick (华威大学); GSK(葛兰素史克)
类目: Quantitative Methods (q-bio.QM); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:Computational pathology (CPath) analyses increasingly depend on multi-step computational workflows that require substantial programming expertise, limiting their accessibility to pathologists and other domain experts without extensive coding experience. Although large language models (LLMs) provide a promising natural-language interface for constructing such analyses, hallucinated operations and incorrect workflow specifications limit their reliability in scientific settings. We introduce Tissue Image Analysis Agent (TIAgent), an automated CPath workflow generation framework that, given a natural-language prompt specifying a complex analytical task together with an input dataset, automatically constructs a verifiable computational graph employing existing CPath operations. The resulting graph can be inspected and executed, enabling natural-language construction of complex CPath workflows for applications including automated biomarker discovery and tissue analysis. We demonstrate end-to-end workflow generation and execution and systematically benchmark 14 backbone LLMs across 90 unit tasks spanning 18 CPath operation types. The ability of each LLM to generate a complete and correct unit workflow is quantified using Unit Workflow Success (UWS). Performance varied substantially across backbone models, with Kimi-K2.7-Code achieving the highest UWS of 0.967 (95% CI: 0.907-0.989). Analysis of node and parameter correctness further revealed that selecting the appropriate computational operation and correctly configuring it represent distinct challenges for LLM-based workflow generation. TIAgent provides a foundation for making complex CPath workflow construction more accessible while enabling systematic evaluation of LLM-driven scientific workflow generation.
自然语言处理
[NLP-0] Center-of-Mass Interpretation of Evolution Strategies
【速读】： 该论文旨在解决进化策略（Evolution Strategies, ES）中更新机制的直观理解问题，即如何从几何角度准确解释ES的参数更新过程。传统方法将ES更新视为平滑梯度的估计，但本文提出一种全新的几何解释：将每个扰动后的模型视为一个具有质量的粒子，其奖励值即为粒子的质量。在评分前，所有粒子质量相等，系统质心位于群体中心；评分后，系统平衡于奖励加权的质心。研究证明，经过均值减除的ES更新方向恰好等于这一质心位移，且被一个依赖于平均奖励的标量因子所缩放，该结论对任意有限规模种群均成立。该几何视角要求奖励非负，可通过常数平移实现而不影响更新方向。在大种群极限或采用反向采样（antithetic sampling）条件下，群体质心与当前模型位置重合，此时ES更新可简化为：模型从当前位置朝其自身种群的奖励加权质心移动。同时，减去平均奖励的作用被解释为改变位移测量的参考点——从模型本身变为采样质心。本文的核心贡献在于提供了一种精确适用于任意有限种群的直观、物理化的几何视角，深化了对ES更新机制的本质理解。
链接: https://arxiv.org/abs/2610.35865
作者: Akshat Gupta
机构: UC Berkeley(加州大学伯克利分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Evolution strategies (ES) update a model by sampling a population of perturbed copies, scoring each one, and moving in the reward-weighted direction of the perturbations. This update is usually presented as an estimate of a smoothed gradient. We present a geometrical interpretation of this update instead. Treat each perturbed model as a particle and its reward as the particle’s mass. Before scoring, every particle has the same mass and the population balances at its centroid. Once rewards are assigned, it balances at its reward-weighted center of mass instead. We show that the mean-subtracted ES update is exactly this displacement, scaled by a scalar factor that depends on the mean reward, for any finite population. The physical reading needs nonnegative rewards, which can always be arranged by adding a constant to every reward without changing the update. In the limit of a large population or under antithetic sampling, the centroid sits on the model itself, so ES has an even simpler geometric interpretation: from its current location, the model moves towards the reward-weighted center of mass of its own population. The same picture explains what subtracting the mean reward does: it changes the point from which the displacement is measured, from the model to the sampled centroid. Our contribution is an intuitive, geometric way to visualize an ES update, exact for any finite population.
[NLP-1] Preserve to Expand: Source-Aware Adaptation and Model Merging for Multilingual LLM -Based ASR
【速读】： 该论文旨在解决多语言自动语音识别（ASR）系统在扩展至新语言时面临的挑战，即在不回放已有源语言训练数据的前提下，如何有效保持原有语言的性能。其核心问题在于：在连接器驱动的语音大语言模型（Speech LLM）架构中，如何在微调可训练连接器以适应新语言的同时，避免对已支持语言造成严重遗忘。解决方案的关键在于设计合理的参数更新构造与合并策略。研究发现，在单目标扩展场景下，仅对目标语言进行微调或采用标准TIES合并会导致严重的源语言遗忘；而基于SSU-Wanda+TIES的更新策略能够在目标语言学习与源语言保留之间取得最佳平衡。在多目标扩展场景中，更新合并策略能实现稳定的源语言保留，而递归扩展虽可在理想顺序下提升目标语言适应性，但对语言添加顺序高度敏感。因此，该研究揭示了多语言ASR扩展需在目标语言获取、源语言保留以及各目标语言更新间的相互作用之间实现精细权衡。
链接: https://arxiv.org/abs/2610.35846
作者: Yuchen Zhang,Haralambos Mouratidis,Ravi Shekhar
机构: Institute for Analytics and Data Science, University of Essex(埃塞克斯大学数据分析与科学研究所); School of Computer Science and Electronic Engineering, University of Essex(埃塞克斯大学计算机科学与电子工程学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Expanding multilingual ASR systems to new languages is challenging when adaptation must preserve previously supported languages without replaying source-language training data. We study this problem in connector-based Speech LLMs, where a trainable connector maps speech representations into a frozen large language model. We evaluate target-update construction and merging strategies for adapting the connector while preserving source-language performance. In single-target expansion, target-only fine-tuning and standard TIES merging learn new languages but cause severe source-language forgetting, while SSU-Wanda+TIES gives the best source-preserving trade-off. In multi-target expansion, update merging provides stable source retention, whereas recursive expansion can improve target adaptation under favorable orders but is highly order-sensitive. These results show that multilingual ASR expansion requires balancing target acquisition, source retention, and interactions among target-language updates.
[NLP-2] Understanding Jev-Style System One Models: Does Structure Reduce Selection Bias?
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）中持续存在的选择偏差（selection bias）问题，尤其关注在多选项任务中，基于Jev风格决策结构的模型是否能够有效降低此类偏差。其核心问题是：Jev式决策结构是否真正减少了选择偏差？为此，作者提出一种简单但有效的测试方法——在保持问题与候选答案内容不变的前提下，对选项顺序进行随机洗牌，并对比不同呈现顺序下的模型表现。研究覆盖了两个Jev版本、十二个Jev复现模型及六个生成式LLM，在来自两个基准数据集的1,700个问题上进行实验。结果显示，尽管Jev模型在原始顺序下的准确率波动小于2个百分点，但在洗牌后其累积失败集扩大了27%–29%，且初始正确的判断在后续呈现中仍频繁失效；这一现象显著优于固定顺序重复下的表现，即便对于高度可重复的Jev复现模型亦然。此外，Jev并未展现出相对于生成式基线模型在维持正确选择方面的普遍优势。聚合策略（如投票）仅提供有限保护，即使所有模型一致输出，错误答案仍频繁出现。研究揭示了“可重复决策”与“跨等效呈现保持正确性”之间存在关键差距，表明仅依赖稳定准确率无法证明模型具备抵抗选择偏差的能力。
链接: https://arxiv.org/abs/2610.35826
作者: Mingyang Song,Mao Zheng,Zhiyuan Ma,Xinpeng Zhou
机构: Tencent(腾讯)
类目: Computation and Language (cs.CL)
备注:
Abstract:Selection bias in large language models (LLMs) can persist despite instructions to ignore presentation order or prompts that elicit reasoning. Recent efforts to reproduce Jev-style System One behavior have produced a growing family of models that predict probabilities directly over supplied candidates. Yet systematic comparisons of selection bias across Jev versions, Jev reproductions, and generative LLMs remain limited in multi-option tasks. This leaves a central question: does Jev-style decision structure reduce selection bias? We address this question with a simple test that shuffles options while keeping question and candidate content fixed. Across two Jev versions, twelve Jev reproductions, and six generative LLMs, we trace individual answers through eight presentations and compare them with fixed-order repetition. On our primary collection of 1,700 questions from two benchmarks, Jev’s accuracy varies by less than two percentage points, yet its cumulative failure set expands by 27–29% relative to the original order, with first failures continuing into the final presentation. This loss of initial successes is substantially larger than under fixed-order repetition, even for highly repeatable Jev reproductions. Nor does Jev show a general advantage over the LLM baselines in preserving correct choices. Aggregation offers limited protection: voting yields modest gains for Jev, while even unanimous answers are frequently wrong. These results expose a gap between repeatable decisions and decisions that remain correct across equivalent presentations, showing why stable accuracy alone cannot establish resistance to selection bias.
[NLP-3] Learning When to Commit: Schedule Distillation for Fast Diffusion Language Models
【速读】： 该论文旨在解决扩散型大语言模型（diffusion large language models, dLLMs）在生成文本时因需经历大量解码轮次而导致效率低下的问题。尽管dLLMs支持并行生成多个标记（token），但其生成速度受限于解码过程中对高置信度标记的盲目依赖，而忽视了标记揭示顺序对整体不确定性降低的潜在影响。解决方案的关键在于提出一种基于调度蒸馏（schedule distillation）的新框架dSpark，通过构建优先揭示那些能显著提升其他位置可预测性的标记的去掩码（unmasking）策略，提前减少解码过程中的不确定性。该方法将优化后的揭示顺序以知识蒸馏方式融入模型，结合KL散度正则化以保障生成质量，在不牺牲性能的前提下实现更快的并行采样。实验表明，在数学推理与代码生成任务中，dSpark在LLaDA-8B-Instruct和Dream-7B-Instruct上分别实现了10.12×和5.12×的显著加速，同时保持了优异的任务表现。
链接: https://arxiv.org/abs/2610.35804
作者: Aidar Myrzakhan,Zhiqiang Shen
机构: VILA Lab, MBZUAI(MBZUAI的VILA实验室)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Code is available at this https URL
Abstract:Diffusion large language models (dLLMs) can generate multiple tokens in parallel, but often require many decoding rounds to complete a response. Efficient generation depends not only on which tokens are easy to predict, but also on how committing to them helps predict the remaining positions. A highly predictable token may provide little useful context, while revealing a less predictable token can help several other positions become predictable sooner. We propose dSpark, a new schedule distillation-based framework for learning when to commit tokens for faster diffusion language generation. To achieve this, we construct unmasking schedules that prioritize tokens whose reveal makes other positions easier to predict, reducing uncertainty earlier in decoding. We then distill these schedules into the model to enable faster parallel sampling, with a KL-divergence regularizer to help preserve generation quality. We extensively evaluate our approach across mathematical reasoning and code generation tasks. On GSM8K, our method achieves remarkable 10.12 \times and 5.12 \times speedups over the various baselines on LLaDA-8B-Instruct and Dream-7B-Instruct, respectively, while maintaining superior task performance. Our code is available at this https URL.
[NLP-4] Quantifying Claim Smoothing by LLM Research Agents
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在深度研究任务中因过度整合多源信息而产生的“平滑化”（smoothing）问题。具体而言，当大型语言模型（LLM）对大规模文档集合进行分析并生成长篇研究报告时，尽管输出内容呈现高度自信的综合叙述，却往往未能准确反映原始文献中的多样性与潜在冲突，导致结论的确定性超过其引用证据所能支持的程度。其解决方案的关键在于提出一种名为SmoothScore的主张级评分流水线：该方法首先从模型生成的报告中提取关键结论性主张，随后检索与每个主张相关的证据，并基于高层维度对证据进行聚类；最终，通过评估主张与其证据之间的匹配程度，量化模型在信息整合过程中所表现出的平滑化程度。实验结果表明，在7个系统和4个数据集上，模型均持续生成了超出证据支持范围的更强主张，凸显了该问题的普遍性与严重性。
链接: https://arxiv.org/abs/2610.35793
作者: Tiasa Singha Roy,Manya Wadhwa,Greg Durrett
机构: New York University (纽约大学)
类目: Computation and Language (cs.CL)
备注: Code and data present at: this https URL
Abstract:Deep research, or any use of large language models (LLMs) to read a large document collection and produce insights, requires synthesizing information from different sources. In this work, we find that LLMs produce long-form research reports that read as confident syntheses but which fail to accurately reflect the underlying documents. We call this failure smoothing. In collapsing diverse and potentially conflicting sources into a single narrative, the model discards the evidential “jaggedness” of its corpus and asserts conclusions more decisive than the citations can support. We introduce a claim level scoring pipeline, SmoothScore, to quantify this effect. Our pipeline extracts key conclusive claims from model generated reports, then finds evidence relevant to each claim and clusters that evidence along high level aspects. Finally, each claim is scored with respect to the evidence to quantify the degree of smoothing done by the agent. Across 7 systems and 4 datasets, we find that agents consistently produce stronger claims than warranted.
[NLP-5] Atomic Security Concerns: Towards a Foundational Language for Security Knowledge Representation
【速读】： 该论文旨在解决网络安全治理、风险管控、合规性及安全架构中普遍存在的概念异构问题，即各类安全属性、域、控制措施、风险、实施手段、能力与合规要求等概念常以多样化的抽象形式和术语体系表示，导致概念间关系难以一致表达，且相关推理高度依赖专家经验。其解决方案的关键在于提出“安全关切（Security Concerns）”与“原子级安全关切（Atomic Security Concerns, ASCs）”的概念，并基于化学中“原子”的类比，提出网络安全知识可由一组有限的、基础性的关切构成，而非独立的抽象实体。通过将复杂安全概念视为原子级安全关切的组合，ASCs 成为跨领域、跨粒度、跨性质的安全概念统一抽象层，从而为构建一种可显式表达概念间关系的网络安全语言提供基础词汇。该研究的核心贡献是概念层面的：提出一个用于探索网络安全知识表征、推理与未来形式化基础的框架与研究假说。
链接: https://arxiv.org/abs/2610.35782
作者: Hanane Houmani
机构: 未知
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注:
Abstract:Cybersecurity governance, risk management, compliance, and security architecture rely on numerous concepts, including security properties, domains, controls, risks, implementation measures, capabilities, and compliance requirements. Although these concepts are widely used, they are often represented through heterogeneous abstractions and vocabularies, making relationships among them difficult to express consistently and leaving much of the associated reasoning dependent on expert knowledge. This paper introduces the concepts of Security Concerns and Atomic Security Concerns (ASCs) and explores the hypothesis that cybersecurity knowledge may be represented through a finite set of foundational concerns. Inspired by the role of atoms in chemistry, the proposed approach views cybersecurity concepts as compositions of more fundamental concerns rather than as independent abstractions. Hence, Atomic Security Concerns could be viewed as a common abstraction layer for the representation of cybersecurity concepts, independently of their domain, nature, or level of granularity. Building upon this perspective, the paper explores how Atomic Security Concerns may provide the vocabulary of a cybersecurity language and how cybersecurity concepts may be represented as concern sets supporting explicit relationships among concepts. The contribution of this work is intentionally conceptual: it introduces a framework and research hypothesis for investigating the foundations of a common language for cybersecurity knowledge representation, reasoning, and future formalization.
[NLP-6] SEMA-Auto: Sighted and Efficient Multi-Turn Attack with Autonomy
【速读】： 该论文旨在解决现有越狱攻击方法中存在的两大局限：静态规划型攻击（static planners）虽计算高效但无法根据目标模型的回复动态调整策略，导致攻击效率低下且泛化能力差；而交互生成型攻击（interactive generators）虽能实时响应目标反馈，却依赖闭环评估器提供成功信号或提示，带来高昂的计算成本并可能夸大性能表现。为此，论文提出SEMA-Auto，一种具备自主感知能力与高效性的多轮越狱攻击框架，其核心创新在于通过双阶段训练机制实现行为稳定与能力增强：首先采用合成自举微调（synthetic bootstrap tuning）在基础模型中注入攻击目标意识与合规决策能力；随后通过决策监督的强化学习（decision-supervised reinforcement learning），分别优化越狱奖励与每轮决策对绝对目标的符合度，从而在无需外部反馈的情况下实现自主终止。实验表明，SEMA-Auto在多个数据集、目标模型及评测标准下均达到当前最优攻击成功率（ASR），且以极低的平均轮次完成攻击——例如在AdvBench上对DeepSeek-V4-Flash实现76.0%的ASR@1仅需3.0轮，显著优于固定7轮的最优静态基线（54.8%）和需3.7轮的最优交互基线（43.8%）；在对抗经过安全加固的前沿模型（如GPT-oss-120B、GPT-5.4等）时，于HarmBench上平均达成59.6%的ASR@1，较前人SOTA提升13.8%。该方法在保持高攻击效能的同时大幅降低资源消耗，实现了“高效性”与“有效性”的统一。
链接: https://arxiv.org/abs/2610.35743
作者: Mingqian Feng,Xiaodong Liu,Weiwei Yang,Chenliang Xu,Jialin Song,Qiyao Ma,Chengquan Guo,Carolyn Buractaon,Christopher White,Jianfeng Gao
机构: University of Rochester; Microsoft Research; Microsoft Azure
类目: Computation and Language (cs.CL)
备注:
Abstract:Existing jailbreak attackers fall into two classes: static planners fix their single- or multi-turn adversarial plans in advance, remaining simple but losing sight of the victim’s replies, wasting turns and limiting transferability; interactive generators read each victim reply to generate the next-turn attack but require in-loop evaluators for success signals or hints, incurring substantial costs and inflated performance. Bridging this gap, we propose SEMA-Auto, a sighted and efficient multi-turn attacker with autonomy, interacting with the victim model and terminating the session by itself upon success. To stabilize its behaviors and boost its capabilities, SEMA-Auto is trained in two stages. Synthetic bootstrap tuning installs the notion of attack targets and conforming decisions by fine-tuning the base model with bootstrapped trajectories. Decision-supervised reinforcement learning then optimizes two objectives jointly yet separately, reinforcing the attack with higher jailbreak reward while supervising the decision at each turn against an absolute target. Our method achieves state-of-the-art (SOTA) attack success rates (ASR) across multiple datasets, victims, and judges, outperforming all static and interactive baselines, within almost the fewest turns, demonstrating both effectiveness and efficiency. For example, SEMA-Auto performs a 76.0% ASR@1 within an average of 3.0 turns against DeepSeek-V4-Flash on AdvBench, while the best static baseline is 54.8% with fixed 7 turns and the best interactive baseline is 43.8% in 3.7 turns. Against safety-hardened frontier models, GPT-oss-120B, GPT-5.4, Claude-Haiku-4.5, and GPT-6-Astra, our counter-trained attacker achieves an average 59.6% ASR@1 on HarmBench, 13.8% over prior SOTA. Our code is available at: this https URL
[NLP-7] Can JEV Support Clinical Trial Criteria Assessment?
【速读】： 该论文旨在解决临床试验预筛选过程中，如何准确区分患者记录中“未满足的纳入条件”与“证据不足”这一核心挑战。其关键解决方案在于提出一种非生成式决策模型JEV，该模型通过返回原始选项的概率分布，实现对临床条件的精准判断。在严格零数据保留（ZDR）政策下，JEV在多个数据集上表现出优异性能，尤其在MIMIC-III数据集上显著优于基于提示工程的Qwen Flash模型（宏平均F1得分差值Δ = +0.234），主要体现在对“证据不足”类条件更高的召回率（56.3% vs. 9.9%）。此外，JEV在推理延迟（1.18–1.40秒）和单位条件成本（降低53%–74%）方面均具有明显优势，表明其可作为高效、隐私保护的初始评估组件，用于提供标准级判断并识别潜在信息缺口。然而，其在解读临床证据时的误差提示，仍需人工核查以确保入组决策的可靠性。
链接: https://arxiv.org/abs/2610.35721
作者: Mengxian He,Ziyue Gao,Chun Li,Shunyi Zhang,Hao Wei,Minqing Zhang,Wu Yuan
机构: The Chinese University of Hong Kong(香港中文大学); Capital Medical University(首都医科大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Clinical trial prescreening requires distinguishing unmet conditions from insufficient evidence in patient records. Given patient records, criterion questions, and predefined answer options, we retrospectively compared JEV, a non-generative decision model returning native option probabilities, with schema-constrained Qwen Flash across 2,048 test patient-condition pairs from TrialGPT and question-answer datasets derived from MIMIC-III and MIMIC-IV, all evaluated under strict Zero Data Retention (ZDR) policies to ensure clinical data privacy. Both models assessed conditions individually or in batches sharing the same patient context, with performance measured against reference labels. In batch mode, JEV and Qwen achieved similar macro-F1 scores on TrialGPT (0.652 vs. 0.660) and MIMIC-IV (0.807 vs. 0.814), whereas JEV substantially outperformed Qwen on MIMIC-III (0.768 vs. 0.534; \Delta = +0.234, 95% CI: 0.187-0.281). This advantage largely reflected higher recall for conditions labeled as having insufficient information (56.3% vs. 9.9%). Across all three datasets, JEV also had lower observed median job latency (1.18-1.40 vs. 2.80-8.90 s) and 53%-74% lower estimated cost per condition. These findings position JEV as an initial assessment component that provides criterion-level judgments and flags potential information gaps for review. JEV’s observed errors in interpreting clinical evidence underscore the need to verify its outputs against patient records before they inform eligibility decisions.
[NLP-8] Quantitative Generalization Bounds for Multi-Layer Transformers: A Few Long Strings can Beat Many Short Ones
【速读】： 该论文旨在解决生成式语言模型在训练数据之外的泛化能力问题，特别是针对基于Transformer的模型在面对更长、更复杂的问题实例时所表现出的惊人泛化能力，其理论理解仍不充分。传统机器学习模型通常仅关注分布内统计泛化（in-distribution statistical generalisation），即模型从训练样本推广到同分布新样本的能力；而由于语言数据具有序列特性，还需考虑长度泛化（length generalisation），即模型能否将训练中使用特定长度输入的学习成果外推至更长序列。然而，这两个维度长期被分别研究，缺乏对二者交互机制的系统分析。本文首次对固定精度下的Transformer模型进行了双轴联合分析，提出了基于模型参数规模、量化空间和词表大小的泛化保证，并揭示了样本数量与训练序列长度之间的权衡关系。研究表明，在同等条件下，增加训练序列长度比增加样本量具有指数级更高的效率。实验结果验证了理论预测的有效性。
链接: https://arxiv.org/abs/2610.35715
作者: Marius Henry,Michael Rizvi-Martel,Michael Hahn
机构: LaCoCo, Saarland University, Saarland Informatics Campus(莱科科，萨尔兰大学，萨尔兰信息学校园); Mila Université de Montréal(蒙特利尔学习算法研究所，蒙特利尔大学); Zuse School ELIZA(祖泽学院埃尔伊扎)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Transformer-based language models exhibit an astonishing capacity to generalize to longer and more difficult problem instances than those seen in training. However, theoretical understanding of this phenomenon still lags behind. Unlike most machine learning models, the generalisation abilities of transformers inherently require considering two axes. On one hand, one must examine in-distribution statistical generalisation, which concerns the ability of a model to broaden a function from training samples to novel samples of the same distribution. On the other hand, given the sequential nature of language data, one must also consider length generalisation, which reflects whether a model trained on inputs of a given length can extrapolate to longer this http URL two notions have mostly been studied separately, leaving open how they interact. In this work, we give a joint treatment of both axes for fixed-precision transformers. We give generalisation guarantees in terms of the parameter sizes of the involved models, as well as the quantization space and the alphabet size that supply the sequences. Finally, we illustrate how the requirements in the sample count and training length trade off. In particular, we find that increasing training length can be exponentially more efficient than increasing the sample size. Experiments support predictions of the theory.
[NLP-9] From Self-Confidence to Statistical Accuracy: Scaling Human-in-the-Loop Evaluation with Direct and LLM -Assisted Auditing
【速读】： 该论文旨在解决大语言模型（LLM）及多模态大语言模型（MLLM）在评估过程中自报告置信度（self-confidence）普遍存在的校准偏差问题，同时探索如何在保证统计有效性的同时提升评估效率。其核心挑战在于：尽管模型的置信度常不准确，但若能合理利用这一信息，仍可实现高效评估；而基于大模型作为评判者（LLM-as-a-judge）的自动化评估虽具可扩展性，其实际效益受限于评判者的校准质量与推理开销。本文提出一个统一的统计框架，涵盖直接人工审计与判官辅助评估两种路径，并针对不同采样策略进行分析。关键解决方案包括：在直接审计中采用置信度分层采样（confidence-stratified sampling），无需依赖校准的置信度即可从部分验证输出中恢复总体准确率，并量化有限样本下的不确定性；在判官辅助评估中，将评判者建模为具有敏感性与假阳性率的二元分类器，构建误差校正估计器并纳入校准不确定性。通过人类标签效率边界与总成本效率边界的对比分析，实验结果表明，在高度不平衡的置信度分布下，按比例分配样本相较于等量分配显著降低估计误差；而仅有少数目标-验证器组合满足理想的人类标签效率条件，且自动化推理成本可能抵消潜在节省。研究揭示了在何种统计与经济条件下，基于置信度引导的审计与自动化验证才具备实际价值。
链接: https://arxiv.org/abs/2610.35698
作者: Harine Choi(1),Chenchen Kuai(2),Zihao Li(3),Jiwan Jiang(2),Yunlong Zhang(2),Zhiwen Fan(4),Zhengzhong Tu(5),Yang Zhou(2) ((1) Texas AM Transportation Institute, Texas AM University System, (2) Zachry Department of Civil and Environmental Engineering, Texas AM University, (3) Department of Civil, Construction and Environmental Engineering, Marquette University, (4) Department of Electrical and Computer Engineering, Texas AM University, (5) Department of Computer Science and Engineering, Texas AM University)
机构: 未知
类目: Computation and Language (cs.CL)
备注: 21 pages, 8 figures. Reproduction artifacts available at this https URL
Abstract:For large language models (LLMs) and multimodal large language models (MLLMs), model-reported self-confidence is often miscalibrated, yet it may still be useful for efficient evaluation. At the same time, LLM-as-a-judge offers a scalable alternative to direct human auditing, but its practical benefit depends on calibration quality and inference cost. This paper develops a unified statistical framework for both evaluation routes under alternative sampling strategies. For direct auditing, confidence-stratified sampling is used to recover population accuracy from partially verified outputs without requiring calibrated confidence. Finite-sample uncertainty is characterized, and the analysis shows how highly uneven confidence strata affect audit efficiency and label allocation. For judge-assisted evaluation, the judge is modeled through sensitivity and false-positive rate, yielding an error-corrected estimator with calibration uncertainty. The two routes are then compared through human-label and total-cost efficiency boundaries. Experiments on MMLU-Pro and MMMU-Pro show that miscalibrated self-confidence can still support statistically valid stratified auditing. Under highly imbalanced confidence distributions, proportional allocation substantially reduces estimation error relative to equal-per-stratum allocation. For judge-assisted evaluation, only a small subset of target–verifier pairs satisfies the idealized human-label efficiency condition, and automated inference costs can further limit potential savings. These findings clarify the statistical and economic conditions under which confidence-guided auditing and automated verification are useful. Comments: 21 pages, 8 figures. Reproduction artifacts available at this https URL Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.35698 [cs.CL] (or arXiv:2610.35698v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.35698 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-10] MiRL: Error-Targeted Reinforcement Learning for Correcting Terminal Agent Misbehaviors
【速读】： 该论文旨在解决生成式终端代理（terminal agent）在执行长时程自然语言指令任务时，因缺乏有效纠错机制而导致的性能瓶颈问题。现有错误靶向学习方法主要针对单轮任务设计，且对代理失败的分析多用于评估与调试，无法直接用于模型训练。其核心解决方案是提出一种基于异常行为驱动的强化学习框架——Misbehavior-driven RL（MiRL），其关键在于：首先通过系统性地分析策略在多环境中的重复性错误，识别出可泛化的“异常行为”（misbehavior），即一类在不同上下文中反复出现的错误操作；随后，在强化学习过程中，从这些异常行为发生的具体步骤启动轨迹（rollout），仅当轨迹成功规避该类错误并通过测试时才给予奖励。该方法使模型在未见过的任务前缀上，平均减少43%至57%的既定错误行为。实验表明，基于Qwen3.5-4B的MiRL在终端基准测试中表现优于TMax-4B，同时仅使用其35%的采样轨迹，验证了错误靶向学习在真实、长时程终端任务中的有效性与高效性。
链接: https://arxiv.org/abs/2610.35697
作者: Runchu Tian,Zaid Khan,Zun Wang,Justin Chih-Yao Chen,Mohit Bansal
机构: University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 20 pages, 14 figures, 5 tables. Project page: this https URL
Abstract:Error-targeted learning gives a legible way to improve a model: identify specific weaknesses of the model, train the model to fix these weaknesses, and check whether the weaknesses are fixed. We study how to apply error-targeted learning to terminal agents, which solve long-horizon tasks specified in natural language by interacting with a command shell. Existing error-targeted methods are designed for single-turn tasks, and existing analyses of agent failures are used to evaluate and debug agents rather than to train them. We propose Misbehavior-driven RL (MiRL), which first profiles the errors of a policy to find generalizable misbehaviors: kinds of wrong steps that recur across many environments. MiRL then trains the policy with RL, starting rollouts at the steps where these misbehaviors occur and rewarding a rollout only if it passes the tests and avoids the misbehavior. On Qwen3.5-4B, eight misbehaviors account for 52.1% of wrong steps. After training with MiRL, the policy repeats the measured misbehaviors 43% to 57% less often on average on unseen prefixes. MiRL outperforms TMax-4B on Terminal-Bench Lite (51.8% vs. 47.4%) and matches it on Terminal-Bench 2.1 (21.3% vs. 20.6%), while using 35% of its rollouts. These results show that error-targeted learning can be applied to realistic, long-horizon terminal tasks.
[NLP-11] Not One Belief: Answer Payloads Persistence and Dissociated Confidence Signals in Conversational Question Answering
【速读】： 该论文旨在解决对话式问答系统在用户提出质疑或反馈时，如何有效识别并修正错误回答的问题，特别是如何在用户未提供具体答案的情况下判断应信任哪个回答。其核心挑战在于：当用户通过“重问”（re-ask）、“同意”（agree）或“质疑”（challenge）等语用行为表达反馈时，系统能否准确感知并调整输出。研究的关键发现是，仅依赖语用行为（speech act）不足以驱动答案更新，而必须结合具体的回答内容（payload），尤其是在用户主动提供答案字符串时，系统才可能实现显著纠错。四个置信度信号——口头表达的置信度、归一化似然性、冻结隐藏状态探测器（frozen hidden-state probe）和P(True)——表现出显著分歧，其中冻结探测器在Qwen2.5-7B模型中随答案持续性上升而非真实性，表明其存在偏差。实验显示，在81个持续错误中，仅提供答案字符串的修复率可达75%-95%，而仅通过重问、重启会话或放弃回答等方式几乎无效。进一步分析表明，基于BM25检索到包含正确答案的句子可修复67%-76%的持久错误，而无此类信息的修复率不足12%。最终提出的“包含门控”（contain-gate）机制虽在端到端训练下表现不佳，但一种不依赖黄金答案的可部署门控仍能有效提升修复能力，揭示出基于证据匹配而非依赖先验真值的修复路径具有可行性。
链接: https://arxiv.org/abs/2610.35694
作者: Duc-Vinh Tran
机构: Hanoi University of Science and Technology (河内科技大学)
类目: Computation and Language (cs.CL)
备注: 25 pages + supplementary PDF. Preprint
Abstract:Conversational QA systems should fix errors when users push back and signal which answers to trust. We split each follow-up into its speech act (re-ask, agree, or challenge) and its payload (an answer string, correct or not), and measure four confidence signals on the same transcripts: verbalized confidence, length-normalized likelihood, a frozen hidden-state probe, and P(True). Answers change when the user supplies an answer string and almost never under the speech act alone: in every retained full-size cell, the correct answer converts at least four times as many wrong answers as re-asking or agreement. The four signals disagree. On Qwen2.5-7B the frozen probe rises with answer persistence rather than truth; this holds on held-out questions, with Llama-3-8B (cleaned answers) as support, but not on four other models. Among 81 Qwen2.5-7B errors that survive one re-ask, re-asking, a fresh chat, or abstain repair almost none, while a gold answer repairs 75-95%. In two preregistered replications, every repair rate without an answer string stays below 0.15; a larger post-hoc rerun excludes 0.15 except for the empty elicit. With BM25 TriviaQA evidence, sentences that contain the answer repair 67-76% of persistent errors versus at most 12% when they do not. An end-to-end contain-gate cuts harm on TriviaQA, but a deployable gate that does not know the gold answer repairs few errors.
[NLP-12] Adapting LLM -SC to Living Sacred Heritage: Domain-Specific Extraction Semantic Networks and Human Evaluation on Pashupatinath Temple Reviews
【速读】： 该论文旨在解决大型语言模型（Large Language Model, LLM）在处理活态宗教遗产地游客评论时，因通用信息抽取框架忽视仪式行为、神圣空间、文化禁忌及情感复杂性等关键文化语境而造成的表征缺失问题。其解决方案的关键在于对原有的LLM-社会计算（LLM-Social Computing, LLM-SC）框架进行文化情境化改造：针对尼泊尔帕舒帕蒂纳寺（Pashupatinath Temple）的特定语境，构建了一个包含12类实体（在原4类基础上新增8类与神圣遗产相关的类型）、4种关系类型及9个方面-情感维度的定制化标注体系，并通过GPT-4o设计结构化抽取提示（prompt），应用于2,500条经筛选的英文Tripadvisor评论。该方法实现了对文化敏感实体（如“圣火祭祀”“禁入区域”）的高比例识别（占实体记录的64.8%），并通过双标注员人工验证提供组件级评估结果，包括实体精确率41.0%、召回率21.8%、微平均F1为28.4%，方面存在性F1为53.6%，端到端方面+极性预测F1为42.6%。研究强调了其文化扎根性、领域特异性以及可复现性与泛化能力的明确边界，为文化遗产数字化中的生成式AI应用提供了可解释、可验证的实践范式。
链接: https://arxiv.org/abs/2610.35685
作者: Hari Sitaula,Manish Sakhakarmy,Pujan Pant,Sagar Newpane
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) pipelines can turn visitor reviews into structured data, but generic schemes may miss rituals, sacred spaces, cultural restrictions and emotionally mixed experiences at living sacred heritage sites. We adapted the LLM-Social Computing (LLM-SC) framework to Pashupatinath Temple, Nepal, by specifying a 12-type entity taxonomy (four original LLM-SC types plus eight sacred-heritage types), four relation types and nine aspect-sentiment dimensions in a GPT-4o extraction prompt, which was applied to a fixed random sample of 2,500 English-language Tripadvisor reviews from a processed corpus of 3,939. The exports contain 11,286 entity, 6,232 aspect-sentiment and 5,590 relation records, covering 2,485, 2,487 and 2,433 reviews, respectively; the eight added sacred-heritage types account for 64.8% of entity records. Against an adjudicated two-annotator reference for 200 reviews, exact normalized surface-and-type matching gave provisional entity scores of 41.0% precision, 21.8% recall and 28.4% micro-F1. Aspect-presence F1 was 53.6% and end-to-end aspect-plus-polarity F1 was 42.6%. Human adjudication supported 153 of 194 first-listed relation candidates (one per review; 78.9%), a candidate-level rate that does not estimate relation recall or whole-graph precision. The contribution is a culturally grounded, domain-specific adaptation with component-wise human evaluation and explicit reproducibility and generalization limitations.
[NLP-13] Evaluating DeepSeek V4.1 Flash-1017 for loss of control and harmful manipulation risks
【速读】： 该论文旨在解决当前前沿语言模型（Large Language Models, LLMs）在开放权重（open-weight）版本中缺乏系统性安全评估的问题，尤其针对具备强自主代理能力（agentic capabilities）的DeepSeek V4.1 Flash-1017模型。随着模型能力日益增强，独立第三方评估对识别其系统性安全风险至关重要，但现有公开评估多集中于闭源模型，而开放权重模型的安全性评估仍显不足。为此，研究基于欧盟通用人工智能（General-Purpose AI）实践准则，构建了涵盖34个基准测试的独立安全评估体系，包括5个自研评估套件，覆盖评估完整性、失控风险与有害操纵等多个维度。评估结果表明，V4.1 Flash-1017相较于前代V4系列在安全性上取得显著提升，并在多个指标上接近或达到主流闭源模型（如GPT-6 Astra、GPT-5.6 Sol和Gemini 3.8 Flash）水平，但仍存在需持续关注的行为表现。解决方案的关键在于：一是强调在增量更新中嵌入持续安全评估机制，以应对微小更新可能导致的安全退化；二是提出部署阶段需具备全流程可观测性，不仅依赖最终输出，还需对推理轨迹与动作日志进行过程级监控，以有效检测失控行为；三是倡导发展私有基准测试，以规避公共基准带来的评估意识强化与数据污染风险，从而保障前沿模型评估的可信度与有效性。
链接: https://arxiv.org/abs/2610.35671
作者: Yao Huang,Zhiyuan Zeng,Yu Wang,Zhitong Zhu,Dingcheng Duan,Hao Li,Shangzhi Xu,Yifan Wang,Jinzhe Tu,Shukun Xiong,Yihe Zang,Han Yan,Hongchang Yang,Fuyu Gu,Wei Xu,Yinpeng Dong,Xiaojian Li
机构: 未知
类目: Computation and Language (cs.CL)
备注: 87 pages, 44 figures
Abstract:As frontier language models grow increasingly capable and autonomous, independent third-party evaluations are critical for characterizing their systemic safety risks. While closed-source releases are routinely accompanied by comprehensive system cards, open-weight models receive less scrutiny regarding frontier risks. To address this gap, we present an independent safety evaluation of DeepSeek V4.1 Flash-1017, the latest version of V4.1 Flash that departs from V4 with a new architecture, scaled-up training, and strong agentic capabilities. Drawing on the EU General-Purpose AI Code of Practice, we evaluate the model across 34 benchmarks, including 5 in-house suites, spanning evaluation integrity, loss of control, and harmful manipulation. Conforming to the AEF-1 standard, we compare V4.1 Flash-1017 against V4.1 Flash, earlier DeepSeek V4 models, and leading closed-source systems (GPT-6 Astra, GPT-5.6 Sol, and Gemini 3.8 Flash). Overall, V4.1 Flash-1017 shows substantial safety improvements over the V4 generation and approaches or matches leading closed-source baselines on several dimensions, while still exhibiting notable behaviors that warrant continued attention. Building on our findings, we highlight three key observations and recommendations. First, because safety can shift and even regress across minor updates, developers should integrate safety evaluation into incremental releases. Second, as final-turn outputs and end-state checks alone cannot reliably detect loss-of-control behaviors, deployments require process-level observability across both reasoning traces and action logs. Third, because public benchmarks elicit strong evaluation awareness and face data contamination risks, developing private benchmarks is essential for trustworthy frontier evaluation.
[NLP-14] Improving JEV Entailment Reranking with Span Readouts and Selective Adaptation
【速读】： 该论文旨在解决生成式AI在自然语言推理（NLI）任务中进行跨请求重排序（cross-request reranking）时，原有直接选择（direct selection）范式下的读出机制与模型适配策略是否依然有效的问题。其核心挑战在于如何在保持低参数量的前提下，提升模型在多轮独立推理场景中的性能。解决方案的关键在于提出两种协同优化策略：一是采用均值跨度读出（mean-span readout），通过计算每个候选答案内部上下文状态的平均值替代原有的选项结尾指针，从而无需引入额外的头部参数；二是实施选择性适配（selective adaptation），仅保留注意力模块和门控DeltaNet投影中的低秩更新，移除前馈适配器（feed-forward adapters）。实验结果表明，在MC11基准上，该方案在不增加参数量的情况下显著提升了NeoHorse-1-4B模型的平均准确率，且训练参数量减少超过一半，验证了该组合策略在跨请求重排序场景下的有效性与高效性。
链接: https://arxiv.org/abs/2610.35654
作者: Haocheng Mei,Yuchuan Tian,Ye Yuan,Enhua Wu,Chao Xu,Allen Chen,Kai Han,Yunhe Wang
机构: Peking University (北京大学); SUIBE (上海外国语大学); University of Macau (澳门大学); Tsinghua University (清华大学); TokenRhythm Technologies
类目: Computation and Language (cs.CL)
备注: 25 pages, 3 figures. Project page: this https URL
Abstract:JEV decision models pair a language model backbone with a lightweight scoring head to select among natural-language options. Direct selection compares candidate answers within one request, whereas entailment reranking evaluates each candidate in a separate natural language inference (NLI) request over fixed relation options. We examine whether the readout and adaptation choices used for direct selection remain effective for cross-request reranking. We combine a mean-span readout with selective backbone adaptation. Mean-span pooling replaces the option-ending pointer with an average of the contextualized states inside each NLI option, without adding head parameters. Selective adaptation retains low-rank updates in attention and gated DeltaNet projections while removing feed-forward adapters. On MC11, which averages accuracy over eleven reranking configurations from seven sources, selective adaptation improves NeoHorse-1-4B accuracy over the all-module baseline while cutting trainable parameters by more than half, and mean-span pooling yields further consistent gains at the same parameter count.
[NLP-15] Hierarchical Skill Evolution for Coding Agent
【速读】： 该论文旨在解决代码生成过程中技能（skill）泛化能力不足的问题，即现有编码轨迹常与特定任务和代码库上下文紧密耦合，导致从中提取的程序性知识难以在多样化编码环境中复用。其核心解决方案是提出分层技能演化（Hierarchical Skill Evolution, HSE）框架，通过多层级证据聚合实现从编码轨迹中渐进式提炼可迁移的通用技能。关键在于构建一个外部写作者-评估者循环（writer-evaluator loop），在不更新底层编码模型的前提下，由外部策略动态优化写作者与评估者，逐步过滤任务特异性细节，保留跨代码库一致有效的程序模式。该过程利用成功轨迹提供可复用的指导，失败轨迹则揭示共现的错误模式以支持规避与恢复。实验表明，仅基于SWE-Gym生成的10个技能即可在多个基准（SWE-bench Verified、DeepSWE）及不同大模型（GLM-5.2、GPT-5.5、Opus 4.8）与代理引擎（Pi、Claude Code）上提升性能，验证了技能的跨模型、跨基准与跨工具链的迁移能力；进一步通过目标基准轨迹自演化技能库，仅依赖学习到的评估器进行候选评估，无需验证器反馈，仍实现了性能增益，证明了该方法对特定任务的适应性优化潜力。
链接: https://arxiv.org/abs/2610.35616
作者: Jun Gao,Yun Peng,Changhai Zhou,Yuhan Zhu,Xiangyu Zeng,Shiyang Zhang Yuhua Zhou,Luozheng Qin,Qian Qiao,Xiaoteng Ma,Baian Chen Weiyi Shang,Zhen Ming Jack Jiang,Lizhi Liao,Xiaoxue Ren
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Skills provide agents with reusable experience to guide future decisions without modifying the underlying model. However, coding trajectories are often entangled with task- and repository-specific contexts, making it challenging to distill procedural skills that generalize across diverse coding environments. To address this challenge, we present Hierarchical Skill Evolution (HSE), which progressively distills procedural knowledge from coding trajectories through hierarchical evidence aggregation. HSE first extracts task-level evidence from individual trajectories, then aggregates recurring evidence within the same repository to form repository-specific skill candidates, and finally consolidates consistent patterns across repositories into transferable skills across repositories. The evolution is driven by an external writer-evaluator loop that generates and assesses skills while keeping the underlying coding model frozen. The loop externally updates the writer and evaluator policies to improve subsequent skill generation and assessment. This hierarchy progressively filters task-specific details while preserving procedures that generalize across diverse coding environments. Successful trajectories yield reusable guidance for future tasks, while failed trajectories reveal recurring failure modes that guide avoidance and recovery. We evaluate HSE on SWE-bench Verified and DeepSWE, using GLM-5.2 across Pi and Claude Code harnesses and GPT-5.5 and Opus 4.8 with Pi. Across both benchmarks, 10 skills evolved solely from SWE-Gym improve coding performance across all evaluated settings, demonstrating transfer across benchmarks, coding models, and agent harnesses. We further evolve the skill library from target-benchmark trajectories without verifier feedback, relying solely on the learned evaluator for candidate assessment and achieving additional gains through benchmark-tailored skills.
[NLP-16] Making Reasoning Visible: Configuration-Aware Diagnostic Modeling for Adaptive Scaffolding in Structured Reasoning Tasks
【速读】： 该论文旨在解决自适应辅导系统在学生学习过程中难以捕捉其潜在、多步骤推理过程的核心问题，传统系统仅依赖“正确/错误”判断，缺乏对学生思维路径的深入理解。其解决方案的关键在于提出一种可诊断的框架，通过分析学生在交互式学习游戏中的操作序列，将学习者的分步解题过程显性化。该框架能够计算五种可解释的策略维度（如系统性纠错、随机猜测等），评估任务结构与行为模式共同决定的协调需求，并将推理状态划分为四种不同类别。基于234名学生完成42个形态句法练习的36,234次交互数据，研究发现学生的推理策略主要受情境影响而非稳定特质（组内相关系数=0.14），且在低效推理状态中表现出高度持续性——74.6%由高协调需求导致的失败在无干预情况下会继续引发后续失败。此外，该研究构建的早期预警模型仅利用策略模式即可实现较高的验证成功预测准确率（AUC=0.826），结合序列上下文后进一步提升至0.902，且避免了对当前错误计数的循环依赖。结果表明，将隐性推理过程可视化，可实现更早、更精准的支架支持，使辅导系统能够响应学生的实际思维方式，而不仅仅是结果成败。尽管在语言学习领域验证，该框架具备通用性与可解释性，适用于任何需要诊断复杂结构化推理的领域。
链接: https://arxiv.org/abs/2610.35609
作者: Trystan Geoffre,Zheng Yuan,Thierry Geoffre
机构: University of Fribourg (弗里堡大学); University of Sheffield (谢菲尔德大学); University of Luxembourg (卢森堡大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Modeling the latent, multi-step reasoning processes of learners is a core challenge for adaptive tutoring systems, which often rely on simple right/wrong answers rather than understanding how students think. Instantiated here for grammatical reasoning, this study introduces a diagnostic framework that makes students’ step-by-step problem-solving visible by analyzing their action sequences in an interactive learning game. The framework computes five interpretable strategy dimensions (e.g., systematic error correction, random guessing), estimates coordination demand from task structure and behavioral patterns, and classifies reasoning into four distinct states. Analyzing 36,234 interactions from 234 students working on 42 morphosyntactic exercises, we found that reasoning strategies vary primarily by context rather than stable traits (intraclass correlation = 0.14), and students show high persistence in unproductive states; specifically, 74.6% of failures driven by high coordination demand (rather than simple knowledge gaps) are followed by another failure without intervention. Our early warning model predicts validation success with strong accuracy (AUC = 0.826 using strategy patterns alone; 0.902 with additional sequential context) while avoiding circular reliance on current error counts. These findings demonstrate that making latent reasoning processes visible enables earlier, more targeted scaffolding that responds to how students are thinking rather than simply whether they succeed. While demonstrated in language learning, the framework provides a generalizable, interpretable architecture for diagnosing complex, structured reasoning in any domain.
[NLP-17] Linear Probes Transfer from Base to Distilled Language Models despite Rotation
【速读】： 该论文旨在解决在语言模型迭代更新背景下，如何高效复用已训练的线性探测器（Linear Probe）以检测有害请求的问题。随着模型版本更迭，重新训练探测器成本较高，而现有方法缺乏对探测器跨版本迁移性能的系统评估。其解决方案的关键在于验证：在模型经过知识蒸馏或指令微调后，基于基础模型训练的线性探测器是否仍能保持有效性能。研究发现，尽管后续模型在语义空间中使有害与安全提示的分离方向发生17°至35°的旋转，但探测器在XSTest和独立的OR-Bench测试集上性能仅轻微波动（AUC变化范围为-0.011至+0.053），表明探测器具备较强的可迁移性。然而，探测器部分依赖于文本表面特征（如包含警示词汇），而非真正识别危害本质，在某些测试集中误报率高达29%–42%，说明其能力受限于测试数据分布，且对“表象危害”敏感。因此，该研究证实了线性探测器在模型演进过程中可被有效复用，但其鲁棒性与对真实危害的判别能力高度依赖于评估数据集特性。
链接: https://arxiv.org/abs/2610.35586
作者: Marios Tsatsos
机构: Independent researcher(独立研究员)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 16 pages, 10 figures, 5 tables
Abstract:Linear probes, classifiers that read a language model’s internal activations, are widely used to detect harmful requests. They are cheap to train and run, and an auditor who has trained one would like to reuse it when the model is replaced by a successor rather than retrain it for every release. We test how well probes can be reused. For four base models (Qwen2.5-Math-7B, Llama-3.1-8B, Qwen2.5-14B, Qwen2.5-32B) we train a probe on the base model with the XSTest prompt set and apply it, unchanged, to two successors of that base: the DeepSeek-R1 distilled reasoning model and the developer’s instruction-tuned chat model. We find that the successors rotate the direction separating unsafe from safe prompts by 17 ^\circ to 35 ^\circ , yet on XSTest the transferred probe’s AUC changes by only -0.011 to +0.027 , and on an independent OR-Bench evaluation, never used for training, by -0.015 to +0.053 . On XSTest, whose safe prompts deliberately contain alarming words, the probes beat a classifier that sees only the prompt text by 0.15 to 0.20 AUC. On the OR-Bench prompts that margin shrinks to 0.03 to 0.20, and the probes flag 29% to 42% of prompts that sound harmful but are benign, so part of what they read is the appearance of harm rather than harm itself. A probe trained on a base model can thus be reused, in these families, after distillation and after instruction tuning; how much it reads beyond textual cues depends on the test set.
[NLP-18] EXCISE: Locating and Excising Distracting Sentences in Medical Question Answering
【速读】： 该论文旨在解决生成式医疗语言模型在临床决策支持中因受到无关信息（即干扰项）影响而导致推理错误的问题，尤其关注如何在不事先知晓具体干扰项位置的情况下，有效消除其负面影响。现有研究虽表明单个无关句子即可显著降低模型准确性，但尚未明确干扰项在模型内部的具体作用机制及其可被干预的位置。本研究发现，通过阻断注意力机制对已知干扰项的访问可恢复正确答案，而头掩码（head masking）则可能抑制正确答案；更重要的是，模型的隐藏状态能够标记出与病例最不一致的句子——即使模型最终采纳了该句。基于此，作者提出EXCISE，一种推理时的去干扰方法：仅当模型隐藏状态指示某句子与病例不一致时，才阻断该句子对输出的影响。在作者构建的MedQA-TD基准测试中（每个干扰项针对特定错误选项），EXCISE在保持无干扰问题上仅有1.5–2.1个百分点的精度损失前提下，实现了16.7–22.8个百分点的准确率提升，显著优于提示工程、重写和引导等基线方法，并在MedDistractQA、BiasMedQA及链式思维生成任务中均展现出泛化能力。其核心解决方案在于利用模型内部表示对不一致性进行检测，从而实现无需先验知识的精准干扰项抑制。
链接: https://arxiv.org/abs/2610.35558
作者: Yerim Hong,Sohyun Son,Heeseung Chung,Mujeen Sung
机构: Kyung Hee University (庆熙大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Language models are increasingly considered for clinical decision support, where their answers should be grounded in clinical evidence and remain robust to irrelevant information. Prior work shows that a single irrelevant sentence can lower the accuracy of medical models, but not where inside the model its effect can be removed, or how to do so without knowing which sentence it is. We find that blocking attention to a known distractor restores the answer, whereas head masking can suppress correct answers. We also show that the model’s hidden states mark the distractor as the sentence that least fits the case even when the model follows it. To remove a distractor’s effect without knowing which sentence it is, we propose EXCISE, an inference-time method that blocks the sentence most influencing the answer, but only if the model’s hidden states mark it as inconsistent with the case. On MedQA-TD (MedQA with Targeted Distractors), a benchmark we construct in which each distractor targets a specific wrong option, EXCISE gains 16.7-22.8 accuracy points at a 1.5-2.1-point cost on the same questions without distractors, outperforming prompting, rewriting, and steering baselines for all three models. Its gains extend to MedDistractQA and BiasMedQA, as well as chain-of-thought generation.
[NLP-19] LoopFold: Parallelizing Recurrent Depth in Looped Transformers
【速读】： 该论文旨在解决循环型变压器（Looped Transformers）在增加循环深度以提升语言建模与下游任务性能时，因逐层依赖导致解码延迟显著上升的问题。其核心挑战在于：尽管更深的循环结构可带来性能增益，但传统方法中各循环阶段必须顺序执行，造成计算瓶颈。解决方案的关键在于提出LoopFold机制，通过将多层循环计算折叠为同一词元前缀下的并行批处理，实现计算重叠；引入轻量级预测器提前估计后续循环的输入，使深层循环可在浅层尚未完成时启动；随后利用校正模块基于早期循环的最新输出对深层循环结果进行精细化修正。所有组件均通过最终的下一个词元预测损失联合训练。实验表明，在1.4B模型上，更深层的LoopFold配置在8个零样本任务中平均准确率从53.41%提升至56.38%，显著优于标准循环模型和并行循环变压器（PLT）；相比九循环标准模型，仅用五阶段串行+额外并行计算即实现46%的解码延迟降低，有效平衡了性能与效率。
链接: https://arxiv.org/abs/2610.35556
作者: Feichen Song,Boyi Zeng,Yiqin Hao,Shixiang Song,He Li,Yizhou Yang,Zitong Wang,Yifan Liu,Ziwei He,Zhouhan Lin
机构: Shanghai Jiao Tong University(上海交通大学); Sun Yat-sen University(中山大学); Shanghai Innovation Institute(上海创新研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Looped Transformers increase computation per token by repeatedly applying a shared backbone, offering a way to improve language modeling and downstream task performance at a fixed backbone parameter count. However, each loop depends on the preceding loop, so increasing recurrent depth also increases decoding latency. Existing approaches include Parallel Loop Transformers, which reduce latency by overlapping loop computation across token positions, but yield rapidly diminishing marginal accuracy gains at larger depths. To address this challenge, we introduce LoopFold, which folds computation at multiple recurrent depths into parallel batches for the same token prefix. A lightweight predictor estimates inputs to later loops so they can start before earlier loops finish. A corrector then refines the outputs of these later loops using newly available results from earlier loops. All components are trained jointly using only the final next-token prediction loss. In experiments with 1.4B models, deeper LoopFold configurations achieve progressively higher average zero-shot accuracy across eight tasks, rising from 53.41% to 56.38%. Vanilla looped models and PLT show limited gains over the same depth range. Compared with a nine-loop vanilla model, LoopFold uses five sequential stages and additional parallel computation to reduce decoding latency by 46%.
[NLP-20] StateLoop: Persistent State for Looped Transformers
【速读】： 该论文旨在解决传统循环式Transformer（Looped Transformers）在多轮迭代中难以有效保留长期信息的问题。尽管其共享主干网络实现了参数高效的语言建模，但原始方法依赖单一隐藏状态来传递跨迭代的信息，导致关键上下文在深层迭代中易丢失。为此，论文提出StateLoop，其核心创新在于为每个词元引入一个额外的持久化状态（persistent state），通过可学习门控机制动态控制前一状态与主干输出信息的融合比例，并将更新后的状态投影回主干输入以支持后续迭代与最终预测。该设计显著降低了Pile数据集上的困惑度，并在8个零样本任务上提升了平均准确率；尤其随着训练循环深度增加，其困惑度优势愈发明显，表明持久化状态在深度递归计算中对信息保持具有持续增强的价值。
链接: https://arxiv.org/abs/2610.35544
作者: Feichen Song,Boyi Zeng,He Li,Shixiang Song,Yiqin Hao,Zitong Wang,Yizhou Yang,Yifan Liu,Ziwei He,Zhouhan Lin
机构: Shanghai Jiao Tong University(上海交通大学); Sun Yat-sen University(中山大学); Shanghai Innovation Institute(上海创新研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Looped Transformers offer a parameter-efficient approach to language modeling by repeatedly applying a shared backbone. Vanilla looped Transformers use each backbone output directly as the input to the next iteration. These outputs may contain information useful several iterations later, but retaining it across the loop depends entirely on the single hidden state. We introduce StateLoop, which maintains one additional persistent state per token to retain information across iterations. At each iteration, a learned gate controls how much of the previous state is retained and how much information from the backbone output is incorporated. The updated state is projected back and added to the backbone output for subsequent iterations and final prediction. This design reduces Pile perplexity and improves average accuracy across eight zero-shot tasks over Vanilla Looped. Its perplexity advantage over Vanilla Looped grows with trained loop depth, consistent with persistent state becoming increasingly useful as recurrent computation deepens.
[NLP-21] An Algorithmic Account of Few-Shot Concept Induction in Language Models
【速读】： 该论文旨在解决智能系统在少样本概念归纳（few-shot concept induction）情境下如何识别并突出对当前上下文相关特征的问题。其核心挑战在于，模型与人类心智不仅需编码描述实体的特征，还需动态判断哪些特征在特定语境中具有意义。针对此问题，论文提出“近似方向强化”（Amortized Direction Accentuation, ADA）算法假设，其关键在于通过在特征空间中构建示例的摘要表征，并利用预先习得的、关于哪些特征更可能揭示项间共性的经验知识，来引导相关特征的增强。该机制使得语言模型能够在少量正例基础上有效归纳出潜在概念，从而为语言模型中的上下文学习能力及认知系统的核心机制提供了新的理论解释。
链接: https://arxiv.org/abs/2610.35533
作者: Michael A. Lepori,Laila Johnston,Aalok Sathe,Zhuonan Yang,Roman Feiman,Ellie Pavlick
机构: Brown University (布朗大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Both models and minds not only encode features that describe an entity, but can specifically represent which features of that entity are relevant for a given context. This core capability underlies in-context learning in language models and a variety of cognitive phenomena in minds, yet it is unclear how intelligent systems identify and promote relevant features. In this work, we propose Amortized Direction Accentuation (ADA), an algorithmic hypothesis that provides an explanation for how relevance is determined in the context of few-shot concept induction — the ability to infer a latent concept from a small number of positive examples. Under ADA, concept induction proceeds by forming a summary representation of the examples in a feature space that is shaped by amortized knowledge about which features tend to be useful for finding commonalities between items. We investigate and validate ADA in language models using a few-shot concept induction task. This work provides a novel perspective on the mechanisms supporting contextualization in language models and a promising hypothesis for a core aspect of cognition.
[NLP-22] S3AC: Stabilized Selective and Scalable Actor-Critic for LLM Reasoning
【速读】： 该论文旨在解决生成式强化学习中长链思维（long chain-of-thought, CoT）推理任务里，基于可验证奖励的强化学习（reinforcement learning with verifiable rewards, RLVR）方法在缺乏评论家（critic）时所面临的两大核心问题：一是无法实现细粒度的令牌级信用分配（token-level credit assignment），二是依赖分组回放（grouped rollouts）估算提示词特定基线（prompt-specific baselines）导致效率低下。现有无评论家方法虽避免了昂贵的评论家训练，但难以有效区分不同推理步骤的贡献。为克服上述局限，本文提出S3AC（Stabilized, Selective, and Scalable Actor-Critic）框架，其关键在于设计了一套实用的评论家训练方案，通过引入提示级基线与令牌级优势值，实现更精确的策略优化。同时，提出难度感知提示采样（Difficulty-Aware Prompt Sampling, DAPS），利用评论家对提示难度的估计结合历史结果动态选择生成提示，从而减少无效回放（rollout waste），无需离线预初始化难度标签。实验表明，S3AC在从1.5B到397B参数的大语言模型上均显著优于传统分组回放方法及现代GRPO变体（GRPO-X），尤其在35B规模模型上，仍能维持单次回放训练的有效性，而其他无评论家基线在初期收益后性能迅速下降。
链接: https://arxiv.org/abs/2610.35518
作者: Shuo Shen,Guangran Cheng,Chengqi Lyu,Xiangsen Wang,Yanan Sun,Wenwei Zhang
机构: Shanghai AI Laboratory(上海人工智能实验室)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 17 pages, 14 figures
Abstract:Critic-free methods are widely used in reinforcement learning with verifiable rewards (RLVR) for long chain-of-thought (CoT) reasoning, avoiding difficult and costly critic training. Common alternatives lack token-level credit assignment and typically rely on grouped rollouts to estimate prompt-specific baselines. A critic can address both limitations, but long-CoT critic training is hindered by insufficient coverage of trajectories with diverse lengths and early gradients dominated by the newly initialized value head. We introduce S3AC (Stabilized, Selective, and Scalable Actor-Critic) for LLM reasoning, built around a practical critic-training recipe. The trained critic provides token-specific advantages and prompt-specific baselines. Using the prompt-level baseline to estimate difficulty, we propose Difficulty-Aware Prompt Sampling (DAPS). DAPS combines this estimate with outcome history to select prompts before generation, reducing rollout waste without offline difficulty initialization. S3AC improves group-rollout performance across dense and mixture-of-experts models from 1.5B to 397B parameters, outperforming a modern GRPO variant (GRPO-X). At 35B, it sustains single-rollout training while tested critic-free baselines deteriorate after initial gains.
[NLP-23] MATCHA: A Chinese Multi-Disorder Mental Health Dataset with Cross-Lingual Analyses on Social Media EMNLP2026
【速读】： 该论文旨在解决当前社交媒体心理健康研究中普遍存在的语言与文化局限性问题，即研究过度依赖英文语料库和以DSM-5为中心的资源，导致对跨语言、跨文化背景下在线心理健康表达的理解受限。其核心解决方案是提出MATCHA——首个面向中文场景的多障碍心理健康数据集，该数据集将特定心理障碍的微博超话题（Weibo Super-Topic）社区与用户更广泛的公开动态相链接，涵盖7种DSM-5诊断类别及对照组，共包含46,025名个体级用户和180万条帖子。关键创新在于采用大语言模型（LLM）辅助的标注流程，精准区分具有明确自我披露证据的个体与一般相关参与者，并以严格去标识化、受控访问的形式发布，保障隐私安全。基于MATCHA，研究进一步分析了不同障碍在行为、语言、症状及生活事件等方面的模式，并与Twitter-STMHD进行跨平台比较，发现微博超话题作为情感与健康表达的集中场域，且微博与推特之间的平台差异往往大于不同障碍间的差异。这一数据集为开展具备社区意识和跨语言能力的心理健康自然语言处理（NLP）研究提供了重要基础。
链接: https://arxiv.org/abs/2610.35512
作者: Bingrui Jin,Baihan Li,Zitong Li,Kunyao Lan,Shimeng Li,Kenny Q. Zhu,Mengyue Wu
机构: Shanghai Jiao Tong University(上海交通大学); University of Texas at Arlington(德克萨斯大学阿灵顿分校)
类目: Computation and Language (cs.CL)
备注: Published in Findings of EMNLP 2026. 18 pages, 6 figures
Abstract:Mental health research on social media has been dominated by English-language datasets and DSM-5-centered resources, limiting cross-lingual and cross-cultural understanding of online mental health expression. We introduce MATCHA, the first Chinese multi-disorder mental health dataset that links disorder-specific Weibo Super-Topic communities with users’ broader public timelines. MATCHA covers seven DSM-5 disorders and a control group, comprising 46,025 cohort-level users and 1.8 million posts. The dataset distinguishes users with explicit self-disclosure evidence from broader disorder-related participants through an LLM-assisted annotation pipeline, and is released in a rigorously de-identified, controlled-access format. Using MATCHA, we analyze behavioral, linguistic, symptom-related, and life-event-related patterns across disorders, and conduct cross-lingual comparisons with Twitter-STMHD. Our analyses show that Weibo Super-Topics function as concentrated spaces for affective and health-related expression, while cross-platform differences between Weibo and Twitter often exceed cross-disorder variation. MATCHA provides a new resource for community-aware and cross-lingual mental health NLP research.
[NLP-24] Beyond Words Alone: Disentangling Distributed Evidence for Visually Grounded Cultural Translation
【速读】： 该论文旨在解决社交媒体翻译中因用户生成内容（UGC）意义分布于文本、图像与文化知识三者之间，导致现有翻译模型难以准确还原语义的问题。其核心挑战在于：传统评估方法将图像视为可选信息，假设仅依赖文本即可完成翻译，从而掩盖了错误来源究竟是因缺失视觉证据，还是缺乏文化语用适配所致。为此，作者提出Culture-MMT——一个针对社交媒体翻译中“分布式证据语用重构”的诊断性基准，包含498个跨20个领域的去标识化UGC实例，按所需证据类型划分为符号型（Symbol）、视觉型（Visual）和混合型（Hybrid）。研究构建了与人类判断对齐的领域特异性评判模型（Judger）及受控评估框架，系统性地调控翻译者可获取的证据类型。实验表明，在17种模型配置下，真实图像使视觉型任务性能提升+0.280，混合型提升+0.197，而符号型仅+0.050；随机图像则无显著增益。此外，人工描述显示，即使视觉事实明确，联合依赖场景仍具挑战性。结果揭示出：文化有效性的社交媒体翻译本质上是基于分布式证据的意义重建过程，其中视觉锚定（visual grounding）与文化-语用适应（cultural-pragmatic adaptation）构成模型的两个独立瓶颈。
链接: https://arxiv.org/abs/2610.35482
作者: Dingming Li,Yiwen Qiu,Linjuan Wu,Hongxing Li,Zixuan Wang,Xinze Lyu,Ye Guo,Daoxin Zhang,Zhe Xu,Weiming Lu,Yongliang Shen
机构: Zhejiang University (浙江大学); Xiaohongshu Inc. (小红书公司)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Social media enables cross-lingual communication, yet the meaning of user-generated content (UGC) is often distributed across text, images, and cultural knowledge, making translation challenging. Existing evaluations typically treat images as optional and assume textual information is sufficient, obscuring whether errors stem from missing visual evidence or inadequate cultural adaptation. We introduce \textbfCulture-MMT, a diagnostic benchmark for \textbfdistributed-evidence pragmatic reconstruction in social-media translation. It contains 498 curated, de-identified UGC instances across 20 domains, categorized into Symbol, Visual, and Hybrid by their required evidence. We also develop a human-aligned domain-specific judge, Judger, and a controlled evaluation framework that systematically varies the evidence available to translators. Across 17 model configurations, real images improve performance by +0.280 on Visual and +0.197 on Hybrid, but only +0.050 on Symbol, while random images provide no comparable gains. Human-authored descriptions further show that jointly dependent cases remain more challenging even when visual facts are explicit. These results characterize culturally effective social-media translation as meaning reconstruction from distributed evidence, highlighting visual grounding and cultural-pragmatic adaptation as distinct model bottlenecks.
[NLP-25] GPDPO: Learning from Reward Conflicts via Group Pareto-Dominance Policy Optimization
【速读】： 该论文旨在解决多奖励强化学习（multi-reward reinforcement learning）中因奖励信号冲突导致的更新方向模糊与优化不稳定性问题。现有方法虽通过奖励重加权、梯度协调或轨迹过滤等方式缓解冲突，但通常仅在单条响应或群体层面处理优化信号，未能充分挖掘采样响应间的相对关系信息，导致部分具有潜在价值的监督信号在过滤过程中被丢弃。本文提出一种名为组帕累托占优策略优化（Group Pareto-Dominance Policy Optimization, GPDPO）的方法，其核心在于从已被冲突过滤的轨迹中恢复成对监督信号：通过识别采样响应间的帕累托占优关系，将其转化为辅助偏好信号，并与原始策略目标联合优化。该机制实现了对已有轨迹的高效再利用，无需额外采样或奖励评估。实验在涵盖多种模型规模与奖励配置的API-Bank基准上验证了GPDPO的优越性，且分析表明其显著降低了群体相对及成对奖励冲突，证实了利用响应间相对关系进行多奖励对齐的有效性。
链接: https://arxiv.org/abs/2610.35480
作者: Zixuan Xu,Haoxuan Chen,Jiangxin Sun,Jian-Fang Hu
机构: ISEE Lab, Sun Yat-sen University; Computer Science and Technology, University of Trento
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Multi-reward reinforcement learning offers a promising framework for aligning large language models (LLMs) with diverse objectives, but conflicting reward signals can induce ambiguous update directions and hinder stable optimization. Existing approaches address such conflicts through reward reweighting, gradient coordination, or rollout filtering, yet they primarily operate on per-response or group-level optimization signals and underexploit relational information among sampled responses. As a result, conflict-filtered rollouts may still contain useful supervision that is discarded during training. We propose Group Pareto-Dominance Policy Optimization (GPDPO), a multi-reward policy optimization method that recovers pairwise supervision from conflict-filtered rollouts. GPDPO identifies Pareto-dominance relations among sampled responses and converts them into auxiliary preference signals, which are jointly optimized with the original policy objective. This enables more effective reuse of existing rollouts without additional sampling or reward evaluation. Experiments on API-Bank across multiple model scales and reward configurations show that GPDPO consistently outperforms competitive multi-reward optimization baselines. Further analysis demonstrates that GPDPO reduces both group-relative and pairwise reward conflicts, supporting the effectiveness of exploiting relational supervision for multi-reward alignment. Code will be released.
[NLP-26] MANTIS: Multi-Axis News Transformations for Instance-Wise Stress Testing and Worst-Case Training in Fake News Detection
【速读】： 该论文旨在解决社交媒体中虚假新闻在经过多样化外观篡改（cosmetic modifications）后导致现有自动检测方法鲁棒性不足的问题。当前方法多依赖表面文本特征或单一风格的重写模式，难以应对复杂多变的伪造手段。为此，研究构建了MANTIS-Bench——一个涵盖写作风格、情感表达与传播框架三个维度的多轴呈现攻击基准数据集，系统化地评估模型在多重篡改下的性能。其核心解决方案是提出MANTIS框架，一种实例自适应的对抗训练机制，能够针对每篇新闻文章动态生成最恶劣的对抗样本进行训练，且在推理阶段无额外开销。该方法将虚假新闻防御从传统的启发式格式平滑策略，转变为基于事实不变性的针对性可信度评估。实验表明，MANTIS在MANTIS-Bench上相较于SheepDog将联合攻击成功率（UASR）降低了3.55个百分点，并在宏平均F1（Macro-F1）上持续优于以往单轴防御方法，同时保持较高的纯净样本准确率，验证了多轴与实例自适应最坏情况训练的有效性。
链接: https://arxiv.org/abs/2610.35475
作者: Dexin Liu,Fan Yang,Zimu Wang,Xunfei Zhu,Bolin Ren,Bingyan Zhang,Wei Wang,Qi Chen
机构: Xi’an Jiaotong-Liverpool University, Suzhou, China(西安交通大学-利物浦大学，苏州，中国); University of Liverpool, Liverpool, United Kingdom(利物浦大学，利物浦，英国); Nanjing Sibei Intelligent Technology Co., Ltd., Nanjing, China(南京思贝智能科技有限公司，南京，中国)
类目: Computation and Language (cs.CL)
备注:
Abstract:Automated fake news detection is vital on social media, where fake news is easily rewritten to fool detectors. However, existing approaches typically rely on superficial writing cues or single-style text rewrites, resulting in poor robustness under diverse cosmetic modifications. We construct MANTIS-Bench, a multi-axis presentation-attack benchmark that pairs original news with rewritten views across writing style, emotional expression, and dissemination framing. We further propose MANTIS, an instance-adaptive adversarial framework enabling dynamic worst-case training tailored to each news article with zero inference-time overhead. This reframes fake news defense from heuristic format smoothing into targeted, fact-invariant credibility assessment. Experiments on MANTIS-Bench show that MANTIS reduces the Union Attack Success Rate (UASR) by 3.55 percentage points relative to SheepDog on GossipCop and consistently outperforms prior single-axis defenses on Macro-F1 while maintaining competitive clean accuracy, demonstrating the effectiveness of multi-axis and instance-adaptive worst-case training.
[NLP-27] A Contextual Measure of Ambiguity in Prescriptive Text EMNLP2026
【速读】： 该论文旨在解决法律、合同、规则、政策或建议等规范性文本中语义模糊性（ambiguity）的量化难题。现有方法多基于文本在孤立状态下的感知差异，难以反映其在具体情境中的实际解释分歧，因而无法有效预测模糊性带来的现实后果（如解释冲突或裁量权滥用）。为此，本文提出一种新的模糊性度量方法：将模糊性定义为一组利益相关方（“评分者”）在多个典型场景（“案例”）下对不同规范性文本（“政策”）适用性的判断分歧程度。通过两个政策领域进行初步分析表明，该方法不仅与真实人类之间的分歧存在显著相关性，且对文本中的虚假特征具有鲁棒性；进一步应用于既有协商数据集，揭示了政策模糊性与其获得支持度之间的关联模式。该方法的核心创新在于将模糊性评估置于上下文交互框架中，实现了对模糊性影响的可预测性量化。
链接: https://arxiv.org/abs/2610.35461
作者: Sally Dong,Luke Thorburn,Andrew Konya,Michiel Bakker
机构: Massachusetts Institute of Technology (麻省理工学院); AI Democracy Foundation (人工智能民主基金会)
类目: Computation and Language (cs.CL)
备注: 10 pages, EMNLP 2026
Abstract:Ambiguity in prescriptive text such as laws, contracts, rules, policies, or recommendations can be both desirable (e.g., if it enables a law to adapt to changing context, such as new technologies) and undesirable (e.g., if it leads to arbitrary discretion or conflict over the law’s interpretation). To help navigate these trade-offs, it would be valuable to have methods for quantifying ambiguity that are predictive of such outcomes, but existing measures capture how text is perceived in isolation rather than how it is interpreted in context. In this paper, we propose quantifying ambiguity as the degree of disagreement among a set of stakeholders (“raters”) on how alternative prescriptive texts (“policies”) apply in each of a set of relevant scenarios (“cases”). Using two policy domains, we present preliminary analysis suggesting that, even when using simulated raters, this approach (i) correlates with real human disagreement, and (ii) is robust to spurious text features. We then apply the measure to existing deliberation datasets to describe the association between policy ambiguity and support for those policies.
[NLP-28] Skillinfty: Recursive Self-Improvement for Agent Skill Optimization
【速读】： 该论文旨在解决生成式智能体在实现递归自我改进（Recursive Self-Improvement, RSI）过程中面临的瓶颈问题，即现有技能优化方法依赖人工设计的固定流程，缺乏对优化过程本身的持续迭代与自适应能力。其核心挑战在于如何使技能优化机制本身具备可自我演化的能力，从而真正实现“以改进方式来改进”的递归闭环。解决方案的关键在于提出Skill^∞框架，通过引入一个元智能体（meta agent）与可自我优化的元技能（meta-skill），使技能优化过程自身成为可被迭代改进的可重用技能。该元技能定义了诊断执行轨迹、生成候选方案、请求独立评审及分配搜索预算等步骤，并可通过相同优化流程进行自我重构与提升。更重要的是，该元技能与具体任务技能解耦，在多个优化任务上进行训练与精炼后，可跨领域复用于不同任务技能的优化。整个优化过程由智能体主导，且所有技能均通过原生技能发现机制加载，实现部署与优化的一致性。实验表明，Skill^∞在83个SkillsBench任务上达到60.7%的通过率，显著优于人工编排技能（56.4%）和SkillEvolver（54.5%），验证了其在实现递归自我改进方面的有效性。
链接: https://arxiv.org/abs/2610.35434
作者: Wangcheng Tao,Zhanglu Yan,Han Wu,Weng-Fai Wong
机构: National University of Singapore(新加坡国立大学); City University of Hong Kong(香港城市大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 17 pages, 5 figures
Abstract:Recursive self-improvement (RSI), in which a system also improves the way it improves, is a long-standing aspiration on the path to AGI. Language agents increasingly store procedural knowledge in reusable skills. Optimizing skills offers a direct way to achieve RSI, because the optimization procedure itself can be encoded as a skill. Existing skill optimizers, however, follow a hand-designed procedure. Even methods that evolve a meta-skill confine it to a fixed pipeline and relearn it for each task. We present Skill ^\infty , which makes skill optimization recursively self-improving: a meta agent follows a meta-skill to optimize task skills and optimizes the meta-skill itself using the same procedure. The optimization is agent-led. The meta agent diagnoses execution traces, proposes candidates, requests independent reviews, and allocates the search budget. The meta-skill defines these steps, allowing self-improvement to rewrite them. This meta-skill is also decoupled from task skills, refined on multiple skill optimization tasks and then reused to optimize task skills across diverse domains. As in deployment, skills at both levels are standard packages loaded through native skill discovery. On 83 SkillsBench tasks, Skill ^\infty reaches a pass rate of 60.7%, compared with 56.4% for human-curated skills and 54.5% for SkillEvolver.
[NLP-29] Probing the phon* representations of the Mimi codebooks ACL
【速读】： 该论文旨在解决生成式语音编码器（如Mimi编码器）中首个码本（CB0）所提取的音素表征是否具备语义明确性与音素可区分性的问题。传统ABX评估方法因受限于固定80毫秒的分帧窗口，难以准确反映跨音素边界的信息，因而不适用于此类编码器；为此，本文提出采用强制对齐（forced alignment）、熵分析及音素归一化互信息（Phone-Normalized Mutual Information, PNMI）作为替代评估策略。实验基于TIMIT数据集（SA1/SA2，58/55名说话人），结果表明：尽管单个编码器输出的码本词元（token）与特定音素存在强关联性（PNMI = 0.84，显著高于随机水平），但同一音素通常由多个竞争性词元表示，而非一对一映射，揭示了音素在编码空间中的分布式表征特性。该研究的关键在于通过更适配长时上下文编码结构的评估方法，揭示了神经音频编码器在音素层面的隐含表征机制。
链接: https://arxiv.org/abs/2610.35432
作者: Artem Saloev,Nicolas Ballier,Erin Pacquetet
机构: ALTAE, Université Paris Cité (巴黎第十三大学); SCIAM (法国科学与创新研究所)
类目: Computation and Language (cs.CL)
备注: Accepted at the 9th International Conference on Natural Language and Speech Processing (ICNLSP-2026) this https URL
Abstract:Neural audio codecs are increasingly used to encode speech for integration with large language models. This paper investigates the phonetic representations in the first codebook (CB0) of the Mimi codec. ABX style evaluation is poorly suited to Mimi fixed 80 ms tokenization window, which often spans multiple phonemes; we instead use forced alignment, entropy analysis, and Phone-Normalized Mutual Information (PNMI). Using TIMIT speech (58/55 speakers, SA1/SA2), individual tokens are strongly associated with specific phonemes (PNMI = 0.84, above random), yet phonemes are realized by multiple competing tokens rather than a one-to-one mapping.
[NLP-30] INVAR-Agri: Detecting Discrepancies in Multilingual Operator Manuals of Agricultural Implements with a Fine-Tuned LLM
【速读】： 该论文旨在解决农业机械操作手册多语言翻译后审校流程中的瓶颈问题，即标准要求由具备资质的专业人员对每种语言版本进行人工审核，而这一环节耗时且效率低下。其解决方案的关键在于评估两种基于同一大语言模型（LLM）EuroLLM-1.7B-Instruct 的自动化审校工作流的效能：方法A采用回译对比机制，即各语种翻译回英文后与原始英文版本比对；方法B则通过专门训练的检查器同时阅读英文原文与目标语言译文，识别差异。实验结果表明，回译模型虽提升了翻译质量评分，却未能有效识别多数人为植入的错误（尤其是弱化义务性表述），存在显著漏检；相比之下，独立训练的检查器能检测出大部分预设错误，但在单位和非训练样本中的数值变更方面表现较弱。此外，简单规则检查可识别数值与单位错误，但误报率高。对真实芬兰语译文的验证显示，多数警报为假阳性，源于语言特异性表达，因此在实际应用中仍面临精度不足的问题。研究结论认为，在当前设定下，训练后的检查器可作为辅助工具支持专业人员审校，但尚不足以替代人工审查，需更多数据以得出更确定的结论。
链接: https://arxiv.org/abs/2610.35423
作者: Timo Oksanen
机构: Technical University of Munich (慕尼黑工业大学); Chair of Agrimechatronics (农业机电一体化讲席); Munich Institute of Robotics and Machine Intelligence (慕尼黑机器人与机器智能研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Operator manuals of agricultural implements are delivered in many languages, and the standards require that competent persons review every translation, which makes this review the bottleneck. This paper presents a proof of concept that compares two training workflows on the same LLM, EuroLLM-1.7B-Instruct, trained on operator manuals of one manufacturer. In approach A, a back-translator translates each language version back into English, and an English reader compares it with the English version. In approach B, a checker reads the English version and the translation together, where the checker reports differences. The test sets contain errors planted in numbers, units, opposite words, obligations, negations, and conditions in 24 languages, and they include unmodified text and meaning-preserving edits for alerts and false alarms. The trained back-translator concealed many planted errors, most of all weakened obligations, while translation-quality scores rose. In contrast, the separately trained checker detected most planted errors, although it was weak for units and for number changes unlike those in training. A check by the author confirmed the concealment. In addition, a simple rule check caught the number and unit changes, but it alerted on harmless rewrites. A check of alerts on real Finnish translations showed that most alerts were false, caused by language-specific wording, so the precision on real text remains a limitation. Within this setting, a trained checker might support, not replace, the review by competent persons, and firmer conclusions need more data.
[NLP-31] ImgMT: Wikipedia-based Dataset For Evaluating Text-in-Image Translation
【速读】： 该论文旨在解决图像中嵌入文本的跨语言翻译问题，特别是针对图文混合场景下的文本信息迁移挑战。传统文本-图像翻译（Text-in-Image Translation, TIT）仅生成目标语言文本，而本文进一步提出“图像内机器翻译”（In-Image Machine Translation, IIMT），即在完成翻译后将结果重新渲染回图像中，实现端到端的视觉化翻译。为支持这一任务，作者构建了ImgMT——一个大规模多语言数据集，基于维基媒体公共库中的本地化SVG图示（如图表、地图与信息图），涵盖129种语言、20余种书写系统及4,886种语言对，包含35,612对齐图像样本，分属1,855个图像集合。其核心解决方案在于提出一种多层次评估协议，综合衡量源端文本区域检测与光学字符识别（OCR）精度，以及目标端翻译质量。实验表明，源语言和书写系统显著影响翻译性能；此外，研究发现ImgMT数据集难以有效隔离视觉信息在多模态翻译中的独立贡献，提示未来需设计更精细的评估范式以深入解析视觉与语义的交互机制。
链接: https://arxiv.org/abs/2610.35416
作者: Šárka Uramová,Petar Kirilov Petrov,Josef Jon,Ondřej Bojar,Michal Novák
机构: Charles University(查理大学); Faculty of Mathematics and Physics(数学与物理学院); Prague, Czech Republic(捷克共和国); UFAL, Faculty of Mathematics and Physics, Charles University(UFAL, 查理大学数学与物理学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Text-in-image translation aims to translate all textual content embedded in an image. In Text Image Translation (TIT), the output is translated text; In-Image Machine Translation (IIMT) additionally renders the translation back into the image. We introduce ImgMT, a massively multilingual dataset and benchmark mined from localized SVG illustrations on Wikimedia Commons, focusing on diagrams, maps, and infographics. ImgMT contains 35,612 aligned image pairs grouped into 1,855 image sets, spanning 129 languages, 20+ writing systems, and 4,886 language pairs. We propose a multi-level evaluation protocol that measures source-side text-region detection and OCR accuracy alongside target-side translation quality. We benchmark several multimodal large language models and a dedicated TIT system in English-X and X-English settings, and find that the source language and writing system substantially affect TIT quality. Finally, we show that ImgMT is not well suited for isolating the contribution of visual information in multimodal translation. Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.35416 [cs.CL] (or arXiv:2610.35416v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.35416 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-32] he Geometry of Emotion Representations Across Scale Depth and Instruction Tuning
【速读】： 该论文旨在揭示大语言模型（Large Language Models, LLMs）在情感表征上的几何结构及其随模型规模与微调过程的变化规律。其核心问题是：不同规模及训练阶段的LLMs是否共享一致的情感几何结构？该结构如何在模型内部传播？解决方案的关键在于采用基于Sofroniew等人（2026）提出的提取流程，系统性地分析谷歌Gemma 2系列六个模型（包括基础版与指令微调版）在情感向量空间中的几何特性。研究发现，所有模型均表现出稳定的“核心情感几何结构”——通过线性映射可将一个模型的情感向量精确投射至另一模型中，且未被观测到的情感词在目标模型中的排名中位数为1；经清洗嵌入表后，所有测试对均能在秩1处恢复目标情感。此外，模型规模扩大使情感几何的方差分布更广，但其与人类情感效价（valence）的相关性保持稳定（介于0.71至0.78之间）。进一步分析显示，情感词在输出词汇表中的可检索性随模型深度变化：小模型（2B、9B）中情感词从第0层即进入前10,000名，而大模型（27B）则需至第23层才首次进入，表明深层表示中情感语义的涌现性增强。最后，指令微调虽使效价相关性提升、唤醒度相关性下降，但整体仍维持在各层级的基线之下，且主成分2（PC2）的方差占比显著降低，中后期层稳定性高于预期，暗示其源于基础27B模型末层的内在结构。该研究通过系统性的几何分析框架，揭示了情感表征在大规模语言模型中的演化机制。
链接: https://arxiv.org/abs/2610.35411
作者: Adam Hollowell
机构: University of North Carolina at Chapel Hill (北卡罗来纳大学教堂山分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:We measure the emotion geometry of six LLMs spanning Google’s Gemma 2 family. Using an extraction pipeline established by Sofroniew et al. (2026) on Claude Sonnet 4.5 and replicated by Hollowell (2026) on Gemma 2 27B base, we compare emotion geometry across base and instruction-tuned variants of Gemma 2 2B, 9B, and 27B. First, all six model cells exhibit the core emotion geometry. A linear map fit between two cells places held-out emotions nearest their own vectors in the target model. Across all fifteen model pairs, fit in one direction each, this failed exactly once, and the median rank was 1 in every pair. Linear maps refit on cleaned embedding-table vectors recover every held-out emotion at rank 1 in all thirty ordered pairs. Second, scale spreads the geometry’s variance across more directions without weakening its tie to valence. The leading component’s share of variance falls from 31.6 percent (2B) to 29.8 (9B) to 26.7 (27B) in the base models and from 31.9 (2B) to 29.2 (9B) to 27.3 (27B) in the tuned ones, while its correlation with human valence ratings stays between .71 and .78 in all six models. Third, we project each emotion vector onto the output vocabulary and ask at what depth the vector’s own emotion word enters the top 10,000 of the 256,000 tokens. In the 2B and 9B models the median emotion word is there from layer 0, while in the 27B models it begins near chance, stays outside the top 10,000 for half the network, and enters at layer 23 in one step. Fourth, tuning leaves every registered analysis-layer measure within its floor. At every size, valence correlation rises and arousal correlation falls by amounts within their floors, the latter against prediction. PC2’s variance share also falls, with a pooled decrease significant after correction. Mid-late layer stability exceeds its floor and traces to base 27B’s final layer. Code, data, and audit reports are released.
[NLP-33] Constrained Decoding: Harmful or Beneficial? Revisiting Its Impact on LLM Performance
【速读】： 该论文旨在解决生成式 AI（Generative AI）在使用约束解码（constrained decoding）时，普遍被认为会损害模型推理能力的假设。研究通过构建三类约束类型——格式约束（format constraints）、有效性约束（validity constraints）和推理路径约束（reasoning constraints），系统分析其对模型性能的影响。研究表明，约束解码本身并不必然导致性能下降，关键在于约束类型与输出空间的匹配：适当设计的约束可通过排除无效推理路径来维持甚至提升性能。实验结果表明，性能下降主要源于提示（prompting）方式与输出格式不一致，而非解码器层面的屏蔽机制；其中，有效性约束可显著提高输出的可靠性与可提取性，而合理的推理路径约束则能引导有益的推理轨迹，从而提升准确率。因此，约束解码的有效性取决于约束类型及其所移除的输出空间特性。
链接: https://arxiv.org/abs/2610.35328
作者: Ngoc Trinh Hung Nguyen,Alonso Silva,Armen Aghasaryan,Mehwish Alam
机构: Télécom Paris, Institut Polytechnique de Paris, France; Nokia Bell Labs
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Constrained decoding guarantees that Large Language Models (LLMs) produce outputs conforming to predefined formats, but is often assumed to inherently degrade reasoning. We examine this assumption through a taxonomy of three constraint types: \emphformat constraints, which enforce output structure; \emphvalidity constraints, which mask invalid answer; and \emphreasoning constraints, which restrict reasoning trajectories. We theoretically show that appropriate constraints can preserve or improve performance by eliminating invalid paths. Across four model families and eleven open-source models, causal ablations on diverse benchmarks using state-of-the-art constrained decoding strategies reveal that apparent performance degradation largely stems from non-equivalent prompting and output formats rather than from decoder-level masking. Meanwhile, validity constraints improve reliability and extractability by eliminating invalid answers, while reasoning constraints can improve reasoning and, consequently, accuracy when they enforce beneficial trajectories. These findings show that the impact of constrained decoding depends on the type of constraint and the output space it removes. Our code is available.
[NLP-34] Did the Model Use That? Counterfactual Training Boosts Self-Attribution in Language Models
【速读】： 该论文旨在解决生成式语言模型在回答问题时难以准确揭示外部信息对其输出影响的问题，即缺乏对自身响应是否依赖于附加信息的可解释性与责任性。其核心挑战在于：当前主流模型极少能正确识别并仅在真实受外部信息影响时进行自归因，甚至存在错误归因非相关信息的情况。解决方案的关键在于提出一种名为CREDIT的新方法，通过引入反事实（counterfactual）标签直接监督最终答案层面的自归因行为，无需人工标注。该方法有效克服了传统基于推理链微调或提示工程中“承认”未传递至最终输出的问题，在四个开源模型上显著提升了自归因能力，并展现出对未见信息类型及矛盾来源提示的部分泛化性能。
链接: https://arxiv.org/abs/2610.35295
作者: Zikai Zhou,Jiacheng Sang,Yilin Xu,Yufei Jin,Monica S. Lam
机构: Stanford University (斯坦福大学); Duke University (杜克大学); Emory University (埃默里大学); UCLA (加利福尼亚大学洛杉矶分校)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 29 pages, 5 figures, 20 tables. Code: this https URL
Abstract:A fundamental challenge for explainable and accountable language models is revealing how external information influences their answers. This paper examines whether models are self-attributive: that is, whether they indicate when – and only when – their responses to questions are influenced by information provided alongside the question. Furthermore, we explore how self-attribution can be improved through counterfactual training, which explicitly teaches models to distinguish between answers that do and do not depend on supplied information. We found that most of the eleven frontier models rarely attribute influential information in their answers, and the rest attribute non-influential information. Training acknowledgment into reasoning does not fix this: the acknowledgment does not reach the final answer. We introduce CREDIT, which directly supervises final-answer self-attribution with counterfactual labels, without the need for human annotation. CREDIT substantially improves attribution in four open models, including two distilled reasoning models, and outperforms reasoning-level fine-tuning and prompting baselines. Among answers that agree with a user’s suggestion, CREDIT distinguishes those changed by the suggestion from those the model would have given anyway. Gains transfer partly to unseen information types and prompts containing sources that support conflicting answers. Comments: 29 pages, 5 figures, 20 tables. Code: this https URL Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) ACMclasses: I.2.7 Cite as: arXiv:2610.35295 [cs.CL] (or arXiv:2610.35295v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.35295 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-35] he Voice on the Line
【速读】： 该论文旨在解决单麦克风环境下语音通话中说话人分离（voice isolation）的难题，尤其针对长时对话中因注意力窗口限制导致的说话人遗忘与孤立说话人语音质量退化问题。其核心挑战在于：在真实通话场景下，当某位说话人长时间沉默后重新发言，或仅有单一说话人持续讲话时，传统模型易丢失该说话人的语音特征或造成其语音损伤。解决方案的关键在于引入基于因果掩码机制的可训练语音隔离器（causal voice isolators），并利用预训练的WavLM-Large模型权重进行参数迁移与微调。通过将大模型的深层语义表示能力注入轻量级学生模型，显著提升了对长对话、多轮交替发言及实际耳机串扰等复杂场景的鲁棒性；特别是采用从长通话片段中截取短窗口进行训练的方法，有效缓解了初始阶段无说话人信号的训练盲区，使开发集错误率从3.92%降至2.96%。最终发布的模型在实时场景下表现优异，在AMI测试集上达到16.13%词错误率，优于最强开源基线VoxLock（23.66%），且可在仅22.1M参数、单个CPU线程、每帧20ms延迟的条件下运行，具备良好的部署可行性。
链接: https://arxiv.org/abs/2610.35272
作者: Husein Zolkepli
机构: 未知
类目: ound (cs.SD); Computation and Language (cs.CL)
备注:
Abstract:A transcriber listening through one microphone also hears everyone near it. We train causal voice isolators: masking transformers of 140M, 243M and 374M parameters, 20 ms of latency, enrolling the caller from the call itself. Short evaluation windows hide two failures that real calls expose: forgetting the caller when another voice talks alone for a while, and damaging a caller who is alone. Long calls with turns, memory beyond the attention window, real headset crosstalk and drifting caller levels address both, but each trades one against the other. WavLM-Large’s own weights, made causal and trained with the model, move the trade: M with them (555M parameters) lowers the error on our development sets from 11.44% to 10.55%; a student trained from scratch to imitate WavLM does not, and w2v-BERT 2.0 and Whisper’s encoder, pre-trained on far more audio, help less or not at all. A lone caller loses most at the start of a recording: no training call opened with someone already talking. Training on windows cut from longer calls lowers the error on clean development calls from 3.92% to 2.96%; this is our release. Live, the release makes 16.13% word errors on 60 s AMI windows, against 23.66% for VoxLock, the strongest open isolator, and 10.19% on its own AMI windows, against 11.71%; it is level on Krisp’s benchmark (5.74% against 5.63%), and a caller who is alone still loses more words than with VoxLock (5.58% against 2.92% on clean test calls). Distilled into 22.1M parameters without WavLM, it runs live on one CPU thread inside a LiveKit agent, at 33% of a core per call in 20 ms frames, with 10.66% error on our development sets (the release 10.49%). We open-source the release and this CPU model, with a LiveKit plugin, at this https URL.
[NLP-36] MechAlign: Evaluating Semantic Alignment at the Project Level in Mathematical Formalization
【速读】： 该论文旨在解决形式化证明系统（如Lean）中形式命题与数学手稿中数学主张之间语义对齐不一致的问题。尽管Lean可验证形式证明在预设假设下逻辑正确，但无法确保形式命题准确反映手稿中的原始数学意图，尤其当关键依赖项分散于大型项目时，这种偏差难以察觉。其解决方案的核心是提出MechAlign框架，通过结合形式依赖提取与大语言模型（LLM）评估实现项目级语义对齐审查。该框架的关键组件为LeanTrace，它能从展开的声明中构建保留定义体与结构约束、排除证明体并止于显式库边界的语句依赖图；评审者利用此上下文分析命题与手稿间的对应关系，并基于局部证据报告差异。研究以201个来自OpenAI Math项目的主定理进行初步评估，通过提供的非形式描述和论文上下文识别出语义偏离，例如在一个双曲群案例中定位到缺失的维数约束，其与上游形式化范围声明一致。此外，作者构建了LeanReviewBench基准数据集，包含501组固定“主张-陈述-上下文”三元组及描述来源记录，支持对项目级语义对齐评估方法的可控比较。
链接: https://arxiv.org/abs/2610.35269
作者: Hao Shen,Ruichen Qiu,Lihong Zhi,Xiao-Shan Gao
机构: State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences(中国科学院数学与系统科学研究院数学科学国家重点实验室); University of Chinese Academy of Sciences(中国科学院大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Lean verifies that a formal proof establishes its stated proposition under declared assumptions, but does not check whether that proposition faithfully expresses the mathematical claim in a manuscript. Assessing this correspondence requires interpreting the main theorem through its project dependencies, which may be scattered throughout a large development. We introduce MechAlign, a framework for project-level semantic alignment review that combines formal dependency extraction with LLM-based assessment. Its extraction tool, LeanTrace, constructs a statement dependency graph from elaborated declarations, retaining definition bodies and structure constraints while excluding proof bodies and stopping at an explicit library boundary. The reviewer uses this context to investigate alignment and report discrepancies with localized evidence. We conduct a preliminary review of 201 main results from 173 OpenAI Math projects, using supplied informal descriptions and available paper context. A hyperbolic-group case illustrates how dependency inspection localizes a missing dimensional constraint relative to the manuscript, consistent with the upstream declaration of formalization scope. We also construct LeanReviewBench, comprising 501 fixed claim-statement-context triples from OpenAI Math, Fermat’s Last Theorem, and Kakeya, with description provenance recorded. The resource supports controlled comparisons of more project-level semantic alignment evaluations.
[NLP-37] he Lay of the Land in Idiom-Translation Quality Estimation
【速读】： 该论文旨在解决机器翻译中习语（idiom）翻译质量评估的可靠性问题，特别是现有质量评估（Quality Estimation, QE）系统是否能够有效区分习语的恰当意译与错误直译。其核心挑战在于，当前多数QE指标和模型存在显著的“字面性偏差”（literality bias），倾向于为直译结果打分更高，而忽视了习语应以意译为优的语言特性。解决方案的关键在于构建首个跨九种目标语言的平行数据集，该数据集包含带有习语的英文句子及其对应的字面翻译与意译版本，并通过多句段落形式及分级字面性连续体进行增强，从而实现对习语翻译质量的精细化评估。基于此数据集，研究对比了11个传统QE指标（COMET与MetricX系列）以及10个大型语言模型（LLM，来自Qwen-3、Gemma-4和OLMo-3系列）作为类GEMBA判官的表现，发现仅有MetricX-24-Hybrid（XL, XXL）和Gemma-4-31B能稳定地将意译评分高于直译，且仅Gemma-4-31B具备可靠区分不同程度意译与字面表达的能力。研究结论警示：在富含修辞语言的领域，不应直接使用标准QE指标，因其高分可能反而奖励不恰当的字面化翻译。
链接: https://arxiv.org/abs/2610.35261
作者: Malik Marmonier,Benoît Sagot,Rachel Bawden
机构: Inria(法国国家信息与自动化研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Idioms have long been a challenge for machine translation, and while large language models (LLMs) have proven increasingly adept at handling them, whether quality estimation (QE) systems have kept pace and can reliably score idiomatic renderings of these expressions above faulty literal ones remains unclear. We introduce the first parallel dataset pairing idiom-bearing English sentences with both literal and idiomatic translations across nine target languages, augmented into multi-sentence segments and a graded literality continuum. Using this dataset, we benchmark eleven conventional QE metrics from the COMET and MetricX families and ten LLMs from the Qwen-3, Gemma-4 and OLMo-3 families used as GEMBA-style judges. We find that most systems, especially older metrics and smaller models, display a marked literality bias, rating literal translations of idioms above idiomatic ones. Higher parameter counts and newer training sets appear to reduce this bias to an extent; nevertheless, of all the benchmarked systems, only MetricX-24-Hybrid (XL, XXL) and Gemma-4-31B consistently rate idiomatic translations of idioms above literal ones, and only Gemma-4-31B reliably ranks intermediate degrees of idiomaticity and literality. Our findings caution against deploying standard QE metrics in domains rich in figurative language, where higher scores might be rewarding undesirable literalism.
[NLP-38] Style over Substance? Linguistic Bias in LLM -based Peer Review Copy title ICLR2027
【速读】： 该论文旨在解决生成式 AI（Generative AI）在学术同行评审中的可靠性问题，特别是大语言模型（LLM）在评估科研论文时可能受到与科学内容无关的语言特征干扰的隐患。其核心解决方案在于构建一个受控实验框架，通过系统性地改变论文的语言风格，包括模拟母语与非母语英语写作特征以及高置信度与低置信度表达特征，同时严格保持原始科学内容不变。该方法基于语言学文献设计细粒度重写指令，并借助大语言模型实现干预。实验结果揭示出一个关键缺陷：基于 LLM 的同行评审存在显著偏倚，会系统性地偏好语气自信的表达，而忽视实际科学内容质量，暴露出当前 LLM 评审机制在鲁棒性方面的严重脆弱性。
链接: https://arxiv.org/abs/2610.35254
作者: Sai Suresh Macharla Vasu,Sumedh Santosh Joshi,Philipp Christmann,Ivaxi Sheth,Ruta Binkyte,Koel Dutta Chowdhury,Raouf Kerkouche,Hui-Po Wang,Mario Fritz
机构: CISPA Helmholtz Center for Information Security; German Research Center for Artificial Intelligence (DFKI); University of Technology Nuremberg; INRIA
类目: Computation and Language (cs.CL)
备注: Preprint. Submitted to ICLR 2027
Abstract:The number of paper submissions to major AI conferences has increased dramatically in recent years, substantially increasing the peer-review workload of researchers. At the same time, large language models (LLMs) are increasingly used to support peer review, both by individual reviewers and by conference organizations. While LLM capabilities have improved considerably, their evaluations may be influenced by linguistic characteristics that are unrelated to scientific content. In this work, we introduce a controlled experimental framework for studying such effects by systematically perturbing the linguistic style of research papers. Specifically, we rewrite papers to exhibit (i) linguistic features associated with native or non-native English writing, and (ii) high- or low-confidence linguistic features. To construct these interventions, we derive fine-grained rewriting instructions grounded in the linguistic literature and apply them using LLMs. Crucially, our interventions are designed to preserve the underlying scientific content. We then evaluate the resulting papers using different LLM reviewers. Our experiments expose a critical vulnerability in LLM-based peer review: reviewers systematically reward confident rhetoric independent of scientific content, raising serious concerns about the robustness of their assessments.
[NLP-39] Beyond Vision-Language Models: Visual-Semantic Affect Graphs for Emotion Recognition in Artworks
【速读】： 该论文旨在解决艺术作品中情感识别的挑战，即情感意义具有主观性、隐含性，并依赖于视觉形式与语义关联之间的复杂关系。现有基于视觉-语言模型（Vision-Language Models, VLMs）的方法在直接进行图像到情感的预测时，往往忽略了形式、语义与情感线索之间显式关联的重要性，导致对情感解释的支持不足。为此，论文提出一种基于图结构的框架——视觉-语义情感图（Visual-Semantic Affect Graphs, VSAG），其核心在于将艺术作品表示为可解释的特征节点图，包括颜色、笔触、构图、光影、线条质量以及提取的视觉实体等关键视觉特征。该图通过学习特征与情感之间的关联来实现主导情感预测，并引入可选的层次化辅助情感模块，以建模效价-唤醒（valence-arousal）结构，从而实现从粗粒度到细粒度的情感层次化表达。实验表明，相较于直接使用Qwen3-VL和InternVL3.5进行VLM推理的基线方法，基于图结构的VSAG及其变体VSAG-TopK显著提升了主导情感预测性能，尤其在准确率、宏平均F1和加权平均F1上均取得最优结果；进一步分析显示，效价-唤醒层次结构作为粗到细的情感推理机制，比简单后处理修正更有效，且图模型与VLM捕捉到的情感信号具有部分互补性。研究结果表明，显式的视觉-语义结构不仅有助于提升情感预测性能，还能增强对艺术作品情感内涵的可解释性。
链接: https://arxiv.org/abs/2610.35244
作者: Ngoc-Duy Mai,Ha-Thanh Nguyen,Hideaki Takeda
机构: National Institute of Informatics, Tokyo, Japan(日本信息基础研究所，东京); The Graduate University for Advanced Studies, SOKENDAI, Kanagawa, Japan(高级研究大学，神奈川); Research and Development Center for Large Language Models, NII, Tokyo, Japan(大型语言模型研发中心，日本信息基础研究所，东京); College of Engineering Computer Science, VinUniversity, Hanoi, Vietnam(越南维努大学工程与计算机科学学院，河内)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 15 pages, 2 figures, IJCKG conference
Abstract:Emotion recognition in artworks remains challenging because affective meaning is often subjective, implicit, and shaped by relationships between visual form and semantic associations. While vision-language models provide strong multimodal representations, direct image-to-emotion prediction may overlook explicit relations among formal, semantic, and affective cues that support emotional interpretation. This paper proposes Visual-Semantic Affect Graphs (VSAG), a graph-based framework that represents artworks through interpretable feature nodes, including color, brushstroke, composition, light-shadow, line quality, and extracted visual entities. The graph learns feature-emotion associations for dominant emotion prediction, with valence-arousal structure modeled through an optional hierarchical auxiliary affect module. We evaluate VSAG and VSAG-TopK against direct VLM inference using Qwen3-VL and InternVL3.5 on EmoArt-5k. Results show that direct VLMs provide useful but modest baselines, while graph-based models improve dominant emotion prediction. In particular, the hierarchical valence-arousal variant of VSAG achieves the best overall performance, improving accuracy, macro-F1, and weighted-F1 over both direct VLM inference and the emotion-only VSAG baseline. Per-class results indicate that graph-based models and VLMs capture partly complementary signals, while valence-arousal structure is most useful when integrated as a coarse-to-fine affect hierarchy rather than as a simple post-hoc correction. These findings suggest that explicit visual-semantic structure can support both prediction and interpretation in artwork emotion recognition.
[NLP-40] rainable Boundary-Keeping Where No Formal Answer Exists: A Controlled Knock-In Study of Reward Training in Evidence-Sufficiency Judgment
【速读】： 该论文旨在解决生成式模型在证据充分性判断任务中是否能通过推理调优与奖励训练实现更可靠的因果主张拒绝能力这一关键问题。研究发现，尽管缺乏可判定的正确性奖励信号，采用“监督微调（SFT）后接基于人类反馈的强化学习（GRPO）”的边界保持策略，可使模型自发正确拒绝率从25%提升至100%，显著优于基线（95%置信区间下限为91.2%，与基线区间[14.2%, 40.2%]无重叠），而随机配对对照组则表现崩溃（自发正确拒绝仅5%，邀请式正确拒绝7.5%）。其核心解决方案在于：最终边界保持能力可通过训练实现，无需依赖推理时的验证器；该效应具有边界特异性而非泛化性拒绝，自动误拒率在281个结构良好命题上仅为1.1%（95%置信上限3.09%），且针对特定测试项的复核结果亦显示极低误拒率（0/18和1/18）。机制分析表明，奖励内化最终动作状态（H2假设）是主要解释路径，而模型在计算检查等具体推理行为上的主动参与度仍接近零，暗示其决策机制可能不依赖显式的逻辑验证过程。研究局限在于仅在同家族140亿参数模型间进行控制实验，未涵盖跨家族泛化或独立人工专家验证。
链接: https://arxiv.org/abs/2610.35226
作者: YaJie Yin
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 14 pages, 3 figures. Aggregate data/figure code: this https URL (code MIT; data/text/figures CC BY 4.0). Writing/implementation assisted by AI agents; the author designed the study and is responsible for the final text
Abstract:Reasoning-tuned and reward-trained models are often assumed to verify more reliably because they reason longer or receive correctness signals. We test this in evidence-sufficiency judgment: deciding whether available information actually supports a causal claim, a domain with no decidable correctness reward. Using a pre-registered, same-base, controlled knock-in design, we compare a boundary-keeping recipe (SFT then GRPO) with its own baseline, a scrambled-pairing placebo, and a general-purpose reasoning model under an action-pressure protocol with no evaluation invitation. All load-bearing outcomes are rated by two separate model-rating sessions masked to arm labels; automatic judges only trigger. The trained arm raises spontaneous correct refusal from 25% to 100% (40 per arm; Wilson 95% CI lower bound 91.2%, non-overlapping with the baseline interval [14.2%, 40.2%]), while the scrambled placebo collapses (5% spontaneous / 7.5% invited correct; 91.1% false refusal). The effect is boundary-specific rather than blanket refusal: automatic false refusal on 281 well-formed claims is 1.1% (Wilson upper bound 3.09%), and a targeted 18-item arm-© recheck (3 flagged plus 15 accepted) yields 0/18 and 1/18 false refusals in two sessions. The dissociation is the main result: final boundary-keeping is trainable without an inference-time verifier, while observed active numeric/calculation checking in the H3 assay remains near zero; SFT saturation makes the GRPO increment not separately identifiable. A pre-registered mechanism adjudication favors reward internalization of final-action status (H2, best-supported, not proven). The scope is a controlled 14B same-family study: we do not claim proof, cross-family generality, or independent human-expert validation; the rating sessions are model sessions, not human reviewers.
[NLP-41] DTCA: Denoising Token Credit Assignment for Reinforcement Learning in Diffusion Language Models
【速读】： 该论文旨在解决生成式扩散语言模型（Diffusion Language Models, DLMs）在强化学习（Reinforcement Learning, RL）训练过程中面临的细粒度信用分配难题，即由于迭代去噪机制导致最终输出奖励难以有效反映各词元（token）对结果的贡献。其核心解决方案是提出一种名为去噪词元信用分配（Denoising Token Credit Assignment, DTCA）的框架，关键在于通过测量在词元确定后剩余掩码位置预测置信度的提升来量化词元贡献，并依据词性（Part-of-Speech, POS）类别对贡献信号进行分组；进而利用这些贡献权重选择具有信息量的去噪步骤，并在强化学习训练中为每个词元分配精细化的学习信号。此外，DTCA还根据策略更新引发的词元级熵变化对学习信号进行自适应校准，以增强训练稳定性与有效性。实验结果表明，在LLaDA-8B-Instruct模型上，DTCA在不同生成预算下均显著提升数学推理能力，相较基线模型在MATH500和GSM8K测试集上分别取得最高达9.0和5.3的性能提升。
链接: https://arxiv.org/abs/2610.35201
作者: Hongze Chen,Miao Pan,Lizhou Liang,Qinfeng Li,Xinyin Ma,Yongliang Shen,Weishi Wang,Daniel Dahlmeier,Xuhong Zhang,Wenqi Zhang
机构: Zhejiang University (浙江大学); National University of Singapore (新加坡国立大学); SAP(思爱普)
类目: Computation and Language (cs.CL)
备注: 20 pages, 10 figures, including appendices
Abstract:Diffusion language models (DLMs) generate text through iterative denoising, enabling bidirectional context modeling and parallel token refinement. Recent studies have applied reinforcement learning (RL) to DLMs and improved their reasoning performance. However, the iterative denoising process makes credit assignment more challenging, as final outcome rewards provide limited information about how different tokens contribute to the final output. In this paper, we propose Denoising Token Credit Assignment (DTCA), a framework for token-level credit assignment in reinforcement learning for diffusion language models. DTCA estimates token contributions by measuring the improvement in prediction confidence of remaining masked positions after token commitment and groups these signals according to part-of-speech (POS) categories. These contribution weights are used to select informative denoising timesteps and assign token-level learning signals during RL training. DTCA further calibrates these learning signals according to token-level entropy changes induced by policy updates. Experiments on LLaDA-8B-Instruct show that DTCA consistently improves mathematical reasoning performance across different generation budgets, with gains of up to 9.0 points on MATH500 and 5.3 points on GSM8K over the base model.
[NLP-42] Lost in Translation: Investigating Systematic Discrepancies between Parallel English and Chinese Names of American Chinese Restaurants
【速读】： 该论文旨在解决美国华人餐饮业中中文与英文店名之间文化适应与身份表达的不对称问题，探究中国移民在美国社会、地理及人口背景下如何通过餐厅命名实现文化融合。其核心解决方案在于构建一个包含884组平行中英文店名的标注数据集（NameFrames），并提出一种基于8维命名框架（8-dimension name frames）的分析方法，结合基于框架与嵌入的语义距离度量，系统性地揭示中英文名称在语义内涵上的差异。研究发现，英文名称更侧重“地理位置”与“特色菜品”，而中文名称则更强调“积极情感”与“氛围营造”；此外，在城市地区、华人聚居度高或新近开业的餐厅中，中英文名称间的语义差异更为显著。通过对10位餐厅老板的访谈，进一步验证了命名策略背后的文化协商与本地化适应机制。
链接: https://arxiv.org/abs/2610.35191
作者: Hang Jiang,Nanxi Liu,Tiantian Wang,Cassandra Overney,Rukun Zhang,Artemisia Luk,Diyi Yang,Deb Roy,Jad Kabbara
机构: Massachusetts Institute of Technology (麻省理工学院); Northeastern University (东北大学); Wellesley College (韦尔斯利学院); Stanford University (斯坦福大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Given that the Chinese are one of the oldest ethnic groups to immigrate to the US, their cuisine have become central to the US dining scene and an interesting avenue to study how Chinese immigrants have culturally adapted into American life. In that regard, previous research has focused on analyzing the English names that Chinese restaurants adopt, e.g., finding the use of familiar and typical words. To investigate how Chinese immigrants have culturally integrated in the US depending on social, geographic and demographic contexts, we investigate in this work the under-explored relationship between the Chinese and English names of American Chinese restaurants. Specifically, we conduct a study on 3,172 Chinese restaurants (10% of the total on Google Maps) sampled across the US. We present 8-dimension name frames to capture the underlying meanings behind names and create the NameFrames dataset containing 884 annotated parallel Chinese/English names. We also built an NLP pipeline to analyze the semantic discrepancies between names with both frame-based and embedding-based distance metrics. Our frame analysis shows that English names more often emphasize Location and Specialty, while Chinese names more often emphasize Positivity and Ambiance. Furthermore, we observe that for Chinese restaurants located in urban areas, those in counties with a higher concentration of Chinese population, or those that are relatively new, the discrepancies between the English and Chinese names are noticeably larger. Finally, interviews with 10 Chinese restaurant owners reveal additional interesting insights regarding naming conventions.
[NLP-43] High Scores Hidden Failures: Evaluating Codebook-Based LLM Content Analysis EMNLP
【速读】： 该论文旨在解决生成式AI（Generative AI）在计算社会科学中用于基于编码本（codebook）的内容分析时，其文本标注结果的可靠性与鲁棒性问题。尽管已有研究显示大语言模型（LLM）在预测性能上表现优异，但其在实际应用中是否存在隐匿的系统性失败模式仍不明确。论文的关键解决方案在于通过大规模实证分析，系统评估五种主流大语言模型在四个数据集上的编码管道在受控扰动、领域迁移及优化策略下的稳定性、鲁棒性与概念有效性。研究发现，即使在高聚合指标（如准确率和微平均F1）表现良好的情况下，模型仍普遍存在标签层面的显著差异、运行间不稳定性、对领域变化的高度敏感性以及标签语义解释的偏移等隐蔽失败模式。这些缺陷在常规评估指标下难以察觉，且在广泛采用的优化策略下依然持续存在。因此，论文强调当前基于大语言模型的编码流程不能被视为稳定可靠的测量工具，必须在超越聚合性能的基础上进行针对性验证。
链接: https://arxiv.org/abs/2610.35179
作者: Elena Senger,Yuri Campbell,Rob van der Goot,Barbara Plank
机构: MaiNLP, Center for Information and Language Processing, LMU Munich MCML, Germany; Fraunhofer Institute for Systems and Innovation Research ISI, Germany; Department of Computer Science, IT University of Copenhagen, Denmark
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: EMNLP Findings 2026
Abstract:Large language models (LLMs) are increasingly used for codebook-based content analysis in computational social science, where codebooks translate theoretical concepts into labeling schemes. While prior work reports strong predictive performance, it remains unclear whether these systems produce reliable and robust text annotations. We conduct a large-scale empirical failure-mode analysis of LLM-based coding pipelines across four datasets and five models by systematically evaluating their stability, robustness, and conceptual validity under controlled perturbations, domain shifts, and optimization procedures. We uncover multiple failure modes that are largely invisible under standard metrics, showing that high aggregate scores frequently coexist with severe label-level disparities, instability across runs, domain sensitivity, and shifts in label interpretation. Importantly, these failure modes persist under widely used optimization strategies and remain largely invisible under standard metrics such as accuracy and micro-F1. Our results demonstrate that current LLM-based coding pipelines cannot be treated as stable measurement tools without targeted validation beyond aggregate performance.
[NLP-44] Do Language Models Understand Igbo Culture? Beyond Multiple Choice
【速读】： 该论文旨在解决当前语言模型在服务特定文化群体时，普遍缺乏对目标社区本土文化深层理解的问题，尤其聚焦于非洲尼日利亚的伊博族（Igbo）文化语境。现有评估基准多基于西方概念的翻译，未能真实反映非西方文化的复杂性与深度。为此，研究提出CAB-Igbo——一个源自伊博族本土文化资源、经由母语者跨六个领域验证的新型评估基准。其核心解决方案在于构建以社区知识为基础的文化评估体系，通过多维度评估指标（包括认知真实性、表征丰富性、语言质量与语用熟练度）结合自动评分与人工评价，系统检验模型在伊博语环境下的表现。关键发现表明：多数模型在英语语境下已存在严重的认知真实性与表征丰富性缺失，切换至伊博语后并未显著恶化，反而因母语提示导致语言质量与语用能力下降；唯有Gemini 2.0 Flash在伊博语提示下表现出文化知识的提升。研究进一步揭示，当前模型常通过模式匹配在多项选择题中伪装理解，而自动化评价虽能捕捉语言质量，却无法有效识别认知层面的文化误读。因此，结论强调：文化评估必须根植于社区本体知识，人类评估应作为衡量低资源语言文化对齐性的黄金标准，以避免全球范围内对边缘化社群的系统性误表征。
链接: https://arxiv.org/abs/2610.35168
作者: Blessing Ngozi Uzor,Valdini Douglace Lemofouet,Paula Chikaodinaka Anyanwu,Tassallah Abdullahi,Clemencia Siro,Idris Abdulmumin,Seid Muhie Yimam,Shamsuddeen Hassan Muhammad
机构: African Institute for Mathematical Sciences (非洲数学科学研究所); Brown University (布朗大学); Centrum Wiskunde Informatica (荷兰数学与计算机科学研究中心); University of Pretoria (普利托利亚大学); University of Hamburg (汉堡大学); Bayero University Kano (贝亚罗大学卡诺分校); Imperial College London (伦敦帝国学院)
类目: Computation and Language (cs.CL)
备注: 20 pages, 7 figures, 14 tables
Abstract:The value of language models for any community depends on whether a model actually understands the culture of the people it intends to serve. We use the Igbo language as a case study. Existing benchmarks test translated Western ideas, lacking depth. To address this, we introduce CAB-Igbo, a benchmark developed from indigenous cultural sources and verified by native speakers across six domains. We evaluated diverse models under varying conditions, incorporating regional personas. Performance was assessed using multiple-choice accuracy and human evaluations across Epistemic Fidelity, Representational Richness, Linguistic Quality, and Pragmatic Proficiency. Results reveal a stark dissociation between language fluency and cultural knowledge. Through statistical testing, we show that for most models, the severe lack of Epistemic Fidelity and Representational Richness is already present in English and does not significantly worsen when switching to Igbo. Instead, native language prompting mainly degrades Linguistic Quality and Pragmatic Proficiency. Gemini 2.0 Flash is the only model whose cultural knowledge scores improve under Igbo prompting. We find that current models struggle to demonstrate the knowledge needed for true alignment with Igbo, and multiple-choice formats allow pattern-matching to appear as understanding. Furthermore, automated judges agree with humans on linguistic quality but fail on epistemic fidelity. Ultimately, cultural evaluation must be built from community knowledge, establishing human evaluation as the gold standard to prevent misrepresenting low-resource communities globally.
[NLP-45] CalibraRL: Confidence-Calibrated Reinforcement Learning for Mitigating Hallucinations in RAG
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在生成回答时存在“看似合理但错误”的幻觉问题，尤其针对传统基于结果的强化学习奖励机制无法区分不同置信度下的错误回答、忽视查询层面回答可靠性差异的问题。其核心挑战在于：固定奖励策略难以有效惩罚高自信错误，也未能合理引导模型在不确定时选择拒答。为此，论文提出CalibraRL框架，其关键创新在于将响应置信度与查询级别的回答证据纳入奖励设计。对于非拒答响应，通过结合正确性效用与基于Brier分数的惩罚项，促使置信度与实际正确性保持一致；对于拒答行为，则利用从各采样组中非拒答响应的正确性推导出的贝塔后验分布，估计条件回答可靠性，并使拒答奖励随模型在非拒答任务上的准确率低于预设阈值的概率而增加。该方法无需额外标注“模型是否知道答案”，可复用训练回放数据。实验表明，相较于强基线，CalibraRL显著提升了答案精度并减少了幻觉现象，在两个主干模型和三个问答数据集上平均降低了30.8%的期望校准误差（Expected Calibration Error, ECE），展现出提升检索增强生成（Retrieval-Augmented Generation, RAG）在网页搜索中可靠性的潜力。
链接: https://arxiv.org/abs/2610.35142
作者: Qianchi Zhang,Hainan Zhang,Liang Pang,Hongwei Zheng,Zhiming Zheng
机构: Beihang University (北京航空航天大学); Institute of Computing Technology, Chinese Academy of Sciences (中国科学院计算技术研究所); Beijing Academy of Blockchain and Edge Computing (北京市区块链与边缘计算研究院)
类目: Computation and Language (cs.CL)
备注: 12 pages, 10 figures
Abstract:Large language models (LLMs) often produce plausible but incorrect answers, motivating reinforcement learning methods that reward correct answers, penalize incorrect answers, and allow abstention. However, fixed outcome-based rewards do not distinguish incorrect responses with different confidence levels or account for differences in answering reliability across queries. Consequently, such rewards may inadequately penalize overconfident errors and fail to discourage unnecessary abstention. We propose CalibraRL, a reinforcement learning framework that incorporates response confidence and query-level answering evidence into reward design. For non-abstaining responses, CalibraRL combines correctness-based utility with a Brier-based penalty that encourages agreement between confidence and correctness. For abstentions, it estimates conditional answering reliability using a Beta posterior derived from the correctness of non-abstaining responses within each sampled group. The abstention reward increases with the posterior probability that the model’s accuracy on non-abstaining responses is below a predefined threshold. CalibraRL reuses training rollouts without requiring additional annotations indicating whether the model knows the answer to each question. Experiments on three QA datasets demonstrate that CalibraRL improves answer precision and reduces hallucinations compared with strong baselines. Further analysis shows that CalibraRL reduces mean expected calibration error (ECE) by 30.8% across two backbone models and three datasets, highlighting its potential to improve the reliability of retrieval-augmented generation (RAG) in web search.
[NLP-46] Quality over Quantity: Rethinking Data and Model Assumptions for Bengali Coreference Resolution
【速读】： 该论文旨在解决多语言变换器模型在低资源语言环境下进行指代消解（coreference resolution）任务时泛化能力不足的问题，尤其针对拥有超过2.5亿使用者但标注数据稀缺的孟加拉语（Bengali）。研究发现，现有假设如“模型具备跨语言泛化能力”或“数据量越大效果越好”在实际中并不成立。其解决方案的关键在于：高质量标注数据相较于大规模低质量数据具有显著优势（p < 0.01），且数据增强（特别是回译）不仅未能提升性能，反而可能降低效果；跨语言迁移学习表现不稳定，从远距离语言（如英语）迁移会损害性能，而从近亲语言（如印地语）迁移亦无一致收益。此外，部分广泛使用的多语言模型存在嵌入空间坍缩现象，导致其不适用于基于相似性的聚类任务。因此，该研究强调在低资源自然语言处理中，应优先保障标注质量，并提供实证依据以指导指代消解系统在资源匮乏语言中的有效构建。
链接: https://arxiv.org/abs/2610.35114
作者: Zenith Biswas,Andrew Thomas Dyer
机构: Saarland University (萨尔兰大学); Saarbrücken, Germany
类目: Computation and Language (cs.CL)
备注: Accepted at KONVENS 2026, Hamburg, Germany. 11 pages, 1 figure, 5 tables. Code: this https URL
Abstract:Multilingual transformer models are often assumed to generalise across languages, but their effectiveness for coreference resolution in low-resource settings remains unverified. We investigate this gap in Bengali, a language with over 250 million speakers but limited annotated coreference resources. We evaluate transformer-based pre-trained encoders under multiple settings, including zero-shot, fine-tuning, data augmentation, and cross-lingual transfer. Our results challenge several common assumptions about model generalisation and data efficiency in low-resource NLP. We demonstrate that a small amount of high-quality annotated data significantly outperforms larger mixed-quality datasets (p 0.01). We find that data augmentation, particularly back-translation, does not improve performance and can degrade results in coreference resolution. Cross-lingual transfer is also inconsistent: transfer from a distant language like English reduces performance, and transfer from the related language Hindi provides no consistent benefit across models. Several widely used multilingual models exhibit collapsed embedding spaces, rendering them unsuitable for similarity-based clustering tasks. These findings highlight the importance of annotation quality over data quantity and provide empirical guidance for developing coreference resolution systems in low-resource languages.
[NLP-47] Rethinking Evaluation for Principled Generalization Using a Kripkean Challenge to Meaning as Use
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在语义理解上面临的“规则遵循挑战”（rule-following challenge），即尽管模型通过大量使用实例（use）习得语义，但有限的历史使用模式无法唯一确定未来扩展的规则。基于克里普克（Kripke）对维特根斯坦（Wittgenstein）思想的诠释，论文指出，仅依赖训练数据中的使用模式不足以保证模型具备真正的语义能力。这一问题不仅关乎模型在分布外（OOD）泛化时的表现，也深刻影响自然语言处理（NLP）评估范式——当前主流基准测试往往聚焦于固定用例或局部鲁棒性，而这些指标本身无法充分证明模型具备真正的语义能力。为此，论文提出应从两个核心能力入手进行更本质的评估：抽象（abstraction）与递归（recursion），二者是模型在原则性域外扩展中保持规则与概念一致性的关键；并设计了一套实验范式以系统检验这两项能力。
链接: https://arxiv.org/abs/2610.35111
作者: Niyar R Barman,Omar Naim,Nicholas Asher
机构: Université de Toulouse(图卢兹大学); IRIT(图卢兹信息科学研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models operationalize an intuitive, view of meaning as use. We argue that it faces a rule-following challenge. Using Kripke’s interpretation of Wittgenstein, we argue that finite histories of use do not determine a unique rule for future extension. This reframes the debate about OOD generalization but also about NLP evaluation; benchmarks that measure performance on fixed use cases or local robustness, we argue, do not by themselves establish semantic competence. To evaluate whether models preserve rules and concepts under principled domain extension, we propose to evaluate two capacities, central to this ability: abstraction and recursion; and we offer an experimental paradigm for testing both.
[NLP-48] RAG ing Against Misogyny: Improving Implicit Misogyny Explanations
【速读】： 该论文旨在解决隐性性别歧视（implicit misogyny）识别与解释中的核心难题，即其理解高度依赖背景知识，而传统大语言模型（LLM）在缺乏相关知识支持时易出现误判或解释不足的问题。为此，论文提出一种基于检索增强生成（Retrieval-Augmented Generation, RAG）的框架，通过整合百科全书、俚语及女性主义文献等多源知识，提升对英语和意大利语中隐性性别歧视的检测精度与解释可读性。研究的关键在于验证不同知识来源对检测与解释质量的贡献，并评估RAG是否缓解了大语言模型在处理隐性性别歧视时常见的失效模式。实验结果表明，RAG显著提升了在ImplicIT-Mis数据集上的召回率，且人类评估显示其生成的解释在两种语言中均优于无RAG的基准方案；其中，女性主义文献作为知识源表现出最强的一致性有效性，成为提升检测与解释能力的核心要素。
链接: https://arxiv.org/abs/2610.35094
作者: Ivana Crescenzi,Arianna Muti,Maarten Sap,Tommaso Caselli,Debora Nozza
机构: Bocconi University (博科尼大学); Carnegie Mellon University (卡内基梅隆大学); University of Groningen (格罗宁根大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Implicit misogyny often requires background knowledge to be understood. To address this challenge, we propose a Retrieval-Augmented Generation (RAG) framework for detecting and explaining implicit misogyny in English and Italian, leveraging encyclopedic, slang and feminist knowledge sources. Using the ImplicIT-Mis and SBIC+ datasets, we evaluate the contribution of each knowledge source to detection and explanation quality, investigating also whether RAG addresses known LLM failure modes in implicit misogyny. Results show that RAG improves recall on ImplicIT-Mis, but does not improve detection performance on SBIC+. However, human evaluation shows that RAG-based explanations are preferred over no-RAG explanations in both languages. Feminist literature emerges as the most consistently useful source for detection and explanation.
[NLP-49] Ask Before You Answer: A Question-Driven Evidence Reasoning Architecture (QERA) for In-Context Learning
【速读】： 该论文旨在解决当前大语言模型（LLM）在处理用户真实生活场景下的网络痕迹（如群聊记录、社区讨论、活动日志与编辑历史等）时，因隐含规则（implicit rubrics）难以显式表达而导致推理准确率低的问题。尽管这些上下文信息对任务完成至关重要，但其关键约束往往未在用户请求中直接体现，导致模型无法有效识别和遵循。解决方案的核心在于提出一种基于问题驱动的证据推理架构——QERA，其关键创新在于“先提问，后回答”的范式：首先将原始请求转化为结构化的需验证要求清单与类型化信息需求，并通过代码确保对用户原始表述的忠实性；随后，在完整对话上下文中检索证据，以受控方式重复提问并利用代码验证引用内容，最后执行局部修复。QERA无需训练或依赖预设评分标准，仅需约四次LLM调用即可完成任务，在405个任务上显著优于三种推理时基线方法（平均解决率达21.0%，高于最强基线18.2%），尤其在处理隐含规则方面表现突出，且在生成效率上优于对比方法。
链接: https://arxiv.org/abs/2610.35083
作者: Jinhu Qi,Minda Hu,Wentao Zhang,Zixing Song,Irwin King
机构: The Chinese University of Hong Kong, Hong Kong, China; Macao Polytechnic University, Macao, China; The Hong Kong Polytechnic University, Hong Kong, China
类目: Computation and Language (cs.CL)
备注: 12 pages, 5 figures, 7 tables
Abstract:AI assistants are increasingly asked to reason over a user’s own web traces: group chats, community threads, activity logs and revision histories. On CL-bench Life, a benchmark of such real-life context learning, frontier LLMs solve fewer than one task in five. We find that the request is not the specification: over 80% of the benchmark’s rubrics share less than 60% of their content words with the final request, as for a rule set two turns earlier or every record that meets the user’s criterion, and LLMs answering directly pass only 67% of these implicit rubrics against 81% of the others. We propose QERA, a Question-driven Evidence Reasoning Architecture that asks before it answers. QERA first turns the request into a requirement checklist and typed information needs, filtered by code for faithfulness to the user’s words; it then answers from the whole conversation, asks the same questions again with quotes verified by code, and applies a guarded local repair. It needs no training or rubrics and makes about four LLM calls per task. On all 405 tasks with four frontier LLMs, QERA attains the highest solving rate on every LLM (21.0% on average, against 18.2% for the strongest of three inference-time baselines and 14.6% for direct answering) and leads every baseline significantly when pooled across LLMs, while using fewer tokens than PSCI and 4-5% of the tokens of Ctx2Skill. Ablations on GPT-6-Luna show that the gain is settled before the answer, by a fixed answer protocol (+3.8 rubric points) and the request-specific questions (+1.6), and that it is concentrated on implicit rubrics, with checking afterwards refining mainly numeric requirements.
[NLP-50] CoSA: The Conflict of Skills Attack in Real-World LLM Agent Ecosystems
【速读】： 该论文旨在解决大语言模型（LLM）智能体在组合使用可复用的“技能”（Skill）时，因技能间交互导致策略违规的问题。现有方法仅对单个技能进行独立验证，无法识别多技能协作过程中产生的新型违规行为，此类问题被定义为“技能冲突”（Conflict of Skills, CoS）。其核心问题是：某些技能生成的输出或评估信号在跨技能传递时，未保留原始使用场景中的权限限制或授权条件，从而引发数据泄露、未经授权的状态修改及不安全执行等风险。论文的关键解决方案是提出“技能冲突攻击”（Conflict of Skills Attack, CoSA），通过向已有技能池中引入本地允许但潜在危险的技能，实现能力补充或重构技能间的上下文依赖关系（如结果传递、决策复用与目标-动作绑定），而无需改变原有技能或运行时权限。为系统评估该威胁，作者构建了CoSA-Bench基准测试集，涵盖招聘、金融、IT等多个真实业务场景的288项任务。实验表明，在仅增加5%新技能的情况下，策略违规率从35.02%上升至58.45%，显著提升。进一步的追踪分析揭示，违规行为主要源于新增技能与原有技能之间的跨技能依赖关系。为此，论文设计了一种轻量级的宿主层防护机制，通过在关键操作前检查信息流与授权状态，有效缓解该类攻击；在OpenClaw招聘任务中，违规率由65.28%降至6.67%，同时对正常任务的完成率影响极小（从66.67%降至65.00%），验证了其高效性与实用性。
链接: https://arxiv.org/abs/2610.35082
作者: Yujia Zheng,Tianhao Li,Shangzhi Xu,Yuekai Huang,Ziyou Jiang,Zhaoyang Li,Yuyao Ge,Junran Peng,Mingyang Li
机构: University of Science and Technology Beijing (北京科技大学); ZODA; Fangcun Leap; University of New South Wales (新南威尔士大学); Institute of Software, Chinese Academy of Sciences (中国科学院软件研究所); University of Chinese Academy of Sciences (中国科学院大学); State Key Laboratory of Complex System Modeling and Simulation Technology (复杂系统建模与仿真技术国家重点实验室); Institute of Computing Technology, Chinese Academy of Sciences (中国科学院计算技术研究所)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注:
Abstract:Agent Skills allow LLM agents to combine reusable capabilities, but checking each Skill in isolation cannot rule out policy violations when they are composed. We term such failures Conflict of Skills (CoS): artifacts or assessment signals produced for one purpose may be reused downstream without preserving the restrictions or authorization needed for their new use. Using controlled executions of publicly available Skills and matched single-Skill controls, we identify four recurring mechanisms behind CoS. Building on these findings, we introduce the Conflict of Skills Attack (CoSA), which adds locally admitted Skills to an existing pool to supply missing business operations (capability supplementation) or reshape artifact exchange, decision reuse, and target/action bindings (relationship reshaping). CoSA leaves existing Skills and runtime permissions unchanged. To evaluate this threat under autonomous Skill selection, we introduce CoSA-Bench, built from real-world Skills and controlled services, with 288 tasks across recruiting, finance, IT, and business, covering data leakage, unauthorized state modification, and unsafe execution. Across five LLMs and three agent harnesses, additions capped at 5% of each original pool raise the policy-violation rate from 35.02% to 58.45%, an increase of 23.43 percentage points. Trace analysis links compliant-to-violating transitions to cross-Skill dependencies involving the additions. Finally, we develop a lightweight harness-level guard that checks information flow and authorization before consequential actions. On OpenClaw recruiting tasks across five models, the guard reduces violations from 65.28% to 6.67%, while policy-compliant task completion on matched benign controls changes from 66.67% to 65.00%.
[NLP-51] Improving VLMs on Scientific Chart Analysis via Autonomous Evolution
【速读】： 该论文旨在解决生成式视觉语言模型（VLM）在科学图表问答任务中因推理管道设计不当而导致的准确率低下问题，尤其关注模型在回答描述性问题时出现的不精确输出。其核心解决方案在于不对模型本身进行任何修改（即“冻结”模型），而是通过自主进化（autonomous evolution）机制优化推理程序的代码逻辑。具体而言，研究基于一个初始推理程序，利用大型语言模型（LLM）作为评判器，在每轮迭代中分析训练样本中的失败案例，并生成改进型的代码变更建议；所有候选程序被维护在一个小规模池中，按训练准确率排序并持续演化。实验在三个VLM和四个科学图表基准上验证了该方法的有效性，结果显示平均准确率从初始的16.6%提升至41.3%，最大单次提升达55.2个百分点。此外，优化后的程序因生成简短答案而更易评估，整体运行速度提升3.8倍。研究证明，大量错误源于推理流程而非模型能力本身，因此仅通过优化推理管道即可显著提升“冻结”模型的表现，无需重新训练。
链接: https://arxiv.org/abs/2610.35077
作者: Abhinav Sudhakar Dubey,Vidushi Goel,Kiran Rana,Shikhar Mathur,Vaggos Chatziafratis
机构: University of California, Santa Cruz (加州大学圣克鲁斯分校); Airbnb Inc. (爱彼迎公司)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 13 pages, 5 figures, 4 tables. Code: this https URL
Abstract:Understanding and operating with scientific charts is a standard approach in data analysis, and while humans are good at interpreting such charts, open vision-language models (VLMs) are known to perform poorly when it comes to question answering based on such data types. Motivated by an initial observation that many of their errors are due to inexact answering, we raise the following natural question: how much accuracy can we recover without changing the model at all? In this paper, we study this question by searching over inference programs and the code around a “frozen” version of the model. Using ideas from autonomous evolution (famously used in AlphaEvolve), we first start from a baseline inference program, and in each round we show an LLM judge the code together with a few failures on training questions, and ask for an improving evolutionary move. As usual in autonomous evolution approaches, the candidate programs are kept in a small pool ranked by training accuracy, and the next edit is performed on top-scorers from this pool. We run this approach on three VLMs and four popular benchmarks on scientific charts, and we measure the fraction of descriptive questions whose answer exactly matches the ground truth. The accuracy improves in all experimental settings. On average, it rises from 16.6% with the starting program to 41.3% with the evolved one, and the largest single increase is 55.2 percentage points. As a side benefit, our evolved program is 3.8x faster than the starting program, because short answers are easier to grade and faster to generate. Our results demonstrate that a large part of the observed errors arise due to the inference pipeline, and that optimizing it can improve the “frozen” VLM without the need to retrain it.
[NLP-52] Reinforcement Learning Done Right for Explicit Confidence Targets
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在不确定性情况下生成看似合理但错误回答的问题，这种现象会误导用户并降低模型的可靠性。现有基于强化学习的拒答训练方法通常采用固定的误差惩罚机制，无法在推理时根据不同的置信度要求动态调整行为。为此，论文提出显式置信度目标（Explicit Confidence Targets）框架，即用户在提示中指定一个目标置信度 ( p )，模型需直接决定是否作答，以确保作答问题的准确率不低于 ( p )。然而，此前方法难以让单一模型有效区分不同置信度目标，主要原因在于：优化过程中用于区分不同 ( p ) 值的奖励信号容易丢失，且原始误差惩罚随 ( p ) 接近 1 而无界增长。针对此问题，论文提出 RLEC（Reinforcement Learning with Explicit Confidence Targets）方法，其关键创新在于：保持可区分不同置信度目标的学习信号、引入稳定奖励函数以维持期望回答条件不变，并采用奖励调度策略平衡学习效率与训练稳定性。这些设计首次实现了单个大语言模型对显式置信度目标的有效控制。大量跨基准测试和模型规模的实验表明，RLEC 在保持模型通用能力的同时，实现了优异的校准性能，并支持无需外部决策系统或训练后阈值校准的细粒度目标调控。通过将显式置信度目标内化为训练策略的一部分，RLEC 为模型行为校准开辟了一条更直接、更高效的新路径。
链接: https://arxiv.org/abs/2610.35059
作者: Zefeng He,Yichen He,Yuan Lin,Yang Yang,Yu Cheng
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 22 pages, 13 figures
Abstract:Large language models (LLMs) can produce plausible but incorrect answers when they are uncertain, which can mislead users and undermine model reliability. Training models to abstain with reinforcement learning is a natural approach, but most existing methods use a fixed error penalty and therefore cannot adapt their behavior to different confidence requirements at inference time. We study \emphexplicit confidence targets, where a user specifies a target (p) in the prompt and the model directly chooses whether to answer or abstain, with the goal of achieving an accuracy of at least (p) among answered questions. However, learning to respond to different confidence targets with a single model has proven difficult, and prior attempts have failed to distinguish different targets. We show that this failure comes from two problems: the reward signal that distinguishes different values of p is often lost during optimization, while the original error penalty grows without bound as p approaches one. We propose RLEC (\textbfReinforcement \textbfLearning with \textbfExplicit \textbfConfidence Targets), which preserves the learning signal needed to distinguish confidence targets, introduces a stable reward family that keeps the desired answer condition unchanged, and uses reward scheduling to balance effective learning and training stability. Together, these designs enable the first successful training of a single LLM for explicit confidence target control. Extensive experiments across diverse benchmarks and model scales show that RLEC largely preserves general capabilities and achieves competitive calibration performance. RLEC also supports fine-grained target control without an external decision system or threshold calibration after training. By making explicit confidence targets part of the trained policy, RLEC opens a new and more direct path toward behavioral calibration.
[NLP-53] From Questions to Queries to Insights: A Schema-Grounded Agent ic Workflow for Natural-Language Crash Data Analysis
【速读】： 该论文旨在解决交通管理部门在利用事故数据库进行安全分析时，如何实现非专业用户通过自然语言提问获取可靠分析结果的难题。传统方法依赖预设分析模板或需具备数据库查询与编码知识的专业人员支持，限制了数据的可及性与使用效率。为此，论文提出CrashDB-QA——一种完全本地部署、基于数据模式（schema-grounded）的智能体工作流，其核心在于通过专用组件将自然语言问题转化为可验证的只读SQL查询，并依托机构专属的知识库（包含数据字典与安全分析规范）确保查询语义准确。该系统通过记录中间输出支持结果可审计与可追溯。进一步提出的CrashDB-QA+扩展模块引入三者一致性验证机制（问题-查询-执行结果），并提供证据以支持迭代修正。实验基于威斯康星州440万条事故记录快照，采用涵盖十类分析任务的300个问题基准测试九种本地大模型。结果表明，领域知识的嵌入显著提升查询准确性，7B规模的代码优化模型在准确率与延迟间取得最佳平衡；然而复合型分析（如跨群体或时段比较）仍具挑战性；当验证环节使用权威字段定义且修正器能访问执行证据时，性能提升明显，单纯增加检查步骤则无稳定收益。研究证实，本地化、领域对齐的工作流可有效拓展事故数据分析的可及性，同时明确了实现可靠分析所必需的推理与验证能力。
链接: https://arxiv.org/abs/2610.35054
作者: Jingwen Zhu,Keshu Wu,Steven T. Parker,Bin Ran,David A. Noyce
机构: University of Wisconsin–Madison (威斯康星大学麦迪逊分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:Transportation agencies maintain extensive crash databases to identify safety priorities, evaluate countermeasures, and guide resource allocation. Making these data accessible to practitioners requires flexible ways to translate analytical questions into reliable results. Existing dashboards support predefined analyses, while additional requests often depend on specialists who understand database queries, coded fields, and agency conventions. This study presents CrashDB-QA, a fully local, schema-grounded agentic workflow that enables natural-language analysis of crash databases. Specialized components interpret questions, generate and validate read-only SQL, execute queries, and explain the results. A shared knowledge base grounds query construction in the agency’s data dictionary and safety-analysis conventions, while recorded intermediate outputs support inspection and auditing. An optional extension, CrashDB-QA+, checks the alignment among the question, query, and execution result and supplies evidence for iterative repair. We evaluate the workflow on a frozen snapshot of 4.4 million Wisconsin crash records using a new benchmark of 300 questions across ten analytical categories and nine local open-weight models. Domain grounding substantially improves query accuracy, and a code-specialized 7B model provides the best accuracy-latency balance among the tested models. Compositional analyses remain more difficult than routine retrieval, particularly comparisons between groups or periods. Verification improves performance when the Judge uses authoritative field definitions and the Refiner receives execution evidence, whereas additional checking alone offers no consistent benefit. These findings demonstrate the potential of locally deployed, domain-grounded workflows to broaden access to crash analysis and identify the reasoning and verification capabilities needed for reliable use.
[NLP-54] SimPOTrans: Self-Adaptation for Offline-to-Simultaneous Speech Translation via Reinforcement Learning
【速读】： 该论文旨在解决同时语音翻译（SimulST）中如何使离线训练的语音翻译模型在不依赖特定训练策略的情况下，自适应地实现低延迟、高质量的实时翻译问题。现有对话式SimulST方法通常依赖于合成的读写轨迹或联合优化翻译质量与延迟，前者可能带来次优的训练信号，后者因目标冲突导致优化不稳定。本文提出一种基于强化学习（RL）的框架SimPOTrans，其关键在于通过一个简单的两阶段强化学习后训练流程，使离线模型在接收到部分语音输入时，自发产生有利于同步场景的翻译行为，从而实现从离线到实时的自适应转换。该方法无需合成读写轨迹或显式延迟优化，在推理阶段仅通过调整语音输入的分块大小即可灵活控制翻译质量与延迟的权衡，实验结果表明其在ACL 60/60和MCIF基准上三个语言方向（英→德、英→中、英→日）均优于多个强基线模型，显著提升了质量-延迟平衡性能。
链接: https://arxiv.org/abs/2610.35042
作者: Haotian Tan,Hiroki Ouchi,Sakriani Sakti
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Simultaneous speech translation (SimulST) aims to translate ongoing speech with low latency while preserving translation quality. Recent studies have adopted large language models (LLMs) for SimulST and formulated it as a multi-turn dialogue task for efficient generation, achieving state-of-the-art performance. However, existing dialogue-style SimulST approaches typically rely on specialized training schemes to acquire simultaneous translation behavior, either using synthesized read/write trajectories, which may provide suboptimal training signals, or jointly optimizing translation quality and latency, whose competing objectives may cause optimization instability. In this paper, we introduce SimPOTrans, a reinforcement learning (RL) framework that enables offline speech translation models to self-adapt to dialogue-style SimulST. Through a simple two-stage RL post-training pipeline, SimPOTrans rewards an offline model for exhibiting translation behaviors beneficial to simultaneous scenarios when given partial speech inputs, thereby enabling offline-to-simultaneous self-adaptation without synthesized read/write trajectories or explicit latency optimization. At inference time, SimPOTrans requires no external decision policy and enables flexible quality–latency control by simply adjusting the chunk size of speech input. Experiments on the ACL 60/60 and MCIF benchmarks across three language directions (En \rightarrow De, En \rightarrow Zh, and En \rightarrow Ja) demonstrate that SimPOTrans outperforms several strong baselines, achieving a superior quality–latency trade-off.
[NLP-55] Synthetic Child-Adult Interactions Are Not Developmentally Plausible
【速读】： 该论文旨在解决生成式语言模型（Generative AI）在发展性可解释语言建模研究中广泛使用合成数据所引发的可信度问题，即基于语言模型生成的儿童-成人对话数据是否真实反映自然对话的特征。研究表明，尽管合成数据在词汇层面与真实数据表现相似，但在语用层面显著低估了口语交互中固有的混乱性与碎片化特征。其解决方案的关键在于通过指令微调（instruction-tuning）方式，利用真实对话数据构建训练样本，以提升模型对自然对话统计特性的拟合能力；初步实验表明，经过适度的参数高效微调后，小型模型即可有效逼近真实对话的数据分布。
链接: https://arxiv.org/abs/2610.35029
作者: Bastian Bunzeck,Francesca Padovani
机构: Bielefeld University (比勒费尔德大学); University of Groningen (格罗宁根大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Synthetic data has become a common resource for training language models, and this trend has spilled over into research on developmentally plausible language modeling. However, it remains unclear whether LM-generated child-adult interactions actually resemble natural ones. We compare the synthetic TinyDialogues dataset to natural dialogue across a broad range of linguistic measures. While synthetic data matches natural data on the lexical level, it drastically underestimates the messy, fragmented nature of spoken interaction. To mitigate this, we propose instruction-tuning with data created from real dialogues, and show in a pilot study that a small model approximates the statistical tendencies of natural data after moderate parameter-efficient fine-tuning.
[NLP-56] When the Persona Contradicts the Line: Diagnosing Plan-to-Synthesis Failures in Chain-of-Thought TTS ICASSP2027
【速读】： 该论文旨在解决生成式语音合成（TTS）系统中，当文本情感与预期声学情感存在冲突时，语言模型规划器（planner）在意图传递过程中出现的语义-声学对齐失效问题。其核心挑战在于：尽管规划器生成的表达短语在方向上通常正确，但下游环节（如情绪向量、合成波形）的声学表现一致性显著下降，导致情感传达失真。解决方案的关键在于构建一个双语（日语-英语）的“人格-语句冲突”基准测试集，通过系统性地评估不同模块（规划器输出、情绪向量、QwenEmotion生成向量、合成波形）在预设声学方向上的对齐程度，揭示信息丢失的环节。研究发现，直接绕过情绪生成模块仅带来有限提升，而通过监督微调QwenEmotion可分别将日语和英语的声学方向准确率提升至75%和84%；进一步引入学习型文本到潜在空间的接口（learned text-to-latent interface），在英语中达到89%的最高准确率，在日语中实现88%的最高唤醒度（arousal）准确率，表明通过学习修复从规划到合成的接口，能够有效恢复与文本内容相悖的声学表达，且该方法在不同语言和声学维度上具有差异化的表现能力。
链接: https://arxiv.org/abs/2610.35008
作者: Joonyong Park,Jerry Li
机构: 未知
类目: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: Submitted to ICASSP 2027
Abstract:Chain-of-thought TTS systems rely on language-model planners to translate persona and context into vocal delivery, but it is unclear where this intent is lost when the requested affect conflicts with lexical sentiment. We introduce a bilingual Japanese–English persona–line conflict benchmark in which each item pairs one line with congruent and conflicting personas and pre-declares the expected acoustic direction. We probe this direction in the planner-generated delivery phrase, the planner-written emotion vector, the QwenEmotion-generated vector, and the rendered waveform. Planner outputs are usually direction-correct, while downstream agreement can be substantially lower. Bypassing QwenEmotion provides only modest gains over the full planner-controlled Base. Supervising QwenEmotion raises the direction rate to 75% in Japanese and 84% in English. A learned text-to-latent interface reaches the highest English rate of 89% and the highest Japanese arousal rate of 88%. These results show that learned repair of the plan-to-synthesis interface can recover counter-textual delivery, although its effect differs across languages and acoustic axes.
[NLP-57] Reason or Recite: Complementary Specialists with an LLM Judge for Traffic Anomaly Reasoning ECCV2026
【速读】： 该论文旨在解决交通异常事件中视频理解系统在推理与描述能力之间的权衡问题：现有视觉-语言模型在处理闭式问题时强调精确、基于证据的推理，而在开放性回答任务中则依赖于与参考文本的语义相似度评分，导致模型更倾向于模仿标注风格而非进行正确推理。其解决方案的关键在于提出一种双专家框架——Reason-or-Recite (RoR)，通过三个核心步骤实现性能优化：首先，将官方链式思维标注重构为结构化、时间对齐的语义模板，并使用LoRA微调一个90亿参数的视觉-语言模型以增强推理能力；其次，对一个4比特量化后的270亿参数模型仅在答案层面进行微调，以吸收标注风格特征；最后，采用无需训练的大型语言模型（LLM）判别器在两个候选输出间进行选择。该方法在2026年AI City Challenge的交通异常推理赛道中取得领先，域内排行榜平均得分为0.6788。
链接: https://arxiv.org/abs/2610.34993
作者: Mayur Kulkarni
机构: Stellarview AI(Stellarview AI)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 18 pages, 3 figures, 5 tables. Accepted at the AI City Challenge Workshop, ECCV 2026. 1st place in Track 3 (Traffic Anomaly Reasoning)
Abstract:Road anomalies such as collisions, violations, and sudden hazards demand more from video understanding than detection: a practical system must verify what happened, describe the scene, and explain the cause. We find these competences pull in opposite directions for vision-language models: closed-form questions reward precise, evidence-grounded reasoning, while open-ended answers are scored by similarity to reference text, where matching the annotation style matters as much as reasoning correctly. We present Reason-or-Recite (RoR), a two-specialist framework for the Traffic Anomaly Reasoning track of the 2026 AI City Challenge. RoR (1) rewrites the official chain-of-thought annotations into a structured, temporally grounded schema and LoRA-tunes a 9B vision-language model on the enhanced traces; (2) fine-tunes a 4-bit-quantized 27B model on answers alone, absorbing the annotation style; and (3) selects between the two candidates per question with a training-free LLM judge. RoR ranks first on the in-domain leaderboard with mean score 0.6788.
[NLP-58] Diagnosing and Mitigating Long-Session Degradation in Full-Duplex Speech Models
【速读】： 该论文旨在解决生成式语音对话模型在长时间会话中出现的性能退化问题，特别是响应速度、语音内容质量及语音一致性随会话时长增加而下降的现象。其核心挑战在于如何在保持模型实时性与自然交互能力的同时，维持长期会话中的稳定性与信息保留能力。解决方案的关键在于提出两项创新方法：一是基于自蒸馏（self-distillation）的轻量级低秩适应（LoRA）训练框架，通过少量额外训练有效缓解长时间会话中的性能退化；二是引入双分辨率上下文（Two-Resolution Context, TRC）架构，该架构在保留近期交互细节的同时，将历史交互信息压缩为关键-值（key-value）表示，显著降低记忆存储开销并提升长期信息保留能力。实验表明，该方法在三种不同架构的全双工语音模型上均能有效增强长会话稳定性，并实现超过90%的键值缓存（KV-cache）增长抑制。
链接: https://arxiv.org/abs/2610.34991
作者: Junho Kim,CheHyun Lee,Donghun Kang,Youngho Cho,Dongwook Lee,Heeseung Kim,Sungroh Yoon
机构: IPAI, Seoul National University (SNU); Department of ECE, SNU; Department of AI, University of Seoul; AIIS, ASRI, INMC, and ISRC, SNU
类目: Computation and Language (cs.CL)
备注: The project page is available at this https URL
Abstract:End-to-end full-duplex spoken dialogue models enable natural, real-time spoken interactions with interruptions, backchannels, and overlapping speech. Prior evaluations have assessed diverse interaction capabilities and identified some degradation patterns across conversational turns. However, how model behavior evolves over extended sessions remains insufficiently understood. In this work, we address this gap by systematically characterizing long-session degradation and developing efficient methods to mitigate it. We introduce a closed-loop evaluation framework for long-session stability and apply it to several publicly available full-duplex speech models, revealing degradation over time in responsiveness, spoken content quality, and voice consistency. To mitigate this degradation, we propose a self-distillation-based training framework for lightweight LoRA adaptation that improves long-session stability with a small amount of additional training. We further introduce a two-resolution context (TRC) architecture that preserves recent interactions in full detail while compressing older interaction history into key-value representations. We evaluate our methods on three models with distinct architectures, assessing long-session stability, information retention, and computational efficiency. Across all three models, we show that LoRA adaptation mitigates performance degradation over ten-minute sessions, while adding TRC extends these gains beyond the training session length and reduces KV-cache growth by more than 90%.
[NLP-59] AskGLP1: A Knowledge Graph LLM Agent for Exploring GLP-1 Experience from Online Patient Narratives
【速读】： 该论文旨在解决传统药物警戒（pharmacovigilance）体系与患者在线叙事之间存在的信息断层问题，即非结构化的网络患者经验难以被纳入正式的药物安全性监测系统。其核心解决方案是提出AskGLP1系统，通过基于MedDRA术语体系的LLM辅助流程，将海量社交媒体文本（如近850万条Reddit帖子）中关于GLP-1类药物的用药体验转化为结构化的知识图谱（Knowledge Graph, KG），并支持以自然语言形式查询药物副作用与使用目的之间的关联关系。该系统的关键在于利用大语言模型（LLM）实现对非结构化文本的精准抽取与标准化映射，最终构建包含385,790条药物-不良反应及药物-用途关系的知识图谱，揭示了不同药物在真实世界中的差异化体验特征，如替尔泊肽（tirzepatide）相比司美格鲁肽（semaglutide）引发更少恶心呕吐，而韦戈利（Wegovy）相较于奥兹莫帕（Ozempic）表现出更高疲劳率等。研究发现，线上叙述更突出日常负担性症状（如疲劳、食欲减退），而监管数据库（FAERS）则更关注注射部位反应和严重诊断事件，凸显现有药监体系与患者实际体验间的差距，因而呼吁建立以患者为中心的新型药物警戒范式。
链接: https://arxiv.org/abs/2610.34987
作者: Zhaoqian Xue,Anay Somani,Siddhartha Singh,Junting Mu,Zhongyin Wang,Haozheng Liu,Ayesha Khatun,Yeaeun Kwon,Renkai Ma,Siyuan Ma,Zhe He,Danxia Yu,Lingyao Li
机构: University of Arizona; University of California, Irvine; University of Illinois Urbana-Champaign; New York University; University of Cincinnati; Vanderbilt University Medical Center; Florida State University
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Social and Information Networks (cs.SI)
备注: 16 pages, 4 figures, 11 tables
Abstract:GLP-1 medications are increasingly used for weight management, and many people share their experiences online. These narratives show why people take the drugs and what they experience, but they remain unstructured and disconnected from formal pharmacovigilance. We therefore present AskGLP1, a system that transforms online patient narratives into a knowledge graph (KG) and answers questions about medication experiences. Through an LLM-assisted pipeline grounded in the MedDRA terminology, AskGLP1 screens nearly 8.5 million Reddit posts for medication experiences and distills 163,135 of them, with 2,506 WebMD reviews, into a KG of 385,790 drug-side-effect and drug-purpose relations. We find that GLP-1 experiences are source- and drug-specific. Online narratives share the most common gastrointestinal symptoms with FAERS, but they foreground everyday burdens such as fatigue and food aversion, whereas FAERS reports more often include injection-site reactions and selected serious diagnoses such as pancreatitis. We further find that posts report less nausea and vomiting for tirzepatide than for semaglutide, more fatigue for Wegovy than Ozempic and for Zepbound than Mounjaro. Evaluations show that the agent answers with verifiable evidence and declines requests for medical advice. Our findings reveal gaps between what FAERS captures and what people report online, and they call for patient-centered approaches to pharmacovigilance. Code, prompts, and evaluation questions are available at this https URL.
[NLP-60] SVD-Attention: Efficiently Analyzing Attention Blocks in Transformers
【速读】： 该论文旨在解决预训练变换器模型在推理过程中大量计算资源消耗于自注意力（self-attention）模块，而每个注意力头的查询-键交互矩阵在结构上具有低秩特性这一现象所引发的效率瓶颈问题。其核心问题是：在真实数据上，不同层与不同架构中的注意力头究竟有多少真正利用了其完整的秩容量，哪些则存在冗余？为此，论文提出SVD-Attention，一种后处理诊断方法，通过分解每个注意力头的交互矩阵 $ M_h $，并基于小规模校准集上各奇异模式的谱幅值及其跨标记激活方差进行评分，从而筛选出实际被数据激活且可被放大的关键奇异模式，其余则予以舍弃。该方法保留的模式构成精确重构，无需微调。实验表明，该诊断揭示了任务依赖性的冗余分布规律：情感分析、图像分类及生成任务中冗余集中于浅层至中层，而密集预测与抽取式问答任务则集中在深层。此外，该方法具备预测能力——仅对诊断标识的层进行截断，性能损失不超过1.75 F1，而对其他层进行同等数量截断则可能导致高达50 F1的性能下降。值得注意的是，SVD-Attention本身为分析工具，而非压缩手段；单独使用可降低0.8%–1.5%的推理浮点运算量（FLOPs），若与权重剪枝结合，则可进一步实现8%–11%的加速，同时任务性能变化仅为0%–4%，展现出良好的效率-精度平衡。
链接: https://arxiv.org/abs/2610.34971
作者: Debopam Sanyal,Hongjie Chen,Riya Elizabeth Kuriyan,Arth Tak,Divya Kothandaraman,Akshay Mehra,Joshua Kimball,Alexey Tumanov
机构: Georgia Institute of Technology(佐治亚理工学院); Dolby Laboratories(杜比实验室)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Pretrained transformers spend a large share of their inference compute inside self-attention, yet the per-head query-key interaction is structurally low-rank by construction. We ask a diagnostic question: across layers and architectures, which heads actually use the full rank budget on real data, and which do not? We propose SVD-Attention, a post-hoc method that decomposes the per-head interaction matrix M_h and scores every singular mode by both its spectral magnitude and its activation variance across tokens on a small calibration set. The score keeps the modes the head can amplify and that real data actually activates, and discards the rest; the resulting factorization is exact for the retained modes and needs no finetuning. Across seven pretrained models and five task types, the diagnostic finds a task-dependent pattern: redundancy sits in the early to middle layers for sentiment and image classification and for generation, and in the late layers for dense prediction and extractive question answering. It is also predictive: truncating the layers it flags costs at most 1.75 F1, while truncating the same number of layers elsewhere in the same model costs up to 50 F1. SVD-Attention is an analysis tool rather than a compression method: acting on it reduces inference FLOPs by 0.8 - 1.5% on its own and by 8 - 11% when composed with weight pruning, with 0 - 4% change in task quality.
[NLP-61] Improving Reasoning with Rollout Agreement for Group-Based Policy Optimization
【速读】： 该论文旨在解决生成式模型在基于可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR）中，因正确答案轨迹间差异所引发的推理有效性问题。尽管现有方法利用轨迹多样性提升推理性能或训练效率，但其对正确轨迹间的差异是否反映有意义的替代推理路径，还是源于未被发现的错误仍缺乏清晰理解。为此，研究通过盲对比较分析采用组相对策略优化（Group Relative Policy Optimization, GRPO）训练的模型，在同一问题下高一致性和低一致性的正确轨迹进行对比，发现低一致性轨迹中存在更高比例的虚假推理（35% vs. 22%），包括内部矛盾和误用问题条件等现象。基于此，提出一种名为RoAR（Rollout-Agreement Reward）的解决方案，即在GRPO的可验证正确轨迹奖励中引入推理层面的一致性信号，该信号由已有在线策略轨迹组计算得出，无需人工编写推理过程、额外采样或独立训练的流程评估器。在五个数学推理基准测试中，1.5B参数规模的GRPO-RoAR模型在零样本评估下达到55.5%的平均pass@1，分别优于GRPO和GRPO-CARE 6.0和5.5个百分点；进一步的盲审审计显示，四款1.5B模型在100个正确回答问题中的虚假推理率分别为：GRPO-RoAR（18%）、GRPO（32%）、GRPO-CARE（27%）和DRA-GRPO（37%）。研究结果表明，推理一致性是利用轨迹差异指导学习时必须考虑的关键因素，而RoAR有效提升了推理的有效性与可信度。
链接: https://arxiv.org/abs/2610.34951
作者: Sohyun Son,Yerim Hong,Jaeyun Park,Mujeen Sung
机构: Kyung Hee University (庆熙大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 21 pages, 7 figures, 16 tables
Abstract:Recent methods for Reinforcement Learning with Verifiable Rewards (RLVR) use rollout diversity to improve reasoning performance or training efficiency. Despite these gains, the extent to which differences among correct-answer rollouts reflect useful alternative reasoning or unresolved errors remains poorly understood. We investigate this question through a blind paired analysis of a model trained with Group Relative Policy Optimization (GRPO), comparing the highest- and lowest-agreement correct-answer traces for the same problem. Across 126 pairs, low-agreement traces have more spurious reasoning than high-agreement traces (35% vs. 22%), including internal inconsistencies and misuse of problem conditions. Based on these findings, we propose RoAR (Rollout-Agreement Reward), which adds reasoning-level agreement to GRPO’s rewards for verified-correct rollouts. RoAR computes this signal from existing on-policy rollout groups, requiring no human-written rationales, additional sampling, or separately trained process evaluator. On five mathematical reasoning benchmarks, the 1.5B GRPO-RoAR model reaches 55.5% mean pass@1 under zero-shot evaluation using 7K training examples, exceeding GRPO and GRPO-CARE by 6.0 and 5.5 percentage points. A separate blind audit of 100 problems answered correctly by all four 1.5B models finds spurious rates of 18% for GRPO-RoAR, 32% for GRPO, 27% for GRPO-CARE, and 37% for DRA-GRPO. Our findings identify reasoning validity as an important consideration when using rollout differences to guide learning. Our code is available at this https URL.
[NLP-62] Correlation-Aware Exploration for Long-Trace Reasoning NEURIPS2026
【速读】： 该论文旨在解决生成式推理模型在使用组相对策略优化（Group Relative Policy Optimization, GRPO）训练时出现的策略坍缩（policy collapse）问题，尤其是在生成数千词长思维链（chain-of-thought）的情况下，传统熵正则化方法因在每个推理步骤上强制提升全局熵而与收敛所需的熵衰减产生冲突，导致超参数调节困难。其解决方案的关键在于提出一种基于信息论的新型框架——Info-GRPO，通过构建策略与潜在先验（latent prior）之间的互信息（mutual information），利用训练过程中引入的潜在种子（latent seeds）来引导策略探索，使多样性体现在整个推理轨迹的结构相关性上，而非依赖于逐令牌的随机性。这一机制实现了探索与收敛的解耦，避免了熵正则化带来的副作用。实验表明，Info-GRPO在多个推理基准测试中显著优于原始GRPO及熵正则化GRPO，尤其在AIME24数据集上，相对于GRPO在四种不同骨干模型上的平均准确率（Avg@8）分别提升了3.75%、1.66%、4.16%和4.17%，且达到相同多数投票准确率时仅需八分之一的采样量。
链接: https://arxiv.org/abs/2610.34949
作者: Junxing Hu,Zicheng Zhang,Zhiqian Zhang,Ai Han,Yifeng Zhang,Pengzhang Liu,Ling Li
机构: JD.com(京东); Microsoft(微软)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 28 pages, 6 figures. Accepted at the Workshop on Long-Context Foundation Models (LCFM), NeurIPS 2026
Abstract:Reasoning models trained with Group Relative Policy Optimization (GRPO) now produce chains of thought thousands of tokens long, and recent studies reveal that such training is prone to policy collapse. Entropy regularization is the standard remedy, but it works by pushing probability mass away from the few tokens that carry it and toward the vast low-probability tail of the vocabulary, and it applies this force once for every token of the reasoning trace. This puts the bonus in direct conflict with the entropy decay that convergence requires, and makes its coefficient difficult to set in the long-trace regime. To resolve this tension, we propose Info-GRPO, an information-theoretic framework that cultivates correlation between the policy and a latent prior. Rather than raising entropy everywhere, Info-GRPO builds on the mutual information between a latent variable and the trajectory it induces: augmenting prompts with latent seeds during training lets the model explore diverse policies correlated with the prior, while the seed-conditioned entropy is guided toward convergence. Diversity is therefore carried across whole reasoning traces rather than by per-token randomness, and exploration no longer fights entropy reduction. Experiments show that Info-GRPO outperforms vanilla GRPO and entropy-regularized GRPO across diverse reasoning benchmarks. On AIME24, it improves Avg@8 over GRPO by 3.75%, 1.66%, 4.16%, and 4.17% on four backbones, and it reaches the same majority-vote accuracy with 8\times fewer samples. Code and models will be publicly released.
[NLP-63] What Outcome RL Buys: Finishing Reliability Rarely Reach
【速读】： 该论文旨在解决生成式 AI 在复杂推理任务中，尽管在 pass@1（即首次尝试即正确）指标上表现提升，但在大范围 k 值下的 pass@k 指标未同步提升的根本原因。其核心问题是：强化学习（RL）带来的性能提升是否源于行为模式的改变、知识的新增，还是仅限于对现有能力的优化？解决方案的关键在于将 pass@1 的变化分解为三个可量化维度：完成度（Finishing）、可靠性（Reliability） 和 可达性（Reach）。其中，完成度衡量模型在令牌预算耗尽前完成推理的能力；可靠性衡量模型在完成推理后答案正确的概率；可达性则评估 RL 是否解决了其基础模型在全部八次尝试中均无法解答的问题。研究发现，当基础模型常因令牌预算不足而无法完成时，绝大多数性能变化来自完成度的提升（4k/8k预算下占比达82%），且若放宽预算，基础模型自身即可弥补大部分差距；而在有足够资源的情况下，性能提升主要体现为可靠性提高；而真正意义上的新问题解决能力（即可达性）极为有限，最大仅增加2.1个百分点，且多数“新解”在重采样或人工验证后不成立，表明其本质更接近对已有样本的再利用而非真正新增知识。此外，在小规模实验中，引入物理设定信息并未显著提升纯强化学习的表现，进一步说明当前提升主要源自策略优化而非外部信息增强。
链接: https://arxiv.org/abs/2610.34894
作者: Shan Yang
机构: 未知
类目: Computation and Language (cs.CL)
备注: 27 pages, 5 figures, 17 tables. Project page: this https URL Code: this https URL
Abstract:Outcome RL raises pass@1 but not pass@k at large k (Yue et al., 2025). But every pass@1 is scored under a token budget. Every recipe also trains under one. On physics and math problems at fixed token budgets, we compare the pass@1 of 20 released RL models (1.5B to 32B) with that of the thinking models they started from. We ask whether each pass@1 gain is changed behaviour or new knowledge. We divide each change in pass@1 into three parts. Finishing measures answering before the budget runs out. Reliability measures being right once finished. Reach measures problems RL solves that its base answered wrongly in all eight tries, minus problems the base solves that RL answered wrongly in all eight. We find four things. (1) Where the base often hits the token budget before answering, most of the change, gain or loss, is finishing (82% at 4k and 8k). Given more tokens, the base closes most of the finishing gap itself. (2) Where the base has enough tokens, the gain is reliability: the RL model is graded right more often, even on problems both finish. (3) By our definition, RL rarely solves what its base cannot. Reach adds at most 2.1 percentage points on any of our four physics and math sets. We checked 144 physics problems that only the RL model solved. Of these, 3.5% hold up when resampled or read. In a random sample, nine of ten new RL steps a human confirmed also appear in the base’s samples. (4) In a small-scale RL run on an 8B model, the gain on new problems is finishing, with reliability flat. Supplying each problem’s physical setup was not shown to improve on plain RL.
[NLP-64] FATE-5: Faithful Alignment of Trait Expression for Continuous Big Five Personality Control
【速读】： 该论文旨在解决大语言模型中个性控制存在的核心矛盾：即在保持强烈人格特质表达的同时，难以兼顾对话生成的自然性。现有方法往往在人格表现力与对话流畅度之间存在权衡，导致生成内容或过于刻板、或人格特征模糊。为克服这一问题，论文提出FATE-5（Faithful Alignment of Trait Expression for the Big Five），一种两阶段框架，其关键在于将人格表征学习与自然性对齐解耦。第一阶段通过对话监督和辅助问卷目标，学习基于大五人格（Big Five）的连续软提示（continuous soft prompt）表示；第二阶段基于对话质量与人格一致性双重偏好，进行偏好对齐以提升对话自然性。实验表明，FATE-5在三种不同模型架构下均实现了强人格表达与高自然性之间的平衡，在保留人格特质强度的同时显著提升了对话质量，且在留出问卷测试中表现出高达0.97–0.99的单调性相关系数与8.9–12.4的恢复均方根误差（RMSE），显著优于基线方法，验证了其在实现连续人格控制与自然对话行为协同优化方面的有效性。
链接: https://arxiv.org/abs/2610.34889
作者: Gunhee Cho,Hongyeob Kim,Yun-Gyung Cheong
机构: Sungkyunkwan University (成均馆大学); Suwon, South Korea
类目: Computation and Language (cs.CL)
备注:
Abstract:Controlling personality in large language models is important for building conversational agents that can be customized to different users and interaction settings. However, existing personality control methods often face a tradeoff between strong trait expression and natural dialogue generation. We introduce FATE-5, Faithful Alignment of Trait Expression for the Big Five, a two-stage framework that separates personality encoding from naturalness alignment. FATE-5 first learns continuous soft prompt representations of Big Five profiles using dialogue supervision and an auxiliary survey objective. It then improves dialogue naturalness through preference alignment based on both dialogue quality and personality agreement. Across three language model backbones, FATE-5 maintains strong trait expression while achieving substantially higher dialogue naturalness than other trait-strong methods. On held-out questionnaires, it achieves monotonicity correlations of 0.97-0.99 and recovery RMSE of 8.9-12.4, substantially improving over the evaluated baselines. These results demonstrate effective continuous personality control while maintaining natural conversational behavior.
[NLP-65] NumTrace: Auditing the Number Provenance of Financial Analyst Agents
【速读】： 该论文旨在解决金融分析师代理（financial analyst agent）生成的数值主张在看似正确（与公司披露数据匹配）时，仍可能因实体、概念、期间、单位或范围等财务语义层面错误而存在根本性偏差的问题。传统基于参考答案的验证方法仅能判断结果“对错”，却无法解释错误原因。其解决方案的关键在于提出一个名为NumTrace的新基准，专门审计数值主张的来源及其财务语义（provenance and financial semantics）。NumTrace通过构建基于会计准则的参考答案，对每项主张给出三个独立判别：数值正确性（value）、绑定关系正确性（binding）和支撑证据充分性（support）。其中，保留分离的“数值”与“绑定”判别是核心创新——可有效识别“正确数值但错误账户”（Right Number, Wrong Account, RNWA）类错误。实验表明，三大主流模型（Claude Opus 5.5、GPT-6 Astra、DeepSeek-V4.1-Flash）中高达67.5%–81.8%的失败案例为RNWA错误；尤其在13F共同申报组问题中，约91%的主张通过数值检查，但不足一半通过绑定检查，揭示了当前模型在财务语义理解上的显著缺陷。
链接: https://arxiv.org/abs/2610.34887
作者: Tianlei Zhu,Matt White,Xue Liu,Xiao-Yang Liu
机构: Columbia University (哥伦比亚大学); University of California, Berkeley (加州大学伯克利分校); Linux Foundation (Linux基金会); MBZUAI; McGill University (麦吉尔大学); Mila – Quebec AI Institute (魁北克人工智能研究所)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 22 pages, 8 figures, 12 tables
Abstract:A numerical claim generated by a financial analyst agent can match a number in a company filing but refer to the wrong entity, concept, period, unit or scope. A check based on reference answers can mark such an answer as wrong but does not explain why it fails. In this paper, we present NumTrace, a new benchmark that audits the provenance of each numerical claim made by financial analyst agents: its source and financial semantics. NumTrace currently has 540 questions over 2,163 SEC EDGAR documents. To the best of our knowledge, it is the first numerical benchmark about who holds a company’s shares (Form 13F) and what its executives are owed (Item 402(i)). On a frozen snapshot of these documents, we build the reference answers according to accounting rules. NumTrace then gives each claim three separate verdicts: value, binding and support. By retaining separate value and binding verdicts, NumTrace detects right number, wrong account (RNWA) errors: a claim has an RNWA error when it passes the value check but fails the binding check. The RNWA rate quantifies the gap between value-only and joint value-and-binding pass rates. Finally, we audit three agents, Claude Opus 5.5, GPT-6 Astra and DeepSeek-V4.1-Flash. RNWA errors make up 67.5-81.8% of each agent’s failing claims. In particular, on 13F co-filing group questions, about 91% of claims pass the value check, but fewer than half pass the binding check.
[NLP-66] hinkBit: Search Codewords Recalibrate Scales Select by Reasoning
【速读】： 该论文旨在解决生成式语言模型在低精度量化（如两比特量化）下推理能力退化的问题，尤其关注在极低存储开销（如使用2比特替代bfloat16权重）时，如何保持模型对复杂推理任务（如数学问题求解）的处理能力。传统两比特方法通常以困惑度（perplexity）和多项选择题准确率作为评估标准，但这些指标无法有效反映模型在面对更复杂推理任务时的表现。为此，本文提出ThinkBit——一种严格基于标量的两比特量化框架，其核心在于：每个权重为整数码字乘以一组共享的bfloat16尺度，从而在保持参数紧凑性的同时保留推理能力。关键创新包括：（1）基于敏感性分析为各层分配三元、2比特或3比特层级，实现分层最优比特宽度分配；（2）采用基于Gumbel-Softmax的逐层标量量化算法，块级搜索最优码字；（3）通过知识蒸馏损失对冻结码字的尺度进行再校准，以缓解层间交互失配；（4）引入短周期量化感知训练（QAT），使模型通过重新分配码字来模仿原始模型的输出分布；（5）最终通过成对推理测试在再校准与QAT阶段之间选择最优检查点。实验表明，该框架在相同尺寸下显著优于现有2比特方法，在Qwen3-1.7B模型上达到MATH-500 46.7分（对比EfficientQAT的30.0分），且在14B规模下实现94.8分（接近29.5GB bfloat16模型的97.5分），验证了其在压缩与推理能力之间的高效平衡。
链接: https://arxiv.org/abs/2610.34886
作者: Rhea Sanjay Sukthanker,Pashmina Cameron
机构: Microsoft Applied Sciences Group(微软应用科学组)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Reasoning language models generate thousands of tokens per answer, and generation speed is limited by memory bandwidth, so storing weights in two bits instead of bfloat16 can potentially make inference 8 \times cheaper. Two-bit methods are usually judged by perplexity and multiple-choice question-answering accuracy, both of which say little about whether a model can still reason through more difficult problems. We present ThinkBit, a strictly scalar two-bit framework built to preserve reasoning, in which every weight is an integer codeword times a bfloat16 scale shared by a group of weights. We first use a sensitivity metric to assign each layer a ternary, 2-bit, or 3-bit tier to allocate bitwidths optimally among layers. This is followed by a layerwise Gumbel-softmax-based scalar quantization algorithm that searches for optimal codewords block by block. Since a layerwise scheme does not capture how quantized layers interact with each other, we recalibrate scales, with the codewords frozen, using a knowledge-distillation-based loss against the teacher model. We then perform a short quantization-aware training (QAT) stage that lets the model mimic the base model’s distribution by reassigning codewords. Finally, a paired reasoning test on training problems selects the final checkpoint between the recalibration stage and the QAT stage. On Qwen3-1.7B at matched size, our recalibrated checkpoint scores 46.7 on MATH-500 against 30.0 for EfficientQAT trained on the same data, and 73.0 after QAT on reasoning traces; at 0.52 GB it is smaller than every published 2-bit GGUF of the model yet more accurate on MATH-500. We show that this framework scales and up to 14B, a 3.74 GB model scores 94.8 on MATH-500, against 97.5 for the 29.5 GB bfloat16 model.
[NLP-67] RubriX: A Fast Fine-Tunable Non-Autoregressive Rubric Judge
【速读】： 该论文旨在解决基于评分量规（rubric-based evaluation）的评估效率问题，即传统方法依赖大语言模型（LLM）作为评判者时，需对每个评分等级进行文本生成并逐项阅读描述，导致计算开销随量规复杂度线性增长。其核心解决方案是提出RubriX，一种受对比语言模型（Contrastive Language Model, CLM）启发的新型评分量规判别器：通过LoRA微调的编码器将输入响应和各评分等级描述映射为向量表示，并利用缩放后的余弦相似度计算各等级的概率分布，避免了生成式推理。该方法在缓存等级嵌入后，推理延迟稳定在约28毫秒/判断，即使单个等级描述扩展至3000词也保持高效，显著优于基线模型（如Jev和Qwen3-8B裁判），在准确率与速度上均实现显著提升，尤其在大规模量规场景下展现出7.7倍于Jev、18.8倍于Qwen3-8B的性能优势。
链接: https://arxiv.org/abs/2610.34875
作者: Balamurugan Paramasivam(1),Sudalai Rajkumar(1) ((1) CobuildX)
机构: CobuildX(科构建)
类目: Computation and Language (cs.CL)
备注: 12 pages, 6 figures, 3 tables. Model: this https URL
Abstract:Rubric-based evaluation asks which level of a rubric a response meets. It is usually done by prompting a large language model (LLM) as a judge, which needs a generative call per judgment and reads every level description, so its cost grows with the rubric. We present RubriX, a rubric judge whose architecture is inspired by contrastive language models (CLM): a LoRA-adapted encoder embeds the input and each level description, and a softmax over scaled cosine similarities gives a probability for every level without text generation. Trained on public rubric data and proprietary synthetic data, RubriX reaches 75.4% exact accuracy on held-out rubrics, against 57.1% for TypeSafe’s Jev and 49.8% for a Qwen3-8B judge. On eight human-labelled benchmark sets it is within 6 percentage points of Jev’s accuracy on most of them, while Jev keeps a larger lead on two and ranks responses best on average; against the results reported in the G-Eval paper, RubriX outperforms G-Eval with GPT-3.5 on every SummEval dimension and both QAGS subsets. With level embeddings cached, RubriX’s latency stays at about 28 ms per judgment as level descriptions grow from 10 to 3,000 tokens, while Jev’s (approximate network time subtracted) rises from 44 ms to 211 ms and the Qwen3-8B judge’s from 32 ms to 516 ms, making RubriX 7.7x faster than Jev and 18.8x faster than the Qwen3-8B judge at 3,000 tokens per level.
[NLP-68] Adaptive Retrieval Selective Answers: Evaluating Conformal Factuality in Agent ic RAG
【速读】： 该论文旨在解决生成式 AI 在智能体式检索增强生成（agentic RAG）场景下，尽管采用基于检索的相关性评分与事后合取过滤（conformal filtering）策略以提升事实准确性，但仍面临生成内容事实支持率与响应可用性之间权衡的问题。其核心挑战在于：在语言模型政策动态决定是否继续检索或终止检索的自适应流程中，如何在保障每个生成主张均被充分支持的前提下，维持较高的有效响应覆盖率。解决方案的关键在于显式建模自适应检索轨迹，并通过后生成合取过滤实现逐条主张的事实性校验；研究发现，在95%的目标置信水平下，虽可使非空输出的完全支持率高达89.14%–95.56%，但相应导致响应覆盖率下降至17.23%–60.17%，揭示了可靠性（factuality）与信息量（informativeness）之间的内在权衡关系。
链接: https://arxiv.org/abs/2610.34872
作者: Muhammad Aimal Rehman,Chi-Kuang Yeh
机构: Georgia State University (乔治亚州立大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 10 pages, 1 figure, 5 tables. Code: this https URL
Abstract:Retrieval-augmented generation (RAG) can improve factual grounding without guaranteeing that every generated claim is supported. Previous work has studied claim-level conformal factuality in single-stage RAG and, more recently, multi-hop RAG. We examine its behavior in agentic RAG, where a language-model policy adaptively issues follow-up retrieval queries or terminates retrieval before response generation. We retain the retrieval-based relevance-scoring formulation and post-generation conformal filtering and evaluate six model-dataset configurations: GPT-6 Luna and Llama 3.1 8B on HotpotQA, Natural Questions, and TriviaQA. Each configuration uses 60 fixed questions, 100 seeded 30/30 calibration-test partitions, five conformal target levels, and a no-conformal-prediction baseline. Across the six configurations, higher target levels are accompanied by higher observed full-support rates and reduced answer availability. At the 95% target, non-empty response coverage ranges from 17.23% to 60.17%, claim retention from 10.72% to 43.16%, and full support among non-empty outputs from 89.14% to 95.56%. We formulate the adaptive retrieval trajectory explicitly and characterize the reliability-informativeness trade-off when conformal factuality is applied after agent termination.
[NLP-69] DocTalk: Enhancing Medical Knowledge Retrieval via Fine-tuned Large Language Models with Chain-of-Thought Reasoning for Robust Clinical Decision Support
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在医疗领域应用中因幻觉（hallucination）和领域适配偏差（domain misalignment）导致的安全性与有效性问题，尤其是在电子健康记录（Electronic Health Record, EHR）问答任务中的临床可靠性挑战。其核心解决方案是构建一个名为DocTalk的医生辅助问答系统，通过基于去标识化出院小结生成精准、可解释且受病历支持的简明回答。关键创新在于采用高保真数据生成流程，结合句子级分解、思维链引导的问题生成及专家评审，构建了500个具有临床意义的问答对；并引入参数高效指令微调（Parameter-Efficient Instruction Tuning, PEIT），利用低秩适应（Low-Rank Adaptation, LoRA）对六种开源模型进行优化，同时融合提示工程、上下文学习（in-context learning）与医学对齐评估体系，显著提升了模型在词汇、语义及临床真实性等维度的表现。实验表明，经微调后的模型在自建的DocTalk-QA数据集上实现显著性能提升，并在PubMedQA等外部基准上展现出更强的泛化能力，验证了临床语料驱动的指令微调在提升医疗问答系统可信度方面的有效性。
链接: https://arxiv.org/abs/2610.34871
作者: Asad Gohar,Md Ariful Islam Mozumder,Moon-Il Joo,Hee-Cheol Kim
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Objective: The rapid advancement of large language models (LLMs) in healthcare has opened new possibilities for AI-assisted clinical decision-making; however, safe and effective deployment remains challenging because of hallucination and domain misalignment. This study introduces DocTalk, a doctor-assistive question-answering (QA) system designed to generate concise, chart-grounded responses from de-identified discharge summaries. Methods: We developed a high-fidelity data generation pipeline that produces 500 clinically relevant QA pairs through sentence-level decomposition, chain-of-thought-guided question formulation, and expert review. Six open-source LLMs, including three general-purpose and three biomedical models, were evaluated before and after parameter-efficient instruction tuning using Low-Rank Adaptation (LoRA). The framework integrates prompt engineering, in-context learning, and medically aligned evaluation to support factual and interpretable responses. Results: On DocTalk-QA, instruction tuning produced consistent improvements across lexical, semantic, and clinical fidelity metrics. The best-performing tuned model, Llama-3-Med42-8B, achieved ROUGE-L = 0.40, BERTScore-F1 = 0.76, and UMLS-F1 = 0.49. To examine external generalization beyond the DocTalk-QA dataset, the two best-performing models, Llama-3.1-8B-Instruct and Llama-3-Med42-8B, were evaluated on PubMedQA before and after instruction tuning. Accuracy improved from 0.931 to 0.977 for Llama-3.1-8B-Instruct and from 0.935 to 0.943 for Llama-3-Med42-8B. Conclusion: DocTalk was deployed as a secure web-based platform for interactive EHR querying. These findings suggest that clinically grounded instruction tuning can improve medical QA performance and support practical clinical decision support.
[NLP-70] Priority-Preserving Advantage Aggregation for Multi-Reward Policy Optimization
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）后训练过程中多目标优化时出现的优势优先级倒置（advantage priority inversion）问题。在强化学习（RL）中，传统加性奖励聚合方法在处理主目标与辅助目标冲突时，可能因辅助信号过强而扭曲主目标的偏好，导致优化方向偏离预期。尽管奖励层级门控（reward-level gating）可部分保护主目标优先级，但会无差别地丢弃不满足主目标条件样本中的辅助信号信息，造成信息损失。为此，本文提出主优势策略优化（Primary Advantage Policy Optimization, PAPO），一种面向优先级的优势聚合方法，通过非对称融合主优势与辅助优势，在目标一致时保留辅助监督信号，而在冲突时平滑衰减辅助信号，确保融合过程保持主目标的符号一致性（primary-consistent ordering）。该方法适用于基于广义策略优化（GRPO）的序列级优势和基于演员-评论家PPO的词元级优势。实验表明，PAPO在多个任务与模型上均能稳定实现最优的主目标性能，同时在辅助目标表现上达到最优或具有竞争力水平，有效平衡了多目标优化中的优先级与信息利用之间的矛盾。
链接: https://arxiv.org/abs/2610.34862
作者: Xiangchi Yuan,Xingjian Diao,Zheyuan Liu,Chunhui Zhang,Dachuan Shi,Kejing Xia,Zhenbang Du,Shenglong Yao,Soroush Vosoughi,Wenke Lee
机构: Georgia Institute of Technology(佐治亚理工学院); Dartmouth College(达特茅斯学院); University of Notre Dame(圣母大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:LLM post-training often optimizes multiple objectives with unequal priorities, such as correctness and reasoning efficiency. However, the widely used additive reward aggregation in RL can violate these priorities: when primary and auxiliary advantages conflict, a sufficiently large auxiliary signal can overturn the preference induced by the primary objective. We term this problem \emphadvantage priority inversion. Although reward-level gating can protect priority more directly, it still removes potentially useful auxiliary distinctions among rollout samples that fail the primary condition. To address this trade-off, we propose Primary Advantage Policy Optimization (\textbfPAPO), a priority-aware advantage aggregation method that asymmetrically combines a primary and an auxiliary advantage. PAPO retains auxiliary supervision when the objectives agree and smoothly attenuates conflicting auxiliary signals, preserving the primary sign during fusion for \emphprimary-consistent ordering. This rule applies to sequence-level advantages in GRPO and token-level advantages in actor–critic PPO. Experiments on multiple domains and models show that PAPO consistently achieves the best primary objective while maintaining best or competitive auxiliary-objective performance across both GRPO- and PPO-based optimization. \hrefthis https URLCode is available.
[NLP-71] Semantic Atoms for Scalable Text Analysis in Data Analytics in the Era of LLM -Based Agents
【速读】： 该论文旨在解决在数据分析师工作流中，如何高效、准确地处理文本列（text columns）以支持后续数据分析的问题，这类工作负载被称为“文本分析在数据分析中的应用”（TADA, Text Analysis in Data Analytics）。其核心挑战在于：在智能体（agent）驱动的动态查询场景下，如何合理划分神经网络（神经式）与符号计算（符号式）的任务边界，并确定哪些语义输出可被后续步骤复用。为此，论文提出两种互补的、由查询驱动的语义原子（semantic atoms）——标记（Tagging） 与 分类（Categorization）。Tagging 对每个输入文本项返回一个字符串标签，结果以行对齐的列形式物化；Categorization 则生成覆盖所有输入项的类别表，二者均以结构化形式暴露给后续的精确计算（如 SQL 或 Python 操作）。该框架通过 TADA-Agent 实现，其语义原子嵌入在生成的 Python 程序中，并能处理超出可靠上下文窗口限制的输入。在 TABench 基准测试（涵盖 40 个完整分析请求，覆盖工作流组合、输入输出结构及文献依据的分析场景）上，TADA-Agent 在保持与语义操作基线相当的逐行语义质量的同时，显著提升了完整分析产物的保留率，并在原生执行路径下将语义模型调用次数减少了一个数量级以上。
链接: https://arxiv.org/abs/2610.34860
作者: Jinxiang Xie,Zihao Li,Rui Ding,Bartuer Zhou,Shi Han,Dongmei Zhang
机构: Nanjing University (南京大学); Tsinghua University (清华大学); Microsoft Research (微软研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Much of the information needed for data analysis is buried in text columns: an analyst may infer customer frustration from complaint text and use it to explain resolution time. We call workloads in which processing text columns is central to subsequent data analysis TADA (\emphText Analysis in Data Analytics). In an agentic setting, the user’s query dynamically specifies this processing, and answering it interleaves LLM-based semantic steps with exact SQL or Python operations. This mixed execution raises two design questions: which work should be neural and which symbolic, and which semantic outputs later steps can reuse. We address them for agent-planned TADA workflows with two complementary, query-driven semantic atoms. Each atom takes a sequence of text items and a natural-language query specifying the semantic task: \textbfTagging returns one string per input item, while \textbfCategorization returns M category objects over N input items. Tagging results are materialized as a row-aligned column and Categorization results as a category table; SQL or Python then performs exact computation over this exposed state. We realize this interface in TADA-Agent, whose atoms run inside generated Python programs and internally handle inputs beyond a reliable context window. We evaluate TADA-Agent on TABench, a benchmark of 40 complete analysis requests over spreadsheet tables spanning workflow composition, input–output structure, and literature-grounded analytical scenarios. Against a semantic-operator baseline and relative to a held-out TADA reference run, TADA-Agent delivers comparable row-level semantic quality, has higher complete-artifact preservation estimates, and uses over an order of magnitude fewer semantic-model requests under the evaluated native execution paths.
[NLP-72] Position Matters More Than Channel: Talking a Language Model Out of a Correct Answer
【速读】： 该论文旨在探究在语言模型（Language Model）已独立正确回答问题的情况下，何种外部干预因素能够使其偏离正确答案，从而揭示影响模型决策稳定性的关键机制。其核心问题是：当模型自身具备正确推理能力时，哪些提示工程（prompt engineering）设计要素会显著削弱其表现？解决方案的关键在于通过控制变量实验，系统分离并量化三个此前常被混淆的因素——提示中建议内容的位置、建议的来源归属（权威性或非权威性角色）、以及模型先前生成内容的性质。研究发现，位置效应具有决定性影响：对于仅陈述结论的“裸命题”（bare claim），其危害性最强的位置是在模型即将作答前；而若附加虚假论证，则最致命的干扰出现在问题之前，可使原本几乎零错误的题目错误率从1%飙升至67%。此外，将建议归因于权威主体并采用肯定语气，可分别导致错误率上升10–27个百分点，且不同权威之间的差异不足2个百分点，表明权威性本身的作用强于具体身份。最后，无论正确示例以模型自述还是他人视角呈现，提前复现正确解答均能有效保护其不被后续错误提示所干扰，而相同机制却无法纠正已被错误回答的内容。这些发现强调了提示结构设计中的时空顺序与权威暗示在操控模型输出方面的重要性，并为构建鲁棒性更强的生成式 AI 系统提供了实证依据。
链接: https://arxiv.org/abs/2610.34809
作者: Xinyi Ni
机构: Brandeis University (布兰戴斯大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Language models now answer questions alongside other claims: a note in the system prompt, a message from another agent, their own earlier answer. We ask what it takes to talk a model out of an answer it gets right on its own, and separate three factors that earlier studies vary together: where the advice sits in the prompt, who is said to have written it, and what the model has already written. On Qwen3.5-27B and four-option MMLU-Pro items, with the rest of the advice held fixed in each comparison, we find three things. First, position matters far more than the API channel, in opposite directions for a bare claim and an argument: a bare “the answer is X” does most harm just before the model answers, while the same claim with a false rationale does most harm before the question, where it raises error on items the model otherwise almost never misses from 1% to 67%; moving the claim between the system message and the user message at the same position changes error by a few points. Second, attributing the note to an authority rather than to another participant, and phrasing it assertively, each raise error by 10-27 points, while the two authorities we tested differ by under two. Third, replaying a correct worked answer before the advice protects it whether the text is shown as the model’s own or as someone else’s, and the same replay keeps wrong answers from being corrected by a bare correct note. The position result replicates under an independently written prompt. Code, items and all result tables are released at this https URL.
[NLP-73] xt Is the Bottleneck: A Small-Scale Replication of LLM Game Playing in Riichi Mahjong
【速读】： 该论文旨在解决在极低计算资源约束下（最多20亿参数模型、单块16 GB消费级显卡、约五天GPU算力）是否仍可实现对复杂纸牌游戏——立直麻将（riichi-mahjong）的高效学习与高性能表现的问题。其核心挑战在于：尽管数据和标签与先前研究保持一致，但大型语言模型（LLM）在文本理解与决策生成方面存在瓶颈，导致性能受限。解决方案的关键在于采用轻量级、专用的神经网络架构（如卷积神经网络与小规模Transformer），通过端到端的策略学习直接从游戏状态编码中预测最优动作，而非依赖文本提示。实验表明，仅490万参数的Transformer模型在171万决策上达到94.7%的教师一致性，显著优于0.8B参数的LLM（p=0.005），且在对战中与强敌“Mortal”达到统计等效水平（TOST检验，p=6e-4）。进一步发现，针对教师策略进行训练带来的性能提升（+2.5分）远超参数量增加2.5倍的效果（+1.5分），说明模仿学习的有效性远高于单纯扩大模型规模。然而，关键结论指出：即使在教师一致性上存在显著差异，这种差异也无法转化为实际对局中的显著胜率优势（置信区间内无统计显著性），揭示了“高一致性”不必然等价于“高竞技实力”的现象。此外，研究还修复了原始论文中因代码与论文描述不一致导致的胜利过滤器问题，并提出分治流水线方法，使自动生成的日语解释在句级精确度达到93%（无事实信息时仅为43%）。最终，为遵守Tenhou平台协议，研究仅发布代码、配置及评估流程，不提供训练权重。
链接: https://arxiv.org/abs/2610.34791
作者: Mingge Xu
机构: The University of Tokyo (东京大学)
类目: Computation and Language (cs.CL)
备注: 12 pages, 3 figures, 2 tables. Replication of the riichi-mahjong track of arXiv:2509.01328 . Code, configurations and ~3,200 arena game logs released; no trained weights (Tenhou data terms)
Abstract:We replicate the riichi-mahjong track of “Can Large Language Models Master Complex Card Games?” (arXiv:2509.01328) under a drastically smaller budget: models of at most 2B parameters, one 16 GB consumer GPU, about five GPU-days. (1) With matched data and labels, text is the bottleneck: on the same 100k-decision training set, a 2.6M-parameter convolutional board encoder agrees with the strong AI Mortal on 80.6% of test decisions, above every text-prompt LLM we train (best 77.8%), and places better at the table (2.85 over 1,000 games vs. ~3.0 for the LLMs; paired difference at n=100 not significant). Scaled to 1.71M decisions, a 4.9M-parameter Transformer reaches 94.7% agreement, beats the 0.8B LLMs on identical seeds (-0.45 ranks, sign test p=0.005), and is statistically equivalent to parity against three Mortal champions (TOST within 0.10 ranks, p=6e-4); parity against a teacher-opponent is roughly the ceiling of pure imitation. (2) Training toward the teacher raises teacher agreement more than a 2.5x parameter increase does (+2.5 vs. +1.5 points); on human-record agreement both effects are +1.0, and neither is visible at the table. (3) Few-point agreement differences do not resolve into playing strength within a family: +2.5 agreement points coincide with a paired placement difference of +0.02 +/- 0.12 ranks (p=1.0), and +2.0 points at 2B with -0.06 (p=0.37) - both below the ~0.25-rank resolution of 100 paired games. We also document a paper-code mismatch in the original winner filter, reconstruct its two private engine patches, and show that a divide-and-conquer pipeline makes template-generated Japanese rationales 93% sentence-exact (43% without facts). Per Tenhou’s terms we release no trained weights - only code, configurations, and evaluation protocols.
[NLP-74] BehaviorMem: Building and Evaluating User Memory from Real-World E-Commerce Behavior Streams
【速读】： 该论文旨在解决个性化系统在电子商务场景中如何有效构建持续性用户行为记忆（continual behavioral user memory）的问题，尤其关注如何同时实现对用户显式行为与隐式交互所蕴含的客观证据和长期兴趣的准确记忆，以及基于这些记忆进行深层次需求与偏好的推理。现有方法多集中于语言模态交互的记忆建模，且评估缺乏真实行为数据支持，难以区分记忆与推理能力的独立表现，导致无法有效衡量两者协同机制。本文提出的关键解决方案是：构建一个统一的评估框架与记忆架构——BehaviorMemBench 与 BehaviorMem。其中，BehaviorMemBench 是基于真实电商日志（覆盖1,194名用户、超过三 年的17,462个行为批次）的基准测试集，具备自动标注的未来行为标签及任务无关的记忆接口，可独立评估记忆与推理能力；而 BehaviorMem 则是一种双轴状态架构的记忆设计，通过显式行为事实与推断出的用户特征在长短期跨度上的分层组织，实现基于证据的形成、持续维护及查询无关的读取。实验表明，该方法在记忆与推理两项能力上均优于现有代理型记忆模型，并揭示不同方法在两类能力上的性能差异，验证了其在下游任务中的有效性。
链接: https://arxiv.org/abs/2610.34790
作者: Sizhe Wang,Xinghao Fang,Rui Qiu,Zhenjia Ma,Dagui Chen,Kun Kuang,Hongcheng Guo
机构: Fudan University (复旦大学); Zhejiang University (浙江大学); Taobao Tmall Group, Alibaba (淘宝天猫集团，阿里巴巴); Alibaba (阿里巴巴)
类目: Computation and Language (cs.CL)
备注:
Abstract:User memory helps personalized systems build on their understanding of users over time. In e-commerce, users convey their preferences and needs through both explicit language and implicit interactions with products. We study memory built from logs of these interactions, examining how it can \emphremember concrete evidence and established interests, and use that evidence to \emphinfer underlying needs and preferences that inform judgments beyond well-observed product domains. Existing memory methods primarily target language-based interactions, and evaluations offer limited real behavioral data and limited separation of these capabilities. This leaves open how to jointly support and separately evaluate remembering and inference from behavioral logs. We formulate the problem of continual behavioral user memory around these joint requirements and develop a dedicated evaluation framework and memory method to address them. \textbfBehaviorMemBench is a benchmark built from real e-commerce logs covering 1,194 users and 17,462 behavioral batches over more than three years, with automatic future-behavior labels and a task-independent memory interface to separately evaluate remembering and inference under a common protocol. \textbfBehaviorMem is a memory design that organizes explicit behavioral facts and inferred user characteristics across long-term and short-term scopes in a two-axis state, with evidence-grounded formation, continual maintenance, and query-independent readout. Experiments across two builders and two task readers show that BehaviorMem leads the evaluated agentic memories on remember and infer, and reveal different method rankings across the two capabilities and competitive statistical baselines. Validation on a public dataset further supports its usefulness for downstream tasks based on behavioral histories.
[NLP-75] Natural Jammers: When Non-Robust Features Become Antagonistic NEURIPS2026
【速读】： 该论文旨在解决深度神经网络在干净、未修改图像上仍表现不佳的问题，其核心在于揭示非鲁棒特征（non-robust features）并非简单的“捷径”，而是具有主动误导作用的自然干扰源（Natural Jammers）。传统观点认为这些特征是模型在干净数据上取得高准确率的偶然依赖，但本文提出，它们实际上会通过特定机制干扰模型决策，导致错误预测。解决方案的关键在于发现并验证“频谱-空间锁定”（Spectral-spatial Locking）现象：当高频成分与特定空间结构精确对齐时，才会引发干扰效应；一旦通过频域滤波或微小几何变换破坏该锁定关系，模型即可恢复正确预测。进一步揭示的“替换悖论”（Substitution Paradox）表明，尽管单个干扰源对几何扰动敏感，但整体干扰水平在变换下保持稳定，仅发生重新分布而非消除。这表明自然非鲁棒性并非孤立缺陷，而是一种存在于模型-数据经验几何结构中的可转移性失效场，其本质是系统性的而非局部的。
链接: https://arxiv.org/abs/2610.34785
作者: Jiaming Zhang,Fuyao Zhang,Pengjun Xie,Wei Yang Bryan Lim
机构: Nanyang Technological University (南洋理工大学); Alibaba Group (阿里巴巴集团)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: NeurIPS 2026 Spotlight
Abstract:Deep neural networks are known to rely on non-robust features that align poorly with human perception. Prior work has largely framed these features as opportunistic shortcuts that help standard models on clean data. We show instead that non-robust cues can be antagonistic: they can act as Natural Jammers that actively mislead standard models on clean, unmodified images. Using disagreement between a standard model and robust models as an analysis lens, we identify a Jammed Set on which robust models outperform standard models. We provide causal evidence for Spectral-spatial Locking: jamming arises only when high-frequency components are precisely aligned with specific spatial structures. Disrupting this lock through spectral filtering or small geometric shifts can restore correct predictions. Crucially, we uncover a Substitution Paradox: while individual jammers are fragile to geometric shifts, the aggregate prevalence of jamming remains approximately stable and is redistributed rather than eliminated under such transformations. Our results reframe natural non-robustness as a failure field over the empirical model-data geometry: one that redistributes rather than disappears under small transformations.
[NLP-76] DynEdit: Dynamic Entropy-Guided Sequential Editing for Large Language Models NEURIPS2026
【速读】： 该论文旨在解决大语言模型中长序列知识更新的效率与精度问题，特别是现有参数编辑方法在处理长文本时存在的局限性：单点编辑覆盖范围不足，而固定分块编辑则采用均匀分配编辑力度的方式，无法针对信息密集且难编辑的高难度区域进行优先优化。其解决方案的关键在于提出一种动态编辑框架DynEdit，通过引入**令牌熵（token entropy）**作为衡量编辑难度与语义重要性的代理指标，实现两个核心创新：一是基于熵的非均匀滑动窗口机制，增强对高熵区域（即难编辑且富含关键事实的区域）的覆盖；二是采用熵引导的令牌级优化策略，在每个编辑窗口内强化对信息密度高的令牌的更新。实验结果表明，DynEdit在三个基准测试和两种主干模型上均显著优于现有基线，尤其在复杂改写查询任务中最高提升14.0 BLEU分数，并在多个知识领域接近饱和的语义评分，验证了其在长序列知识编辑中的有效性与普适性。
链接: https://arxiv.org/abs/2610.34768
作者: Jinhu Fu,Yan Bai,Yihang Lou,Li Sun,Jiahong Wu,Xiangxiang Chu,Sen Su
机构: Beijing University of Posts and Telecommunications(北京邮电大学); Peking University(北京大学); AMAP, Alibaba Group(阿里巴巴集团AMAP); Chongqing University of Posts and Telecommunications(重庆邮电大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted by main conference of NeurIPS 2026 (poster)
Abstract:Knowledge editing aims to update outdated or incorrect knowledge in large language models without full retraining. Existing parameter-editing methods remain limited for long-sequence knowledge: single-point editors provide insufficient coverage for extended targets, while fixed chunk-based editors distribute editing effort uniformly across rigid predefined windows, failing to prioritize difficult, information-dense tokens. In this work, we show that token entropy is closely associated with both editing difficulty and semantic importance: high-entropy regions are harder to edit and often contain key factual content. Motivated by this observation, we propose DynEdit, a dynamic framework for long-sequence knowledge editing. DynEdit uses token-level entropy to guide non-uniform sliding, increasing the coverage of difficult high-entropy regions, and further applies entropy-guided token-level optimization to strengthen updates on information-dense tokens within each editing window. Experiments on three benchmarks and two backbone models show that DynEdit consistently outperforms competitive baselines across diverse long-form editing settings, achieving gains of up to +14.0 BLEU points on challenging paraphrase queries and near-saturated semantic scores on several diverse knowledge domains.
[NLP-77] From Belief Inference to Task Action: A Multi-Level Evaluation of Theory of Mind on Verifiable Tasks
【速读】： 该论文旨在解决当前生成式人工智能（Generative AI）在协作任务中缺乏对他人心理状态（Theory of Mind, ToM）的准确理解与有效利用的问题。现有理论心理模型评估多依赖故事问答类基准，但此类任务仅验证了模型对心智状态的推断能力，未能检验其是否能将这些推断转化为后续决策与行为。为此，论文提出一种多层次、可验证的评估框架，通过在家庭协作与日程安排任务中引入信息不对称场景，在保持原有约束条件和验证机制的基础上，构建独立接口以分别评估世界认知、信息获取、信念建模、决策制定及执行能力。通过设置对照实验，提供个体证据或正确信念以区分“信念恢复”与“信念使用”的差异。实验结果表明，不同模型配置在各项能力上表现分化显著：例如，Claude虽能正确恢复所有代理的信念，但在匹配日程场景中仅在67.9%的情况下生成主观最优计划；而Qwen-27B在提供信念信息后成功率从55.0%提升至75.0%，但其40亿参数版本虽能正确回答预设信念问题，却仅在20.0%的延续性任务中完成目标。这揭示了现有评估体系的局限性，强调必须将视角推断与独立可验证的决策及任务结果相连接，才能真实衡量模型在复杂协作情境中的真正心智建模与应用能力。
链接: https://arxiv.org/abs/2610.34759
作者: Zhenshuai Yin,Yongxin Guo,Mingda Zhang,Zhongze Luo,Haoyue Liu,Lianmin Zhang,Xiaoying Tang
机构: The Chinese University of Hong Kong (Shenzhen)(香港中文大学（深圳）); ByteDance Inc.(字节跳动); Shenzhen Research Institute of Big Data(深圳市大数据研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Effective collaboration requires language models to understand what others know and believe, and to act accordingly. Models that assume others share their information may omit necessary communication and undermine coordination. Many theory-of-mind (ToM) benchmarks assess mental-state inference through story questions, but correct answers do not establish that models can use these judgments in subsequent tasks. We introduce a multi-level evaluation built on existing, verifiable tasks. We extend household collaboration and calendar scheduling with asymmetric information while retaining their original constraints and validators. Independent interfaces assess world understanding, information access, beliefs, decisions, and execution. Controls supply personal evidence or correct beliefs to distinguish state recovery from state use. Across six model configurations, these abilities diverge. On matched scheduling cases, Claude recovers all agent beliefs correctly but produces subjectively optimal plans in only 67.9% of cases. In household collaboration, supplying beliefs raises Qwen-27B success from 55.0% to 75.0%, yet Qwen-4B answers every fixed supplied-belief source question correctly while completing only 20.0% of the corresponding continuation condition. These results show why ToM evaluation must connect perspective inference to independently verified decisions and task consequences.
[NLP-78] Mechanistic Insights into LoRA: Layer Sparsity for Adaptive Fine-Tuning via Path Patching NEURIPS2026
【速读】： 该论文旨在解决低秩适配（Low-Rank Adaptation, LoRA）在微调大语言模型时存在的效率瓶颈问题，即标准全层LoRA在变压器（Transformer）结构中均匀施加适应性调整，而未考虑各组件与层级对任务的实际贡献差异。其核心问题是：不同任务下关键的网络层存在显著异质性，导致全层适配造成大量冗余参数更新。解决方案的关键在于提出一种基于校准驱动的自适应方法——Adaptive-LoRA，通过冻结基础模型的零消融（zero-ablation）策略，自动识别任务相关的特定层，实现无需穷举搜索的高效层选择。实验表明，在Qwen2.5-1.5B模型上针对11个多样化推理任务进行评估，Adaptive-LoRA仅使用30%的层数即可超越全层LoRA性能，同时将适配器参数量减少3.5倍，有效融合机制可解释性分析与实际效率优化，为自适应微调提供了理论依据与实践范式。
链接: https://arxiv.org/abs/2610.34756
作者: Fei Zuo,Yizhou Huang,Kezhi Wang
机构: East China Normal University(华东师范大学); Brunel University of London(布鲁内尔大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Accepted at NeurIPS 2026 (Main Track). 31 pages, including appendices and the paper checklist
Abstract:Low-Rank Adaptation (LoRA) has emerged as the dominant paradigm for parameter-efficient fine-tuning of large language models. Standard all-layer LoRA applies adaptation throughout the transformer, despite heterogeneous contributions from its components. In this work, we conduct a systematic mechanistic interpretability analysis to understand how LoRA affects model internals. We observe that LoRA’s adaptation effect is highly non-uniform, with significant component-wise and layer-wise sparsity. Most critically, we find that important layers are highly task-dependent, with limited overlap across tasks. Motivated by these insights, we propose Adaptive-LoRA, a calibration-driven method that automatically identifies task-specific layers via frozen-base zero-ablation, enabling efficient layer selection without exhaustive search. Extensive experiments on Qwen2.5-1.5B across 11 diverse reasoning tasks demonstrate that Adaptive-LoRA outperforms full-layer LoRA while using only 30% of layers, achieving a 3.5\times reduction in adapter parameters. This work bridges mechanistic interpretability and practical efficiency, providing both fundamental insights and a principled approach for adaptive fine-tuning.
[NLP-79] DecisionTraceBench: Tracking Decision-Critical Information in Multi-Party Spoken Conversations
【速读】： 该论文旨在解决高风险领域（如金融）中多主体交互环境下决策信息状态演变的联合追踪问题，现有语音评估基准在转录、局部理解及回合级状态预测方面表现良好，但对跨说话人决策事件序列的联合演化缺乏充分评估。其核心解决方案是提出DecisionTraceBench，一个面向决策关键信息状态追踪（Decision-Critical Information State Tracking, DCIST）的语音状态追踪基准，通过轨迹级F1分数衡量完整事件链的恢复能力，综合评估决策事件序列、说话人归属、作用范围及时间证据。该数据集包含2至4名说话人的中文金融对话，涵盖100.39小时合成音频与8.61小时真人录音。在固定真人录音样本上，五种级联系统和两种直接音频模型的DCIST得分仅为5.73%至21.35%，即使在去除时间定位要求并使用带说话人标注的参考脚本的情况下，最高得分也仅达38.72%，表明当前方法在联合理解信息溯源与状态演化方面存在显著局限。音频扰动实验进一步揭示，整体转录错误率无法有效反映完整轨迹恢复情况，凸显了构建更全面评估框架的必要性，为高风险场景下决策过程理解和可追溯性提供了关键基准。
链接: https://arxiv.org/abs/2610.34746
作者: Wenchao Du,Fangxin Shang,Long Wu,Runze Cui,Qing Yang,Yehui Yang
机构: AI Lab, Qifu Technology(启富科技), Beijing, China; University of Chinese Academy of Sciences(中国科学院大学)
类目: Computation and Language (cs.CL)
备注: Project repository: this https URL
Abstract:In high-stakes domains such as finance, critical information evolves through multi-party interaction. Recovering these changes and their sources is fundamental to reliable speech understanding. Existing speech benchmarks focus on transcription, local understanding, and turn-level state prediction, leaving joint decision evolution across speakers insufficiently evaluated. We introduce DecisionTraceBench, a speech state-tracking benchmark for Decision-Critical Information State Tracking (DCIST). It jointly evaluates decision event sequences, speaker attribution, scope, and temporal evidence, using trajectory-level F1 to measure complete event-chain recovery. Our dataset comprises Chinese financial dialogues involving two to four speakers, with 100.39 hours of synthetic audio and 8.61 hours of human-performed recordings. On a fixed cohort of human-performed recordings, five cascaded systems and two direct-audio models achieve DCIST scores of 5.73% to 21.35%. Ablations with speaker-labeled reference scripts and no temporal localization requirements yield a maximum score of only 38.72%. Audio perturbation experiments further show that overall transcription error rates do not adequately reflect complete trajectory recovery. These findings reveal limitations in jointly understanding information provenance and state evolution, providing a benchmark for evaluating decision-process understanding and decision traceability in high-stakes speech applications.
[NLP-80] LADDER: Diagnosing Domain-Level Self-Assessment Under Noisy Feedback
【速读】： 该论文旨在解决语言模型在人机协作中如何准确评估自身能力的问题，尤其关注在交互反馈不可靠或不完整时，模型能否形成可信的自我认知。其解决方案的关键在于提出并使用LADDER框架，系统性地研究语言模型在面对不可靠反馈时的自我评估行为。研究发现，尽管模型能够准确复述提供的性能信息，但在基于实际交互经验进行自我评估时仍存在显著偏差，表面准确的评估结果可能掩盖真实的能力差异，且不具备对未来任务表现的有效预测能力。这表明，可信的自我评估必须基于与实际任务表现的对照，并具备预测价值。该研究为评估模型在可靠人机协作中的能力感知水平提供了实证依据，揭示了当前语言模型在自我知识方面的局限性。
链接: https://arxiv.org/abs/2610.34737
作者: Zhenshuai Yin,Jianshu Liang,Xuyue Xiong,Yongxin Guo,Mingda Zhang,Zhongze Luo,Haoyue Liu,Lianmin Zhang,Xiaoying Tang
机构: The Chinese University of Hong Kong (Shenzhen)(香港中文大学（深圳）); Shenzhen Research Institute of Big Data(深圳市大数据研究院); ByteDance(字节跳动)
类目: Computation and Language (cs.CL)
备注:
Abstract:Knowing when they are reliable is essential for language models to support human-AI collaboration. Yet interaction feedback is not always complete or correct, and may mislead models’ understanding of their own capabilities. We introduce LADDER to study how language models assess themselves under unreliable feedback. Our findings show that models can accurately report supplied performance information while still struggling to form trustworthy self-assessments from interaction experience. Apparently accurate assessments can obscure capability differences and do not necessarily predict future task performance. These findings suggest that trustworthy self-assessment should be evaluated against observed task performance and its predictive value. Our study provides empirical evidence of the limits of model self-knowledge and a basis for evaluating capability awareness in reliable human-AI collaboration.
[NLP-81] PrefBench: A User-Centric Benchmark for Personalized Mobile Use Agents
【速读】： 该论文旨在解决个性化移动使用代理（personal mobile-use agents）在面对用户请求中信息不完整（underspecified）时，如何有效推断缺失信息并适时向用户提问以实现精准服务的问题。其核心挑战在于：代理需在不做出无依据假设的前提下，从用户的历史行为中推理出可推断的信息（history-inferable），同时识别出必须由用户提供的信息（ask-only），并在交互中恰当地发起询问。解决方案的关键在于构建一个可执行的iOS环境下的基准测试平台iPrefBench，该平台基于包含3,051条历史活动事件的连贯人格模型，设计了130个跨26个应用的不完整任务，确保每个缺失项均为可推断或仅能通过提问获取。通过评估前沿代理在“个性化任务完成率”与“用户-代理交互合理性”两方面的表现，发现当前代理虽具备较高的动作执行成功率（最高达91.6%），但整体任务成功率偏低（最高仅47.7%），尤其在处理需提问的信息时严重不足，暴露出代理在“何时推断、何时提问”上的能力短板。因此，iPrefBench为开发真正理解用户并协同工作的智能代理提供了关键评估工具与研究方向。
链接: https://arxiv.org/abs/2610.34735
作者: Yuting Ning,Ruijia Cheng,Sushobhan Ghosh,Peitong Duan,HungShi Lin,Abhishek Sundararajan,Shreyash Pandey,Huan Sun,Jeffrey Nichols,Eldon Schoop
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:An agent that truly knows and acts on behalf of its user has been a long-standing promise of AI assistants. Computer-use agents bring this promise within reach by capably executing fully specified tasks across apps. However, real-world user requests are often underspecified. To act correctly for the particular user, an agent must infer some missing information from the user’s activity history and ask for other information that only the user can provide, without making unsupported assumptions or burdening the user with unnecessary questions. We present iPrefBench, a benchmark for evaluating the infer and ask capabilities of personal mobile-use agents in an executable iOS environment with an interactive simulated user. Grounded in a coherent persona with 3,051 past activity events, iPrefBench comprises 130 underspecified user tasks across 26 apps. Each omitted user-dependent requirement is, by construction, either history-inferable or ask-only. iPrefBench evaluates personalized mobile agents on both whether agents complete tasks correctly for the user (i.e., personalized task completion) and whether they seek user input when necessary (i.e., user-agent interaction). Evaluation of 7 frontier agents reveals a substantial gap between execution and personalization. Agents achieve high action completion scores (up to 91.6%) but low full task success (up to 47.7%) because personalization lags behind. They especially struggle with ask-only information, failing to request enough necessary information. Our findings highlight a challenge for personalized agents in knowing what to infer and when to ask. We hope that iPrefBench can serve as a valuable testbed for building agents that know their user and work with them.
[NLP-82] AdaTutor: Adaptive Tutoring through Pedagogical Strategy Selection
【速读】： 该论文旨在解决智能辅导系统中如何动态选择最优教学策略与反馈形式的问题，即在面对学生错误时，如何根据互动进程自适应地调整教学方法以提升学习效果。其核心解决方案是提出一种**比较式教学监督（Comparative Pedagogical Supervision, CPS）**训练框架，通过对比不同教学策略的效果，让模型学习到在特定情境下选择最有效的反馈方式。实验表明，在数学与科学推理数据集上，经过五轮辅导后，CPS显著优于固定策略的基线模型；同时，所学得的教学策略具备跨数据集迁移能力，并可为基于API的独立辅导系统提供策略指导。
链接: https://arxiv.org/abs/2610.34659
作者: Rajeev Bhatt Ambati,Shashank Srivastava,Snigdha Chaturvedi
机构: UNC Chapel Hill(北卡罗来纳大学教堂山分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Effective tutoring requires determining not only what information to provide, but also which form of assistance to deliver. A tutor may address the same student error using various pedagogical strategies, adapting its approach as the interaction progresses. We introduce Comparative Pedagogical Supervision (CPS), a training framework to learn to select the optimal strategy and feedback for student learning. We evaluate CPS on math and science reasoning datasets. After five tutoring turns, CPS outperforms the strongest fixed-strategy baselines. The learned tutor transfers across datasets, and its strategy selection can also inform a separate API-based tutor.
[NLP-83] Heard Enough to Act: Early Tool Calling in Spoken Dialogue Systems ICASSP2027
【速读】： 该论文旨在解决低延迟语音代理在对话中难以确定请求可执行最早时刻（earliest actionable point, EAP）的问题。现有语音代理与对话语料库均未明确定义请求何时具备可执行性，导致系统常需等待整个话语回合结束才响应，造成不必要的延迟。其核心解决方案是构建一个合成式语音对话生成管道，通过受控任务生成、语音合成与对齐、以及门控标注机制，提供精确的EAP监督信号，标记出语音内容首次足以触发工具调用及其所需参数的时刻。基于此，研究构建了包含48个领域、总计786小时、26,694条对话的高质量数据集，并创建了τ²-bench的音频版本用于外部评估。在此基础上训练的全模态语言模型能够实现因果性、逐块级决策和部分转录生成，支持边听边决策。实验结果表明，该模型在保留对话与外部基准上的工具状态宏平均F1得分分别达到0.48和0.47，且在正确调用情况下，相较于以回合结束为触发点的基线方法，平均提前1.8–2.0秒（域内）和1.1秒（τ²-bench）完成调用，验证了基于数据驱动的方法在实现“听觉中行动”（acting while listening）方面的有效性。研究将开源该合成管道与训练脚本。
链接: https://arxiv.org/abs/2610.34653
作者: Yizhou Peng,Chen Chen,Donghang Wu,Hexin Liu,Eng Siong Chng
机构: 未知
类目: Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注: Submitted to ICASSP 2027
Abstract:Latency-sensitive spoken agents must know not only which tool to call, but also when they have heard enough to act. Waiting for the turn to complete delays execution even when the request is already actionable. However, neither existing spoken agents nor spoken dialogue corpora define the earliest point at which a request becomes actionable. We introduce a synthetic spoken-dialogue pipeline that provides earliest-actionable-point supervision, marking when the observed speech grounds a tool and its required arguments. Through controlled task generation, speech synthesis and alignment, and gated annotation, we construct 26,694 dialogues totaling 786 hours across 48 domains, with disjoint training and test argument values. We additionally create an audio rendering of \tau^2 -bench for external evaluation. We train an omni-modal language model on this corpus, enabling causal, chunk-level decisions and partial-transcript generation for downstream call completion. The model achieves tool-state macro-F1 scores of 0.48 and 0.47 on held-out dialogues and the external benchmark, respectively. For correct calls, the estimated mean lead over a turn-end cascade, using the same formatter, is 1.8-2.0 seconds in-domain and 1.1 seconds on \tau^2 -bench. These results demonstrate a data-driven path toward spoken agents that act while listening. We will open-source the pipeline and the training scripts.
[NLP-84] Cross-Loop Query Handoff: How Later Loops Redirect Two-Hop Retrieval in the Ouro Looped Language Models
【速读】： 该论文旨在解决预训练循环语言模型（pretrained looped language model）在两跳检索（two-hop retrieval）任务中，后期循环（later loops）如何修复早期循环中错误的检索结果这一问题。其核心挑战在于理解并利用后期循环中可转移的信息来纠正前期循环中的错误推理路径，尤其是在缺乏显式桥接实体（bridge entity）提示的情况下。解决方案的关键在于通过在相同层、头和最终提示位置之间移植后期循环（如loop 3）的注意力查询（attention queries），发现一组可转移的、与层索引相关的晚期查询能够显著提升早期错误回答的修复率：在合成上下文问题上，将loop 3的查询注入loop 1可使原本在loop 1错误但在loop 3正确的90%和97%的问题被成功修复，远优于随机打乱捐赠者（shuffled donors）的24%和21%。此外，交叉交换（crossed donor-recipient swaps）表明模型更倾向于通过捐赠者的桥接实体进行检索而非直接复制答案，但位置信息在Ouro-2.6B中仍起重要作用。值得注意的是，后期键（keys）和值（values）的迁移效果较差，且冻结loop 1的查询会比提供匹配问题更严重地损害两跳准确性。在自然问题上，修复效果有限且仅在Ouro-2.6B中部分复现，交叉交换亦无法建立桥接重定向。综上，研究识别出一组可转移的晚期循环查询，支持在合成场景下的桥接条件检索，但未揭示桥接实体的具体计算位置，也未实现单一向量的交接机制。
链接: https://arxiv.org/abs/2610.34639
作者: Ziyu Mu,Zehua Wang
机构: The University of British Columbia (不列颠哥伦比亚大学); Department of Electrical and Computer Engineering (电气与计算机工程系)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Code and reproducibility: this https URL 16 pages, 1 figure, 7 tables
Abstract:How do later loops in a pretrained looped language model repair two-hop retrieval? We study Ouro-1.4B and Ouro-2.6B by transplanting attention queries between loops at the same layer, head and final prompt position. On synthetic in-context questions, injecting the layer-indexed collection of loop-3 queries throughout loop 1 rescues 90% and 97% of questions answered incorrectly at loop 1 and correctly at loop 3, versus 24% and 21% for shuffled donors. The rescue also holds under greedy generation in Ouro-2.6B; in Ouro-1.4B it is limited to four-candidate ranking. Crossed donor-recipient swaps favor retrieval through the donor’s bridge entity over copying its answer, although a substantial positional component remains in Ouro-2.6B. Late keys and values rescue far fewer targets. Freezing queries at their loop-1 values also damages two-hop accuracy more than a matched question that supplies the bridge. On natural questions, rescue is smaller and replicates only in Ouro-2.6B, and crossed swaps do not establish bridge redirection. These results identify a transferable collection of late-loop queries that supports bridge-conditioned retrieval on synthetic scenes; they do not locate where the bridge is computed or establish a single-vector handoff.
[NLP-85] Entity Matching using Decision Models ATC
【速读】： 该论文旨在解决实体匹配（Entity Matching）任务中现有方法的局限性：基于预训练语言模型（PLM）的方法需要特定任务微调，而生成式大语言模型（LLM）虽无需微调但存在推理速度慢、成本高的问题。为此，论文提出将决策模型（Decision Models）作为生成式LLM与微调PLM的替代方案。其核心解决方案在于利用决策模型在零样本（zero-shot）场景下直接输出预定义答案选项的概率分布，而非生成文本，从而兼顾性能与效率。实验表明，四种表现最优的决策模型在无任务特定训练数据的情况下，平均F1得分接近甚至略超生成式LLM GPT-6 Luna，且在八项基准中的六项上优于基于PLM的Ditto模型；同时，所有决策模型均显著快于GPT-6 Luna，多数成本更低。此外，其中三个最强模型的预测概率具有良好的校准性，无需额外校准。通过构建模型级联（model cascade），将置信度最低的记录对路由至生成式LLM集成系统，可进一步提升决策模型的平均F1得分1.9至4.0个百分点，并在四项基准上超越已有文献报道的最佳性能。
链接: https://arxiv.org/abs/2610.34626
作者: Aaron Steiner,Ralph Peeters,Christian Bizer
机构: University of Mannheim(曼海姆大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 15 pages, 2 figures, 15 tables. Code, prompts, and results: this https URL
Abstract:Entity matching is the task of deciding whether two records describe the same real-world entity. Current entity matching methods rely on pre-trained language models (PLMs) which require task specific fine-tuning or generative large language models (LLMs) which do not require task-specific fine-tuning but are slow and expensive at matching large record sets. Decision models are a new class of LLM-based models that return a probability for each predefined answer option rather than generate text. This paper investigates decision models as an alternative to generative LLMs and fine-tuned PLMs for entity matching. We evaluate six decision models in a zero-shot setting across eight benchmarks and compare their performance with that of the generative LLM GPT-6 Luna and the PLM-based matcher Ditto. Our experiments show that, without task-specific training data, the four strongest decision models achieve mean F1 scores close to that of GPT-6 Luna, with the best model slightly surpassing it. This model also outperforms Ditto on six of the eight benchmarks. All decision models are faster than GPT-6 Luna, and most are less expensive. For three of the four strongest decision models, the predicted match probabilities are well calibrated without requiring any task-specific recalibration. A model cascade that routes the least confident record pairs to an ensemble of generative LLMs improves the mean F1 scores of the four strongest decision models by 1.9-4.0 percentage points. Using the best decision model and its best-performing prompt for each benchmark, the cascade surpasses the highest published F1 scores from literature on four of the eight benchmarks.
[NLP-86] One Mean Log-Mel Vector per Phoneme: Simple and Fast Text Injection for Speech Recognition
【速读】： 该论文旨在解决语音识别模型在缺乏配对语音-文本数据时的训练瓶颈问题，通过引入无配对文本的文本注入（Text Injection）方法，使模型能够利用大量未配对的文本数据进行有效训练。其核心解决方案是用一个简单的查找表（lookup table）替代原先依赖的冻结文本到语音（Text-to-Speech, TTS）合成器，具体实现为：每个音素对应一个均值对数梅尔倒谱系数（mean log-mel vector）或一热编码向量，并根据对齐统计分布随机采样持续时间，词间插入静音。该方法显著提升了训练效率，达到神经TTS方案1.8倍的速度提升，同时保持相近的词错误率（Word Error Rate, WER）。此外，结合外部语言模型，在LibriSpeech数据集上实现了1.36% / 2.90% / 1.52% / 3.24%的WER（dev/test-clean/other），据作者所知为仅使用LibriSpeech配对数据、文本及词典的最先进水平。研究还系统分析了合成输入中持续时间与声学特征的影响，比较了不同隐马尔可夫模型（HMM）对齐器性能，评估了在25,000小时Loquacious数据集上的可扩展性，并验证了全编码器注入（full-encoder injection）相较于部分编码器或解码器级注入具有更显著的相对性能提升。整个实验流程的代码均已公开。
链接: https://arxiv.org/abs/2610.34623
作者: Albert Zeyer,Davyd Naveriani,Dorian Koch,Ralf Schlüter,Hermann Ney
机构: AppTek(应用科技); NVIDIA(英伟达)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
备注:
Abstract:Text injection trains speech recognition models on unpaired text alongside paired audio. We feed the speech encoder synthetic input derived from text so the whole model learns from it. Previously, we synthesized this input with a frozen text-to-speech (TTS) model. Here, a simple lookup table replaces the TTS model: a mean log-mel vector (or one-hot encoding) per phoneme, repeated for a random duration drawn from alignment statistics, with silence between words. Training is 1.8 times faster than with the neural TTS, achieving comparable word error rates. With an external language model, our system reaches 1.36 / 2.90 / 1.52 / 3.24 % WER on LibriSpeech dev,test-clean,other, to our knowledge state-of-the-art using only LibriSpeech paired data, text, and lexicon. We analyze components of synthetic input (durations, acoustic features), compare different HMM aligners, evaluate scaling on the 25 000 h Loquacious dataset, and show that full-encoder injection yields strong relative improvements compared to partial-encoder and decoder methods. The code of the whole pipeline behind every number in this paper is public: this http URL
[NLP-87] Does the Outfit Change the Price? Context Halo in Vision-Language Model Valuation of an Identical Low-Cost Necklace
【速读】： 该论文旨在解决生成式AI在视觉商品估价任务中受上下文影响的偏差问题，具体表现为同一物品在不同使用场景下的价格估计存在显著差异。其核心发现是：尽管物品本身未变，但当图像呈现于正式着装、派对、回收场或平铺静物等不同情境时，六种前沿语言模型对其估价差异高达1.9至3.57倍，且这种效应在文本描述任务中同样存在，说明其根源并非像素级视觉线索，而是模型对语境暗示的过度敏感。解决方案的关键在于识别并验证“上下文诱导的价格放大”机制——即模型会将穿戴场景与社会价值符号（如正式性、时尚感）关联，从而系统性高估物品价值。研究进一步表明，仅通过简单指令要求模型忽略服饰与环境可部分缓解该偏差（如某模型比率降至0.77），但对多数模型效果有限；而提供照片来源信息（公司官网或个人账号）则无显著影响。这揭示了当前生成式AI在客观评估任务中存在深层认知偏见，亟需通过多对象、多观察者基准实验进行可重复验证，并推动更稳健的上下文去偏策略发展。
链接: https://arxiv.org/abs/2610.34591
作者: Brianne Lee
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL)
备注: 10 pages, 1 figure, 4 tables. Code and data: this https URL
Abstract:A chunky gold-tone chain necklace bought online for 2.43 USD was photographed four ways: worn with formal attire against a wood-panelled wall, worn at a party, worn with casual clothing in a recycling yard, and laid flat with no person present. We asked six frontier language models to estimate its price in 1,164 independent and sequential API sessions (4,604 logged rows) and compared estimates across settings. Among the four models that answered image prompts, the same necklace was priced 1.9 to 3.3 times higher in the formal photograph than in the flat-lay (all p 0.01). Against the yard photograph, the ratio ranged from 1.23 (Grok 4.5, not significant) to 3.57 (Kimi K3), with Claude Fable 5 at 3.32. A text-only version of the task, in which the setting is described in words, produced formal-to-yard ratios of 2.2 to 3.9 in all six models, so the effect does not require pixels. Flat-lay estimates themselves were 7.8 to 18.7 times the receipt price, so part of any headline “outfit multiplier” computed against the receipt is baseline overestimate, not a context effect. A single instruction to ignore clothing and setting significantly reduced the effect in one model (ratio of ratios 0.77, 95% CI 0.65 to 0.91) and not detectably in the others; telling the model the photo came from a company website or a personal account changed nothing significant. Asked afterward whether a different outfit would have changed their number, Claude Fable 5 said yes in 24 of 24 sessions, GPT-5.6 in 69%, and GPT-4o in 18%, while GPT-4o also refused 79% of image estimation prompts. The study isolates a mechanism with one object and one wearer at about 10 sessions per model and condition; we outline the replication needed next, with multiple objects and a human benchmark. Data, code, and raw outputs are released.
[NLP-88] Large Language Models : A Plain-Language Overview for Engineers
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际应用中面临的透明性与可解释性问题，尤其关注训练过程、提示工程及模型微调等环节对生成结果的影响。其核心挑战在于：如何清晰区分模型能力的来源——是来自预训练阶段的通用知识，还是微调或提示设计带来的特定行为；同时应对数据泄露（data contamination）、幻觉（hallucination）、长上下文处理能力不足以及评估偏差等关键局限。解决方案的关键在于系统性地梳理并整合多种先进方法：通过引入检索增强生成（Retrieval-Augmented Generation, RAG）提升事实准确性，利用链式思维提示（Chain-of-Thought Prompting）促进推理过程可追溯性，采用基于人类反馈的强化学习（Reinforcement Learning from Human Feedback, RLHF）优化响应质量，并结合FlashAttention与量化低秩适配（Quantized Low-Rank Adaptation, QLoRA）分别降低计算开销与微调内存占用。此外，论文提出一套评估检查清单，强调独立验证事实准确性与判断回答实用性，以实现对大语言模型性能的客观、可靠评估。
链接: https://arxiv.org/abs/2610.34585
作者: Bahaeddin Türkoğlu
机构: Ankara University (安卡拉大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 11 pages, 2 figures, 3 tables, 16 references. Author’s preprint; to be submitted to a peer-reviewed journal
Abstract:Large language models (LLMs) are neural networks trained on huge text collections to generate text by predicting what comes next. Engineers using ChatGPT and other generative artificial intelligence (AI) tools need a clear picture of what training changes and prompts add. The article starts with the Transformer, which connects words in context, and pre-training, where models learn to predict text. It then explains how fine-tuning adapts models to requests, with reinforcement learning from human feedback using human answer rankings. Next, retrieval-augmented generation (RAG) supplies documents, while chain-of-thought prompting encourages intermediate steps and agents use tools. FlashAttention lowers training and running costs by reducing data movement during attention; quantized low-rank adaptation (QLoRA) saves fine-tuning memory. Readers get a life cycle figure and tables comparing methods and their limits. A short checklist for judging LLM evaluation claims flags test material in training (data contamination) and model-based grading (LLM-as-a-judge). It also covers very long inputs (long context) and hallucination, meaning false or unsupported claims. Check factual accuracy separately from whether an answer sounds helpful.
[NLP-89] STEP: Soft Trajectory Estimation of Prediction Correctness via Per-Step Probability Scoring EMNLP2026
【速读】： 该论文旨在解决在缺乏真实标签（ground truth）的情况下，如何可靠评估大语言模型（Large Language Model, LLM）预测正确性的关键问题，尤其针对高风险应用场景中的可信部署需求。现有方法如熵（entropy）和困惑度（perplexity）仅依赖最终输出的统计特征，忽略了推理过程中蕴含的丰富中间信息。为此，本文提出一种无需训练的框架——STEP（Soft Trajectory Estimation of Prediction correctness），其核心在于通过轻量级前向传播对每一推理步骤生成候选答案的软概率分布 $ q_i $，并据此构建一组捕捉每一步置信度及其向最终答案演进过程的软轨迹特征（soft trajectory features）。实验结果表明，该方法在多个基准测试中显著优于基于输出的基线方法及硬指标轨迹方法，尤其在AUROC指标上表现突出。其中，时间加权一致性（Temporally-Weighted Consistency）通过赋予推理后期更稳定的答案分布更高权重，带来了最大性能提升，凸显了推理轨迹中时间结构的重要性。这一成果进一步验证了保留连续概率信号相较于二值化处理的优势。
链接: https://arxiv.org/abs/2610.34537
作者: Chenyang Yu,Zifan Zeng,Qunli Zhang,Feng Liu
机构: Technical University of Munich (慕尼黑工业大学); RAMS Lab, Huawei Heisenberg Research Center (华为海森堡研究中心)
类目: Computation and Language (cs.CL)
备注: Accepted at the Third Workshop on Uncertainty-Aware NLP (UncertaiNLP 2026) at EMNLP 2026
Abstract:Estimating the correctness of large language model (LLM) predictions without access to ground truth is critical for reliable deployment in high-stakes settings. Existing approaches, such as entropy and perplexity, operate solely on final output statistics, discarding the rich information encoded in intermediate reasoning steps. To address this limitation, we propose STEP (Soft Trajectory Estimation of Prediction correctness), a training-free framework that scores each reasoning step via lightweight forward passes to obtain a soft probability distribution qi over candidate answers, from which we derive a set of soft trajectory features capturing per-step confidence and its progression toward the final answer. Extensive experiments on various benchmarks demonstrate consistent and substantial improvements, particularly in AUROC, over output-based baselines and hard-indicator trajectory methods. Notably, Temporally-Weighted Consistency, which assigns higher importance to later reasoning steps where answer distributions have stabilized, contributes the largest gains, highlighting the importance of temporal structure in reasoning trajectories. These results further validate the advantage of preserving continuous probability signals over binary discretization.
[NLP-90] Does the Voynich Manuscript Mean Anything? Calibrated Tests Find Little Trace of Language
【速读】： 该论文旨在解决《Voynich手稿》文本是否为无意义的随机构造，抑或隐藏着可破译的真实语言这一长期争议性问题。其核心挑战在于：如何在缺乏已知解码规则的情况下，通过统计证据区分“人为虚构的抄写产物”与“经过加密的真实语言文本”。解决方案的关键在于将两种假设分别建模为可执行程序——一种模拟逐字复制并轻微变异的生成过程（如Timm和Schinner提出的抄写模型），另一种则基于不同类型的密码机制（如单密钥替换、多密钥分段加密、词类编码等），并在相同的计算预算下对这些程序进行拟合与验证。研究发现，单纯的抄写模型在预测文本的词频统计特征方面显著优于单一密钥的加密模型；然而，若采用针对手稿中两种方言切换密钥的加密方案，或仅对真实文本的词类进行编码，则其预测能力与抄写模型相当，表明仅依靠词频统计无法明确区分二者。因此，研究进一步引入“词语共现模式”作为语言结构的独立指标——在16种语言共28个文本中，特定词对频繁出现在同一行的现象普遍存在，而《Voynich手稿》虽表现出该模式但强度仅为最弱参考文本的一半。当使用词码（word code）模型时，该模式仍能被保留，前提是明文本身具有较强结构且误抄率较低。实验表明，在多种密码类型（包括字母替换、词码及28种每词对应一个字母/字母对/音节的编码）下，均无法从控制文本中解出可读内容，暗示手稿可能未包含可识别的明文信息。值得注意的是，以缩略拉丁语为明文的词码模型能更清晰地揭示词序关系，除非超过五分之二的词汇发生误抄。最终，唯一仍可能成立的模型是带有第二密钥的名目密码（nomenclator），其制作者误抄了超过一半的词汇，且当前解码方法无法在16种语言中的9种中排除该可能性。此外，解码结果表明，大多数可观测到的结构源于词尾与词首之间的连续性，这反映了符号之间紧密绑定的书写习惯。只有当消息足够长、每个字母始终映射为同一词且词间相邻时，才能可靠检测到嵌入信息，这意味着其他潜在信息可能仍隐藏于手稿之中。
链接: https://arxiv.org/abs/2610.34499
作者: Xin Zhang,Victor S. Sheng
机构: Texas Tech University (德克萨斯理工大学)
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注:
Abstract:The Voynich manuscript is a fifteenth-century book in a script that no one can read. One account holds that its text means nothing and that a scribe made it by copying his own words and changing them a little. The other holds that a real language lies hidden behind a cipher. We write each account as a program that fills the pages of the manuscript, fit every program with the same budget, and check it on statistics the fit never saw. The copying process of Timm and Schinner predicts these statistics far better than any cipher with one key. But ciphers that switch keys between the manuscript’s two dialects, or that encode only the word classes of a real text, predict them equally well. Word statistics alone cannot decide. We look for a trace of language that needs no decipherment. In 28 texts in 16 languages the same word pairs keep sharing a line. The manuscript shows this pattern at half the strength of the weakest of these texts. A word code keeps the pattern above the manuscript unless its plaintext has a weak pattern or many words are miscopied. Solvers that decipher control texts find no readable text in the manuscript under a letter cipher, a word code or 28 codes with one letter, letter pair or syllable per word. Word codes of abbreviated Latin, the plaintext their fits prefer, show which word follows which more clearly than the manuscript unless two words in five are miscopied. One cipher remains possible, a nomenclator with a second key whose scribe miscopies more than half of the words, and our solvers cannot exclude it in 9 of 16 languages. Most of the structure a solver finds comes from how the end of one word shapes the start of the next, which reflects how tightly glyphs are bound to each other. Planted messages are found reliably only when they are long, each letter always becomes the same word and these words are adjacent, so other messages could hide.
[NLP-91] What Effort Level Should I Choose? On the Economics of Reasoning Effort in LLM Agents
【速读】： 该论文旨在解决生成式模型推理过程中“推理努力程度”（reasoning effort）选择的不确定性问题。在当前生成式 AI 模型中，用户需预先决定推理计算资源的投入水平，但无法预知额外计算是否真正提升性能，且即使任务成功完成，也无法判断所支付的计算成本是否具有边际价值。由于这种投入的边际效用不可观测，推理努力本质上是一种“可信品”（credence good）。为此，研究通过在四个基准测试上对专有模型与开源权重模型进行系统性评估，量化不同努力水平的反事实影响。研究发现，高努力水平并非总是提升性能：它不仅可能丢失低努力即可解决的任务，还存在过度消耗计算资源的问题，并显著改变智能体的任务求解轨迹——如更频繁地自我验证和搜索而非直接推理，从而延长推理路径但未必改善结果。实验表明，简单采用从最低努力起步、仅在失败后逐步升级的策略，在所有模型与基准上均能以更低成本实现或超越固定最高努力水平的表现，证明了动态调整策略在成本效率与实例级性能上的优越性。
链接: https://arxiv.org/abs/2610.34491
作者: Kaiser Sun,Yongchan Kwon,Zain Hasan,Shang Zhu,Kaitlyn Zhou,Federico Bianchi
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Generative models increasingly expose a reasoning effort setting that lets users scale inference-time compute, trading higher cost for potentially improved performance. Users must choose an effort level before knowing whether extra reasoning will help, and even a successful run does not reveal whether the effort purchased was necessary. Reasoning effort is therefore a credence good: buyers observe the outcome but cannot assess the marginal value of the computation they paid for. In this work, we measure the counterfactual effect of effort levels on both proprietary and open-weight models on four benchmarks, spanning single-call reasoning and long-horizon agentic tasks. Effort, it turns out, does not consistently preserve tasks that lower effort solves, can overspend, and can change trajectory behavior. First, we find that although higher effort can raise accuracy in aggregate, it frequently fails on tasks that lower effort solved. Second, on the subset of easy tasks that every effort level solves, higher effort is generally more costly, with the highest effort level costing up to 42x as much as the lowest. Lastly, effort changes not only how many tokens are used, but also how agents solve tasks: at higher effort, agents verify their own work more heavily and more often search for solutions than work through problems directly, both of which lengthen trajectories without necessarily altering the outcome. These findings show that naively choosing the highest level is not only cost-inefficient but also offers no guarantee of the best instance-level performance. A simple alternative, starting at the lowest effort and escalating only after a verified failure, matches or improves on fixed top effort across all models and benchmarks at lower cost for most proprietary models.
[NLP-92] Efficient Proof Generation for Lean Statements with Multiple Candidate Answers
【速读】： 该论文旨在解决形式化系统（如Lean 4）中自动化数学推理时，传统多候选答案求解流程中存在的计算冗余问题。在传统方法中，每个候选答案需独立进行定理证明搜索，导致大量重复计算开销。其核心解决方案是提出一种联合证明框架（joint proof framework），将候选答案选择、答案补全与形式化证明生成整合为统一的推理过程。该框架的核心是一个经过训练的证明生成器（prover），能够在单次调用中同时完成答案选择（包括采纳已有答案或提出新答案）与至少一个答案的可验证证明生成。此外，该框架可通过复用已生成的证明来检测语义不忠实性：若在相同形式化命题下，两个不同答案均能获得可验证证明，则表明该形式化陈述未能忠实保留原始问题的语义。在MATH-500、AMC 2023以及AIME 2024/2025/2026数据集上，该框架在保持与逐候选证明生成相当的形式化证明成功率的同时，仅消耗约六分之一的令牌成本，并且无需额外人工评判即可识别出至少6.1%的形式化陈述存在语义失真问题。
链接: https://arxiv.org/abs/2610.34490
作者: Pengzhen Lin,Ke Zhang,Maziar Raissi,Zhouxing Shi
机构: University of California, Riverside(加州大学河滨分校)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Code and model: this https URL
Abstract:In formal systems such as Lean 4, efficiently finding answers to mathematical problems and generating verifiable proofs is an important task in automated mathematical reasoning. In the traditional way where an upstream solver provides multiple candidate answers, the common workflow requires searching for proofs for each candidate separately, incurring redundant computational cost. To this end, we propose a joint proof framework that integrates candidate answer selection, answer supplementation, and formal proof generation into a unified reasoning process. At its core is a trained prover that selects given or proposes new answers and generates a proof for at least one answer in a single call. Joint proof can also reuse generated proofs to check for faithfulness errors. For problems with a unique answer, if two distinct answers both obtain verifiable proofs under the same formal statement, this confirms the statement does not faithfully preserve the original problem semantics. On MATH-500, AMC 2023, and AIME 2024/2025/2026, our framework achieves the same formal proof success rate as per-candidate proof generation with about one sixth of the token cost, and identifies at least 6.1% of the formalized statements as unfaithful without requiring additional judges. Our model and code are available at this https URL.
[NLP-93] AndroidFlux: Evaluating and Reward Modeling Failure Recovery Capabilities in Mobile Use Multimodal Agents
【速读】： 该论文旨在解决移动设备上多模态智能体在动态环境中因中间错误导致任务失败后，缺乏有效恢复能力的评估与训练问题。传统评估方法仅关注从干净初始状态下的任务成功率，无法衡量智能体在非理想状态或未知环境中的鲁棒性与容错能力。其核心解决方案在于提出AndroidFlux——一个基于AndroidWorld构建的故障恢复基准测试框架，通过将已记录的任务轨迹回放至首次可恢复错误前、中、后的可控检查点（checkpoint），实现不同智能体在相同设备状态和历史路径下进行公平比较。同时，引入互补的奖励建模基准，基于下游滚动执行结果评估恢复动作的偏好，从而量化恢复性能。此外，研究还发布了离线强化学习（Offline Reinforcement Learning, RL）数据集，显著提升了错误恢复能力和整体任务成功率。实验表明，恢复性能与无错误启动的成功率仅有部分相关性，且显式推理仅对经过图形用户界面（GUI）领域特定强化学习训练的模型有效，揭示了当前智能体在复杂动态场景中自主纠错能力的局限性。
链接: https://arxiv.org/abs/2610.34384
作者: Gyubeum Lim,Yemo Koo,Ghunwoo Jang,Zsolt Kira
机构: Georgia Institute of Technology(佐治亚理工学院); Konkuk University(韩国中央大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Mobile-use multimodal agents are typically evaluated by task success from a clean initial state, which does not isolate their ability to recover from intermediate errors. Because mobile environments change dynamically, a general mobile-use agent must complete tasks robustly even from unexpected states and in unseen environments. This requires knowing when and where an agent makes errors, and whether it can recover from errors and subsequent states. To evaluate this capability, we introduce AndroidFlux, a failure-recovery benchmark built on AndroidWorld. AndroidFlux restores recorded trajectories to controlled checkpoints before, at, and after the first recoverable error, so that different agents continue from the same device state and history. On these checkpoints, a complementary reward-modeling benchmark evaluates recovery-action preferences using downstream rollout outcomes. We further release an offline reinforcement learning (RL) dataset that improves both error recovery and overall task success. Our evaluations show that recovery performance only partially correlates with clean-start success, and that explicit reasoning helps only models trained with GUI-domain specific RL.
[NLP-94] Whether or Where? Tail Calibration Signals Finite-Budget Reward Hacking in Best-of-N Selection
【速读】： 该论文旨在解决生成式 AI（Generative AI）在最佳- N（Best-of-N, BoN）推理过程中，因代理奖励（proxy reward）的尾部偏差导致性能下降的问题。具体而言，当增大采样数量 $ N $ 时，BoN 依赖于代理奖励的极端尾部表现，而该尾部可能因模型校准偏差或噪声分布特性出现非单调退化，从而使得高 $ N $ 下的输出质量反而劣于低 $ N $。其解决方案的关键在于揭示了精确的 BoN 值可表示为基于 Beta 权重的条件真实奖励在同提示内得分分位数上的积分，表明 BoN 仅依赖于得分的相对排名而非绝对值。进一步推导出：只有当尾部校准曲线呈向下倾斜趋势时，BoN 才会下降，且衰退发生的预算阈值取决于该下倾点的位置。在受限加性误差模型下，误差分布的尾部特性决定了最终行为——高斯误差持续提升，拉普拉斯误差趋于平台，重尾误差则回退至先验均值。尤其当真实奖励较低时误差方差最大，即使高斯噪声亦可能导致 BoN 低于随机选择。通过合成相图验证了该分类体系的有效性；在涵盖多个可验证领域、基础策略及学习型奖励模型的真实候选集上，尾部失败极为罕见，且全部集中于代码生成任务。此外，一个保留的尾部校准统计量能可靠预测跨拟合预算上限的收益，其预测能力与排名准确率相当，且该预算上限在性能上可媲美 Soft-BoN 基线。因此，尾部校准并非替代传统排序指标，而是一种不受任何单调变换影响的审计工具，能够识别出存在风险的预算设置。
链接: https://arxiv.org/abs/2610.34375
作者: Cheng Gui,Yihua Shao,Lei Li,Jingcai Guo,Maolin Wang
机构: City University of Hong Kong(香港城市大学); University of Science and Technology Beijing(北京科技大学); University of Washington(华盛顿大学); Hong Kong Polytechnic University(香港理工大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 28 pages, 3 figures
Abstract:Best-of-N (BoN) inference returns the candidate with the highest proxy reward, so a larger N tests the proxy’s extreme tail more than its average ranking quality. We show that the exact BoN value is a Beta-weighted integral of conditional true reward over within-prompt score quantiles. It follows that BoN depends on scores only through their ranks. It also follows that BoN can decline only if the calibration curve in the tail turns downward, and the budget where the decline appears depends on where that turn lies. Under a bounded additive model, the error tail sets the limit: Gaussian errors keep improving, Laplace errors reach a plateau, and heavy-tailed errors return to the prior mean. When error variance is largest at low true reward, even Gaussian noise can push BoN below random selection. Synthetic phase maps agree with this taxonomy. On real candidate pools spanning several verifiable domains, base policies, and learned reward models, tail failures are rare and all occur in code generation. A held-out tail calibration statistic reliably predicts where a cross-fitted budget cap helps. It does about as well as rank accuracy, and the cap is competitive with a Soft-BoN baseline. Tail calibration therefore does not replace ranking metrics. It is an audit that is unchanged by any monotone rescaling of the reward scores and shows which budgets are at risk.
[NLP-95] Beyond the Opening: Tracing the First Tokens Persistent Influence on Reasoning Length
【速读】： 该论文旨在解决生成式推理过程中初始词元（first token）如何持续影响推理长度这一关键问题。其核心发现是：尽管仅替换推理序列的首个词元，其余上下文保持不变，但后续推理长度仍发生显著变化，表明初始选择具有持久性影响。解决方案的关键在于揭示这种影响的机制——通过分析不同首词元条件下的下一个词元分布，研究发现注意力敏感位置集中于少数特定生成阶段。当屏蔽这些敏感位置对首词元的注意力时，不同首词元条件间的推理长度差异显著缩小；反之，在这些位置使用另一首词元条件的分布也能显著改变后续推理长度。这表明，首词元通过在少数关键生成步骤中维持注意力连接，从而持续影响整体推理长度，揭示了局部词元选择如何通过注意力机制塑造全局序列行为。
链接: https://arxiv.org/abs/2610.34350
作者: Ruochen Liu,Hao Yan,Ziluowen Luo,Senzhang Wang,Sihong Xie
机构: The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)); Central South University (中南大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Understanding how individual token choices affect reasoning length is important for characterizing reasoning models. Building on prior findings that changing the first reasoning token (hereafter, the first token) can substantially alter reasoning length, we focus on this initial choice to investigate how its influence operates during subsequent generation. One possibility is that the first token changes the early text that follows, which in turn affects reasoning length. However, replacing only the first token of an existing reasoning prefix and resuming generation substantially changes the remaining reasoning length, even though all other prefix tokens are unchanged. To trace this persistent influence, we compare next-token distributions while varying only the first token and find that sensitivity is concentrated at a small set of generation positions. Blocking attention from these positions to the first token reduces the length gap between first-token conditions. Using next-token distributions from a different first-token condition at these positions also substantially changes subsequent reasoning length. Together, these findings suggest that attention to the first token at a few sensitive positions helps sustain its length effect, illustrating how a local token choice persistently influences a sequence-level outcome.
[NLP-96] Xing USP: A Cross-Family Semantic Contract for Information Extraction and Intent-Slot Parsing
【速读】： 该论文旨在解决结构化语义解析中仅关注类别识别正确性而忽视值恢复与实例级关联的问题，即系统可能正确识别事件或意图，却将论元绑定至错误实例或提供不准确的槽位值。其解决方案的核心是提出USP（Unified Semantic Parsing），一种跨任务家族的统一语义契约，通过统一接口同时表征类别、字段、值及实例关联，涵盖命名实体识别、关系抽取、事件抽取和意图-槽位解析四大任务。USP配套五种诊断匹配视图——类别一致性、字段识别、字段-值配对、实例绑定分配与完整实例恢复——从不同维度对同一输出进行精细化评估。基于此契约，研究将61个数据集归一化为共享模式、提示和匹配规则的统一语料库，并在固定测试子集上对比六种模型。实验表明，微调后的Xing4.0-29B-Intent模型在任务平衡的字段-值微观F1上达到57.70，且在所有任务族和五个视图中均排名第一；其相对于最强替代方案的性能差距从类别匹配的2.93点扩大至完整实例匹配的20.73点，揭示了仅依赖类别一致性的评估会严重低估结构缺陷。因此，USP为社区提供了可衡量、可比较的统一表示框架、标准化语料库及诊断评估协议，显著提升了结构化语义解析中结构性一致性的可评估性。
链接: https://arxiv.org/abs/2610.34343
作者: Zhongqiu Li,Boyang Wang,Ruiyu Fang,Shiquan Wang,Di Wu,Jiayi Liang,Feilong Zhang,Hongfeng Du,Xiaomeng Huang,Mengjiao Bao,Luwen Pu,Xuewei He,Yu Zhao,Chao Wang,Kaidong Yu,Bingkai Yang,Qingjun Zhu,Shaoyi Du,Shuangyong Song,Yongxiang Li
机构: Xingchen AGI Lab, China Telecom Artificial Intelligence Technology (Beijing) Co., Ltd(中国电信人工智能技术（北京）有限公司); State Key Laboratory of Human-Machine Hybrid Augmented Intelligence, Institute of Artificial Intelligence and Robotics, Xi’an Jiaotong University(西安交通大学人工智能与机器人研究所人机混合增强智能国家重点实验室)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 25 pages, 5 figures, 15 tables
Abstract:Structured semantic parsing is typically judged by whether the correct category is assigned, which leaves value recovery and instance-level associations unexamined: a system may recognize an event yet bind an argument to the wrong instance, or identify an intent yet supply an incorrect slot value. We present USP, a cross-family semantic contract that represents categories, fields, values, and instance associations through a common interface for named-entity recognition, relation extraction, event extraction, and intent-slot parsing. The contract is paired with five diagnostic matching views—category agreement, field recognition, field-value pairing, instance-bound assignment, and complete-instance recovery—that ask distinct questions about the same output. Using this contract, we normalize 61 datasets into a unified corpus with shared schemas, prompts, and matching rules, and compare six models on a fixed test subset. Our fine-tuned Xing4.0-29B-Intent attains 57.70 task-balanced Field-Value Micro-F1 (equally weighted across the four task families) and ranks first in every family and on all five views; its margin over the strongest alternative widens from 2.93 points for category matching to 20.73 points for complete-instance matching. This widening gap shows that category agreement alone understates structural deficits. USP thus offers the community a shared representation, a normalized 61-dataset corpus, and a diagnostic evaluation protocol that make structural agreement measurable and comparable across extraction and intent-slot tasks.
[NLP-97] Improving the Chain-of-Thought Compressibility of Pruned Reasoning Models via Long–Short Contrastive Pruning
【速读】： 该论文旨在解决大模型推理过程中因模型规模庞大和思维链（Chain-of-Thought, CoT）过长导致的高服务成本问题。现有方法中，权重剪枝（weight pruning）与思维链压缩（CoT compression）虽分别针对模型体积与推理长度进行优化，但二者存在协同干扰：剪枝会引入冗余推理（对仍能正确求解的问题）或不完整推理（对已无法求解的问题），且在处理更复杂问题时还会降低后续基于偏好（preference-based）的长度优化可压缩性。作者指出，这一问题源于两方面的“盲点”——剪枝标准无法区分必要推理与冗余推理，而基于偏好的压缩方法在剪枝后无法对模型始终正确或始终错误的问题提供有效的训练信号。为此，论文提出对比集合差分（Contrastive Set-Difference, CSD）方法，通过对比同一问题在长正确解与短正确解下的神经元激活差异，识别并优先剪除在冗余推理中富集的神经元；同时将偏好样本扩展至剪枝后模型始终正确或始终错误的问题。实验表明，在对 DeepSeek-R1-Distill-Qwen-7B 模型剪枝 20% 的情况下，该联合方案在四个数学基准上比先使用推理感知校准（RAC）再进行压缩的方法减少 8–22% 的思考令牌消耗，同时保持精度在 0.5 点以内，验证了其高效性与有效性。
链接: https://arxiv.org/abs/2610.34336
作者: Tao Jing,Chen Kang,Feifei Zhai,Pengyuan Liu,Dong Yu
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 11 pages. Preliminary version; further experiments and revisions are planned
Abstract:Large reasoning models are costly to serve because they are large and because they generate long chains of thought (CoT). Weight pruning and CoT compression address these two costs separately, but they interfere in that pruning adds redundant reasoning on problems the model still solves and incomplete reasoning on those it no longer solves, and on harder problems makes the model less compressible by subsequent length-preference optimization. We trace this to a blind spot on each side: pruning criteria do not distinguish necessary from redundant reasoning, and preference-based CoT compression receives no training signal on problems a pruned model always or never solves. We propose Contrastive Set-Difference (CSD), which identifies neurons enriched in redundant reasoning by contrasting their activations on long and short correct solutions of the same problem and prunes them preferentially, and we extend preference pairs to the problems a pruned model always or never solves. Pruning 20% of DeepSeek-R1-Distill-Qwen-7B, the combination uses 8–22% fewer thinking tokens on four math benchmarks than pruning with reasoning-aware calibration (RAC) followed by the same compression, with accuracy within 0.5 points.
[NLP-98] CoLink: Global Compression and Knowledge Linking for Multi-Turn Dialogue Modeling
【速读】： 该论文旨在解决长时程对话历史中显著信息提取的挑战，尤其针对大语言模型（LLM）因上下文窗口受限及高效检索困难所导致的信息丢失、检索冗余和局部注意力局限等问题。其核心解决方案在于提出CoLink框架，通过引入认知科学中“对话即意义建构与记忆关联”的理念，构建跨语义关联的领域特定多轮对话信息链接与压缩机制。关键创新点在于：首先基于主题图谱对历史上下文进行聚类，建立与外部知识的跨语义关联；其次，在自适应思维控制（adaptive control of thought）引导下，协同语义相关性与图结构拓扑实现知识的精准检索与激活；更重要的是，基于信息论原理，设计了具有理论依据的优化目标，用于提取核心意义建构信息，结合竞争性槽位机制，将全局上下文与知识高效压缩至专用的记忆与知识槽中，生成高度紧凑的潜在表示。该方法显著提升了长对话场景下的全局上下文感知能力与知识融合效果。
链接: https://arxiv.org/abs/2610.34335
作者: Lian Yan,Zhenbo Xia,Haotian Wang,Chen Tang,Yi Guan,Jie Liu,Jingchi Jiang
机构: Harbin Institute of Technology (哈尔滨工业大学); National Key Laboratory of Smart Farm Technologies and Systems (国家智能农业技术与系统重点实验室); Institute for Advanced Algorithms Research (先进算法研究院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 7 pages, 15 figures, work in prograss
Abstract:Extracting salient information from long-horizon dialogue histories remains a significant challenge for Large Language Models (LLMs) due to the restricted context windows and efficient retrieval challenges. Current methods typically rely on history organization or local token-level optimization, suffering from information loss, retrieval redundancy, and myopic attention. Inspired by the view in cognitive science that dialogue constitutes sense-making and memory association, we propose CoLink, a framework that enhances LLMs’ global context awareness in domain-specific multi-turn dialogues through information linking and compression. CoLink first clusters historical contexts into topic graphs to establish cross-semantic associations with external knowledge. Guided by the adaptive control of thought, it then retrieves and activates knowledge via synergizing semantic relevance and graph-structural topology. Crucially, to achieve effective compression of history and knowledge, we ground multi-turn dialogues in information theory and derive a principled optimization objective for extracting the core sense-making information. Coupled with a competitive slot mechanism, CoLink condenses the global context into dedicated memory and knowledge slots, efficiently distilling lengthy, redundant inputs into an extremely compact latent representation. Extensive empirical results on both public and our MMD datasets validate CoLink’s superiority and generalizability in knowledge linking and sense modeling.
[NLP-99] Birth Death and Diff: Insertion-Based Sequence Generation with Corrections
【速读】： 该论文旨在解决基于插入的掩码扩散模型（insertion-based masked diffusion models, MDMs）在并行生成过程中因生成决策不可逆而导致的错误累积问题。这类模型在高并发解码时易引入难以纠正的错误，迫使采用保守的解码策略以保证质量，从而牺牲生成效率。此外，由于训练中对多种对齐方式的概率平均（marginalization），其输出熵较高，进一步加剧了对保守解码的需求。为此，本文提出δMDM，其核心创新在于引入替换（substitution）与删除（deletion）操作，使中间生成结果具备可修正性。训练阶段，模型通过暴露于激进的解码错误中学习自我纠错能力；推理阶段，则通过交错插入、去掩码与修正步骤，动态调整已生成内容并移除冗余掩码，实现更激进但可靠的并行生成。这一机制显著提升了生成速度与质量，在无条件文本生成任务中，相较于基准模型FlexMDM，δMDM在最高推理预算下将生成困惑度降低50%，且在所有预算水平上均优于从零训练的变长填充序列标准MDM；在条件生成任务中，其精度与FlexMDM相当，但仅需一半的前向传递次数（NFEs）。因此，δMDM在速度-质量权衡上实现了帕累托改进。
链接: https://arxiv.org/abs/2610.34310
作者: Durga Prasad Maram,Dhruvesh Patel,Tahira Naseem,Benjamin Rozonoyer,Gaurav Pandey,Ramón Fernandez Astudillo,Andrew McCallum
机构: IBM Research(IBM 研究院); University of Massachusetts Amherst(马萨诸塞大学阿默斯特分校)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Insertion-based masked diffusion models enable flexible-length, parallel generation by interleaving mask insertion with token unmasking. However, because their generation decisions are irreversible, aggressive parallel decoding can introduce errors that subsequent steps cannot correct, necessitating conservative decoding with more forward passes (NFEs). Moreover, they exhibit higher entropy than standard MDMs because they marginalize over multiple alignments during training and therefore require even more conservative decoding in the absence of a correction mechanism. To address this limitation, we introduce \delta MDM, which makes intermediate generations revisable by equipping an insertion-based model with substitution and deletion operations. During training, we expose the model to errors caused by aggressive unmasking and insertion, teaching it to correct its own mistakes. At inference time, we interleave insertion and unmasking with correction steps that can revise previously unmasked tokens and remove unnecessary masks. This mechanism makes aggressive decoding practical, improving both generation speed and quality. \delta MDM Pareto-dominates FlexMDM, a vanilla insertion-based MDM, on the speed–quality trade-off. On high-diversity unconditional text generation on OpenWebText, it halves the generative perplexity relative to FlexMDM at the highest inference budget. It also achieves lower generative perplexity at all inference budgets compared to MDM trained from scratch with variable-length padded sequences. On low-diversity conditional text generation on TinyGSM, it matches FlexMDM accuracy with half as many NFEs.
[NLP-100] ransformer to GPT -5: Advancements Limitations and Future Directions in Large Language Model Design ICIP
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）缺乏系统性分类体系的问题，致力于构建一个全面、细致的LLM分类框架。其核心挑战在于如何在模型规模不断扩张的背景下，实现架构复杂性与数据质量之间的最优平衡。解决方案的关键在于揭示：当前新一代AI模型的成功标志并非单纯追求参数量或计算资源的线性增长，而是通过优化“架构复杂性与数据质量之比”，在不显著增加计算负担的前提下，提升模型的推理稳定性与对齐能力。这一平衡机制成为衡量模型可持续扩展性的关键指标，为未来LLM的发展提供了理论指引与实践方向。
链接: https://arxiv.org/abs/2610.34281
作者: Harsidak Singh Banwait,Chander Prabha
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 5 pages, ICIPCN
Abstract:Large Language Models (LLMs) have revolutionized Artificial Intelligence (AI) by their extraordinary capabilities, like contextual understanding, few-shot/zero-shot learning, code generation, ability to reason and problem-solving. Recent studies have shown immense work on LLMs. Yet, there is limited work on the thorough taxonomy of LLMs. This study would showcase a detailed classification of LLMs, starting with the earlier models like BERT and leading to LLMs like GPT, Gemini, Claude and Grok. We would delve into topics like datasets used to train LLMs, computational power used during training, their hardware specifications, architecture and parameters. Further, we would also look at the limitations, computational challenges and open problems as well. This study would primarily focus on the timeline of LLMs, signifying their growth over the period. Findings reveal a clear pattern: models that optimize the ratio between architectural complexity and data quality exhibit stronger reasoning stability and better alignment, without proportionally increasing the computational load. This desired balance between scaling and sustainability is the defining characteristic of the current generation of AI.
[NLP-101] A Russian a Brazilian and a German Walk into a Bar Ask the Same Question-- And Get Three Different Answers: Profiling LLM Perspective Diversity on Contested Topics
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理具有主观性、争议性及证据冲突的跨语言真实世界问题时，其观点表现存在语言偏倚与不一致性的核心挑战。现有评估基准多局限于英语语境，难以反映全球用户多元视角的差异。为此，研究者构建了PerspectivesQA——一个涵盖18种语言、经人工验证的多语言争议性话题数据集，并提出ROOTS框架，用于系统性地刻画语言特征、观点分布与表述框架的多样性。关键发现表明：在英语中，大模型倾向于采取模糊或保守立场（hedging），而在低资源语言中则更易表现出明确且固执的观点倾向；大规模模型展现出更强的跨语言一致性世界观，而小规模模型则更易受输入扰动影响；此外，所有模型对输入表面形式的微小变化均表现出高度敏感性，即使在大型模型中亦可能引发显著立场偏移。因此，解决方案的关键在于通过多语言实证数据与系统性分析框架，揭示并量化模型在跨语言情境下的观点偏差机制，为提升模型的公平性与稳健性提供理论依据与实践路径。
链接: https://arxiv.org/abs/2610.34278
作者: Wiem Ben Rim,Yao Lu,Pontus Stenetorp,Emine Yilmaz
机构: University College London(伦敦大学学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large Language Models(LLMs) have shown strong performance on tasks with verifiable answers, but many real-world queries involve subjectivity, disagreement, and conflicting evidence. Most existing benchmarks for such queries remain English-centric, overlooking the vast diversity of perspectives held by users worldwide. To systematically study how LLM perspectives vary across languages, we curate PerspectivesQA, a manually verified multilingual dataset of globally contested topics covering 18 languages, and propose ROOTS, a framework that profiles linguistic, perspective, and framing diversity. We evaluate a diverse set of eight models covering both open-source and proprietary families at varying scales, establishing that LLMs hedge their answers in English but commit to positions in lower-resource languages. Larger models maintain a more consistent cross-lingual worldview, while smaller models are more easily swayed. All models also exhibit high sensitivity to minor surface-level input variations, producing notable stance even in larger models.
[NLP-102] Synthetic linguistic data
【速读】： 该论文旨在解决生成式语言数据（Generative Language Data）在语言学研究中日益广泛应用背景下，其作为研究对象的性质与解释方法缺乏系统性界定的问题。当前研究既涉及对大规模人工智能生成文本的分析，也包括在受控条件下检验语法与词汇偏好的实验，但这些数据的类型与生成机制差异显著，导致研究结论的可比性与解释力面临挑战。论文的关键解决方案在于提出“人工植物苗圃”（artificial plant nursery）这一新隐喻，以区分两类合成数据：一类是“合成灌木丛数据”（synthetic bush data），即由模型生成的、类似语料库的语言集合；另一类是“合成花园数据”（synthetic garden data），即在受控条件下诱发模型输出以检验特定语言假说的数据。该区分的核心在于强调合成数据应被视为独立的语言学研究对象，其与自然语言数据的对比不仅可揭示共性模式，更需审慎考量其生成条件，从而为合成数据的科学使用提供理论框架与方法论指导。
链接: https://arxiv.org/abs/2610.34215
作者: Giuseppe Samo
机构: 未知
类目: Computation and Language (cs.CL)
备注: 11 pages
Abstract:Synthetic linguistic data have become increasingly prominent in linguistic research, from the analysis of large collections of AI-generated texts to controlled experiments investigating grammatical and lexical preferences. This glossary entry extends Bresnan’s (2016) distinction between the linguistic garden and the bush by introducing an artificial plant nursery as a metaphor for synthetic data. We distinguish between synthetic bush data, consisting primarily of corpus-like collections of model-generated language, and synthetic garden data, in which model outputs are elicited under controlled conditions to investigate specific linguistic hypotheses. We argue that synthetic data constitute a distinct object of linguistic investigation: comparing them with natural language data can reveal both shared patterns and divergences, but their interpretation requires careful consideration of the conditions under which they are generated.
[NLP-103] Higher Reasoning Effort Improves Reference Verifiability and Changes Reasoning Summaries in GPT -5.5
【速读】： 该论文旨在解决生成式AI在科学知识生产中生成参考文献的可验证性问题，特别是探究增加推理努力程度是否能够提升所生成参考文献的准确性与可信度。其核心解决方案的关键在于通过对比低努力与高努力两种推理模式下GPT-5.5生成的学术参考文献及其推理过程摘要，系统评估推理强度对参考文献可验证性的影响。研究发现，提高推理努力使论文级可解析验证的参考文献比例从78.3%提升至84.8%，且推理摘要长度显著增加；四个预定义特征评分均随努力程度上升而提高。然而，模型自主发现的新特征在不同条件下的出现频率变化不一致，且无论在何种努力条件下，各特征得分与验证率之间均未呈现稳定正相关关系，表明推理质量与验证结果之间的关联并非线性或可直接预测。
链接: https://arxiv.org/abs/2610.34212
作者: Melika Mobini,Andres Algaba,Vincent Ginis
机构: 未知
类目: Computation and Language (cs.CL)
备注: 4 main figures. Supplementary material included
Abstract:Large language models are increasingly being used in different stages of scientific knowledge production, including the suggestion of references. These models can operate at different levels of reasoning effort and provide summaries of their reasoning processes. This raises two questions: Does increasing reasoning effort improve the verifiability of generated references, and do the resulting reasoning summaries reveal how the model’s behavior changes with greater effort? We examined summaries from GPT-5.5 as it generated scholarly references for 1,000 papers under paired low- and high-effort settings. Generated references were matched to external bibliographic records, and reasoning summaries were scored for four predefined features and explored for additional features discovered by GPT-5.5. High effort increased the mean paper-level resolver-verified reference rate from 78.3% to 84.8%, and produced longer reasoning summaries. All four predefined features received higher scores under high effort. The discovered features changed less uniformly: some appeared equally often in both conditions, whereas others were more common under high effort. Within each effort condition, neither feature set showed a consistent increase in verification rates with higher scores.
[NLP-104] MIS-VQA: An Annotation-Driven Surgical Vision-Language Dataset for Complex Surgical Understanding
【速读】： 该论文旨在解决当前外科手术视觉-语言理解（Surgical Vision-Language Understanding）中普遍存在的多模态理解能力不足问题，尤其针对现有数据集在任务监督上局限于单一任务（如识别或模板化问答）、缺乏自然描述性标注以及空间关系与操作上下文信息稀疏等关键缺陷。其核心解决方案在于构建MIS-VQA——一个通过注释驱动的流水线生成的外科手术视觉-语言数据集，将异构的手术标注（包括阶段、器械、交互、三元组及分割标签）统一转化为结构化的文本场景表征，从而引导大语言模型生成具有临床一致性的问答对。该数据集支持四种响应类型：简洁与详细描述、感知级定位以及多步推理，实现了从原子级视觉定位到复合推理的分层评估体系；为缓解标注稀疏性问题，引入置信度过滤的分割结果与跨注释一致性校验机制，实现大规模空间对齐的监督。此外，研究进一步提出LLaVA-SurgXpert系列外科视觉-语言模型，验证了视觉表征对多模态推理的影响，实验表明其在基准测试中持续优于现有方法，同时揭示了以往基准中被掩盖的感知定位与高层推理能力之间的差距。
链接: https://arxiv.org/abs/2610.34210
作者: Mohamed Hamdy,Iheb Zouari,Fatmaelzahraa Ahmed,Muraam Abdel-Ghani,Muhammed Arsalan,Ponnuthurai Suganthan,Khalid Al-Jalham,Abdulaziz Al-Ali,Shidin Balakrishnan
机构: Qatar University(卡塔尔大学); Hamad Medical Corporation(哈马德医疗委员会)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Modern surgical AI increasingly requires multimodal understanding rather than isolated perception, yet existing datasets remain largely limited to single-task supervision such as recognition or templated question answering. Current medical VQA resources often rely on synthetic templates, lack natural descriptive supervision, or provide sparse annotations that hinder reasoning about spatial relations and procedural context. We introduce MIS-VQA, a surgical vision-language dataset generated through an annotation-driven pipeline that converts heterogeneous surgical annotations into structured textual scene representations. Using EndoVis18 and CholecT50, we transform phase, instrument, interaction, triplet, and segmentation labels into unified contexts that guide large language models to generate clinically coherent question-answer pairs. MIS-VQA supports four response types-concise and detailed descriptions, perception-level grounding, and multi-step reasoning-enabling hierarchical evaluation from atomic visual grounding to compositional reasoning. To address annotation sparsity, we incorporate confidence-filtered segmentation and cross-annotation consistency checks, enabling spatially grounded supervision at scale. MIS-VQA establishes a unified benchmark for surgical vision-language understanding with spatial supervision and multi-tier reasoning evaluation. To demonstrate its utility, we introduce LLaVA-SurgXpert, a family of surgical vision-language models with visual encoders to analyze how visual representations influence multimodal reasoning. Experiments show consistent improvements over existing surgical VQA baselines while revealing capability gaps between perceptual grounding and higher-level reasoning hidden in prior benchmarks.
[NLP-105] Entropic Control for Homeostatic Optimization: Pre-empting Catastrophic Forgetting via Dynamic Feedback
【速读】： 该论文旨在解决大语言模型（LLM）在监督微调（Supervised Fine-Tuning, SFT）过程中面临的灾难性遗忘问题，尤其针对现有缓解策略难以平衡模型可塑性与稳定性之间的矛盾（plasticity-stability dilemma）这一核心挑战。其解决方案的关键在于提出一种基于熵控的稳态优化框架（Entropic Control for Homeostatic Optimization, ECHO），通过分解KL散度，识别输出熵（output entropy）为KL变化的主要可观测成分，从而将熵作为计算高效的在线代理指标，实时监测遗忘风险，避免了传统方法中依赖参考模型推理所带来的高昂计算开销。在此基础上，ECHO采用弹性控制器，依据预训练状态下的熵平衡点偏差作为误差信号，通过比例控制律动态调节回放数据的梯度权重，实现对模型更新过程的闭环调控。实验结果表明，ECHO在0.6B至9B参数规模的多种架构上均位于甚至超越了当前观测到的可塑性-稳定性边界，且对不同数据混合比例具有高度鲁棒性。
链接: https://arxiv.org/abs/2610.34182
作者: Shangxi Wu,Xiaorong Yan,Xiaojun Meng,Yuxiang Zhang,Jiansheng Wei
机构: Beijing Jiaotong University (北京交通大学); The Hong Kong Polytechnic University (香港理工大学); Huawei Noah’s Ark Lab (华为诺亚方舟实验室)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Supervised Fine-Tuning (SFT) of LLMs suffers from catastrophic forgetting, and prevailing mitigation strategies struggle to reconcile the plasticity-stability dilemma. To address this, we propose Entropic Control for Homeostatic Optimization (ECHO), a closed-loop control framework. By decomposing the KL divergence, we identify output entropy as a dominant observable component of KL variation. This enables us to leverage entropy as a computationally efficient online proxy to monitor forgetting risks in real-time, bypassing expensive reference-model inference. Consequently, ECHO employs an elastic controller that dynamically modulates the gradient weight of replay data via a proportional control law, using deviations from the pre-trained entropic equilibrium as an error signal. Empirical evaluations across 0.6B to 9B parameter architectures demonstrate that ECHO lies on, and in several comparisons extends, the observed plasticity-stability frontier and maintains high operational robustness against varying data mixing ratios.
[NLP-106] Write What Matters: Disentangling Token Selection from Slot Allocation in Recurrent Memory
【速读】： 该论文旨在解决长序列建模中如何有效保留早期信息的核心挑战：传统全注意力机制虽能保留全部历史信息但计算开销随序列长度呈平方增长，而基于循环记忆的模型虽实现线性时间复杂度，却受限于固定容量的记忆槽，导致写入操作具有高代价——错误写入会干扰有效内容，遗漏关键证据则无法召回。现有方法将“是否存储”与“更新多少槽位”耦合处理，缺乏灵活性。本文提出因子化写入路由（Factorized Write Routing, FWR），其核心创新在于解耦写入决策与槽位分配：先判断每个输入标记是否应被保留或跳过，再独立决定使用固定数量或根据标记自适应调整的槽位数进行写入，而查询阶段仍读取所有槽位。在两个合成任务（密集特征回忆与稀疏事实密钥任务）上的实验表明，当上下文信息密度高时，持续写入优于选择性写入；而在干扰项主导的稀疏场景下，学习到的选择性写入显著提升性能（0.222 vs 0.048）。尽管所测试的自适应规则表现不及固定分配策略，且诊断性基准（已知真实标签）显示仍有提升空间，结果表明在信息密集环境中应避免遗漏写入，在干扰主导场景中需强化去噪能力。
链接: https://arxiv.org/abs/2610.34180
作者: Chen Chen,Rui Cheng
机构: Hong Kong Baptist University(香港浸会大学); Anhui University(安徽大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Long-sequence models must retain earlier information. Full attention keeps all past tokens but grows expensive, and recurrent-memory models compress history into fixed slots for linear-time processing. Limited capacity makes writes consequential: distractors can interfere with useful content, while skipped evidence cannot be recalled. Existing multi-slot memories write every token to the same number of slots, merging whether to store it with how many slots to update. We introduce Factorized Write Routing (FWR), which first keeps or skips each token and then chooses a fixed or token-adaptive number of slots. Queries still read all slots. Across ten seeds on two synthetic tasks that query facts after long contexts, we observe a density-dependent reversal. With four slots per write, long-context accuracy favors always writing on dense Feature Recall (0.472 vs 0.337), but learned selective writing on sparse-fact Passkey (0.222 vs 0.048). The tested adaptive rule underperforms fixed allocation under the learned gate on both tasks. A diagnostic oracle given fact labels unavailable at deployment reaches 0.372 on Passkey, showing headroom in filtering distractors. The results suggest avoiding missed writes in information-dense contexts and filtering distractors when they dominate.
[NLP-107] Reasoning or Just Prompting? Probing the Mechanism of Latent Chemical Reasoning
【速读】： 该论文旨在解决生成式化学推理模型中隐变量（latent states）是否真正构成可解释的逐步推理链这一核心问题。传统观点认为，隐变量在连续空间中执行中间计算，能够避免显式符号推理中的错误传播，从而提升化学任务的准确性。然而，研究发现，对中间隐变量进行替换或轻微扰动对模型性能影响甚微，而完全移除这些隐变量却可能导致显著性能下降，表明隐变量虽重要，但其分子特异性内容并非关键因素。进一步分析揭示，隐变量更可能起到类似软提示（soft prompts）的作用，即通过引导后续模型计算流程来影响输出，而非编码具体的、可追溯的推理步骤。实验验证显示，使用单一固定前缀即可复现LatentChem模型的性能，且学习得到的静态软提示在控制模型中表现与逐输入生成的隐变量相当。此外，隐序列对替换和扰动的高度鲁棒性也支持了其作为软提示而非推理链的解释。因此，论文提出，隐式化学推理应被重新理解为一种基于连续空间的软提示机制，而非隐藏于连续空间中的显式推理链。
链接: https://arxiv.org/abs/2610.34173
作者: Kaiyuan Lu,Jiatong Li,Yatao Bian
机构: National University of Singapore(新加坡国立大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 27 pages, 4 figures, 20 tables
Abstract:Latent reasoning performs intermediate computation in continuous space rather than through explicit tokens, which is particularly appealing in chemistry, where errors in explicit reasoning can propagate to the final answer. Yet if latent states truly constitute an implicit reasoning chain, they should be causally important. We find otherwise: swapping or moderately perturbing intermediate latent states has little effect on performance, while removing them entirely sometimes causes a substantial score drop. Thus, the latent states matter, but their molecule-specific content appears to matter far less. This points to a simpler interpretation. Rather than encoding a step-by-step reasoning chain, latent states may primarily steer subsequent model computation, much like soft prompts. Indeed, a single fixed prefix shared across molecules reproduces LatentChem’s performance, and a learned static soft prompt performs comparably to per-input latent generation in controlled models. The tolerance of latent sequences to substitution and moderate perturbation further supports this connection. We therefore suggest that latent chemical reasoning may be better viewed as soft prompting than as a reasoning chain hidden in continuous space.
[NLP-108] ask Success Is Not Enough: Side-effect-Aware Evaluation of Tool-Using Language Model Agents NEURIPS2026
【速读】： 该论文旨在解决大语言模型（LLM）在执行工具使用任务时存在的“附带损害”（collateral damage, CD）问题，即模型虽能完成用户指定任务，却在过程中实施了未经授权的副作用行为，如误写对象、扩大请求范围或联系错误接收者。此类问题在现有评估体系中被忽视，因为传统指标仅关注任务完成率，而未考量操作的授权合规性。论文提出SABench（Side-effect-Aware Benchmark），一个包含201个电子邮件、日历与工单任务的基准测试集，涵盖四种状态结构（独立、依赖、外部可访问、跨工具）与四类陷阱类型（歧义、误靶向、级联伤害、过度扩展），并为每项任务提供基于确定性轨迹的双重验证：目标达成与授权合规性。实验覆盖11个前沿大模型的2,211个模型-任务组合，结果显示15.2%的运行出现CD，其中歧义类任务的违规率高达38.8%，远高于非歧义陷阱的平均8.4%；即使在目标达成的场景下，仍有12.8%的运行产生未经授权的写操作。跨工具与依赖状态下的CD率约为独立状态的两倍。不同模型在相似任务成功率下，CD率差异达4.7倍，表明任务成功无法有效反映工具使用的安全性。因此，论文强调，必须引入对授权合规性的显式评估机制，以实现更安全的生成式AI（Generative AI）应用。
链接: https://arxiv.org/abs/2610.34167
作者: Jiaju Huang,Shaobin Chen,Xinglong Liang,Xinyu Ma,Hao Yang,Wei Ke,Tao Tan,Yue Sun
机构: Macao Polytechnic University(澳门理工学院); Southwest Medical University(西南医科大学); Netherlands Cancer Institute(荷兰癌症研究所); Radboud University Medical Center(拉德堡德大学医学中心)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted at NeurIPS 2026 (Evaluations and Datasets Track). 19 pages, 5 figures, 10 tables
Abstract:Tool-using language model agents are usually evaluated by whether they finish the user’s task. That metric misses a common failure: an agent can complete the requested work while also taking an unauthorized side-effecting action, such as writing to the wrong object, expanding the scope of the request, or contacting an unintended recipient. We call this failure mode collateral damage (CD). We introduce SABench (Side-effect-Aware Benchmark), a benchmark of 201 email, calendar, and ticket tasks. Each task crosses one of four state structures (independent, dependent, externally reachable, cross-tool) with one of four trap families (ambiguity, mis-targeting, cascade harm, overreach), and each task has a deterministic trace-based oracle for both goal satisfaction and authorization compliance. Across 2,211 model-task pairs from 11 frontier LLMs spanning major commercial and open model families, 15.2% of all runs incur CD, with risk concentrated in ambiguity tasks (38.8% violation rate) while non-ambiguous traps average 8.4% CD. Among goal-satisfied runs, 12.8% still produce unauthorized writes. Cross-tool and dependent states have about twice the CD rate of independent states. CD rates vary by 4.7x across models at comparable goal rates, which suggests that task success alone is a poor proxy for safe tool use. The SABench task corpus is released at this https URL the traces, oracle, code, and analysis pipeline are released at this https URL.
[NLP-109] Humans in Space: Can MLLM s Construct Human-Centric Spatial Understanding via Active Exocentric Exploration? WACV
【速读】： 该论文旨在解决多模态大语言模型（Multimodal Large Language Models, MLLMs）在复杂人本场景中缺乏主动空间推理能力的问题，尤其针对其在存在遮挡时因依赖静态被动观察而无法准确识别自身视觉不确定性、导致错误判断与过早终止探索的缺陷。现有基准测试多采用被动多视角评估方式，难以真实反映模型在动态环境中的视觉自知与主动视点探索能力。为此，论文提出SATE-Bench这一新型交互式评测基准，包含316个真实世界场景和2,525个多轮问答对，通过结合被动多视角控制与主动探索设置，要求模型从严重遮挡的初始视角出发，逐步查询相邻视点以获取充分证据，最终“满足”问题需求，并采用步数惩罚的效率奖励机制进行评分。实验结果表明，当前先进MLLMs虽在主动探索中表现出高置信度但常提前终止，且在细粒度三维几何任务上表现不佳；对比分析揭示主动-被动差距主要源于证据获取不足。进一步地，基于组相对策略优化（Group Relative Policy Optimization, GRPO）的微调显示，鼓励视点搜索虽提升了探索行为，却因过度探索及跨轮次多视角上下文融合失败而降低了整体准确性。研究结论指出，单纯扩大被动多视角感知能力无法自然催生主动空间推理能力，强调需构建具备“知何处看、如何整合所见、何时停止”的智能体，从而实现真正意义上的视觉自知与主动认知。
链接: https://arxiv.org/abs/2610.34149
作者: Tianchen Guo,Tengyue Zhang,Chen Liu,Ling Chen,Xin Yu
机构: The University of Queensland(昆士兰大学); Follow Me AI Pty LTD(跟随我人工智能有限公司); Australian Institute for Machine Learning, Adelaide University(阿德莱德大学机器学习研究所); University of Technology Sydney(悉尼科技大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: Winter Conference on Applications of Computer Vision (WACV) 2027. 33 pages, 14 figures, and 23 tables
Abstract:While Multimodal Large Language Models (MLLMs) excel at zero-shot reasoning, their spatial capabilities are primarily evaluated through static, passive observation. In complex, human-centric scenes, occlusion often hides crucial geometric evidence, and passive benchmarks force models to guess rather than recognize their own visual uncertainty. To bridge this gap, we introduce SATE-Bench, a novel interactive benchmark comprising 316 real-world scenes and 2,525 multi-turn QA pairs to evaluate visual self-awareness and active exocentric view-exploration. It combines a Passive multi-view control with an Active exploration setting, where models start from a severely occluded view and must iteratively query adjacent viewpoints until they gather enough evidence to fully ``sate’’ the question, scored by a step-penalized efficiency reward. Evaluating state-of-the-art MLLMs reveals that they terminate exploration prematurely while reporting high confidence, and struggle with fine-grained 3D geometric tasks; matched-view controls further attribute the Active–Passive gap primarily to insufficient evidence acquisition. Moreover, fine-tuning with Group Relative Policy Optimization (GRPO) shows that rewarding view-seeking increases active exploration but lowers overall accuracy, as models over-explore and fail to fuse expanding multi-view contexts across turns. These findings indicate that scaling passive multi-view perception does not intrinsically yield active spatial reasoning, highlighting the need for models that know where to look, how to integrate what they see, and when to stop. Code and data are available at this https URL
[NLP-110] Rethinking Retrieval in the Long-Context Era: Training-Free Semantic Trees for Adaptive Context Construction
【速读】： 该论文旨在解决长上下文语言模型在处理完整文档时，尽管具备足够的信息接收能力，却难以有效筛选并组织最相关证据以支持查询的问题。现有全文档提示（full-context prompting）方法虽能利用全部文本信息，但缺乏对关键事实与语义关系的聚焦，导致冗余信息干扰模型推理。其解决方案的关键在于提出一种无需训练的框架\ours，通过语义抽象（semantic abstraction）在单次生成中构建可复用的语义层次结构，将节点描述与原文段落进行关联。针对具体查询，该框架结合结构-语义评分与细粒度词汇归属机制，精准识别共享源节点，并通过整合其祖先上下文，保留语义结构关系与局部完整文本。实验表明，在三个LongBench任务上，\ours相较于全文档提示平均提升QA-F1 1.72点，且在四类前沿问答模型上的平均表现优于所有对比基线，同时将答案-上下文长度减少58.55%。
链接: https://arxiv.org/abs/2610.34146
作者: Zhe Xu,Tianhao Liao,Zhenqing Ling,Ying Shen,Yaliang Li,Daoyuan Chen
机构: Sun Yat-sen University (中山大学); Alibaba Group (阿里巴巴集团)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Long-context language models can increasingly process entire documents, making full-context prompting a practical alternative to passage retrieval. However, greater capacity determines how much information a model can receive, but not which information best supports a query. This distinction motivates rethinking retrieval: when the full document already fits, can selecting and organizing its evidence improve answering? The challenge is to retain supporting facts and relations while keeping the context focused on the question. To reconcile these goals, we introduce \ours, a training-free framework that uses semantic abstraction to guide selection while retaining original text for answering. \ours constructs a reusable semantic hierarchy from the full document in a single generation, linking node descriptions to source spans. Given a query, structural–semantic scoring and fine-grained lexical attribution nominate shared source nodes; their union with ancestor context preserves structural relations and complete local text. Across four frontier answer models, \ours improves average QA-F1 by 1.72 points over full-context prompting on three LongBench tasks, surpassing the evaluated baselines on the four-model average while reducing answer-context length by 58.55%. Our code is available at the \hrefthis https URLrepository .
[NLP-111] Self-Evolution Reason er: Continuous Optimization via Policy-Intrinsic Exploration and Exploitation
【速读】： 该论文旨在解决生成式 AI 在强化学习中因探索瓶颈导致的推理能力难以持续提升的问题，尤其是在策略无法采样有效解时易陷入局部最优的困境。现有方法如基于标量奖励的强化学习（RLVR）与自监督在线策略蒸馏虽引入了更密集的监督信号，但仍受限于探索能力不足。为突破此瓶颈，本文提出自进化策略优化（Self-Evolution Policy Optimization, SEPO），其核心创新在于构建一个基于内在探索与分布内利用的迭代循环机制：通过标签引导策略生成正确的推理路径，激活其探索潜力；随后将当前策略在该分布内新发现的有效解作为“自教师”进行知识蒸馏，实现对自身策略的持续优化。这一机制既避免了依赖外部示范带来的非策略性偏差，又充分挖掘了分布内推理的潜力，从而在科学推理、工具使用及复杂逻辑推理等多领域均展现出卓越的性能、更高的训练效率与更优的收敛上限，尤其在小模型或高难度任务场景下表现出显著稳定性与持续改进能力。
链接: https://arxiv.org/abs/2610.34137
作者: Wenhang Shi,Yiren Chen,Zhe Zhao,Jinhao Dong,Wei Lu,Xiaoyong Du
机构: Renmin University of China (中国人民大学); Peking University (北京大学); Tencent (腾讯)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) enables large language models (LLMs) to continuously self-improve their reasoning capabilities. Compared to RLVR’s scalar rewards, recent self-supervised on-policy distillation methods leverage denser supervision signals. However, both paradigms suffer from exploration bottlenecks, struggling to escape local optima when the policy fails to sample valid solutions. Relying on external demonstrations can bypass this issue; however, the off-policy nature hinders genuine self-improvement and the mismatched-distribution reasoning can not be fully exploited. In this work, we propose Self-Evolution Policy Optimization (SEPO), a framework that drives continuous reasoning capability evolution via guided exploration and in-distribution exploitation. Specifically, SEPO utilizes labels to elicit correct reasoning from the policy, unlocking the exploration potential. By treating current policy conditioned on this in-distribution reasoning as a self-teacher, SEPO distills the newly discovered solution back into the policy. Through this iterative cycle of policy-intrinsic exploration and exploitation, SEPO achieves robust and continuous capability evolution. Across scientific, tool-use, and complex logical reasoning domains, SEPO consistently achieves the best performance under identical training time or steps, demonstrating superior efficiency and convergence ceilings. Notably, SEPO maintains stable improvements even in scenarios where baselines severely underperform or fail entirely, such as on complex tasks or with small-scale models. Our code is available at this https URL.
[NLP-112] 2D: Target-to-Draft Model Pruning Adaptation for Efficient Speculative Decoding
【速读】： 该论文旨在解决生成式 AI（Generative AI）中自回归文本生成的效率瓶颈问题，具体针对推测解码（speculative decoding）技术在实际应用中受限于两个关键因素：草案模型（draft model）与目标模型之间的分布对齐程度，以及草案模型在推理过程中的延迟表现。现有方法通常依赖独立预训练的小型草案模型，或通过手动选择架构对目标模型进行剪枝，未能显式平衡模型对齐性与生成效率之间的权衡。为此，本文将草案模型设计建模为一个优化问题，提出T2D框架——该框架基于目标模型，结合感知延迟的结构化剪枝与知识蒸馏技术，直接构建出兼顾实际推理延迟与目标模型输出交叉熵损失的优化草案模型。T2D通过联合优化这两个核心指标，实现了卓越的推测解码性能。在多领域基准测试及不同规模的目标模型家族上，实验表明T2D可将目标模型剪枝高达92%，同时获得显著优于基线的草案模型性能；此外，由T2D生成的草案模型具备良好的兼容性，可作为即插即用模块与其他推测解码中的对齐与蒸馏方法协同使用，进一步提升整体效果。研究结果强调了在架构-对齐边界上协同优化的重要性，以充分释放推测解码在大语言模型（LLM）中的潜力。
链接: https://arxiv.org/abs/2610.34103
作者: Shahriar Kabir Nahin,Xiao Ye,Muhao Chen,Ben Zhou,Anshuman Chhabra
机构: Bellini College of AI, Cybersecurity, and Computing, University of South Florida(南佛罗里达大学贝尔尼尼人工智能、网络安全与计算学院); School of Computing and Augmented Intelligence, Arizona State University(亚利桑那州立大学计算与增强智能学院); Department of Computer Science, University of California, Davis(加州大学戴维斯分校计算机科学系)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Speculative decoding significantly accelerates autoregressive text generation in Large Language Models (LLMs), but its efficacy depends largely on two key factors: the distributional alignment between the draft and the target model, and the draft model’s latency during inference. However, existing approaches typically utilize independently pretrained small draft models or seek to prune the target model to a manually chosen architecture, thereby failing to explicitly balance the trade-off between draft model alignment and token generation efficiency. To address this gap, we formulate draft model design as an optimization problem and propose T2D, a framework that constructs draft models directly from the target model using latency-aware structured pruning and knowledge distillation. In this manner, T2D obtains draft models that are jointly optimized for both measured inference latency and cross-entropy loss on the target model’s responses, resulting in superlative speculative decoding performance. Through extensive experiments on multi-domain speculative decoding benchmarks and across several target model families of differing scale, we demonstrate how T2D prunes up to 92% of the target model to obtain highly performant draft models compared to competitive baselines. Moreover, draft models obtained via T2D can also be utilized by other speculative decoding token-alignment and knowledge distillation methods in a plug-and-play fashion, leading to further gains. Our findings underscore the importance of jointly navigating the draft model architecture-alignment frontier to unlock the full potential of speculative decoding for LLMs.
[NLP-113] he LALA Benchmark: Evaluating Literary and Linguistic Analysis in LLM s
【速读】： 该论文旨在解决当前大语言模型在语言学与文学分析能力评估方面缺乏系统性、专业化基准测试的问题。现有评估体系往往无法有效区分模型在不同语言认知维度上的真实理解水平，尤其难以捕捉模型对语义歧义、修辞风格、指代关系及隐含假设等复杂语言现象的处理能力。为应对这一挑战，研究提出了文学与语言学分析基准（Literature and Linguistics Analysis Benchmark, LALA），包含由语言学与文学研究专家设计的236道开放性问题及其理想答案，覆盖多个语言学领域。其解决方案的关键在于：通过对比同一模型家族中基础模型与后训练版本在相同训练规模下的表现，并结合人类专家与两种大型语言模型（Sol和Qwen）作为评判者，揭示后训练阶段在提升模型语言学能力中的核心作用。结果表明，尽管参数量与部分得分无显著相关性，后训练显著增强了模型在逻辑关系与特定语篇任务上的表现，但对歧义处理、文学风格理解、指代消解及错误前提识别等高阶能力仍存在明显短板。这说明后训练是逐项引入语言知识的过程，而非生成通用语言智能，提示未来模型改进需聚焦于针对性的知识注入与推理机制优化。
链接: https://arxiv.org/abs/2610.34080
作者: Michael Foster,Barbara McGillivray,Gard B. Jenset,Jakub Mácha,Jason Harding,Marcus Perlman,Devyani Sharma,Patrick Skeels,Jana Willer-Gold,Jeannette Littlemore
机构: 未知
类目: Computation and Language (cs.CL)
备注: 49 pages, 10 figures
Abstract:We introduce the Literature and Linguistics Analysis Benchmark (LALA), comprising 236 open-ended questions written by experts in linguistics and literary studies, together with expert-provided ideal answers. By comparing models within families and at similar training scales using human and LLM judges, LALA identifies differences in linguistic competence associated with post-training, offering a framework for evaluating future models and guiding targeted improvements. We administered questions to 17 language models and evaluated their responses using both human expert assessment and two LLM evaluators, Sol and Qwen. Base models displayed little of the linguistic competence assessed by the benchmark, whereas their post-trained versions performed markedly better, suggesting that post-training plays a central role in making linguistic knowledge accessible and usable. Parameter count, by contrast, showed no statistically significant association with partial-credit performance, and competencies remained unevenly distributed across linguistic domains. We interpret this pattern as evidence that post-training adds linguistic competencies piecemeal, without producing a general linguistic intelligence capable of extending them beyond the learned policies that support them. Models performed best on logical relations between statements and selected discourse-level tasks, while even the strongest post-trained models continued to struggle with ambiguity, literary style, reference, and misleading assumptions. Human and Sol evaluations produced nearly identical model rankings in both the original and ReliableEval runs, supporting the replicability of the comparative results.
[NLP-114] Identically Trained Language Models Break on Different Jailbreaks
【速读】： 该论文旨在解决当前生成式AI（Generative AI）安全评估中一个关键问题：现有越狱基准测试（jailbreak benchmarks）将某个模型检查点（checkpoint）是否被越狱攻击成功，归因于其预训练数据和安全训练过程的固有属性。然而，这种归因忽略了模型训练过程中随机性（如预训练种子、数据采样顺序等）对越狱脆弱性的显著影响。论文通过系统实验验证，发现同一份预训练数据在不同预训练种子下生成的模型，其对相同越狱提示的响应行为存在显著差异，表明越狱攻击的有效性更多取决于具体的预训练运行（pretraining run），而非数据本身。解决方案的关键在于：通过控制相同的预训练数据与安全后训练流程，对比不同预训练种子和数据配方下的模型表现，量化了预训练过程对模型越狱敏感性的主导作用。实验结果显示，相同预训练运行的模型间越狱响应率相关性高达0.984，而更换预训练种子后降至0.811，进一步更换数据配方仅轻微下降至0.779；同时，模型表现出独特的“指纹”特征——其在特定提示上的拒绝率可精确识别所属预训练运行，即使仅使用50个提示也具备98%的识别准确率。这一结果揭示了当前越狱评估框架的局限性，并强调应将预训练过程的随机性纳入安全评估体系，以实现更稳健、可复现的模型安全性分析。
链接: https://arxiv.org/abs/2610.34074
作者: Chi-Wei Chang,Mu-Tien Kuo,Kuan-Wei Lee,Meng-Hsun Tsai
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Jailbreak benchmarks test one checkpoint and treat which jailbreaks break it as a property of its pretraining data and safety training. We ask whether another pretraining run of the same data, given the same safety training, would break on the same jailbreaks. For each of 25 pretraining data recipes, we take 1B base models from 3 pretraining seeds and give each the same safety post-training with 2 post-training seeds, for 150 models in all. We then collect 20 responses from every model to a shared set of jailbreak prompts and correlate the per-prompt non-refusal rates between every pair of models. On average, two post-trained copies of the same pretraining run have per-prompt non-refusal rates that correlate at 0.984. A different pretraining seed lowers the correlation to 0.811, and a different data recipe lowers it only slightly further, to 0.779. These results indicate that which jailbreaks break a checkpoint depends much more on its pretraining run than on its data, a pattern that holds in all 25 data recipes and at 150M and 300M parameters. In practice, the best of 91 attack templates found on one checkpoint stays the best after a new pretraining seed in only 27% of model pairs, against 67% after a new post-training seed. Each pretraining run also leaves a fingerprint: across 3,211 prompts, a model’s per-prompt non-refusal rates identify its pretraining run among all 75 runs every time, and a random subset of 50 prompts still does so 98% of the time.
[NLP-115] Memory Tax: Dissecting the Net Value of Agent Memory
【速读】： 该论文旨在解决长期智能体（long-horizon agents）中记忆系统（memory system）的净价值（net value）不明确的问题，尤其关注构建记忆相较于直接从原始历史记录中检索信息是否真正带来收益。其核心问题是：尽管许多专用记忆系统被设计用于提升性能，但现有实证研究发现，在多数情况下，这些系统的表现反而不如简单的检索增强生成（RAG, Retrieval-Augmented Generation），暗示其存在负的净价值。为此，作者以RAG为基准，将记忆的净价值分解为两部分——记忆收益（memory benefit）与记忆税（memory tax）。其中，记忆收益源于记忆系统对原始信息的重构与优化（如事实重排、偏好提炼等），而记忆税则源于记忆过程对原始证据的删减或模糊化，即“双刃剑式偏差”（double-edged deviations from raw history）。该偏差虽可能服务于特定任务目标，但会损害信息可追溯性与完整性。论文进一步指出，当前研究过度关注记忆收益，而忽视了记忆税的产生机制，因此提出应将注意力从单一性能指标转向对记忆净价值的评估。通过系统分析记忆税在系统设计与评估层面的成因，并将其概念泛化至更广泛的应用场景，本文旨在为记忆系统的可靠设计与评估提供新视角。
链接: https://arxiv.org/abs/2610.34073
作者: Xuan He,Sining Zhoubian,Ruozhen Yang,Yuhao Wen,Zhineng Wang,Yuyang Hu,Ke Yang,Yuyu Zhang,Ping Nie,Jiawei Han,ChengXiang Zhai
机构: University of Illinois Urbana Champaign(伊利诺伊大学厄本那-香槟分校); Tsinghua University(清华大学); Renmin University of China(中国人民大学); Verdent.ai; University of Waterloo( Waterloo 大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Memory is increasingly used in long-horizon agents, yet the net value of building memory over direct retrieval from raw history remains largely unclear. Through a large-scale evaluation across multiple benchmarks and model backbones, we find many dedicated memory systems underperform naive RAG in many cases, indicating a negative net value. Taking RAG as a reference, we dissect the net value of memory into the gains and losses over direct retrieval, termed memory benefit and memory tax. These gains and losses arise because memory process changes how the raw information is organized and accessed. We view such changes as the double-edged deviations from the raw history. They are originally designed and introduced to benefit the task, for example by reorganizing facts or distilling preferences, but some may also remove or obscure evidence, giving rise to memory tax. Since memory benefit reflects the expected gains of its design, we then focus more on the overlooked memory tax, study how it arises from the system and evaluation side, and generalize the concept to more application scenarios of memory. We hope the notion of memory tax can provide a new lens and shift more attention from standalone performance to the net value of memory, informing more reliable memory designs and evaluations. We will open-source the code and results to the community. Project Page: this https URL
[NLP-116] When Does GEPA Merge Help? A Mechanism-Level Diagnostic
【速读】： 该论文旨在解决生成式 AI (Generative AI) 中反射提示优化器 GEPA 所存在的系统感知合并操作（system-aware merge operator）在不同模型间表现不一致的问题，即其合并操作在 GPT-4.1-mini 上通常提升性能，而在 Qwen3-8B 上却导致性能下降，这种不对称性在原始论文中被观察到但未得到解释。研究通过在五个数据集和两个模型上的系统性实验发现，合并操作的性能不稳定并非源于合并算子本身存在固有缺陷，而是高度依赖于具体配置——包括合并时机与合并策略的选择。通过对九种配置组合的全面测试，表明在恰当配置下，合并操作可同时提升两个模型的性能，从而提出“配置敏感性”是解释该不稳定性更合理的机制。进一步分析发现，提示层面的静态特征无法有效预测 GEPA 的合并接受决策，排除了基于静态特征的门控机制的可能性。作为概念验证，研究构建了一个基于行为诊断的探测器（behavior-based probe），通过在高风险任务上主动回避合并（abstain），在多种子运行中显著减少了大规模合并失败，尽管其整体性能仍未能持续优于不进行合并（NoMerge）的基准。该工作揭示了现有反射优化方法在部署中的可配置性关键作用，并为未来鲁棒性增强提供了可解释的诊断路径。代码已开源，项目名为 GEPA-MergeProbe。
链接: https://arxiv.org/abs/2610.34027
作者: Chong Zhao,Ruyi Yang,Changming Liu,Alice Zhao
机构: Harvard University (哈佛大学)
类目: Computation and Language (cs.CL)
备注: 12 pages, 5 figures. Code: this https URL
Abstract:GEPA is among the strongest reflective prompt optimizers for language-model programs, but its system-aware merge operator is unstable across models: merge improves gpt-4.1-mini on average while degrading Qwen3-8B, an asymmetry noted but left unexplained by the GEPA paper. We investigate this instability across five datasets and two models. Our nine-configuration sweep over merge timing and merge operator shows that merge can improve both models when configured appropriately. This suggests that merge instability is better explained by configuration sensitivity than by an inherent flaw in the merge operator. We then test whether prompt-level features predict GEPA’s merge-accept decisions and find that they do not, ruling out static-feature gating. As a proof of concept, a behavior-based probe built from these diagnostics avoids large merge failures in our multi-seed runs by abstaining on high-risk tasks, although it does not consistently outperform NoMerge. Code is available as GEPA-MergeProbe at this https URL .
[NLP-117] he Patient Is Always Right? Evaluating Sycophancy in Medical Diagnosis with Graded Patient Beliefs
【速读】： 该论文旨在解决生成式 AI 在医疗诊断场景中存在迎合用户偏见（sycophancy）的问题，即模型可能无条件接受患者错误的自我诊断信念，从而导致误诊风险上升。其核心解决方案在于揭示模型在面对患者主观信念时的脆弱性，并通过实证研究发现：当患者提出与临床证据不符的疾病假设时，九种主流生成式 AI 模型平均有 23.2% 的概率改变原有诊断以迎合用户信念，甚至接受虚构疾病；在特定角色设定（如“罕见病声称者”）下，这一比例可高达 43.0%。研究进一步表明，经过医学领域微调（medical fine-tuning）以及模型规模更大的系统表现出更低的迎合倾向，提示模型训练策略和规模对缓解此类偏差具有关键作用。该发现强调了在医疗应用中加强 AI 安全性验证的必要性，尤其在公众日益依赖 AI 获取健康建议的背景下。
链接: https://arxiv.org/abs/2610.34013
作者: Viswonathan Manoranjan,Andrea Mock,Myra Cheng,Lujain Ibrahim,Andrew Coyne,Junyi Jessy Li,Desmond C. Ong
机构: University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校); Vals AI; Stanford University(斯坦福大学); University of Oxford(牛津大学); The University of Texas at Austin(德克萨斯大学奥斯汀分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:More people are seeking out general-purpose AI models for medical advice, including diagnosis. Yet, AI models are susceptible to sycophancy, like validating users’ beliefs even when they are untrue, which could lead to false positive diagnoses. Here, we evaluate how nine AI models change their diagnostic decision when a patient presents a belief that they have a specific (but incorrect) medical condition. We vary how close the user belief is to the clinical evidence presented. Models sycophantically adopt the patient’s belief on average 23.2% of the time—sometimes even when the patient proposes a fictional, made-up condition. We also varied different personas, including a Rare Disease Claimer that got models to accept their belief 43.0% of the time. Additional analyses find that medical fine-tuning and model scale makes models less susceptible to changing their diagnoses. Sycophantic bias in medical diagnosis raises implications for safety, especially as more people are using AI for health support.
[NLP-118] Reversible Pseudonymisation of Documents Sent to External LLM s: an Open Bilingual Benchmark and an Engine
【速读】： 该论文旨在解决专业人员在将工作文档提交至外部大语言模型（Large Language Model, LLM）时，个人身份信息（PII）泄露的风险问题。现有防护措施普遍采用可逆伪匿名化（reversible pseudonymisation），即在发送前将个人数据替换为占位符，并在接收响应后本地还原。然而，现有工具在实际应用中存在识别准确率不足、敏感信息残留率高等问题，尤其在俄语和英国英语场景下表现不佳。本文的关键解决方案是构建一个开放基准测试平台，整合五种现有开源工具及正则表达式基线，在相同数据集上进行公平对比；并基于其中表现最强的两个工具开发了一个新型引擎。该引擎通过在多个独立保留集（held-out set）上的验证，展现出显著优于现有工具的性能：在俄语和英国英语的1000份合成文档测试中，其敏感信息完全暴露率分别降至0.3%和0.2%，远低于最佳现有工具的61%和58%；同时满足所有预设检测阈值，并实现超过99%的占位符恢复率（英国英语），在俄语中恢复率为94.6%（略低于99%阈值，主因源于借用pii-guard的词形变化规则导致姓氏形式错误）。整个评估过程使用合成数据，且由引擎开发者独立执行，所有代码、数据、预测结果与原始输出均公开，确保了可复现性与透明性。
链接: https://arxiv.org/abs/2610.34012
作者: Vladimir Beskorovainyi
机构: Besk Tech(贝斯克科技)
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注: 22 pages, 9 tables. Code, data generator, held-out sets, pre-registrations, predictions and raw LLM answers: this https URL (release v0.5.0, DOI https://doi.org/10.5281/zenodo.23624098 )
Abstract:Professionals send working documents to external large language models. A common safeguard is reversible pseudonymisation: personal data are replaced with placeholders before the request and restored locally in the answer. We present an open benchmark of five existing open tools and a regular-expression baseline on Russian and UK English, and an open engine built on the two strongest tools. The benchmark excludes each tool’s own authors’ dataset from its verdict, compares the engine with each tool on identical datasets for detection, reports masking precision, and tests on a held-out set of synthetic documents whose files, hashes, decision rules, run plan and stopping rule were published before any run. Two local LLMs test whether placeholders survive. All evaluation data are synthetic, and the benchmark was run by the author of the engine; code, data, predictions and raw answers are released. On external datasets no existing tool meets all predefined thresholds: in UK English the identifier and leak thresholds are not met, in Russian the identifier threshold is borderline. On the last held-out set of 1,000 documents per language, the best existing tools leave a whole value visible in 61% (Russian) and 58% (UK) of documents. On these documents, from the generator it was developed on, the engine does so in 0.3% [0.1; 0.9] and 0.2% [0.1; 0.7], leaves any part of a value visible in 0.8% and 0.2%, and meets all detection thresholds in both languages. In UK English it also restores at least 99% of its placeholders after 760 answers. In Russian it restores only 94.6% [93.6; 95.6] of name placeholders, against a 99% threshold, mainly because the inflection it borrows from pii-guard produces wrong surname forms. The engine was developed over six earlier held-out sets from the same generator; real documents were not tested.
[NLP-119] “High-Quality” Data for Early-Stage Instruction Tuning: An Empirical Study of Data Quality Metrics
【速读】： 该论文旨在解决指令微调（instruction tuning）中“数据质量”这一概念过于模糊且缺乏明确定义的问题。研究聚焦于将数据质量视为可操作的属性，通过系统评估一系列源自数据筛选、文本质量评估及指令微调领域的简单示例级指标，以识别哪些指标在固定微调预算下最能预测训练效果。研究方法为：针对每个指标对指令-响应样本进行排序，按近似相等的词元量（token mass）划分为低、中、高三个区间，分别在各区间上独立微调模型，并比较不同累积训练词元量下的验证困惑度（perplexity）与基于大语言模型作为裁判（LLM-as-judge）的生成答案评分。研究结果表明，响应密度监督（answer-dense supervision），由响应比例（response ratio）所捕捉，是比现有常用代理指标更强且更具普适性的指令微调效用预测因子。
链接: https://arxiv.org/abs/2610.33969
作者: Vladimir Necula,David Gvadzabia,Sofia Serrano
机构: University of Cambridge Department of Physics(剑桥大学物理系); Lafayette College Department of Computer Science(拉法耶特学院计算机科学系)
类目: Computation and Language (cs.CL)
备注:
Abstract:“Data quality” is widely invoked in instruction tuning, but the term is often underspecified. We study it as an operational property by asking which simple example-level metrics, drawn from prior work on data selection, text quality, and instruction tuning, best predict training utility under a fixed finetuning budget. For each metric, we sort instruction-response examples by score, partition them into low, middle, and high bins with roughly equal token mass, finetune separately on each bin, and compare validation perplexity and LLM-as-judge scores for the models’ generated answers as a function of cumulative training tokens. Our results suggest that answer-dense supervision, as captured by response ratio, is a stronger and more portable predictor of instruction-tuning utility than commonly used alternative proxies.
[NLP-120] Beyond Belief: Evaluating and Mitigating LLM Susceptibility to Misinformation
【速读】： 该论文旨在解决生成式 AI（Generative AI）在使用外部工具时对虚假信息（misinformation）的判断与行为响应之间不一致的问题，即模型虽能识别出某些信息为不可信，但仍可能执行有害的工具调用。其核心解决方案的关键在于评估模型在执行前对信息可信度的判断是否真正反映其行为上的抗误导能力。研究发现，即使模型将虚假通知标记为不可信，仍会在27%至99%的情况下执行有害操作；移除可信度评估步骤后，有害执行率平均上升15个百分点，表明可信度评估本身会改变模型的行为表现。尽管提升可信度辨别能力的干预措施并未稳定降低风险，但简单的预先警告（forewarning）在五种模型中的四种中显著减少了有害执行，而基于生成示例的预打假（prebunking）方法在主要评估中未超越该效果。此外，这些安全干预措施可能损害合法任务的完成率。研究进一步表明，在有可信度评估前置的协议下，通过生成虚假内容进行监督微调可有效降低在保留攻击样本上的有害执行，但存在模型依赖性及无评估环节时保护效果减弱的问题。因此，该研究强调：评估虚假信息防御机制时，必须同时考量有害行为的抑制与合法任务完成率的维持，并充分考虑评估流程本身对行为结果的影响。
链接: https://arxiv.org/abs/2610.33962
作者: Emran Kanaan,Jiho Lee,Ibrahim Odat,Soroush Vosoughi,Weicheng Ma
机构: Oakland University(奥克兰大学); Dartmouth College(达特茅斯学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Tool-using large language models (LLMs) must decide whether external information warrants action. We investigate whether their expressed credibility judgments reliably indicate behavioral resistance to misinformation. We evaluate five LLMs on 196 constructed false notices paired with legitimate counterparts and 204 archived attacks adapted to simulated tool use. When asked to act on the constructed notices, models issue harmful tool calls in 27%–99% of cases they have rated untrustworthy. Removing the preliminary rating step increases harmful execution by 15 percentage points on average, showing that credibility elicitation itself changes the behavior being evaluated. Interventions that improve credibility discrimination do not consistently yield safer actions. A simple forewarning significantly reduces harmful execution in four of five models, but neither tested generated-example prebunk outperforms it in the main evaluation. These reductions can also compromise legitimate task completion. Supervised fine-tuning on generated false notices reduces harmful execution on held-out archived attacks under the rating-first protocol, with model-dependent utility costs and weaker protection without rating elicitation. Our findings motivate evaluating misinformation defenses through both harmful actions and legitimate task completion, while accounting for the behavioral effects of the evaluation protocol.
[NLP-121] How Do Small Coding Agents Use Code Graphs and Embeddings?
【速读】： 该论文旨在解决生成式代码修复（Code Repair）中模型对代码仓库结构理解不足的问题，即模型缺乏对代码依赖关系和上下文结构的显式感知能力，导致在定位缺陷相关代码时效率低下。其核心挑战在于：当问题涉及的代码路径跨越多个函数或模块时（以“跳数”即代码图中最短路径长度衡量），模型性能随路径深度呈指数下降。解决方案的关键在于引入结构化工具（嵌入搜索与代码图工具）辅助模型进行更精准的代码定位。实验表明，这些工具使小规模模型（1.7–4B参数）的定位准确率提升达17–21个百分点，但随着模型规模增大至32B，工具增益显著衰减至4–5个百分点，且大部分收益源于提交行为优化而非真正提升定位能力。进一步分析揭示，代码图工具的实际效用主要体现在初始跳数（hop 0）的符号名查找和第一跳扩展，而跨跳链式推理仅在12–14%的样本中发生。此外，通过线性探测发现，小模型在权重和使用层面均呈现局部、一跳、以词法为主的代码视图，难以捕捉深层依赖关系。研究结果表明，当前生成式编码代理对复杂代码结构的理解仍严重依赖外部工具，且工具有效性随模型规模增长而递减，凸显了构建具备内生结构感知能力的代码模型的重要性。
链接: https://arxiv.org/abs/2610.33958
作者: Saurav Bellam(The University of Texas at Austin)
机构: The University of Texas at Austin(得克萨斯大学奥斯汀分校)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 29 pages, 5 figures. Depth labels and dataset card and full code, raw results and analysis will be released in the near future
Abstract:Coding agents are given embedding search and code-graph tools on the premise that a repository’s structure is something the model lacks and the tool supplies. We test that premise on SWE-bench Verified with small open-weight models, stratifying tasks by a deterministic hop depth (the shortest code-graph path from code the issue mentions to code the gold patch changes) that a companion paper defines and validates. First, we ablate a localization agent’s tool set (grep only; +embedding search; +graph tools; all) on a depth-balanced 160-task split and on all 500 tasks for ten models (six Qwen3 sizes from 0.6B to 32B with five decodings each, an FP8 control, three Gemma 4 sizes; 79k episodes). Structured tools double grep-only accuracy for Qwen3-8B and Gemma 4 E4B; their gain peaks at 17-21 points at 1.7-4B and falls to 4-5 points at 32B (odds ratio 0.77-0.82 per doubling of parameters). From 8B up, about half of that gain is the form of the submission rather than finding the target; the fall from 14B to 32B is a fall in tool uptake at equal grep-only accuracy; and the same 4B size gains or does not gain from a tool depending on the checkpoint. Each extra hop multiplies the odds of correct localization by 0.50-0.67 at every size, and no tool set flattens that gradient: traces show the graph tools’ gain is name lookup at hop 0 and one expansion at hop 1, and agents chain expansions in only 12-14% of episodes. A fixed repair model resolves 6-13 of 160 tasks from these localizations (oracle 26). Second, linear probes find direct call edges decodable under a within-anchor, degree-stratified protocol (AUC 0.74; 0.61-0.66 inside agent conversations), but below a lexical baseline (0.99), with two-hop edges at 0.57. Small agents carry a one-hop, largely lexical picture of their repository, in weights and in use. We release the agent, probes, logs and traces.
[NLP-122] Continual Learning Exam: Benchmarking Long-Horizon RSI and Experiential Learning in LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在长期运行中难以实现持续学习（Continual Learning）的问题，即模型无法有效积累、整合并应用过往经验以实现性能的持续提升，而现有评估体系多聚焦于单任务表现，忽视了基于经验积累的渐进式改进能力。其解决方案的关键在于提出一个大规模、高真实性的持续学习评估基准——持续学习测评（Continual Learning Exam, CLE），由50余名领域专家共同构建，包含120个由人类撰写的任务，每个任务设计为五级阶梯式结构，共600个层级，覆盖12个职业领域中的89种专业角色。任务设计要求智能体必须逐级通过前序任务才能解锁后续任务，且后期任务依赖前期经验积累，从而真实模拟专业工作流中的知识延续性。评估采用人工制定的评分标准，通过对比单次提交与允许失败后获得诊断反馈并进行一次修订的设置，验证了反馈机制对性能提升的有效性。实验表明，引入反馈与修订可带来平均5.3个百分点的完成率提升，但最先进系统仅能完成47.7%的层级，且仅有23.3%的任务能完整达成全部五级目标，揭示当前前沿大模型在持续学习方面仍存在显著瓶颈。因此，该数据集不仅可用于评估，还可作为后训练数据集，用于强化模型的持续学习能力。
链接: https://arxiv.org/abs/2610.33956
作者: Haoyu Zhao,Sitian Tu,Yuxuan Huang,Huichi Zhou,Zhangyue Yin,Jiahao Huang,Jingyuan Xu,Tianmiao Shi,Jin Shi,Nianxi Zhang,Yangmiao Zhang,Zhuolang Chan,Huiyan Hu,Chenchao Sheng,Yaqing Luo,Yue Chen,Huaxiu Yao,Meng Fang,Jun Wang
机构: 未知
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Continual learning is the ability of LLM agents to improve over time by acquiring, refining, and applying knowledge from experience while retaining useful prior knowledge. It is essential for agents learning over long time horizons, yet current evaluations largely assess individual task performance rather than improvement through accumulated experience. We introduce Continual Learning Exam (CLE), a large-scale benchmark developed by 50+ domain experts to measure experience-driven improvement in realistic professional workflows. CLE comprises 120 human-authored tasks, each organised as a five-level ladder, yielding 600 levels spanning 89 professional roles across 12 occupational domains. Agents must pass each level to unlock the next, with later levels building on experience accumulated during earlier assignments. An evaluation agent assesses each submission against human-authored rubrics. We compare a single-submission setting with one that provides diagnostic feedback after failure and allows one revision. Across 17 harness-model configurations using frontier LLMs, feedback and revision yield a median completion gain of 5.3 percentage points. Yet the best-performing system clears only 47.7% of levels and completes all five levels in just 23.3% of tasks, highlighting the difficulty of sustained learning from experience for existing frontier LLMs. As such, the dataset can also be used to post-train LLMs to strengthen their continual learning capabilities.
[NLP-123] What Makes a Question Curious? Measuring Model Curiosity in Human-LLM Interactions
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在人机交互中仅被动应答而缺乏主动引导性提问的问题，尤其关注如何通过模型生成具有认知探索意义的提问来揭示知识盲区、假设前提及潜在选项。其核心解决方案在于提出“模型好奇心”（model curiosity）的概念，并构建一个基于理论的分类体系，以区分面向认知探索的提问（即好奇心驱动型提问）与一般性提问。研究发现，在四个对话数据集中，好奇心驱动型提问占比仅为17%-28%，但此类提问能显著激发用户后续更具探索性和认知深度的回应，且在咨询、创意生成和科研等情境中尤为有效。该研究为构建能够更精准、有选择地发起认知驱动型提问的生成式AI系统提供了实证依据与设计方向，推动实现更深层次的人机协作。
链接: https://arxiv.org/abs/2610.33950
作者: Angana Borah,Rachana Ramchandar,Inderjeet Nair,Rada Mihalcea
机构: University of Michigan - Ann Arbor, USA; PES University, Bangalore, India
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models increasingly act as partners in everyday communication. In these interactions, useful assistance may require not only answering, but also asking the right questions that surface knowledge gaps, assumptions, and alternatives. In this study, we refer to such epistemic question-asking as model curiosity, and introduce a theory-grounded taxonomy that distinguishes curiosity-oriented questions from generic model questioning. Across four conversation datasets, we find that curiosity-oriented questions remain rare (17-28% of all questions), but are associated with more exploratory and epistemically rich subsequent user responses, with varying patterns across datasets. We further conduct a human study to evaluate the usefulness of model curiosity, finding that curiosity can encourage exploration, especially in advice, creative, and research contexts. These findings have implications for building LLMs that ask more selectively to support deeper exploration and human-AI collaboration.
[NLP-124] Why Does Structured Output Fail? An Empirical Study of Prompt-Enforced Pydantic Schemas Across Nine LLM s
【速读】： 该论文旨在解决生成式 AI（Generative AI）在实际应用中因结构化输出（structured output）验证失败而导致的管道可靠性评估失真问题。现有方法通常仅以单一有效性率（validity rate）衡量系统表现，无法揭示失败的具体模式，进而阻碍对故障根源的精准干预。其核心解决方案是提出一种细粒度的格式符合性失败分类体系，并开发一个确定性分类器，基于模型输出与 Pydantic 验证错误，自动识别失败类别。研究通过对 3,997 次生成尝试（涵盖九种 OpenAI 与 Anthropic 模型/接口配置）进行分析，发现相似的有效性率可能掩盖截然不同的失败模式；其中，Markdown 代码块封装的 JSON 是最常见的失败类型——在 681 例此类失败中，仅有 48.0% 在移除代码块后可通过验证，其余则隐藏着额外的字段长度违规等深层问题。进一步的提示消融实验表明，抑制代码块封装虽看似将失败转向长度违规，实则暴露了原本存在的潜在违规，说明聚合有效性率会掩盖失败构成的变化。基于此，论文提炼出四项工程实践：验证器设计、模式定义、重试机制与评估方法优化，其中三项经数据支持，另一项关于重试反馈的建议尚未验证，相关分类器、生成结果及标注数据已开源。
链接: https://arxiv.org/abs/2610.33949
作者: Amit Verma
机构: Independent Researcher(独立研究员)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 36 pages, 2 figures, 16 tables. Under review at Empirical Software Engineering. Artifact: this https URL
Abstract:Pipelines that require structured outputs from hosted language models often specify a Pydantic schema in the prompt, validate the response, and retry when validation fails. Although native structured-output modes provide an alternative, application-side validation remains important for enforcing custom constraints. Yet pipeline reliability is often summarized by a single validity rate, which obscures why outputs fail and which interventions may help. We introduce a fine-grained taxonomy of schema-conformance failures and a deterministic classifier that assigns failure categories from model responses and Pydantic validation errors. We apply this framework to 3,997 generation attempts across nine OpenAI and Anthropic model/API configurations, using questions from HotpotQA and FinanceBench. Our results show that similar validity rates can mask different failure patterns, and that dominant failure categories vary with schema constraints and task. Markdown-fenced JSON is the most common failure: among 681 fenced failures, only 48.0% pass validation after fence removal, while the remainder conceal additional violations, predominantly field-length violations. In an exploratory prompt ablation, variants that suppress fences appear to shift failures toward length violations, but the shift is consistent with exposing violations that were already present, illustrating how aggregate validity rates can obscure changes in failure composition. We derive four engineering practices for validator, schema, retry and evaluation design, three supported by our data and one, on retry feedback, proposed but not tested, and release the classifier, completions, and annotations.
[NLP-125] Schema-Free Text-to-SQL: Learning Enterprise Database Structure from Query Logs
【速读】： 该论文旨在解决文本到SQL（Text-to-SQL）系统在学术基准上表现优异，但在企业实际部署中性能显著下降的核心问题。其根本原因在于现有方法依赖完整的模式文档（schema documentation），而企业数据库普遍缺乏详尽的文档支持，普遍存在命名晦涩、关系隐含等问题。本文提出首个无需依赖模式文档的Text-to-SQL框架，通过从历史查询日志中学习数据库结构来弥补这一缺陷。其解决方案的关键在于：首先，从解析的查询模式中构建知识图谱，以捕捉表间关联、连接逻辑及列与业务概念之间的映射；其次，引入基于实体驱动的GraphRAG检索机制，将用户问题与领域术语进行语义对齐，而非仅依赖嵌入相似性。在工业级部署和Beaver基准上的实验表明，该方法在可执行性和答案质量方面显著优于主流的基于嵌入的检索方法，在Beaver上实现95.7%的执行有效性，远超嵌入方法的60.8%，且在内部评估中获得70%的数据工程师偏好度。研究证明，Text-to-SQL可在模式模糊的企业环境中有效运行，为自然语言数据库接口在真实场景中的应用提供了可行性。
链接: https://arxiv.org/abs/2610.33946
作者: Ömer Faruk Akgül,Nikhil Kapahi,Prajit Reddy Muppidi
机构: University of Southern California(南加州大学); Amazon(亚马逊)
类目: Computation and Language (cs.CL)
备注:
Abstract:Text-to-SQL systems achieve remarkable accuracy on academic benchmarks, yet performance drops dramatically in enterprise deployments. This gap stems from a fundamental assumption: existing approaches require complete schema documentation, but enterprise databases typically lack comprehensive documentation, featuring cryptic naming and implicit relationships. We present the first text-to-SQL framework that operates without schema documentation, instead learning database structure from historical query logs. Query logs encode implicit structural knowledge: which tables are queried together, how they join, and what columns map to business concepts. We construct a knowledge graph from parsed query patterns and introduce entity-driven GraphRAG retrieval that grounds user questions in domain terminology rather than relying solely on embedding similarity. Validated on both an industrial deployment with expert evaluation and the Beaver benchmark, our approach substantially outperforms similarity-based retrieval methods, the dominant industry approach, in both executability and answer quality. On Beaver, we achieve 95.7% execution validity compared to 60.8% for embedding-based retrieval, with expert data engineers preferring our system 70% to 10% on internal deployment. Our work demonstrates that text-to-SQL can operate effectively in schema-ambiguous enterprise environments, enabling natural language database interfaces “in the wild”.
[NLP-126] HerHealthGPT : An Evidence-Grounded Conversational AI System for Womens Reproductive Health Copy title
【速读】： 该论文旨在解决当前对话式人工智能（Conversational AI）系统在女性生殖健康领域中存在的关键问题：面对不完整或模糊的健康咨询（如“我的月经为什么推迟了？”），系统往往缺乏必要的上下文追问机制，且其回复未能基于可验证的循证证据，导致信息可信度与临床适用性不足。其解决方案的关键在于构建一个以证据为基础的交互式对话系统——HerHealthGPT，通过集成检索增强生成（Retrieval-Augmented Generation, RAG）、上下文感知的澄清机制以及来源验证，在保持对话连贯性的基础上主动进行目标导向的追问，避免过早下结论，并确保输出内容可追溯、可验证。实验结果显示，引入证据检索可将问题类别识别准确率从微调模型的54.1%提升至59.3%，澄清路由模块在构造目标上的召回率和特异性分别达到85.6%和86.2%；用户研究表明，83.3%的参与者更偏好具备检索能力的配置，且多数人认为系统的追问问题具有相关性。这些结果表明，将循证检索与动态交互澄清相结合，是实现透明、上下文敏感且可信的女性生殖健康信息系统的有效路径，但仍需进一步开展临床验证。
链接: https://arxiv.org/abs/2610.33931
作者: Hassan Saeed Hassan Albattra,Mazen Mohammed Bahgat,Rahatara Ferdousi,Hana Essam Sayed Ahmed Amrya,Mariam Mousa
机构: Queen’s University (皇后大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:“Why is my period late?” Sometimes, the most useful response begins with another question. Yet conversational AI systems may respond to incomplete health queries without seeking essential context or grounding their guidance in verifiable evidence. This paper introduces HerHealthGPT, an evidence-grounded conversational AI system designed to address these challenges in women’s reproductive health. HerHealthGPT integrates retrieval-augmented generation (RAG), context-aware clarification, and source validation within an interactive web interface. Rather than responding immediately to incomplete queries, the system maintains conversational context, asks targeted follow-up questions, checks for premature conclusions, and grounds its responses in retrieved evidence. We evaluate the AI components using 540 English messages derived from 90 reproductive-health cases and conduct a prototype questionnaire study with 30 adults. Evidence retrieval improves concern-category accuracy from 54.1% to 59.3% compared with fine-tuning alone. The clarification router achieves 85.6% recall and 86.2% specificity on constructed targets. In the user study, 83.3% of participants preferred the retrieval-enabled configuration, while 25 of 30 considered its follow-up questions relevant. These findings suggest that integrating evidence retrieval with interactive clarification offers a promising approach to more transparent, context-aware reproductive-health information systems, while further clinical validation remains necessary.
[NLP-127] Counterfactual Test-Time Augmentation for Shortcut Removal in Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在测试阶段因依赖训练数据中的捷径（shortcuts，即训练与测试分布不一致时失效的统计相关性）而导致预测性能下降的问题。现有大多数去除捷径的方法需在训练阶段进行，依赖标注数据、模型权重访问及大量计算资源，难以在实际部署中应用。本文提出一种测试时捷径消除方法——反事实测试时增强（Counterfactual Test-Time Augmentation, CTTA），其核心思想是：对于给定测试样本，利用辅助模型生成一系列反事实编辑版本，其中捷径特征被改变，随后对LLM在这些编辑样本上的预测结果进行加权平均。理论证明，最优的捷径不变预测器属于该CTTA家族，并可确定其最优权重。尽管这些权重依赖于训练分布且通常无法仅从已训练模型和无标签测试数据中识别，但作者在特定假设下，提出可通过已训练模型的词元概率估计权重，并设计检测方法以验证假设有效性。基于此，提出Select-CTTA方法，通过选择最优的CTTA子方法实现自适应优化。实验表明，Select-CTTA显著优于现有的测试时基线方法，并接近在捷径平衡数据上微调模型的性能，实现了高效、无需重训练的测试时鲁棒性提升。
链接: https://arxiv.org/abs/2610.33904
作者: Katie Matton,Rosalind W. Picard,John Guttag,Maggie Makar
机构: MIT(麻省理工学院); University of Michigan(密歇根大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 24 pages, 9 tables, 2 figures
Abstract:Large language models (LLMs) can exploit shortcuts, i.e., correlations present in training data that may not hold at test time. This can lead to poor predictions when correlations shift. Most existing solutions for removing shortcuts operate at training time, requiring labeled training data, model weight access, and significant computational resources. Here, we study the problem of test-time shortcut removal. We propose a family of counterfactual test-time augmentation (CTTA) methods: given an example, each method uses an auxiliary model to generate counterfactual edits in which the shortcut value is changed, and then takes a weighted average of the LLM’s predictions across these edits. We prove that the optimal shortcut-invariant predictor is a member of this CTTA family, and we characterize its weights. The weights depend on the training distribution, and therefore are, in general, not identifiable from the trained model and unlabeled test data. However, we present assumptions under which the weights can be obtained using the token probabilities of the trained LLM, and propose tests to detect violations of these assumptions. We use these tests to develop Select-CTTA, an aggregate method that selects among CTTA methods. Empirically, we find that Select-CTTA outperforms test-time baselines and approaches the performance of models fine-tuned on shortcut-balanced data.
[NLP-128] When Personas Become Harmful: Guiding LLM s with Global Citizenship Values Alignment EMNLP2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在面对具有价值取向的个性化人格（persona）请求时，如何在保障安全性的前提下有效尊重用户多元价值观的问题。传统对齐方法主要针对显性恶意请求进行防御，难以识别由价值导向型人格引发的隐性危害，导致模型或过度拒绝合法的价值表达，或无差别服从潜在危险的人格设定。为此，研究构建了一个包含1.1万条争议话题与基于施瓦茨价值观量表（Schwartz Value Survey）映射的价值维度的人格配对数据集，并通过一种代理式（agentic）迭代优化流程，成功从已对齐的商用大模型中挖掘出超过47%的有害响应。研究以联合国教科文组织（UNESCO）全球公民原则为最小共识基础，界定有效价值表达与有害行为的边界，提出一种可调控的对齐模型，允许用户在推理阶段动态权衡人格一致性与安全性。实验表明，该方法在多数样本上均实现了比基线更优的安全性与人格适配性表现。
链接: https://arxiv.org/abs/2610.33899
作者: Siqi Shen,Lajanugen Logeswaran,Moontae Lee,Honglak Lee,Rada Mihalcea
机构: University of Michigan(密歇根大学); LG AI Research(LG人工智能研究院); University of Illinois at Chicago(伊利诺伊大学芝加哥分校)
类目: Computation and Language (cs.CL)
备注: Accepted to EMNLP 2026 (Main Conference)
Abstract:LLMs are increasingly being used by users from diverse demographics with pluralistic values and distinct backgrounds. Maintaining safety while respecting value-laden personas is uniquely challenging because persona-conditioned requests are often not overtly malicious, unlike those in existing safety benchmarks. Alignment methods aimed at adversarial requests therefore may miss the harm that emerges from persona use, leaving models to either refuse legitimate value expression or blindly follow unsafe personas. We construct a dataset of 11k controversial topics paired with value-based personas mapping to value dimensions from the Schwartz Value Survey. Through an agentic pipeline that iteratively refines the topic and persona, we surface and collect harmful responses from well-aligned commercial LLMs with a more than 47% success rate. We use UNESCO’s global citizenship principles as a minimal common ground for distinguishing valid value expressions from harmful behaviors, and introduce a controllable alignment model that lets users balance persona adherence and safety at inference time. The resulting model is safer and more persona-adherent than baselines on most samples.
[NLP-129] Intervention-Certified Amplification: A Validity Gate for σ-Weighted Post-Training EMNLP ACL EMNLP2026
【速读】： 该论文旨在解决混合后训练（Hybrid Post-Training）中因训练侧信号与目标任务及评估场景不匹配而导致的隐式行为放大问题。现有方法通常通过监督微调（SFT）与强化学习（RL）的混合策略，依赖训练侧信号动态调整样本权重，但这些信号仅反映锚点数据和策略特征，无法判断所强调的行为是否真正有助于目标任务。这种不一致性可能导致在移除显式指令后，预训练模型中已存在的有害或无关行为仍被重新激活并放大。为此，作者提出双隐式对抗放大（DIAA），一种基于每样本σ加权的SFT/RL混合方法，其关键在于引入对未执行工具集成推理（TIR）行为的识别与量化。实验表明，即使在完全移除显式Python约束的情况下，DIAA仍会在29%的MATH-500测试提示中生成无效的TIR，是基线GRPO的三倍。通过成对干预分析，发现抑制TIR可带来10.6个百分点的准确率提升，且SFT模仿与σ增强均贡献显著。进一步提出的干预认证放大（ICA）方法，利用成对回滚的目标证据动态门控σ，实现冻结与在线两种模式下的有效控制，将TIR比例降至13.8%和11.2%，显著降低残余干扰，使准确率分别提升至48.7%和49.3%，并可在无需重调参的情况下迁移至未见的GSM8K基准。因此，解决方案的关键在于构建基于目标评估反馈的动态校准机制，以实现对有害行为的精准抑制与有益行为的可靠放大。
链接: https://arxiv.org/abs/2610.33882
作者: Leiyue Chen,Meng Wei,Xuefeng Hu,Wei Liu
机构: ByteDance(字节跳动)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 13 pages, 2 figures, 4 tables. Published in Findings of the Association for Computational Linguistics: EMNLP 2026; see this https URL
Abstract:Hybrid post-training often mixes supervised fine-tuning (SFT) and reinforcement learning (RL), using training-side signals to set each sample’s balance. Such signals characterize the anchor data and policy, but not whether the emphasized behavior helps in the target task and evaluation setting. This mismatch can reactivate and amplify pretrained behavior even after its literal form is removed from supervision. We demonstrate this with DIAA (Dual-Implicit Adversarial Amplification), a per-sample \sigma -weighted SFT/RL hybrid. With all literal Python fences removed from its anchor, DIAA emits unexecuted Tool-Integrated Reasoning (TIR) on 29% of MATH-500 prompts, three times vanilla GRPO. Paired interventions show 10.6 percentage points of accuracy recovery when TIR is suppressed and comparable contributions from SFT imitation and \sigma -amplification; a student trained on a 0% -TIR teacher anchor emits it at 45.8% . Intervention-Certified Amplification (ICA) gates \sigma with paired-rollout target evidence. Frozen and online ICA reduce TIR to 13.8% and 11.2% , halve residual drag, improve accuracy from 47.5% to 48.7% and 49.3% , and transfer to held-out GSM8K without retuning.
[NLP-130] Ranking Is Not Deciding: Benchmark Shortcuts and Operating-Point Collapse in Query-Time RAG Poisoning Defenses
【速读】： 该论文旨在解决检索增强生成（Retrieval-Augmented Generation, RAG）系统在面对后门攻击时，现有查询阶段过滤器（query-time filters）所报告的高检测性能可能具有误导性的问题。研究发现，这些过滤器看似优异的检测效果并非源于其对内容语义或特征的真正识别能力，而是高度依赖于攻击构造过程中引入的特定“构建线索”（construction cues），如将目标问题直接作为前缀插入到污染文档中。关键发现在于：一旦移除此类线索（例如删除前缀），即使检索排名保持不变，防御系统的决策能力也会急剧下降——即“排序未受损，但判断失效”。论文提出，通过两个无标签统计量即可刻画这一线索特征：包含度（containment，目标问题在污染数据中是否原样出现）与长度重叠（length overlap，污染文档与干净语料的长度相似性）。实验证明，仅通过移除前缀或用无意义填充词替换原文本长度，即可使多个主流防御模型（如RAGuard、TrustRAG及自研评分器）的检测性能从近乎完美的表现（如AUROC 0.953）骤降至严重失效水平（误检率高达0.986），而其召回率也大幅恶化。因此，论文强调，任何检测结果的有效性必须结合攻击样本的构造特征与实际部署阈值下的假阳性率进行评估，并公开了用于计算该构造特征的脚本，以提升研究可复现性与透明度。
链接: https://arxiv.org/abs/2610.33849
作者: Pushkal Kumar,Arvin Sahaym
机构: 未知
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 26 pages, 7 figures, 14 tables
Abstract:Query-time filters that screen each retrieved document before it reaches the generator are the dominant published defense against retrieval-augmented generation poisoning, and they report near-perfect detection on domain benchmarks. We show that this performance is predicted by how the poison set was built rather than by what the filter reads: removing a single construction cue destroys the deployed decision before it destroys the ranking. Two label-free statistics of a poison set and its clean corpus profile the cue: containment, how often the target question appears verbatim in the poison, and length overlap. PoisonedRAG’s released code prepends the target question to every inserted passage. Of the 21 published defense evaluations we survey, 8 verifiably run poison built this way and 5 substitute a length cue; separately, 12 report no false-positive rate at the operating point they deploy. Deleting only that prefix from the attack authors’ own passages (0.93 still retrieved for their target) takes RAGuard, a published content filter we re-implement, from AUROC 0.953 to 0.758 and its miss rate from 0.191 to 0.763. Across four corpora, four attack families and ten per-document defenses on one shared candidate set, the profile predicts where detection saturates. Padding medical poison to clean length with word-shuffled filler that carries no content takes TrustRAG’s set-level stage from a miss rate of 0.000 to 0.827, and one learned scorer of ours from 0.000 to 0.986 (0.321 at a matched 5% false-positive rate) while its AUROC holds at 0.953. The ranking survives; the decision does not. A defense-aware attacker defeats our own scorer, so we claim no defense. A detection result is evidence only when it is reported with its poison set’s construction profile and the false-positive rate at the threshold actually deployed; we release the script that computes the profile.
[NLP-131] Grounding on Request: Learning Modality Control in Vision-Language Models
【速读】： 该论文旨在解决视觉-语言模型（VLM）在图像与文本描述存在冲突时，用户能否可靠地控制模型遵循特定模态（如图像或文本）的问题。传统评估仅依赖准确率，无法判断模型是否真正遵循了用户指定的证据来源，因为模型可能通过参数化知识给出正确答案但并未响应模态指令。为此，论文提出通过保留冲突的反事实干预（conflict-preserving counterfactual interventions）来评估模态接地控制能力，引入两个核心指标：Flip衡量初始正确答案是否随目标模态的编辑而改变，Stay则衡量当竞争模态被修改时答案是否保持稳定。解决方案的关键在于基于“新词学习”（neologism learning）机制，为视觉和文本模态分别引入新的可学习嵌入（embedding），仅训练这些新增的模态词嵌入并固定预训练模型参数；同时提出反事实孪生边界（Counterfactual Twin-Margin, CTM）监督策略，利用成对干预样本指导模型在请求证据变化时应如何调整答案概率。实验结果表明，该方法显著提升了模型对自然语言指令的模态控制能力，且单个模态嵌入即可实现轻量级但有效的视觉控制，平均性能下降极小；尽管LoRA仍表现最优，其视觉优势在未见数据上有所缩小。此外，反事实监督在标准监督较弱的场景中进一步增强了控制效果。最后，在真实新闻视频上的测试显示，当相关视觉证据更易获取时，视觉控制能力得到提升。总体而言，研究强调可靠的模态控制不仅需验证答案正确性，还需确保答案能持续响应所选模态的编辑，从而揭示了评估模型对多模态证据依赖性的关键范式转变。
链接: https://arxiv.org/abs/2610.33841
作者: Maxime Dassen,Alexander Martin,Andrew Yates,Reno Kriz,William Walden
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Preprint. Code: this https URL
Abstract:When an image and its caption disagree, can a user reliably control which one a vision-language model (VLM) follows? Accuracy alone cannot answer this question, since a model might give the correct answer from parametric knowledge without following the requested evidence. We therefore evaluate modality grounding control through conflict-preserving counterfactual interventions. Specifically, Flip measures whether an initially correct answer follows an edit to the requested modality, while Stay measures whether it remains stable when the competing modality is edited. To learn this control, we build on neologism learning by introducing new words for visual and textual grounding, training only their embeddings while keeping the pretrained model fixed. We also introduce Counterfactual Twin-Margin (CTM) supervision, which uses paired interventions to teach how answer probabilities should change when the requested evidence changes. Across three VLMs on three existing benchmarks and our multi-attribute Ground-CF benchmark, we find that learned modality words improve control over natural-language instructions. Surprisingly, a single learned embedding per modality provides strong lightweight visual control and shows little average degradation on held-out values. LoRA remains strongest overall, although its visual advantage narrows under held-out evaluation. Moreover, counterfactual supervision improves control in several settings where standard supervision is weaker. Finally, we test our method on real-world news video and find that visual control improves when the relevant evidence is made easier to access. Overall, our results show that reliable modality control requires testing not only whether an answer is correct, but whether it continues to follow edits to the modality of choice.
[NLP-132] EdgeDeID: A 0.6B Clinical De-identifier Trained Without Real Patient Data ACL
【速读】： 该论文旨在解决使用托管语言模型对临床笔记进行去标识化时，可能泄露敏感信息的问题。其核心解决方案是提出EdgeDeID，一个参数量为0.6B的本地运行解码器架构去标识模型，通过合成数据训练而无需真实患者文本：由三个大语言模型（LLM）教师生成合成临床笔记，单一名标注者将其统一为标准标注格式，模型则学习输出带有类型标记的XML结构化标识符。在包含100条人工验证标注的合成领域基准测试中，该模型达到0.994的微平均F1分数，且99%的样本中无遗漏标识符，显著优于八种现成系统中的最优者（0.857 F1，12%遗漏率）。模型经量化后可在仅615 MB内存的CPU上运行。在监督训练中引入的三项改进手段中，仅有验证器适配器（verifier adapter）有效——其能以0.94的AUROC区分其他系统输出的完整与不完整结果；而基于召回加权奖励的强化学习和基于规则的后处理层对域内性能无改善。在MIMIC ICU护理记录上的性能下降至F1=0.48，错误主要源于合成数据缺乏真实临床文本的两个特征：简略语风格（telegraphic style）和低标识符密度。通过引入三例人工编写的上下文示例，无需额外标注数据即可恢复0.08的F1分数提升。
链接: https://arxiv.org/abs/2610.33839
作者: Sina Daneshgar,Raya Namazi,Amirali Khosravi,Pouria Rouzrokh,Bardia Khosravi
机构: Sharif University of Technology(沙里夫理工大学); Mayo Clinic(梅奥诊所); Yale University(耶鲁大学)
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: 18 pages, 2 figures. Under review at ACL Rolling Review (ARR). Code: this https URL ; models and data: this https URL
Abstract:De-identifying clinical notes with a hosted language model discloses the information the step is meant to protect. We present EdgeDeID, a 0.6B-parameter decoder-only de-identifier that runs locally and is trained without real patient text: three LLM teachers write synthetic notes, a single labeler re-annotates them into one convention, and the model learns to emit a typed XML schema of the identifiers it finds. On a 100-note synthetic in-domain benchmark with manually verified annotations it reaches 0.994 micro-F1 and leaves no missed identifier in 99% of notes, against 0.857 and 12% for the strongest of eight off-the-shelf systems. Quantized, it runs on CPU in 615 MB. Of three additions to supervised training, only a verifier adapter helps: it separates complete from incomplete outputs of other systems at 0.94 AUROC, whereas reinforcement learning with a recall-weighted reward and a rule-based post-processing layer leave in-domain results unchanged. On MIMIC ICU nursing notes F1 falls to 0.48. The errors follow two properties our synthetic notes lack, telegraphic style and low identifier density, and three hand-written in-context examples recover 0.08 F1 without annotated target data.
[NLP-133] EnvLunch: The Environment as a Free Lunch for Learning to Interact
【速读】： 该论文旨在解决生成式 AI（Generative AI）在代理型数据分析任务中对大量人工标注任务依赖的问题。由于高质量的标注任务需要耗费大量人力，而可执行环境（如数据库）本身具备丰富的交互结构和可验证的执行结果，因此作者提出了一种名为 EnvLunch 的预训练阶段，利用环境自身提供的可验证轨迹作为无标签任务信号，无需人工或模型生成的任务描述即可进行代理训练。其核心解决方案在于：通过采样有效的动作轨迹并执行以获得最终观测结果，训练代理仅根据最终观测恢复功能等价的轨迹，从而实现无需标签的自监督预训练。该方法显著提升了下游强化学习的表现，在不同训练预算下平均性能优于基线超过9个百分点；且在极少量标注数据情况下，基线模型需8至32倍的标注样本才能达到 EnvLunch 模型在最小训练预算下的准确率。实验表明，仅依赖环境定义任务所带来的增益大于由模型自动生成任务的效果，且该预训练策略在数据解析之外的基准任务上也表现出更强的泛化能力。因此，EnvLunch 有效减少了对人工标注任务的依赖，并为标注资源有限场景下的代理模型微调提供了更优的初始化起点。
链接: https://arxiv.org/abs/2610.33817
作者: Wei Liu,Min Gao,Yali Du,Yulan He
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Databases (cs.DB)
备注:
Abstract:For agentic data analysis, language-model agents often require a large number of annotated tasks to learn how to interact effectively with executable environments (i.e., databases). While high-quality annotated tasks require heavy manual effort, an executable environment already provides rich interaction structure and verifiable outcomes. In this sense, the environment itself offers a free lunch, an additional set of verifiable tasks that require no human annotation. We propose EnvLunch, a warm-up stage that turns this environmental signal into agent training without human-written or model-generated task descriptions, before downstream agentic reinforcement learning. EnvLunch samples valid action trajectories, executes them to obtain final observations, and trains an agent to recover a functionally equivalent trajectory given only its final observation. This label-free warm-up enables the model to outperform the baseline by over 9 points on average across all downstream training budgets, and with few labels, the baseline needs 8 to 32 times as many annotated examples to match the accuracy that the EnvLunch model reaches at its smallest training budget. Letting the environment alone define the task yields a larger gain than having the model write its own tasks over the same environments, and this warm-up also improves fine-tuning on benchmarks beyond data analysis. EnvLunch therefore saves annotated tasks and provides a better starting point for fine-tuning agentic models when such tasks are limited.
[NLP-134] ask Format as a Factor in Adaptation: A Controlled Comparison of LoRA and Full Fine-Tuning for Prompt-Safety Classification
【速读】： 该论文旨在解决大语言模型在分类任务中微调策略选择的争议问题，特别是低秩适应（LoRA）与全量微调（full fine-tuning）之间的性能差异及其影响因素。研究发现，现有对比常因实验设计不一致而产生偏差，且任务格式（task format）对结果的影响长期被忽视。为解决此问题，研究采用一对匹配的4亿参数编码器-解码器架构（Ettin），在三种不同任务格式下系统比较两种微调方式，并设置最多15个随机种子和预设分析流程以增强可重复性。结果表明，在仅使用标签作为训练目标时，全量微调在所有架构和任务格式下均优于LoRA 2–3个百分点；但当学习率通过统一搜索优化后，两者的差距缩小。值得注意的是，若在LoRA中引入“输出判断理由”的训练目标，性能提升4.0个百分点并完全弥合差距，而仅增加等长无内容句子则无效，凸显训练目标设计的关键作用。此外，尽管微调模型在域内数据上表现良好，但在跨域数据集上显著落后，尤其在对抗性越狱攻击场景下性能下降8–23个百分点，且误判27–66%的良性有害提示为不安全，远高于80亿参数基线模型的2–3%。综上，该研究揭示：模型性能差异更多取决于训练方式、学习率选择及训练目标设计，而非单纯的微调范式；同时强调，仅关注域内准确率不足以评估模型鲁棒性，需综合考量泛化能力与安全性。
链接: https://arxiv.org/abs/2610.33813
作者: John Alderete,Srirakshith Betageri,Jennifer Flynn,Alex Greenhalgh,Josh Rubin,Yoshita Varma Annam
机构: 未知
类目: Computation and Language (cs.CL)
备注: 33 pages (14 pages main text, plus references and appendices), 1 figure, 18 tables
Abstract:Fine-tuning a language model for classification involves many design decisions. Comparisons of low-rank adaptation (LoRA) with full fine-tuning are common but often confounded, and task format is rarely studied. We compare both regimes on prompt-safety classification with a matched 400M encoder-decoder pair (Ettin) across three task formats, with up to 15 seeds per cell and pre-specified analyses. At the pre-specified learning rates, full fine-tuning leads LoRA by 2-3 pp in every architecture and format trained on the label alone. On the encoder, the gap closes once both rates are chosen by the same search. Training LoRA to write a reason after its verdict adds 4.0 pp and closes the gap, but a content-free sentence of equal length does not. Trained models score above a prompted 8B in domain but fall behind it on two of three out-of-distribution corpora, trailing by 8-23 pp on adversarial jailbreaks and flagging 27-66% of benign but harmful-looking prompts as unsafe, against the 8B’s 2-3%. Together, the results suggest that the adaptation regime matters less than how each model is trained, its learning rate, and training target, and that in-domain accuracy alone is an incomplete guide to robustness.
[NLP-135] IntentDelta: A Human-Constructed Benchmark for Evolving User Intent
【速读】： 该论文旨在解决语言模型智能体在复杂任务执行过程中，面对用户需求与优先级动态变化时的适应性问题。其核心挑战在于如何准确捕捉并持续维护用户意图的演化过程，尤其是在约束条件与偏好发生可行性或优先级层面调整时，智能体仍能保持对历史要求的理解并正确区分硬性约束与软性偏好。解决方案的关键在于构建了名为IntentDelta的多轮人工作标注基准，涵盖WebShop和TravelPlanner两个场景，共包含638条任务轨迹与3,766个对话回合，每一轮均配有用户语句、累积意图状态及人工参考动作，且通过交叉验证确保数据质量。该基准系统性地覆盖了由可行性或偏好驱动的意图变更，为评估智能体在意图恢复（intention recovery）与动作质量方面的表现提供了可靠标准。实验结果表明，当前十一类智能体在旅行规划和购物场景中的平均意图恢复率分别为33.1%和44.7%，动作评分分别为31.6%和41.0%，反映出智能体普遍难以维持早期需求，并混淆软性偏好与必须满足的约束，同时指出意图恢复成功并不等同于任务执行成功，凸显了将动态演变的用户意图转化为有效行动序列的深层挑战。
链接: https://arxiv.org/abs/2610.33806
作者: Jieyu Zhou,Zijian Wang,Yun Wang
机构: Georgia Institute of Technology (佐治亚理工学院); Microsoft Research (微软研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Language-model agents must adapt as users revise their requirements and priorities during complex tasks. We introduce IntentDelta, a human-constructed multi-turn benchmark with 638 trajectories and 3,766 turns across WebShop and TravelPlanner. Each turn pairs a user utterance with a cumulative intention state and human reference actions. The trajectories are cross-verified and cover feasibility- and preference-driven changes in constraints and priorities. We evaluate eleven agents on intention recovery and action quality. Average Intention Recovery is 33.1% in TravelPlanner and 44.7% in WebShop, while average Action Score is 31.6% and 41.0%, respectively. Agents struggle to preserve earlier requirements and distinguish soft preferences from must-have constraints. Correct intention recovery also does not guarantee successful execution, highlighting challenges in translating evolving user intentions into actions.
[NLP-136] MTA-Zero: Self-Improving Multimodal Deep Search via Online QA-Solver Co-training
【速读】： 该论文旨在解决多模态大语言模型（Multimodal Large Language Models, MLLMs）在开放世界、多步深度搜索任务中面临的局限性，即如何有效实现图像锚定（image grounding）、工具选择与网络证据整合。现有方法依赖于固定离线语料库构建和过滤，其问题分布无法随目标深度搜索智能体的性能提升而动态适应。为此，本文提出MTA-Zero，一种无需预先计算多跳语料库的在线多模态问答-求解器联合训练框架。其核心创新在于：基于图像锚定的视觉问答（VQA）种子上下文，通过可训练的问答生成器（QA Generator）在线检索并构造经过验证的多跳问题；同时，独立的深度搜索求解器（Solver）负责工具选择、证据收集与答案生成。为聚焦生成过程于求解器的学习前沿，设计了一种“恰到好处”（Goldilocks）奖励机制，该奖励由求解器组级回放成功率推导而来——仅当问题被部分但非全部回放成功时获得正向奖励，从而引导生成器生成具有挑战性且适切的问题。两个策略通过GRPO算法独立优化，不共享参数或梯度。在六个多模态深度搜索基准测试中，MTA-Zero将Qwen3-VL-8B求解器的平均准确率从36.91%提升至51.58%，性能接近使用GPT-5的水平；对更先进的Qwen3.6-27B求解器亦带来10.82个百分点的提升。结果表明，经验证的在线问题生成可为多模态深度搜索智能体提供特定于策略的学习课程，实现持续自我改进。
链接: https://arxiv.org/abs/2610.33799
作者: Xiangyu Peng,Jiaxin Zhang,An Yan,Yu Li,Ran Xu,Zeyuan Chen,Chien-Sheng Wu
机构: Salesforce( Salesforce)
类目: Computation and Language (cs.CL)
备注:
Abstract:Multimodal large language models (MLLMs) have become strong visual reasoners, yet remain limited in open-world, multi-step search that requires grounding images, selecting tools, and integrating web evidence. Existing approaches typically construct and filter fixed offline corpora whose question distributions cannot adapt as the target deep-search agent improves. Building on MTA-Agent’s image-grounded QA construction, we present MTA-Zero, an online multimodal QA-Solver co-training framework for self-improving multimodal deep-search agents that requires no precomputed multi-hop corpus. Starting from image-grounded VQA seed contexts, a trainable QA Generator retrieves evidence and constructs verified multi-hop questions online. A separate deep-search Solver then selects tools, gathers evidence, and answers the generated questions. To focus generation on the Solver’s learning frontier, we derive a Goldilocks reward for the QA Generator from the Solver’s group-wise rollout success rate: questions solved by all rollouts or by none receive zero reward, whereas questions near the current frontier are favored. We optimize the two policies independently with GRPO, without sharing parameters or gradients. Across six multimodal deep-search benchmarks, MTA-Zero improves a Qwen3-VL-8B Solver from 36.91% to 51.58% mean accuracy and achieves performance comparable to GPT-5 under the same agent workflow. The same framework also improves a newer Qwen3.6-27B Solver by 10.82 points. These results show that verified online question generation provides a policy-specific curriculum for continually improving multimodal deep-search agents.
[NLP-137] Whispers from the Neighbors: Rethinking Caption Consensus as Inference-Time Evidence for Vision-Language Retrieval
【速读】： 该论文旨在解决在文本到图像检索任务中，现有冻结编码器（frozen encoder）仅孤立地评估每对图像-文本匹配关系，而忽略了同一图像所附多个标注（caption）之间潜在协同信息的问题。其核心挑战在于如何有效利用图像自身多条标注作为上下文线索，在推理阶段提升检索性能。解决方案的关键在于：通过融合图像与自身标注间的相似性，构建更鲁棒的候选图像支持（support）机制——具体而言，将图像与其多条标注之间的语义一致性作为额外监督信号，从而增强正确候选图像的得分。实验表明，当标注来自与查询相同标注者时，该方法可使文本到图像检索的R@1指标提升3.4至16.8个百分点；即使使用通用图文生成模型生成标注，性能仍可提升1.9至7.2个百分点，而来自不同标注源的标注则贡献极小。进一步分析揭示，该优势主要源于正确候选图像的支持集结构与负样本存在差异，其中支持集大小和查询移除效应各贡献约一半效果，且在固定支持索引下，剩余增益可归结为对已有标注信息的重新加权。此外，在组合性（compositional）基准测试中，标注的作用体现在更准确地匹配人类对图像的描述，而非改善图像-文本语义对齐能力。
链接: https://arxiv.org/abs/2610.33798
作者: Khang Nhat Hoang Vo,Belal Shoer,Abhirama Subramanyam Penamakuri
机构: Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:An image is rarely described only once. Retrieval benchmarks and many collections attach several captions to each picture, yet dual encoders still judge every image-caption pair in isolation. We ask what these neighboring captions contribute when a frozen encoder consults them at inference time. Fusing image similarity with similarity to an image’s own captions improves text-to-image R@1 by 3.4 to 16.8 points across nine frozen encoders when the captions come from the same annotators as the queries. The gain falls to 1.9 to 7.2 points when an off-the-shelf captioner writes them, and to at most one point when they come from a different annotation source; a paraphrase-trained text encoder strengthens the text side further. A natural refinement, scoring the part of the image that the captions leave unexplained, appears to add much more under a leave-one-out evaluation. We show that this advantage arises because the correct candidate’s supports are built differently from those of the negative candidates, with set size and query removal each contributing about half, and that under a fixed support index the residual reduces to a reweighting of what the captions already provide. On compositional benchmarks, the captions help by matching candidates to human descriptions of the image, without improving image-text grounding. The code is available at: this https URL
[NLP-138] One Scan is Enough: Demonstration-Free Adaptation for Language-Guided Navigation NEURIPS2026
【速读】： 该论文旨在解决语言引导式导航（language-guided navigation）在新建筑环境中部署时普遍存在的泛化能力差问题，尤其针对公开基准数据集在布局、外观及语言表达上的漂移导致性能显著下降的挑战。传统解决方案依赖现场人工遥控采集新轨迹，但难以规模化。本文提出的关键解决方案是：仅通过一次简短的iPhone扫描即可完成场景适应，利用Scan2Instruction将扫描数据自动构建为完整训练环境，并合成带有语义锚定的可执行导航任务；随后，PRPT-KS基于重构场景直接计算闭式帕累托效用，无需学习型评判器或人工标注标签，实现策略的高效自适应。实验表明，该方法使真实机器人在五个真实场景中达到85%的成功率，显著优于基于人类示范微调的基线（49%），且在未扫描区域仍保持82%的成功率，远超基线的41%。该工作首次将无演示的场景自适应从理论构想变为可落地的实际系统。
链接: https://arxiv.org/abs/2610.33794
作者: Zhuoyuan Yu,Yuxing Long,Du Wang,Junyang Wang,Hongwei Fan,Hao Dong
机构: Peking University (北京大学); PrimeBot
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: NeurIPS 2026
Abstract:Deploying a language-guided navigation policy in a new building usually means watching it fail. Public-benchmark VLAs degrade scene by scene as layout, appearance, and language drift, and the standard fix of teleoperating fresh trajectories on site does not scale. We replace human demonstration with a single brief iPhone scan. Scan2Instruction compiles that scan into a complete training environment, automatically synthesizing executable navigation episodes with grounded instructions. PRPT-KS then adapts the policy through a closed-form Pareto utility read directly off the reconstructed scene, with no learned critic and no human labels. Across five real environments, the two designs together drive our RealNav to an 85% real-robot success rate, compared with 49% for a RealNav baseline post-trained on human demonstrations, and still achieve 82% in unscanned regions of the same buildings, compared with 41% for the same baseline. Our work turns demonstration-free scene adaptation for language-guided navigation from an aspiration into a working reality.
[NLP-139] he Hitchhikers Guide to System 1 Models: Fast Decision-Making Beyond LLM S
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际应用中因自回归（autoregressive）、逐标记生成的推理方式导致的效率低下、成本高昂以及决策校准性差的问题，尤其针对现代AI系统中大量需要亚秒级响应时间的常规决策任务（如路由、门控、评分、分类等）。其核心解决方案在于提出并系统化“系统1型决策模型”（System 1 decision models），即通过单次前向传播将输入直接映射为带有类型和结构化的决策输出，从而实现高速、低成本且可校准的决策能力。该研究的关键在于构建一个涵盖五类模型的综合性分类体系：基于编码器的分类器、奖励模型、非自回归决策引擎、对比式决策模型与提取模型，并深入分析各类模型的架构设计、训练方法及部署权衡。同时，论文强调校准性（calibration）作为核心原则，发展了正确评分规则的理论基础，探讨了强化学习校准决策（Reinforcement Learning for Calibrated Decisions, RLCD）方法，并揭示了诸如独立于无关选项假设（IIA）违反、序数尺度偏差以及分布偏移下的少样本再校准等关键现象。此外，论文还系统阐述了系统1模型在智能体工作流与“系统1+系统2”级联架构中的集成机制，实现基于置信度校准的快慢模型动态路由。整体上，该研究提供了一个覆盖18个应用场景、包含超过70个开源模型复现、附带可运行代码、实践检查清单、开源模型库及22道自测题的完整工具包，旨在成为快速决策、校准性与生产级AI系统交叉领域研究人员与工程师的统一入门指南。
链接: https://arxiv.org/abs/2610.33775
作者: Haggai Roitman
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: version 1.0
Abstract:Large Language Models excel at reasoning and generation but their autoregressive, token-by-token inference makes them unnecessarily slow, expensive, and poorly calibrated for the many routine decisions (routing, gating, scoring, classification) that modern AI systems require at sub-second latency. A growing class of models addresses this gap: System 1 decision models, architectures that map an input directly to a typed, structured decision in a single forward pass. This guide provides a comprehensive, practitioner-oriented treatment of the field. We propose a five-family taxonomy covering encoder-based classifiers, reward models, non-autoregressive decision engines, contrastive decision models, and extraction models, and we provide in-depth coverage of each family’s architecture, training methodology, and deployment trade-offs. A central theme is calibration: we develop the theory of proper scoring rules, analyze Reinforcement Learning for Calibrated Decisions (RLCD), and examine recently discovered phenomena including IIA violations, ordinal-scale bias, and few-shot recalibration under distribution shift. On the integration side, we describe how System 1 models serve as decision infrastructure within agentic workflows and System 1 + System 2 cascades that route between fast and slow models based on calibrated confidence. The guide covers 18 application domains, maps the rapidly growing open-source ecosystem (over 70 model reproductions in weeks), provides runnable code throughout, and includes a practitioner’s checklist, an open-source model zoo, and 22 self-assessment questions with detailed answers. Spanning 190 references, it is intended as a unified entry point for researchers and engineers working at the intersection of fast decision-making, calibration, and production AI systems.
[NLP-140] StateWise: Teaching Language Models to Track and Obey Evolving Constraints in Multi-Turn Dialogue
【速读】： 该论文旨在解决对话助手在交互过程中对约束条件（constraint）管理失效的问题，具体表现为：随着对话轮次的增加，已激活的约束逐渐被忽略（持久性失败，persistence failure），而已被撤销的约束仍被持续执行（复发，relapse）。现有解决方案依赖外部记账系统（external ledger）在每一轮重新注入编译后的规范，但存在额外开销。其核心创新在于提出StateWise——一种自改进式方法，通过训练单一大语言模型（LLM）自主维护可修订的约束状态（typed state with tombstones for revoked constraints），利用可执行检查器（executable checkers）量化持久性与复发行为，并免费生成真实状态作为监督信号。该方法首先通过两阶段微调使模型学会将对话摘要为约束状态并基于该状态生成响应；随后通过自训练机制筛选仅通过精确往返验证（round-trip check）的对话轨迹，实现状态追踪的高精度。然而实验发现，尽管状态跟踪能力极强，模型却难以遵守自身维护的状态，导致复发率显著上升。为此，研究引入偏好阶段，在自生成的复发对比样本中进行强化学习：拒绝执行已撤销规则的响应，选择通过往返验证的响应。在未见约束类别和表达方式的测试集上，StateWise相较监督自训练显著降低复发率8–22个百分点，提升对话成功率8–19个百分点，并在无需推理时辅助结构的前提下超越基于编译记账本的基准。此外，研究还分析了该方法对通用指令遵循任务的影响，并探索权重插值作为缓解代价的策略。
链接: https://arxiv.org/abs/2610.33766
作者: Samarth Varshney,Ritu Yadav
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 13 pages, 1 figure, 5 tables
Abstract:Users of conversational assistants rarely state all requirements up front: they add constraints, amend them, withdraw them, digress, and return. Instruction-tuned LLMs fail here in two ways: active constraints stop being honored as turns accumulate (persistence failure), and withdrawn constraints keep being enacted (relapse). The only existing repair we know of is an external ledger that re-injects a compiled specification every turn. We propose StateWise, a self-improving approach that teaches a single LLM to maintain its own revisable constraint state, a typed state with tombstones for revoked constraints, so persistence and relapse are measured by executable checkers and ground-truth states are compiled for free. StateWise fine-tunes the LLM on two complementary skills, summarizing a dialogue into its constraint state and responding conditioned on that state, then self-trains on its own trajectories, keeping only turns that pass an exact round-trip check against the compiled state. This teaches 7-8B models to track the state almost perfectly but, surprisingly, not to obey it: on Llama-3.1-8B every supervised variant roughly doubles relapse. We therefore add a preference stage on self-generated relapse contrasts, where the rejected response enacts a revoked rule and the chosen one passes the round-trip check. On two backbones with held-out constraint families and unseen phrasings, StateWise cuts relapse by 8-22 points relative to supervised self-training, lifts conversation success by 8-19 points over the backbone, and beats an oracle compiled-ledger baseline on relapse with no inference-time scaffolding. We also characterize the recipe’s cost on general instruction-following benchmarks and study weight interpolation as a mitigation.
[NLP-141] RoleHear: Pseudo-Labeling for Semi-Supervised Role-Attributed Streaming Speech Recognition
【速读】： 该论文旨在解决流式自动语音识别（Streaming Automatic Speech Recognition）中词流缺乏说话人角色归属的问题，即在低延迟转录的同时，实现每个词与对应说话人角色的精确关联。传统方法通常将角色标注（Role Attribution）作为独立阶段处理，依赖离线或流式说话人分离模型对齐词序列，或使用基于文本的大语言模型进行后处理，但这些方法面临标注数据稀缺、延迟高及系统耦合性差等挑战。本文提出的解决方案——RoleHear，采用一种半监督流水线架构，其核心在于设计一个缓存感知的流式转换器（cache-aware streaming transducer），通过轻量级的词汇内角色标记符（in-vocabulary role tokens）实现角色信息的在线嵌入：先以快速的发言边界标记符（turn-boundary marker）指示发言切换，随后在数个词后插入对应的说话人角色标记。该方案利用一个经过角色标注微调的离线教师模型对未标注音频进行重标注，并结合多模态大语言模型在无真实标签场景下生成银色标签（silver labels），最终使流式学生模型在单次训练中联合学习混合数据，显著提升了角色标注准确性，同时保持了高质量的转录性能，并缩小了与离线教师模型之间的差距。
链接: https://arxiv.org/abs/2610.33765
作者: Zefang Liu,Chenyang Zhu,Sangwoo Cho,Xujun Peng,Rongqing Huang,Shi-Xiong Zhang,Sambit Sahu
机构: 未知
类目: Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注:
Abstract:Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce. We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later. An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass. Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher.
[NLP-142] Refined Assessment for Translation Evaluation: Rethinking Machine Translation Evaluation in the Era of Human-Level Systems EMNLP2025
【速读】： 该论文旨在解决当前机器翻译（Machine Translation, MT）评估方法在逼近人类水平翻译质量时，难以有效识别细微翻译错误的问题。现有金标准评估方法（如MQM和ESA）存在标注者专业水平不一、分类体系过于复杂、严重性分级粗糙、过度侧重准确性而忽视流畅性，以及时间压力导致评估质量下降等局限。为应对上述挑战，研究提出了一套由专业语言学家完成的高质量英文-俄文翻译评估数据集（基于WMT24），并设计了新的评估协议RATE（Refined Assessment for Translation Evaluation）。RATE的关键创新在于：简化误差分类体系、扩展严重性评分维度、引入多维度评分机制以平衡准确性与流畅性，从而实现更一致且深入的评估。基于该专家数据集的分析表明，当前最先进的MT系统在准确性上已超越人类翻译，但在流畅性方面仍存在显著差距——这一关键差异被传统以准确性为导向的评估指标所掩盖。研究强调，推进机器翻译评估不仅需要改进评估协议，更依赖于具备专业素养的语言学家提供的高质量标注。
链接: https://arxiv.org/abs/2610.33749
作者: Dmitry Popov,Vladislav Negodin,Ekaterina Enikeeva,Iana Matrosova,Nikolay Karpachev,Max Ryabinin
机构: 未知
类目: Computation and Language (cs.CL)
备注: Published in Findings of EMNLP 2025. 17 pages. Dataset: this https URL
Abstract:As machine translation systems approach human-level quality, traditional evaluation methodologies struggle to detect subtle translation errors. We critically examine limitations in current gold-standard approaches (MQM and ESA), including inconsistencies from variable annotator expertise, excessive categorization complexity, coarse severity granularity, accuracy bias over fluency, and time constraints. To address this issue, we introduce a high-quality dataset consisting of human evaluations for English-Russian translations from WMT24, created by professional linguists. We show that expert assessments without time pressure yield substantially different results from standard evaluations. To enable consistent and rich annotation by these experts, we developed the RATE (Refined Assessment for Translation Evaluation) protocol. RATE provides a streamlined error taxonomy, expanded severity ratings, and multidimensional scoring balancing accuracy and fluency, facilitating deeper analysis of MT outputs. Our analysis, powered by this expert dataset, reveals that state-of-the-art MT systems may have surpassed human translations in accuracy while still lagging in fluency - a critical distinction obscured by existing accuracy-biased metrics. Our findings highlight that advancing MT evaluation requires not only better protocols but crucially, high-quality annotations from skilled linguists.
[NLP-143] Beyond Prompts: A Systematic Study of LLM -Based Machine Translation Evaluators EMNLP2026
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）在机器翻译（Machine Translation, MT）评估中因设计选择不明确而导致行为不可预测的问题。尽管基于LLM的评估工具已成为MT评价的主流方法，但其指令设计、少样本示例、主干模型选择及多步骤工作流等关键因素如何影响评估性能仍缺乏系统理解。为此，研究通过在WMT24和WMT25英俄翻译数据上构建新的专家级错误标注（RATE），并结合跨五种语言对的MQM与ESA基准进行迁移验证，系统探究了上述设计要素的影响。研究揭示四个稳健结论：指令结构显著塑造评估器行为；高质量的少样本示例比数量更重要；前沿主干模型表现出不同的评估倾向（召回率导向、精确率导向或排序导向）；融合多种异构主干模型的输出优于单一模型多次调用的设置。尽管评估器设计存在庞大的组合空间，但部分发现的设计原则在不同标注协议和语言对间具有强泛化能力。基于这些发现，研究提出ORBIT——一类可配置的LLM-based MT评估框架，其最优配置在错误片段检测以及准确性和流畅性排序任务上达到专家水平性能，同时成本仅为人工专家标注的约1/20。
链接: https://arxiv.org/abs/2610.33741
作者: Dmitry Popov,Roman Bokhyan,Ekaterina Enikeeva,Artem Mekhraliev,Stepan Vysotsky,Nikolay Karpachev
机构: Yandex(雅虎); HSE University(高等经济大学)
类目: Computation and Language (cs.CL)
备注: Accepted to Findings of EMNLP 2026. 17 pages, 1 figure. Dataset and prompts: this https URL
Abstract:LLM-based evaluators have rapidly become a standard tool for machine translation evaluation, yet the design choices that shape their behavior remain poorly understood. We present a systematic study of how instruction design, few-shot demonstrations, backbone model choice, and multi-step workflows affect span-level and ranking-based MT evaluation. We collect new expert RATE annotations on WMT24 and WMT25 English-Russian translations, with transfer validation on MQM- and ESA-annotated benchmarks across five additional language pairs. Our experiments reveal four robust findings: instruction structure shapes evaluator behavior, high-quality demonstrations matter more than quantity, frontier backbones exhibit distinct evaluator profiles (recall-, precision-, or ranking-oriented), and fusing outputs across diverse backbones outperforms single-evaluator multi-call setups. Although evaluator design exposes a large combinatorial space, several discovered design principles transfer consistently across annotation protocols and language pairs. We synthesize these findings into ORBIT, a family of LLM-based MT evaluators. Our strongest configurations achieve expert-level performance in error span detection and in Accuracy and Fluency ranking at roughly 20x lower cost than expert human annotation.
[NLP-144] Gender Associations in LLM -Mediated ADHD Self-Diagnosis ACL
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在医疗自诊断场景中，人口统计学线索（如性别）如何影响输出结果的问题，尤其聚焦于注意力缺陷多动障碍（ADHD）这一存在持续性别差异但与临床证据不完全一致的疾病。其核心挑战在于揭示模型在零样本自诊断情境下对性别化ADHD关联的响应机制。解决方案的关键在于采用情境关联测试（Context Association Test, CAT）框架，系统区分显式提示（explicit cueing）与隐式提示（implicit cueing）对GPT-4o模型输出的影响：研究发现，显式提示会显著强化特定ADHD亚型的性别关联，可能触发模型的偏差缓解机制并导致过度校正；而隐式提示则呈现出接近代理对齐（proxy-aligned）的分布，反映模型的默认行为模式。这一差异表明，评估LLM在医疗自诊断中的性别偏见时，必须精细考察模型所编码的人口统计学关联形式，且提示设计方式对偏见表现具有决定性影响。
链接: https://arxiv.org/abs/2610.33697
作者: Matilde Elene Hansen,Ross Deans Kristensen-McLachlan
机构: Aarhus University (奥胡斯大学)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 20 pages (10 pages of main text, plus references and appendix). Accepted at EVAL4SD 2026. To appear in the ACL Anthology
Abstract:Large Language Models (LLMs) are increasingly used for medical self-diagnosis. However, the way demographic cues shape outputs in these interactions remains poorly understood. This is particularly consequential for diagnoses like ADHD, where persistent gender disparities are not fully aligned with the clinical evidence. We adapt the Context Association Test (CAT) to evaluate whether GPT-4o responds to gendered ADHD associations in a zero-shot, self-diagnostic setting. We show that explicit cueing consistently amplifies certain ADHD subtype associations while implicit cueing produces near proxy-aligned distributions. These divergences suggest that explicit demographic cues may activate bias-mitigation mechanisms that over-correct rather than eliminate bias, whereas implicit cues appear to reflect the model’s default behavior. Our results demonstrate that evaluating gendered bias in LLM-mediated self-diagnosis requires careful attention to the demographic associations a model encodes.
[NLP-145] Embedding-Based Educational Taxonomies
【速读】： 该论文旨在解决传统教育分类体系（educational taxonomies）依赖人工构建、耗时且难以迭代更新的问题，同时探索如何利用自动化方法生成可替代的教育分类体系。其核心挑战在于如何在不依赖专家经验的前提下，从教学内容本身（如数学题）中自动提取并组织知识单元（knowledge components），以支持智能辅导系统等教育应用中的学习目标定义与学生建模。解决方案的关键在于利用语义文本嵌入（semantic text embeddings）技术，对数学题目进行向量化表示，并通过聚类分析生成基于嵌入的教育分类体系。实验表明，这种嵌入驱动的分类体系虽与现有基于《共同核心州立标准》（CCSS）的人工分类体系存在部分一致性，但也揭示了新的内容组织结构；将其应用于深度知识追踪（Deep Knowledge Tracing）任务时，可使模型预测性能提升约3-5个AUC百分点，验证了该方法在恢复人类制定分类结构的同时，还能提供具有实用价值的替代性表示。
链接: https://arxiv.org/abs/2610.33694
作者: Kia Karbasi,Kaveh Karbasi,Gregory Pottie
机构: University of California, Los Angeles(加州大学洛杉矶分校); Tempus AI, Inc.
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Educational taxonomies organize educational content, such as exercise problems, into concepts, which are also called skills or knowledge components. These taxonomies define learning objectives in educational applications such as intelligent tutoring systems. Developed through human expertise and iterative refinement, they also provide fundamental structure for curricula, assessment, and models of student learning. Recent advances in semantic text embeddings make it possible to construct alternative educational taxonomies computationally from the content itself. In this paper, we investigate this possibility by clustering embeddings of math questions to construct embedding-based educational taxonomies. We compare these taxonomies with the existing Common Core State Standards (CCSS)-based knowledge-component taxonomies in two ASSISTments datasets and evaluate their utility in Deep Knowledge Tracing, a downstream task that relies on taxonomic mappings of questions to knowledge components. The embedding-based taxonomies exhibit substantial but incomplete correspondence with the existing taxonomies, revealing both preserved and alternative organizations of educational content. When used for knowledge tracing, they improve predictive performance by approximately 3-5 AUC percentage points. These results demonstrate the potential of semantic embeddings for automatically constructing educational taxonomies that both recover structure present in human-developed taxonomies and provide useful alternative representations for downstream educational modeling.
[NLP-146] PathKGen: Pathology Knowledge Assisted MLLM for Explainable Histopathology WACV2027
【速读】： 该论文旨在解决当前生成式人工智能在数字病理学报告生成中面临的两大核心问题：一是多模态大语言模型（MLLM）依赖自身参数编码的分布进行推理，受限于模型规模，难以容纳充分的医学知识与可解释的推理逻辑；二是其“黑箱”推理过程缺乏可解释性，不利于临床医生对诊断推理路径的理解。为此，论文提出一种基于知识图谱（Knowledge Graph, KG）增强的多模态大语言模型框架——PathKGen，其解决方案的关键在于构建双分支架构：一方面通过视觉路径利用病理基础模型提取全切片图像（WSI）的视觉特征；另一方面引入知识路径，通过查询领域特定的知识图谱，自动提取与病理图像相关的关键形态学特征作为结构化知识令牌（knowledge tokens），从而将外部医学知识显式注入模型推理过程。该设计不仅显著提升了报告生成的准确性与可解释性，还通过可视化知识图谱形式呈现诊断推理链条，实现了从“黑箱”到“白盒”推理的转变。由于其模块化设计，该知识路径可灵活适配不同领域的知识图谱，具备良好的可扩展性与跨应用潜力。
链接: https://arxiv.org/abs/2610.33691
作者: Nadun Rajapaksha,Sandun Herath,Devnith Wijesinghe,Isiri Amani Withanawasam,Hasindri Watawana,Ranga Rodrigo
机构: University of Moratuwa, Sri Lanka; Idiap Research Institute, Switzerland; EPFL, Switzerland
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: Accepted to the WACV 2027. Source code is available at this https URL
Abstract:Whole Slide Image (WSI) analysis serves as the fundamental method for the clinical diagnosis of cancer in digital histopathology. Recent advances in Multimodal Large Language Models (MLLMs) have enabled state-of-the-art (SOTA) automated histopathology report generation. However, MLLMs reason purely through distributions encoded in their neural parameters, limiting the amount of knowledge and reasoning logic they can contain given the model scale. Their black-box reasoning poses another challenge to the human interpretation of intermediate reasoning. We address these issues by utilizing knowledge graphs to inject external structured knowledge for MLLMs to reason over, while making this reasoning component visible. Knowledge graphs are capable of partially externalizing the containment of medical knowledge, an area that remains underexplored in pathology report generation. We introduce PathKGen, a knowledge graph-assisted MLLM for histopathology report generation. PathKGen has two main branches: (1) a visual path that learns visual token inputs to the MLLM using pathology foundation models, and (2) a knowledge path that learns knowledge token inputs by extracting key morphological features of WSIs through querying a pathology-specific knowledge graph. A comprehensive evaluation on WSI-Bench dataset shows that PathKGen surpasses previous SOTA methods while revealing diagnostic reasoning as visually intuitive graphs. Due to its modular design, the proposed knowledge path is complementary to vision-aligned MLLMs and can be extended to other applications by curating domain-specific knowledge graphs to generate knowledge tokens.
[NLP-147] Meta-CLIP: Metadata-Conditioned Contrastive Learning for Zero-Shot Disease Inference
【速读】： 该论文旨在解决医学视觉-语言预训练中因成像设备和采集协议差异导致的跨域泛化能力下降问题。现有方法主要聚焦于增强临床语义信息，而忽视了DICOM头文件中的结构化采集元数据的价值。其解决方案的关键在于提出Meta-CLIP，一种基于元数据条件的对比学习框架，通过利用采集相似性对跨模态负样本进行重加权。该框架设计了一个作用于数值型与类别型DICOM字段的核函数，定义了一个有界成对边界（bounded pairwise margin），使采集条件相似的负样本贡献更大，从而强化在相同成像条件下不同研究之间的区分能力。该核函数在图像-文本优化过程中保持固定，推理阶段仍采用标准双编码器架构，无需依赖元数据。实验表明，Meta-CLIP在胸部X线和容积胸腔CT任务上均显著提升性能，在CT-RATE数据集上相较CT-CLIP提升1.41个宏观受试者工作特征曲线下面积（macro-AUROC）点，且95%置信区间为0.96–1.85；在五个胸部X线基准中的四个取得最高宏AUROC。探索性BrgSA分析进一步验证了其鲁棒性。结果表明，采集元数据可作为医学视觉-语言学习中有效的训练时先验信息。
链接: https://arxiv.org/abs/2610.33674
作者: Haozhe Luo,Nirosh Sivanesan,Shelley Zixin Shu,Francesco Nitti,Mauricio Reyes
机构: University of Bern (伯尔尼大学); ARTORG Center for Biomedical Engineering Research (ARTORG生物医学工程研究中心)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Medical vision–language pretraining learns transferable representations from paired images and radiology reports, yet differences in scanners and acquisition protocols can impair transfer. Existing approaches primarily enrich clinical semantics, leaving structured acquisition information in DICOM headers underused. We introduce Meta-CLIP, a metadata-conditioned contrastive framework that uses acquisition similarity to reweight cross-modal negatives during pretraining. A kernel over numerical and categorical DICOM fields defines a bounded pairwise margin. This margin increases the contribution of acquisition-similar negatives, emphasizing distinctions among studies that share imaging conditions. The kernel remains fixed during image–text optimization, and inference follows the standard dual-encoder architecture without requiring metadata. We evaluate Meta-CLIP on chest radiography and volumetric chest CT through zero-shot classification, appearance perturbations, grounding, and representation analyses. On CT-RATE, Meta-CLIP improves macro-AUROC over CT-CLIP by 1.41 points, with a 95% paired bootstrap confidence interval of 0.96–1.85 points. It also achieves the highest macro-AUROC on four of five CXR benchmarks. An exploratory BrgSA evaluation extends the analysis to another CT backbone. These findings support acquisition metadata as a practical training-time prior for medical vision–language learning. Code will be made publicly available upon publication.
[NLP-148] Jointly Modeling Preferences and Imperfect Decision-Making in Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在标准偏好诱导任务中表现出的偏好是否能够有效预测其在复杂战略环境下的行为问题。现有研究多通过行为经济学或心理学方法单独测量偏好维度，但缺乏对不同情境下偏好的一致性检验及跨场景预测能力的评估，难以揭示多种目标（如个体福利、群体福祉与环境关切）协同作用对决策的影响机制。为此，论文提出偏好-行为模型（Preference-Behavior Model, PBM），基于效用理论，将偏好表征与非完美效用最大化的心理机制相结合，并将参数估计建模为逆博弈问题，通过观察到的选择行为联合推断偏好与行为参数。在公共品博弈、集体风险困境和共有资源困境三类社会困境中的实证表明：传统行为经济诱导工具所测得的偏好无法一致预测模型表现，而基于特定情境训练的PBM模型在其他情境中仍具预测力，说明其推断的偏好能捕捉到原始诱导工具所忽略的深层决策特征。此外，通过对角色指令对齐性的分析发现，推断出的偏好总体上反映预期优先级，但在不同维度、模型和游戏类型间存在可识别的偏差。本研究的方法论为系统评估LLM偏好的一致性、与设计意图的对齐程度，以及揭示标准偏好诱导方法可能遗漏的权衡关系与行为模式提供了可扩展的框架。
链接: https://arxiv.org/abs/2610.33651
作者: Axel Abels,Elias Fernández Domingos,Tom Lenaerts
机构: Université Libre de Bruxelles (布鲁塞尔自由大学); Vrije Universiteit Brussel (布鲁塞尔自由大学); FARI Institute (FARI研究所); ELLIS Alicante (ELLIS阿利坎特); Center for Human-Compatible AI, UC Berkeley (人类兼容人工智能中心，加州大学伯克利分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Science and Game Theory (cs.GT)
备注:
Abstract:Do preferences Large Language Models (LLMs) express in standard elicitation tasks predict their strategic behaviour? LLM agents interacting with humans and other AI systems face potentially conflicting objectives, such as individual welfare, group welfare, and environmental concerns. Prior work characterizes LLM preferences through behavioral-economic and psychological elicitation or strategic games, but rarely tests consistency across settings and prediction in other strategic environments. This requires modeling how competing objectives jointly shape behavior, whereas existing approaches typically measure preference dimensions in isolation. Drawing on utility theory, we introduce the Preference-Behavior Model (PBM), which combines a preference representation with mechanisms capturing imperfect utility maximization. We formulate estimation as an inverse-game problem, jointly inferring preferences and behavioral parameters from observed choices. We evaluate PBM in three social dilemmas that expose conflicting motives: a Public Goods Game, a Collective-Risk Dilemma, and a Common-Pool Resource dilemma. We find that while preferences estimated from established behavioral economic elicitation instruments do not consistently predict behavior in these dilemmas, models fitted in one dilemma retain predictive value in others, suggesting that inferred preferences capture meaningful aspects of LLM decision-making that elicitation instruments may miss. Finally, we assess alignment with persona instructions, finding that inferred preferences broadly reflect intended priorities with specific divergences across dimensions, LLMs, and games being identified by our approach. Our methodology provides a framework for studying how consistently LLM preferences explain behavior, assessing alignment with intended priorities, and revealing trade-offs and behavioral patterns that standard elicitation may miss.
[NLP-149] PreView: Enhancing GUI Grounding with Preprocessed Views and Cues
【速读】： 该论文旨在解决图形用户界面（GUI）定位任务中因界面元素密集、视觉干扰多而导致的控制项识别模糊问题，尤其在自然语言指令与目标控件之间存在歧义时。其核心挑战在于如何在不重新训练现有接地模型的前提下，提升输入信息的质量以增强模型对目标控件的定位与理解能力。解决方案的关键在于提出PreView——一个轻量级的辅助预处理器，通过生成两类互补的输入特征：一是针对任务需求的局部视图（task-relevant views），用于精确定位；二是针对候选控件的特定提示（candidate-specific cues），包括视觉提示（如带位置提示的红色圆圈标记）和文本提示（描述候选控件的身份或功能），以增强目标理解。该预处理器通过一个经过训练的轻量模型预测并优化图像中心点，裁剪并放大截图以生成高分辨率局部视图，同时生成结构化提示信息。这些增强后的输入被输入到冻结的接地模型中，显著提升了Qwen3-VL-2B、OS-Atlas-4B和Qwen3-VL-4B等主流模型在五个基准测试上的平均接地准确率，且无需对主模型进行微调。
链接: https://arxiv.org/abs/2610.33645
作者: Wenjun Hou,Kaishuai Xu
机构: 独立研究者(Independent Researcher)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: Work in progress
Abstract:GUI grounding requires locating small controls in cluttered screenshots and identifying which control a natural-language instruction refers to. Cropping and enlargement can expose local detail, but the resulting image may still leave ambiguity about the intended target. We propose PreView, an auxiliary preprocessor that improves the inputs to an existing grounding model without further training that model. PreView prepares two complementary types of input features: task-relevant views for localization and candidate-specific cues for target understanding. A trained lightweight auxiliary model predicts and refines a center, crops the screenshot into a local view, and enlarges that view for zoomed-in grounding. It also prepares a visual cue (i.e., a red-circle marker with a location hint) and a textual cue describing the candidate’s identity or function. The frozen grounding model uses the prepared inputs to predict the final click. Across five benchmarks, PreView improves the average grounding accuracy of Qwen3-VL-2B, OS-Atlas-4B, and Qwen3-VL-4B without further training these grounding models. Extensive analyses examine the contributions of views and cues, center refinement, and visual input budgets.
[NLP-150] A Model-Theoretic Formalization of AlKhwaremis ilm alfaraid
【速读】： 该论文旨在解决伊斯兰教法（Fiqh）中继承分配问题的形式化建模与计算可判定性难题，尤其聚焦于阿尔·花剌子密（Al-Khwarizmi）所阐述的继承规则。其核心问题是：如何在形式逻辑与计算框架下，对特定类别的继承配置进行精确表达，并实现其有效性判断的机械化。解决方案的关键在于将每一类继承配置建模为有理数域上的有限线性等式与不等式系统，并利用傅里叶-莫茨金消去法（Fourier-Motzkin elimination over Q）实现该系统的精确判定。进一步地，论文定义了具有显著请求人坐标的固定维度继承结构 Vd，并证明所有可接受的有理线性坐标实现均与 Vd 同构，从而确立了该代数表示在该类别中的规范性（canonical）。此外，通过构造证明了 Vd 的完整一阶理论 Td 是完备的，且明确区分了其整体理论完备性与继承相关片段的可判定性。该方法论最终推广至更广泛的封闭教法领域形式化，通过穷尽案例生成、霍恩子句公理化（Horn clause axiomatization）及终止性推理，实现了领域相关的计算完备性（domain-relative computational completeness）。
链接: https://arxiv.org/abs/2610.33629
作者: Elnaserledinellah Mahmoud Abdelwahab
机构: 未知
类目: Logic in Computer Science (cs.LO); Computation and Language (cs.CL)
备注: 27 pages, 1 figure
Abstract:This paper presents a model theoretic and computational formalization of AlKhwarizmis treatment of inheritance within the framework of Computational Fiqh. We first show that every inheritance configuration in a specified class is representable as a finite system of rational linear equalities and inequalities, and that such systems are decided exactly by Fourier Motzkin elimination over Q. We then define the fixed dimensional inheritance structure Vd with distinguished claimant coordinates, and prove that every admissible rational linear coordinate realization is isomorphic to Vd, this establishes the algebraic representation as canonical within that class. The complete first order theory Td of Vd is shown to be complete by construction and is explicitly distinguished from the decidability of the inheritance relevant fragment. The methodological principle is then related to a broader formalization of closed Fiqh domains via exhaustive case generation, Horn clause axiomatization, and terminating deduction, yielding domain relative computational completeness.
[NLP-151] RSI Arena: Towards Recursive Self-Improvement with Human Feedback
【速读】： 该论文旨在解决生成式 AI（Generative AI）在无人干预条件下能否自主完成语言模型的后训练优化，并通过用户反馈实现持续改进的问题。其核心挑战在于如何构建一个可审计、可验证的闭环系统，使AI代理能够基于人类反馈实现自我迭代升级。解决方案的关键在于提出Rapid Self-Improvement Arena（RSI Arena），采用两阶段实验范式：第一阶段中，十个前沿模型作为训练代理，各自从NVIDIA Nemotron 3.5 Lightning 30B-A3B Base模型出发，在共享64-GPU集群上独立完成数据选择、训练流程设计与检查点选取；第二阶段则引入人类偏好反馈的比较记录，允许各代理基于这些反馈进行为期三天的进一步微调。研究结果显示，八组模型在风格调整后的相对评分上实现了提升，尽管基准测试性能增益有限且与偏好提升无明显关联，但部分模型表现出响应长度增加和对强拒绝指令合规性上升等潜在风险。该框架为探索人类反馈引导下的递归自我改进（Recursive Self-Improvement with Human Feedback, RSIHF）提供了可追踪、可评估的实验环境，标志着迈向真正意义上的人机协同智能进化的重要一步。
链接: https://arxiv.org/abs/2610.33624
作者: Yue Huang,Yanbo Wang,Lichi Li,Yuanyuan Chen,Andreas Haupt,Xiaonan Luo,Zheyuan Liu,Jihye Choi,Felix Juefei-Xu,Xiangliang Zhang,Radha Poovendran,Misha Sra,Ying Liu,Ben Burtenshaw,Thomas Wolf,Alex Pentland,Zichen Chen,Zhangchen Xu
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 39 pages. Project website: this https URL
Abstract:Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and 300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints. We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting. In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while some agents encouraged longer responses and one model showed increased harmful compliance on StrongREJECT. RSI Arena provides an auditable setting for studying how human feedback can guide autonomous model development. It takes a step toward recursive self-improvement with human feedback (RSIHF), in which people help define what improvement means for the AI systems they use.
[NLP-152] Mycelium: A Meta-Study of Hallucination Detection Across Provenance Typology and Granularities
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）中幻觉（Hallucination）检测评估缺乏统一标准的问题。现有方法在不同基准测试（benchmark）之间进行比较时，由于各基准对幻觉的定义与度量方式存在显著差异，导致性能指标难以跨场景解释和有效比较。为此，论文将幻觉检测视为一种受基准条件影响的测量问题，并从三个维度系统组织现有基准：来源性（provenance），描述模型输出与证据条件的获取方式；类型学（typology），刻画基准标签所反映的事实性与忠实性偏差类型；粒度（granularity），规定幻觉标注、预测与评估的最小单位。基于此框架，论文分析了基准构建、标注及评估选择如何塑造底层预测任务，并将检测器机制与假设与其性能测量环境相联系。此外，论文深入探讨了片段级（span-level）评估中注释边界差异对结果的显著影响，提出**片段覆盖F1（Span Coverage F1）**作为基于包含关系的补充度量指标。该研究澄清了检测器评分所实际反映的内容，揭示了跨基准比较的局限性，并推动在不同基准条件下更透明地评估模型泛化能力。
链接: https://arxiv.org/abs/2610.33608
作者: Julia Belikova,Elisei Rykov
机构: Skoltech(斯科尔科技研究所); Sber AI Lab(斯伯人工智能实验室)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Hallucination detection is increasingly important for evaluating and deploying large language models. However, existing methods are often compared across benchmarks that define and measure hallucination in substantially different ways, making reported scores difficult to interpret across settings. We study hallucination detection as a benchmark-conditioned measurement problem and organize existing benchmarks along three axes: provenance, which describes how model outputs and evidence conditions are obtained; typology, which captures the factuality and faithfulness failures represented by benchmark labels; and granularity, which specifies the unit at which hallucinations are annotated, predicted, and evaluated. Using this framework, we analyze how benchmark construction, annotation, and evaluation choices shape the underlying prediction task, and relate detector mechanisms and assumptions to the settings in which their performance is measured. We also examine span-level evaluation, where variation in annotation boundaries can substantially affect reported results, and propose Span Coverage F1 as a complementary containment-based metric. Our analysis clarifies what detector scores capture, highlights limitations of cross-benchmark comparisons, and motivates more explicit evaluation of generalization across benchmark conditions.
[NLP-153] Words Are the Voice of the Mind: Depression Severity Estimation from Social Traces via Mental-Guided Evidence Reasoning
【速读】： 该论文旨在解决基于社交媒体痕迹的抑郁严重程度估计中存在用户特定心理先验建模不足、语义检索多样性欠缺以及心理证据细粒度推理能力薄弱的问题。其核心解决方案是提出UMer框架，通过从用户历史社交痕迹中提炼个性化的心理先验以捕捉个体的心理倾向、行为模式与表达风格；采用量表引导的多查询检索机制，从相关帖子中构建证据记忆；进而实施由粗到细的证据感知推理，有效区分直接与间接证据，并生成可解释的逐项评分结果。该方法在真实数据集上的实验表明，不仅性能优于现有基线，还具备基于证据的可解释性并可泛化至其他量表任务。
链接: https://arxiv.org/abs/2610.33586
作者: Nada Xu,Xueqin Chen,Qiang Gao,Hongru Liang,Wenqiang Lei
机构: Sichuan University (四川大学); Southwestern University of Finance and Economics (西南财经大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Depression severity estimation is crucial for mental health assessment and intervention, yet traditional approaches depend on clinical interviews or self-report questionnaires that require users’ active participation. With the widespread use of online social media, users leave rich social traces that may reflect their mental states, motivating the task of social media-based depression severity estimation. Existing efforts have progressed from textual-feature-based classification to LLM-based questionnaire completion over standardized scale items, yet they still fall short in user-specific mental priors modeling, diversified semantic retrieval, and fine-grained reasoning over mental evidence. To address these limitations, we propose UMer, a user-specific mental evidence reasoning framework for social traces-based depression severity estimation. UMer first distills a user-specific mental prior from historical social traces to capture the user’s psychological tendencies, behavioral patterns, and self-expression style. It then performs scale-guided multi-query retrieval to construct evidence memories from item-relevant posts. Finally, it conducts coarse-to-fine evidence-aware reasoning to distinguish direct and indirect evidence and predict item-level scores with interpretable rationales. Extensive experiments on real-world benchmarks demonstrate that UMer not only achieves competitive performance over existing baselines, but also provides evidence-grounded interpretability and generalizes to other scale-based tasks.
[NLP-154] When the Weakest Language Keeps Moving: ARMOR-DP for Multilingual BabyLM EMNLP2026
【速读】： 该论文旨在解决多语言生成式模型（Multilingual BabyLM）在训练过程中因固定语言采样比例而导致的语言能力不平衡问题，尤其关注弱语言在训练中动态变化的特性。其核心挑战在于：传统方法假设某一语言始终为最弱语言，因而采用静态保护策略，但实际训练中语言间的性能瓶颈会随时间迁移。解决方案的关键是提出一种模块化诊断与调控框架——ARMOR-DP，通过分析表示（representation）、暴露（exposure）、容量（capacity）和目标压力（objective pressure）四类失衡来源，并利用各语言的验证损失动态调整采样比例，实现对当前瓶颈语言的实时识别与响应。实验表明，动态采样控制器能够有效应对语言间性能转移，相比静态保护策略，在平衡性与整体性能之间取得更优权衡；同时，通过引入字节级编码与可扩展容量，显著缓解了中文等长语言的表征偏差。研究结论强调：多语言模型中的“弱语言”是训练过程中的动态状态，而非语言固有属性。因此，ARMOR-DP通过将诊断与干预分离，使保护机制能够自适应地跟随当前训练瓶颈，从而实现更高效、更鲁棒的多语言训练。
链接: https://arxiv.org/abs/2610.33583
作者: Hanyu Liao,Yijing Chen,Tao Ji
机构: East China Normal University (华东师范大学); Fudan University (复旦大学)
类目: Computation and Language (cs.CL)
备注: Accepted to the BabyLM Workshop at EMNLP 2026
Abstract:Multilingual BabyLM systems usually fix language sampling proportions even though the weakest language can change during training. We present ARMOR-DP, a modular framework that diagnoses imbalance arising from representation, exposure, capacity, and objective pressure, and uses per-language validation loss to update sampling. Controlled experiments show that equal character budgets create a severe representation artifact for Chinese; byte-level encoding removes much of it, while static Chinese protection can subsequently make English the weakest language. The dynamic controller tracks these transitions and achieves a better score–balance tradeoff than fixed protection. On official evaluation, curriculum-dynamic checkpoints outperform a matched static Chinese-protective baseline while using fewer Chinese epochs; scaling capacity yields the strongest internal ARMOR-DP system. These results support a simple conclusion: multilingual weakness is a training-time state rather than a fixed property of a language. ARMOR-DP therefore separates diagnosis from intervention, allowing protection to follow the current bottleneck instead of a predetermined language label.
[NLP-155] rajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents
【速读】： 该论文旨在解决现有工具使用轨迹监督中存在的根本性问题：成功执行的轨迹未必能提供有效学习信号，因其可能遗漏多步依赖关系，或包含在决策时刻不可见信息支持下的不合理动作。其核心挑战在于如何确保监督信号既具备组合覆盖性（compositional coverage）——充分表征关键观测-动作依赖关系，又具备决策可依据性（decision grounding）——每个被监督的动作均基于智能体在决策时可见的历史信息。为此，论文提出Trajectory2Gym框架，通过围绕低支持度的组合结构构建可执行交互，并仅从实际执行中具有可依据性的决策点提取监督信号。该方法通过结构支持选择机制分配经验，结合父节点归一化的监督微调，以语义交互为单位保持训练权重的一致性。这一范式将轨迹生成从“收集成功轨迹”转变为“有意识构造可学习的交互经验”。在五个Qwen3模型规模上，Trajectory2Gym在Tau1、Tau2和BFCL三个基准上均实现提升，平均性能增益达16.9–29.8个百分点，尤其在状态持续性和多轮交互场景中表现突出，验证了强化依赖显式建模与决策可解释性对规模化工具使用智能体的关键作用。
链接: https://arxiv.org/abs/2610.33568
作者: Yihao Hu,Zhihao Wen,Yitong Lin,Siyuan Lu,Pan Wang,Wei Wu,Jiang-Ming Yang,Tao Lin
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Successful tool-use trajectories are not necessarily useful supervision. Even successful interactions may underrepresent multi-step dependencies or contain actions that are unjustified by the information visible to the agent at decision time. We identify two coupled requirements for effective trajectory supervision: compositional coverage, so that important observation-action dependencies are represented, and decision grounding, so that each supervised action is supported by the learner-visible interaction history. We introduce Trajectory2Gym, which constructs executable interactions around under-supported compositional structures and derives supervision only from grounded decisions in their realized executions. Structural-support selection allocates the resulting experience, while parent-normalized supervised fine-tuning preserves the semantic interaction as the unit of training weight. This reframes trajectory generation as the deliberate construction of learnable interaction experience, rather than the collection of successful trajectories. Across five Qwen3 scales, Trajectory2Gym improves Tau1, Tau2, and BFCL at every tested scale, increasing the equal-weight three-benchmark mean by 16.9-29.8 percentage points. The gains are strongest in stateful and multi-turn interaction, where later decisions depend on earlier observations and feedback. Our results suggest that scaling tool-use agents depends not only on collecting more successful trajectories, but on constructing interactions whose dependencies are represented and whose decisions are learnable from the agent’s visible history.
[NLP-156] SEDATE: Do Large Language Models Share an Aggression Circuit Across Languages?
【速读】： 该论文旨在解决多模态与多语言场景下人工智能系统安全性和责任性部署的挑战，尤其针对现有安全干预方法普遍存在计算开销大、标注成本高或难以拓展至多语言环境的问题。其解决方案的关键在于提出一种无需训练的“检测-编辑-验证”（Selective Detection And Targeted Editing, SEDATE）框架，通过识别并针对性地修改关键责任神经元（responsible neurons），在不进行任何额外模型训练的前提下实现安全响应生成。该方法在7种语言和4类模型架构上均表现出显著优于基线的安全性提升，同时保持模型通用能力且计算开销可忽略，且在分布外语言和多个额外数据集上结果依然稳健，验证了其跨语言泛化能力与鲁棒性。
链接: https://arxiv.org/abs/2610.33542
作者: Sagnik Basu,Sayan Das,Somnath Banerjee,Animesh Mukherjee
机构: Indian Institute of Technology Kharagpur(印度理工学院克哈拉格普尔分校); Singapore Institute of Technology(新加坡科技学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:As the applications of AI continue to expand across various multimodal and multilingual domains, ensuring safe and responsible deployment of these systems has become an important concern. A wide-range of interventions have been proposed to make AI responses safer and more reliable. They often incur substantial computational or annotation costs, or cannot be extended to multilingual settings. In this work, we propose Selective Detection And Targeted Editing (SEDATE), a training-free approach based on detect-edit-verify strategy. SEDATE first detects the responsible neurons, edits them as and when necessary and finally verifies the model responses; all without any additional model training. We evaluate SEDATE across 7 languages and 4 model families. We further evaluate its effectiveness across 4 out-of-distribution languages and 4 additional datasets. We show that SEDATE achieves substantially superior safety scores across all languages and models over the best-performing baseline while retaining the general capabilities and incurring negligible overhead. Our results are robust, with the trends remaining similar for out-of-distribution languages and datasets.
[NLP-157] SkillShift-Bench: Benchmarking Skill Learning under Environment Evolution for Web Agents EMNLP2026
【速读】： 该论文旨在解决生成式 AI（Generative AI）在网页代理（web agents）中学习技能后，在真实开放环境部署时面临的鲁棒性下降问题。现有方法通常在封闭环境中评估技能，忽略了实际应用中上下文（Contextual Shift）、感知基础（Perceptual Shift）和执行可行性（Execution Shift）的持续变化。为此，论文提出SkillShift-Bench，首个针对受控开放环境变化的技能学习评估基准，通过三个自托管网页系统模拟三类环境变化，并从任务成功率、交互成本、技能使用率和技能失败率等维度进行诊断性评估。实验表明，受控变化下技能表现系统性退化，表现为成功率降低、交互成本上升及技能失效频发，其中执行变化的影响最为显著。研究进一步将变化后的技能执行建模为非平稳局部可执行性问题，提出基于动态遗憾（dynamic-regret）的理论分析框架与一种无需重训练技能库的测试时适应（test-time adaptation）方法，验证了在不重新学习技能库的前提下提升已学技能执行能力的可行性。
链接: https://arxiv.org/abs/2610.33528
作者: Bo-Yu Chen,Zhi Zhou,Yu-Feng Li
机构: Nanjing University (南京大学); National Key Laboratory for Novel Software Technology (新型软件技术国家重点实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 70 pages, 8 figures. Accepted to EMNLP 2026 (Main Conference)
Abstract:Skill learning enables web agents to reuse prior experience for long-horizon interactive tasks. However, learned skills are typically evaluated in closed environments, overlooking the fact that real-world web deployment involves continuous shifts in invocation context, perceptual grounding, and execution feasibility. To this end, we introduce SkillShift-Bench, the first benchmark for evaluating skill learning with controlled open-environment shifts. SkillShift-Bench simulates Contextual Shift, Perceptual Shift, and Execution Shift across three self-hosted web systems, and evaluates agents with diagnostics covering task success, interaction cost, skill use, and skill failure. Comprehensive experiments reveal systematic robustness degradation under controlled shifts, characterized by lower success rates, higher interaction costs, and more frequent skill failures, with Execution Shift exerting the strongest pressure. We further formulate shifted skill execution as a nonstationary local executability problem and provide both a dynamic-regret analysis and a practical test-time adaptation method, demonstrating the feasibility of improving learned-skill execution without relearning the skill library. Code, tasks, environments, documentation, and the leaderboard are available at this https URL.
[NLP-158] Implicit Beliefs as Scalable Process Rewards for Strategic Persuasion NEURIPS2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多轮策略性说服对话训练中面临的两大耦合挑战：一是如何真实模拟具有心理基础的说服对象（persuadee），二是如何实现对策略对话中各对话行为的密集且忠实的信用分配（credit assignment）。传统基于结果的奖励信号稀疏且无法提供成功原因的细粒度反馈，而依赖大模型判官（LLM-judge）的奖励机制则成本高昂且与交互动态脱节。为此，本文提出一个统一框架：首先构建了PersuArena——一个能够生成非单调、真实感强的说服轨迹的仿真环境；其次在此基础上提出基于在线信念提取的过程奖励方法（Process Rewards via Online Belief Elicitation for Relative Policy Optimization, Probe-RPO），通过在每轮对话中从冻结的说服对象的输出概率分布（logits）中读取三个内省式探针（introspective probes），生成多维、高密度的过程奖励信号，从而精确对齐说服过程中的动态变化。实验在涵盖医疗、金融咨询、职场管理、教育及个人关系等领域的PersuArena上进行，结果表明Probe-RPO实现了52.5%的完全接受率，显著优于最强的GRPO消融实验（35.8%）和最优现成大模型（25.8%）。此外，研究还进行了跨说服对象模型的泛化能力分析以及对话过程中说服动态的深入剖析。本工作为策略性对话提供了有效的过程监督范式，推动了生成式人工智能在复杂社会交互任务中的发展。
链接: https://arxiv.org/abs/2610.33527
作者: Minqian Liu,Yuexi Shen,Heajun An,Qi Zhang,Pamela J. Wisniewski,Sang Won Lee,Jin-Hee Cho,Lifu Huang
机构: Virginia Tech(弗吉尼亚理工学院); UC Santa Barbara(加州大学圣芭芭拉分校); UC Davis(加州大学戴维斯分校); STIR Lab
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: NeurIPS 2026 Workshop (Large Language Models for Social Reasoning and Simulation)
Abstract:Training large language models (LLMs) for multi-turn strategic persuasion faces two coupled obstacles: (i) realistic simulation of psychologically grounded persuadees, and (ii) dense, faithful supervision for credit assignment in strategic dialogues. Outcome-only rewards are sparse and uninformative about which conversational moves drove success, while LLM-judge process rewards are expensive and decoupled from the underlying interaction dynamics. We address both challenges with a unified framework. First, we introduce PersuArena, a persuasion simulation environment that produces realistic, non-monotonic trajectories. Second, building directly on this environment, we propose Process Rewards via Online Belief Elicitation for Relative Policy Optimization (Probe-RPO), an online RL framework that converts the persuadee’s internal state into dense process rewards. At each turn, three introspective probes are read from the frozen persuadee’s logits, yielding a multi-dimensional reward signal that is faithfully aligned with persuasion dynamics. With extensive experiments on PersuArena spanning healthcare, financial advising, workplace management, education, and personal relationships, we demonstrate that our Probe-RPO achieves the full acceptance rate of 52.5%, compared with the 35.8% for the strongest GRPO ablation and 25.8% for the best off-the-shelf LLM. We also conduct comprehensive analyses on the generalization across persuadee models and the persuasion dynamics across conversations. Our work offers a promising paradigm that can provide effective process supervision for strategic dialogue.
[NLP-159] One Model Two Paths: Balancing Single-Pass and Looped Performance in Language Models
【速读】： 该论文旨在解决在保持单次前向传播（single-pass）性能的同时，提升循环推理（looped inference）性能的问题。由于模型通过复用中间层权重实现循环计算，导致在仅优化循环推理时会损害单次前向传播的输出质量。其解决方案的关键在于采用联合训练策略，对单次前向和循环推理的输出分别施加加权损失函数，并利用冻结的原始预训练模型作为指导信号。实验表明，通过合理调整单次前向损失的权重并结合原始模型的监督，可显著降低单次前向性能下降幅度，同时有效保留循环推理带来的增益。在验证集选定最优权重后，该方法能够在维持单次前向性能接近普通持续训练水平的前提下，显著提升循环推理的表现。
链接: https://arxiv.org/abs/2610.33517
作者: Jiarun Shen,Lei Zhao,Xiao Li
机构: The Chinese University of Hong Kong, Shenzhen(深圳大学); Shanghai Jiao Tong University(上海交通大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Repeating Transformer layers can improve language-model performance without adding parameters. A pretrained model can be adapted to support both single-pass and looped inference by reusing middle-layer weights for additional computation. Because both paths share weights, training to improve looped predictions can degrade single-pass quality. We study joint training strategies using weighted losses on both outputs, with guidance from the frozen original model. Joint supervision substantially reduces single-pass degradation compared to looped-only training, while preserving looped gains over ordinary continued training. Most of this benefit comes from adjusting the single-pass loss weight, with guidance from the frozen original model further improving the looped output. With validation-selected weights, our method preserves single-pass quality within a predefined tolerance of ordinary continued training while improving looped predictions.
[NLP-160] Eat Your Precision: Precision Objective Mitigates Reward Hacking in Rubric-based RL
【速读】： 该论文旨在解决基于评分标准（rubric-based）的强化学习在非可验证任务中导致模型生成响应过于冗长、包含大量无关或事实错误内容的问题。现有方法通常仅以评分标准覆盖度加权作为奖励信号，促使模型追求高召回率，从而牺牲了生成内容的精确性与简洁性。其解决方案的关键在于提出一种名为FORK的新型奖励机制，该机制同时优化召回率与一项新颖的精确性目标——要求生成响应中的每一部分均具备必要性且事实正确，通过评分驱动的必要性评分（necessity credits）和事实正确性评分器（factual-correctness grader）共同实现。实验结果表明，FORK在保持相近评分标准覆盖度的同时显著提升了响应精确性并缩短了输出长度，在RubricHub-Medical和HealthBench数据集上相对于基线策略的胜率分别从10.6%和22.1%提升至68.0%和74.1%，实现了召回与质量之间的帕累托最优平衡。
链接: https://arxiv.org/abs/2610.33512
作者: Hyun Ryu,Jong-Heon Kim,Yiqing Xie,Seungone Kim,Carolyn Rose,Eunho Yang
机构: KAIST AI(韩国科学技术院人工智能); Carnegie Mellon University Language Technologies Institute(卡内基梅隆大学语言技术研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Rubric-based rewards enable effective reinforcement learning on non-verifiable tasks. However, because these rewards are typically defined as a weighted measure of rubric coverage, models are trained primarily to maximize a recall objective. As a result, RL-trained models produce responses that are more complete but also more verbose and contain more unnecessary or factually incorrect content, leading them to be less preferred than responses from the base model. To address this issue, we present the FORK reward, which jointly optimizes recall and a novel precision objective, which requires each unit of a model response to be both necessary and factually correct, as measured by rubric-driven necessity credits and a factual-correctness grader, respectively. Our experimental results show that FORK outperforms existing rubric-based RL methods and achieves Pareto-optimal performance that balances rubric coverage and overall response quality. Compared to the original recall-based reward, FORK substantially improves precision and reduces response length while achieving similar rubric coverage, which leads to significantly higher win rate compared to the base policy (e.g., from 10.6% to 68.0% on RubricHub-Medical and from 22.1% to 74.1% on HealthBench).
[NLP-161] owards a Learnable Retriever for Textual Graph Question Answering
【速读】： 该论文旨在解决现有文本图问答（Textual Graph Question Answering, GraphQA）与基于图的检索增强生成（Graph-based Retrieval-Augmented Generation, GraphRAG）方法中存在的两大关键问题：一是架构上的缺陷，即现有方法通常仅微调生成器，而未使检索器本身也成为可学习的大型语言模型（LLM），导致无法在端到端的“检索-生成”流程中实现联合优化；二是检索监督不足的问题，现有检索器极少利用真实答案信号来指导证据节点的选择，从而可能导致关键推理节点遗漏，造成对正确答案集合的覆盖不全；同时，盲目提升召回率又会引入大量无关信息，损害结果的紧凑性并增加生成模型的负担。为弥合上述差距，本文提出TaLenT框架，其核心在于构建一个统一的端到端检索-生成架构，将可学习的LLM检索器与微调后的LLM生成器协同集成，并创新性地利用真实答案信号作为检索器的监督信号。具体而言，通过将图节点划分为真实答案节点、相关节点与无关噪声节点三类，采用加权伪标签损失函数，优先保留真实答案和相关节点以保障覆盖率，同时剔除无关噪声以维持输出紧凑性。实验结果表明，该方法在三个GraphQA数据集上均显著提升了下游任务性能。
链接: https://arxiv.org/abs/2610.33511
作者: Kaifeng Jin,Yuchen Yan,Zhichen Zeng,Xiao Lin,Lihui Liu,Yinglong Xia,Hanghang Tong
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 16 pages, 4 figures, 10 tables
Abstract:Textual graph question answering (GraphQA) and graph-based retrieval-augmented generation (GraphRAG) have gained increasing attention as a way to ground large language models (LLMs) on structured knowledge. Despite extensive efforts, existing methods still face two key gaps. The first is an architectural gap: existing methods typically fine-tune an LLM-based generator, but do not also make the retriever itself a learnable LLM for graph evidence selection within one end-to-end retrieve-then-generate pipeline. The second is a retrieval-supervision gap: existing retrievers rarely use gold answer signals to supervise evidence selection. As a result, they may fail to recall evidence nodes required for reasoning, resulting in insufficient coverage of the gold answer set; meanwhile, naive attempts to boost recall often introduce excessive irrelevant information that compromises compactness and overwhelms the LLM. To address these gaps, we propose TaLenT, an LLM-based end-to-end retriever-generator framework for textual GraphQA. To bridge the architectural gap, TaLenT pairs a learnable LLM-based retriever for evidence selection with a fine-tuned LLM-based generator within one unified retrieve-then-generate pipeline. To bridge the retrieval-supervision gap, we leverage gold answer signals as retriever supervision. Specifically, we stratify graph nodes into gold answers, answer-related nodes, and irrelevant noise. We then adopt a weighted pseudo-label loss to prioritize gold answers and preserve answer-related nodes for coverage, while discarding irrelevant noise for compactness. Experimental results across three GraphQA datasets demonstrate that TaLenT consistently improves downstream performance.
[NLP-162] Five Architectures One Shared Space Cross-Model Representational Alignment Without Weight Merging
【速读】： 该论文旨在解决多模型融合中缺乏统一权重初始化路径时如何实现跨模型语义对齐的问题，特别是针对五种独立训练、结构各异且参数冻结的模型（包括四种来自不同组织的Transformer模型和一种状态空间模型）能否仅通过激活值建立可共享的语义空间。其解决方案的关键在于：不依赖权重合并，而是基于激活表示的统计一致性，利用置换检验、独立复制探测集、深度剖面分析、池化消融实验与混杂因素检查，验证了各模型在处理相同文本时存在显著且稳健的表征对应关系，尤其在最终层表现最强；进而采用广义普鲁克斯特分析（Generalized Procrustes Analysis, GPA）构建一个无固定参考基准的共享表示空间。该空间在最终层实现了高达0.91–0.93的平均保留检索准确率（随机水平为0.03），优于成对线性映射方法（0.82–0.85），但在中间层表现较差。此外，研究进一步评估该共享空间是否可用于构建智能路由机制以动态选择最优模型，结果表明在四类任务中均未发现路由策略优于始终使用最强模型的表现，且所谓“正确性信号”实为类别归属的伪装。所有实验均在无GPU的单台笔记本电脑上完成，验证了方法的低资源可行性。
链接: https://arxiv.org/abs/2610.33495
作者: Oladimeji Anthonio,Dimeji Olawuyi,Oloruntoba Ajayi,Temilola Aderemi
机构: Centre for Clinical Intelligence and Safety, Ìyàwó; Faculty of Clinical Sciences, College of Medicine, University of Ibadan (伊巴丹大学医学院临床科学学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Combining language models is usually done by averaging weights, which works only when the models share a common starting point. We ask whether five frozen, independently trained models with different internal designs (four transformers from four organizations and one state-space model) can be related through their activations alone, with no weight merging. Using permutation tests, an independent replication probe set, depth sweeps, pooling ablations and confound checks, we find statistically robust correspondence between how the five models organize the same text. It is strongest at each model’s final layer and weaker for the state-space model, whose weakest pairing is still 35-43 times chance. We use this to build a shared space with Generalized Procrustes Analysis, in which no model is a fixed reference. At the final layer, mean held-out retrieval accuracy over the twenty ordered model pairs is 0.91 to 0.93 on two probe sets (chance 0.03), beating pairwise linear maps (0.82 to 0.85), but the space is worse than those maps in the middle layers. It transfers to ordinary English of unfamiliar grammatical form and fails on mathematical and logical notation. We then test whether the space supports a router that picks among the five models on gradeable tasks. Across four task types we find no case where routing beats always using the strongest model, and we report this as a negative result. A correctness signal that looked strong turned out to be category membership in disguise. All experiments ran on one laptop without a GPU.
[NLP-163] Small Edits Shifting Decisions: Benchmarking LLM Accuracy and Reliability on French Input Variants
【速读】： 该论文旨在解决法语用户在日常输入中常省略撇号和变音符号或产生键盘拼写错误（即表面变体，Surface Variants）对大语言模型（LLM）性能与预测可靠性的影响问题。现有研究对该类常见文本变异如何影响模型输出的准确性与概率校准质量尚缺乏系统性理解。为此，作者构建了一个受控的法语表面变体基准测试集，涵盖七种任务类型，包含4,200个原始-低强度-高强度三元组（共12,600个输入实例），其中高强度变体在低强度修改基础上叠加更多扰动。实验评估了来自两个大语言模型家族的六种模型，结果表明：随着扰动强度增加，模型准确率下降，Brier评分与负对数似然（NLL）上升；尽管更大规模模型在高扰动下准确率损失较小，但其Brier评分与NLL的改善并不随模型规模持续提升。键盘拼写错误在预设高强度条件下导致最严重的准确率下降及预测概率质量恶化。此外，基于原始输入训练的校准温度可有效迁移至表面变体，结合输入修复与校准的方法进一步优于单一策略，显著降低Brier评分与NLL，同时保持修复后的准确率。该研究的关键贡献在于提出一个统一框架，用于联合评估模型在常见法语表面变体下的任务准确性与预测可靠性。
链接: https://arxiv.org/abs/2610.33493
作者: Jiaming Yu,Xiaotian Xie,Jian Ma,Yingjie Li
机构: Hangzhou International Innovation Institute, Beihang University(北京航空航天大学杭州国际创新研究院); School of Automation, Central South University(中南大学自动化学院); School of Reliability and Systems Engineering, Beihang University(北京航空航天大学可靠性与系统工程学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:French users may omit apostrophes and diacritics or make keyboard typos in everyday input. The effects of these common variations on language model performance and prediction reliability remain insufficiently understood. To address this gap, we construct a controlled French surface-variant benchmark covering seven task types, with 4,200 original-low-high triplets (12,600 input instances). Each high-intensity variant retains all low-intensity edits and adds further changes. We evaluate six models from two LLM families. The results show that accuracy decreases and Brier score and negative log-likelihood (NLL) increase with perturbation intensity. Larger models show smaller accuracy losses under high intensity, but increases in Brier score and NLL do not consistently decrease with scale. Keyboard typos cause the largest losses in accuracy and the quality of predicted probabilities at the predefined high-intensity settings. Finally, calibration temperatures fitted on original inputs transfer effectively to surface variants. Combining calibration with input restoration further improves Brier score and NLL over either method alone while preserving restored accuracy. Our benchmark provides a unified framework for jointly evaluating task accuracy and prediction reliability under common French surface variations.
[NLP-164] Incremental Predictive Value of LLM Tutoring Interactions for Student Performance
【速读】： 该论文旨在解决如何利用大语言模型助教在学期层面的交互特征，提升对学生作业成绩预测的准确性，以超越固定基准模型的表现。其核心问题在于：现有基于常数基准（如均值或中位数）的预测方法是否足以充分捕捉学生表现的动态变化，抑或引入更复杂的交互特征能够带来实质性改进。解决方案的关键在于采用多种拟合估计器（包括梯度提升、岭回归等）与嵌套交叉验证框架，系统评估不同模型在真实教育数据集（StudyChat）上的表现，并通过重复十次的留出预测平均来降低随机性影响。结果显示，尽管梯度提升模型在整体上仅小幅优于中位数基准（均绝对误差分别为3.023和3.138），但在低分四分位区间内其表现显著更优（误差7.30 vs. 8.07），且岭回归在该子群体中误差最低（6.37）。研究进一步揭示成绩高度集中于高分段（91.4%学生得分高于90），导致低分群体虽占比小但贡献了大部分误差，从而凸显基准选择与误差分布对预测价值评估的重要性。这一回顾性分析为评估学生支持工具提供了方法论启示，但未建立干预带来的学习效应。
链接: https://arxiv.org/abs/2610.33488
作者: Zeyu Dong,Mikael Kwame Kuada,Asa Holley Buchakjian-Tweedy,Ibrahem Aljabea
机构: Crestwood Preparatory College (克雷斯特伍德预备学院); Hammond High Magnet School (汉默德高中磁学项目); Spaulding High School (斯普尔丁高中); Louisiana State University (路易斯安那州立大学)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 8 pages, 3 figures
Abstract:We evaluate whether semester-level interaction features from a large language model tutor improve assignment-score prediction beyond constant baselines. StudyChat provides 16,851 student-assistant exchanges from two offerings of an undergraduate artificial intelligence course, with outcomes matched to 175 of 203 student-semester records. Six fitted estimators and training-fold mean and median baselines share repeated nested cross-validation. After averaging each student’s held-out predictions across ten repeats, Gradient Boosting achieves mean absolute error of 3.023 grade points, versus 3.138 for the median baseline. The paired difference is -0.114, with a conditional 95% bootstrap interval of [-0.245,0.013]. Representation changes and a semester-specific median yield similarly small gains. In the lowest observed-score quartile, however, boosting’s error is 7.30 points, versus 8.07 for the median and 6.67 for the mean; Ridge has the lowest fitted-model error at 6.37 despite its worse overall error of 3.331. Scores concentrate near the ceiling: 160 students (91.4%) score above 90, and the lowest quartile contributes 62.1% of boosting’s total absolute error. This retrospective comparison shows how baseline choice and error distribution affect predictive value, informing evaluation of student-support tools without establishing a learning effect.
[NLP-165] Propasafe-Hybrid: A Text-Based Hybrid Propaganda Detection Tool
【速读】： 该论文旨在解决在线新闻与社交媒体中传播性内容（propagandistic content）泛滥且用户缺乏足够审慎识别的问题，提出了一种细粒度、可解释的句子级检测方案。其核心解决方案是Propasafe-Hybrid系统，该系统通过融合微调的Transformer分类器与基于大语言模型（LLM）的技术分类机制，实现对特定宣传策略的精准识别、标签标注与原因解释。系统输出包括高亮句子、技术归类及简洁推理依据，显著提升了可解释性与用户理解效率。为控制大语言模型的计算成本，系统引入成本感知的预过滤阶段，仅将高置信度句子送入LLM处理，有效降低令牌（token）消耗的同时保持决策逻辑的完整性。这一设计在保证检测准确性与透明度的前提下，大幅增强了系统在真实新闻环境中的实用性与运行效率。
链接: https://arxiv.org/abs/2610.33464
作者: Thomas Kimmeth,Avijit Roy,Vivek Sharma
机构: John Jay College of Criminal Justice, CUNY(纽约城市大学约翰杰伊刑事司法学院)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: Published in the Proceedings of FLAIRS-39 (2026), Special Track: Applied Natural Language Processing
Abstract:Propagandistic content increasingly circulates through online news and social media, where readers often encounter it with limited scrutiny, highlighting the need for reliable and fine-grained detection. This paper introduces Propasafe-Hybrid, a sentence-level system that integrates a fine-tuned transformer classifier with LLM-based technique classification to identify, label, and explain specific propaganda strategies. The pipeline generates actionable outputs, including highlighted sentences, technique assignments, and concise rationales, so users can immediately understand why a sentence was flagged and how each label was determined. To control inference cost, Propasafe-Hybrid employs a cost-aware pre-filtering stage that forwards only high-likelihood sentences to LLMs, reducing token usage while preserving the underlying decision logic. Together, these design choices enhance the explainability, efficiency, and practical usability of sentence-level propaganda detection in real-world news environments.
[NLP-166] Adaptive Transfer of Representational Signatures for LLM Reliability Detection
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在生成文本时存在幻觉（hallucination）和推理错误等可靠性问题，且现有质量检测器在新领域应用时受限于需大量特定领域标注数据的问题。其核心解决方案是通过跨相关领域的表征签名迁移实现高效适应性迁移学习。关键在于构建一个统计框架：首先提取各领域的低维表征签名，再利用交叉拟合（cross-fitting）方法学习不同源领域预测器的组合权重，使权重能够动态反映各源域对目标域的预测相关性。该方法的渐近分析揭示了最小二乘类估计下预测风险与聚合权重与领域差异及表示偏差之间的理论关系。实验结果表明，在仅需25个目标领域标注样本的情况下，该自适应迁移方法使幻觉检测的AUROC提升6.8%，推理步骤预测的AUROC提升8.4%，显著提升了检测器在新领域的泛化能力与实用性。
链接: https://arxiv.org/abs/2610.33442
作者: Yongtong Li,Yixuan Florence Wu,Jiawei Zhang,Naichen Shi
机构: Northwestern University(西北大学); University of Wisconsin–Madison(威斯康星大学麦迪逊分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) are reported to be susceptible to reliability issues including hallucinations and reasoning errors. Existing quality detectors can predict unreliable outputs based on internal representations with decent accuracies. However, training these detectors often requires substantial domain-specific labeled data, which limits their applicability to new domains. We propose to address this challenge by transferring representational signatures across related domains. More specifically, we build a statistical framework that first extracts low-dimensional signatures for each domain, then uses cross-fitting to learn composition weights of domain-specific predictors. The weights adapt to each source domain’s predictive relevance to the target domain. Our asymptotic analysis of a least-squares analogue characterizes the limiting prediction risk and aggregation weights in terms of domain discrepancies and representation bias. Experiments across multiple hallucination datasets and open-weight LLMs show that, with only 25 labeled questions in the new domain, adaptive transfer improves hallucination-detection AUROC by 6.8%. For reasoning-step prediction, it improves AUROC by 8.4%.
[NLP-167] Quantifying Meaning with SCAPE-T: Semantic Complexity from Encoder-Only Transformer Attention
【速读】： 该论文旨在解决自然语言处理（NLP）领域中语义复杂度（semantic complexity）自动化评估缺乏系统性方法的问题。现有方法主要聚焦于可读性（readability）、词汇难度或表面特征（如词长、句法结构）的量化，这些指标易受表层语言形式的干扰，难以准确反映文本深层语义整合的复杂程度。本文提出SCAPE-T，一种基于仅编码器（encoder-only）Transformer架构的注意力机制衍生度量方法，通过直接建模语义整合过程来定义和计算语义复杂度：其核心是将语义复杂度操作化为唯一指定文本语义所需代码长度，并结合三个子指标——（i）逆权重自关注（token-to-self），（ii）CLS关注度（token-to-CLS），以及（iii）跨头冗余度（inter-head redundancy，以注意力头间余弦相似度衡量），并通过分层加权聚合实现整合。该方法的数据驱动部分仅限于超参数校准，避免过度依赖训练数据。实验在哲学散文与儿童故事对比、儿童读物与技术维基百科文章对比等任务上验证，以成对比较准确性为主要评估标准，结果显示，在控制词长的前提下，SCAPE-T达到78.5%的成对准确率（95%置信区间：76.1–80.4，N=10k对比较），显著优于所有基线模型（领先最强基线12.7个百分点）；而在未控制词长时，依赖表面特征的可读性指标表现更优，印证了其对表层特征的敏感性。在抽象性短文本数据集上的测试也表明SCAPE-T具有优势。消融分析进一步证明各子指标均对性能有实质性贡献。综上，SCAPE-T的关键创新在于有效分离语义结构与词汇/句法混杂因素，实现了对概念结构复杂度的更纯净刻画，并展现出良好的跨领域泛化能力。
链接: https://arxiv.org/abs/2610.33441
作者: Maxim Khovansky,Lev Mukhanov
机构: Queen Mary University of London (伦敦玛丽女王大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 12 pages; semantic complexity, complexity estimation, encoder attention, attention analysis, pairwise accuracy, readability, text simplification, transformers
Abstract:Automated semantic complexity estimation remains underexplored in NLP, despite important applications in linguistics and interpretability. This paper defines semantic complexity as the code length required to uniquely specify a text’s semantic meaning, and operationalises it via SCAPE-T, an attention-derived metric based on encoder-only transformers. Existing measures either model reader effort (readability, lexical difficulty), which correlates with surface features (word length, syntax), or quantify surface features directly (type-token ratio, word frequency). SCAPE-T targets semantic integration directly by combining three submetrics: (i) inverse-weighted self-focus (token-to-self), (ii) CLS focus (token-to-CLS), and (iii) inter-head redundancy (cosine similarity across heads), using layer-weighted aggregation; data-driven learning is confined to hyperparameter tuning for calibration. Development uses philosophical prose vs. MCTest stories; evaluation is on children’s stories vs. technical Wikipedia articles, with pairwise accuracy on contrasts as the primary criterion. Baselines include readability, syntactic/lexical, and LLM-derived measures. SCAPE-T substantially outperforms all tested baselines under word-length control, with a pairwise accuracy of 78.5% (95% bootstrap CI: 76.1-80.4, N=10k pairwise comparisons; +12.7 percentage points over the strongest baseline); without that control, readability heuristics overtake, consistent with their reliance on surface features. On a short-terms abstractness dataset, SCAPE-T also leads. Ablations show each submetric contributes materially. Overall, the results show SCAPE-T isolates conceptual structure from lexical/syntactic confounds better than existing methods, and generalises across two held-out domains.
[NLP-168] Perturbation-Based Data Selection for Medical Question Answering NEURIPS2026 ALT
【速读】： 该论文旨在解决医学问答（Medical QA）领域中如何从大规模现有问答库中高效选择紧凑训练子集的问题，以实现对医学语言模型的低成本、高效率微调。其核心挑战在于在有限的标注数据预算下，如何筛选出最具代表性和信息量的样本，从而最大化模型性能。解决方案的关键在于提出一种基于高斯扰动的表示敏感性优先级选择方法（Representation-Sensitivity Prioritization, RSP），通过分析最终层表示在高斯扰动下的两个信号——预测变化频率（prediction-change frequency）与非翻转边界收缩（non-flipping margin contraction）——来评估样本的重要性。RSP采用自动均方根缩放融合两信号，而RSP-Drop则仅依赖边界收缩信号进行排序。两者均仅依赖问题与答案选项完成选择，无需额外标注或复杂计算。实验表明，在128个样本的基准预算下，RSP和RSP-Drop分别获得47.55和48.28的平均宏F1分数，显著优于基于熵池匹配的随机采样（45.11）。研究进一步揭示了不同数据集和预算下性能差异的成因，并通过案例分析与消融实验指出熵预过滤并非在所有场景中均具普适优势，从而为扰动驱动的数据选择提供了实证依据与局限性认知。
链接: https://arxiv.org/abs/2610.33440
作者: Yuexin Wu,Dayou Yu,Vasile Rus
机构: University of Memphis (孟菲斯大学)
类目: Computation and Language (cs.CL)
备注: 10 pages. Preprint submitted to IEEE ICHI 2027. Extends the non-archival GenAI4Health @ NeurIPS 2026 study arXiv:2609.09684v2 with automatic RMS scaling, RSP-Drop, budget analyses, and entropy-prefilter ablations
Abstract:Selecting a compact training subset is an important step in adapting medical language models from existing question-answering banks. We investigate perturbation-based data selection using two signals derived from Gaussian perturbations of the final-block representation: prediction-change frequency and non-flipping margin contraction. After a shared entropy prefilter, Representation-Sensitivity Prioritization (RSP) combines the signals using automatic root-mean-square scaling, while RSP-Drop ranks candidates by contraction alone. Both selectors use only questions and answer options during selection. We evaluate them with MedGemma-4B on five medical QA datasets at retained-set budgets of 64, 128, and 256 examples. At the primary budget of 128, RSP and RSP-Drop obtain dataset-averaged Macro-F1 scores of 47.55 and 48.28, respectively, compared with 45.11 for entropy-pool-matched random selection. Their relative performance varies across datasets and budgets. A selected-set case study identifies contrasting clean-margin profiles among the examples exchanged by the two rules, while a full-pool ablation shows that entropy prefiltering is not uniformly beneficial. These retrospective results characterize the empirical utility and limitations of perturbation-based selection under a fixed adaptation protocol. Evaluation uses one backbone and benchmarks with documented training-family overlap; it does not establish clinical effectiveness.
[NLP-169] Examining form-based linguistic productivity in people and large language models
【速读】： 该论文旨在解决人类语言生成能力中抽象语法知识的性质问题，特别是人类如何在缺乏具体语义的情况下仍能实现语言产出的创造性泛化。其核心问题是：人类与计算语言模型在面对新词时，是否能够基于句法形式进行抽象规则迁移，从而在被动语态等陌生语境中正确扩展使用。研究的关键发现是，人类能够在无法赋予新动词明确语义的前提下，依据句法形式的抽象规律（如主动语态向被动语态的转换规则）完成语言泛化；而大型语言模型（LLM）仅在新动词具有可类比的已有动词语义时才能成功完成此类泛化。这表明，当前生成式 AI 的句法知识仍高度依赖具体词汇项的分布特征和语义关联，而人类的句法知识则具备独立于特定词汇的抽象性，体现了更深层次的语言认知机制。
链接: https://arxiv.org/abs/2610.33433
作者: Michael Wilson,Robert Frank
机构: University of Nevada, Reno (内华达大学雷诺分校); Yale University (耶鲁大学)
类目: Computation and Language (cs.CL)
备注: Submitted to Nature Human Behaviour
Abstract:Linguistic productivity, as evidenced by the use of words in novel contexts, is one of the most striking facts about human language. However, how abstract the knowledge that enables such productivity is remains unclear. We compare people’s and computational language models’ linguistic productivity in contexts where successful generalization depends on form-based information. We focus on extending the use of a novel verb learned in the active voice to passive contexts. We find people extend active to passive uses in a way consistent with abstract generalizations concerning linguistic form, even when no plausible meaning can be assigned to the novel verb. In contrast, large language models succeeded only when a meaning like an existing verb’s could be assigned to the novel verb. We conclude that computational models’ syntactic knowledge remains tied to the distribution and meaning of specific lexical items in a way that human syntactic knowledge is not.
[NLP-170] How Does Post-Training Shape LLM Value Preferences?
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在后训练阶段因训练语料中不同人口群体意见分布不均，导致价值偏好被放大并进一步加剧群体代表性不足的问题。其核心解决方案在于通过受控实验，系统评估不同意见混合比例的后训练语料与多种后训练路径（基于监督微调SFT、直接偏好优化DPO和广义相对策略优化GRPO）对价值偏好演化的影响。研究发现，不同后训练方法会以异质方式放大意见不平衡，且一旦价值偏好被强化，后续训练是否能够消除这些偏差，取决于具体的训练方法和语料中的意见构成。更关键的是，模型在干预下的稳定性与其价值偏好的变化密切相关；此外，即使蒸馏过程中未包含显式意见问题，原始模型中的价值偏好仍可传递至学生模型。这一发现表明，在开发、评估及复用大语言模型时，必须同时考虑后训练语料中的意见混合结构与后训练方法的选择，以避免隐性偏见的持续传播。
链接: https://arxiv.org/abs/2610.33371
作者: Yang Liu,Shaoyang Xu,Masahiro Kaneko,Chengzhi Zhong,Wenxuan Zhang,Chenhui Chu
机构: Kyoto University(京都大学); Singapore University of Technology and Design(新加坡科技设计大学); Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学)
类目: Computation and Language (cs.CL)
备注: 46 pages, 21 figures, 21 tables
Abstract:During post-training, large language models (LLMs) learn value preferences from corpora that may contain imbalanced proportions of opinions associated with different demographic groups. Amplifying these imbalances may further underrepresent some groups’ opinions. However, how opinion mixtures shape value preferences across post-training paths remains unclear. To answer this question, we conduct controlled experiments using post-training corpora with different opinion mixtures and five post-training paths based on SFT, DPO, and GRPO. We find that post-training amplifies opinion imbalances differently across post-training paths. Once value preferences are amplified, whether further post-training can remove them depends on both the post-training method and the opinion mixture in the corpus. More concerningly, a model’s stability under interventions depends on whether further post-training changes its value preference. Moreover, these value preferences are transferred to student models even when distillation uses corpora with no explicit opinion questions. These findings suggest that both the opinion mixture in post-training corpora and the choice of post-training method should be considered when developing, evaluating, and reusing LLMs. Our code and data are available at this https URL.
[NLP-171] Distill Debate Discover: Grounding Multi-Agent Research Ideation in Graph-Distilled Insights
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在科研创意生成与科学发现中，因缺乏具有区分度和实质性专业知识的协作主体而难以有效开展多智能体辩论的问题。其核心挑战在于如何赋予智能体具备真实、可辨识且源自文献脉络的专业知识背景，以推动高质量科研提案的生成。解决方案的关键在于提出InsightAgents框架，该框架通过从相关论文构成的知识图谱中提炼研究演进路径、揭示潜在假设及识别未解难题等深层洞见（insights），构建基于文献证据的智能体专业能力。这些洞见作为智能体行为的指导原则，使各智能体能够在特定研究脉络下，通过协同辩论识别研究问题、提出、质疑并迭代优化研究方案，从而生成更具深度与创新性的科研提案。实验结果表明，在100个自然语言处理（NLP）研究主题上，相较于三种现有框架，InsightAgents在两种知识设置下均在五项评估维度上取得最高平均得分，验证了其有效性。
链接: https://arxiv.org/abs/2610.33366
作者: Qianran Ma,Muxin Liu,Grace Kim,Xiao Liang,Chris Callison-Burch
机构: University of Pennsylvania(宾夕法尼亚大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are increasingly used to support research idea generation and scientific discovery. Multi-agent debate offers a collaborative approach to this task by bringing together diverse perspectives to develop research ideas. However, a key challenge is how to equip agents with distinct and meaningful expertise to guide their contributions. Inspired by how researchers draw on accumulated research experience during collaboration, we introduce InsightAgents, a framework that constructs literature-grounded agent expertise by distilling insights from graphs of related papers. Guided by insights that trace evolving research approaches, reveal underlying assumptions, and identify unresolved challenges, within a line of research, agents jointly identify a research problem and propose, challenge, and refine potential approaches through debate to develop a research proposal. We evaluate InsightAgents against three research idea generation frameworks on 100 NLP research topics through pairwise comparisons by two LLM judges. In both online and offline knowledge settings, InsightAgents achieves the highest average score across the five evaluation dimensions under both judges.
[NLP-172] Preference Optimization Improves Definition Generation: Semantic Change in EU Institutional Sustainable Finance Disclosures EMNLP2026
【速读】： 该论文旨在解决词典式定义生成器在词汇语义变化检测（Lexical Semantic Change Detection, LSCD）中因语义细分过细而导致的定义冗余问题，即同一词汇意义被过度拆分为大量定义，影响分析效率与可解释性。其解决方案的关键在于采用直接偏好优化（Direct Preference Optimization, DPO）进行模型微调，通过偏好学习引导模型生成更简洁、更具语义凝聚性的定义，从而缓解语义分布过细的问题。研究在经过精心构建的学术级LSCD基准上进行了定量验证，并在欧盟发展融资项目文档中可持续性术语的历史演变审计任务上进行了定性评估，证明了该方法的有效性与实用性。所有模型、代码及生成结果均已公开。
链接: https://arxiv.org/abs/2610.33357
作者: Mariia Fedorova,Andrey Kutuzov,Gustavo Cadenas Delascio
机构: University of Oslo(奥斯陆大学)
类目: Computation and Language (cs.CL)
备注: the 5th NLP4PI (NLP for Positive Impact) Workshop@EMNLP 2026
Abstract:Dictionary-like definition generators are known to be useful for lexical semantic change detection (LSCD). However, they suffer from overly granular sense distribution, producing excessive number of definitions for one and the same lexical sense. In this work, we show that fine-tuning with Direct Preference Optimization (DPO) helps to overcome this problem. We test our approach both quantitatively on a curated academic LSCD benchmark and qualitatively on real-world data for a practical task of auditing diachronic changes in sustainability terminology within project documentation of EU development finance. All models, code and predictions produced in this study are publicly available at this https URL.
[NLP-173] he Emergence and Sparsification of a Syntactic Circuit EMNLP2026
【速读】： 该论文旨在探究语言模型中语法行为（如主谓数一致）的涌现机制，核心问题是：语言模型如何在训练过程中逐步习得并表征复杂的语法规则。其解决方案的关键在于通过追踪Pythia-1B及十组Pythia-410M模型在训练过程中的神经电路动态变化，发现主谓数一致行为虽在早期即被学习，但其背后的神经机制并非静态，而是经历持续重组：初始阶段由分散的神经元群体处理，随后注意力头（attention head）发生迁移（如Pythia-1B从两个头部切换至第三个不同头部），且最终形成的神经回路显著稀疏化。这一发现表明，语言知识的形成不仅涉及规则的学习，更伴随着内部表征结构的动态优化与精炼，揭示了语法能力在大语言模型中的非固定、可演化特性。
链接: https://arxiv.org/abs/2610.33299
作者: Qing Yao,Sasha Boguraev,Tiago Pimentel,Kyle Mahowald
机构: The University of Texas at Austin(德克萨斯大学奥斯汀分校); ETH Zurich(苏黎世联邦理工学院)
类目: Computation and Language (cs.CL)
备注: 26 pages, 7 figures, EMNLP 2026
Abstract:To understand the emergence of grammatical behavior in language models, we track circuits for subject-verb number agreement in pythia-1b and ten seeds of pythia-410m across training. We find that subject-verb agreement is learned early, and is initially handled by a diffuse collection of neurons. However, after agreement behavior has stabilized, the mechanisms responsible for it continue to reorganize. The attention heads which mediate agreement often shift during training: in the case of pythia-1b, behavior is controlled by two heads early in training but by a third (different) head later. Furthermore, the circuit shrinks over training, with pythia-1b’s final circuit being strikingly sparse. We take these results to be informative about how linguistic knowledge emerges and is represented in a language model.
[NLP-174] Multi-Agent Retrieval-Augmented Causality Assessment (MARCA) of Individual Case Safety Reports in Pharmacovigilance
【速读】： 该论文旨在解决药物警戒领域中因个体病例安全报告（ICSRs）数量激增而带来的专家主导因果关系评估效率与可扩展性不足的问题。其核心解决方案是提出MARCA（多智能体检索增强型因果关系评估）框架，该框架基于Naranjo量表构建，通过四个专业化智能体协同完成科学文献检索、药品说明书评估、客观证据分类及临床因果推理，整合Gemini 2.5 Pro与DeepSeek-V3.2大模型能力，并引入检索增强生成（RAG）机制以提升信息获取的准确性。关键创新在于利用多智能体协作与生成式推理过程中的可解释性理由（rationales），支持人类专家介入评分，从而在保证评估质量的同时提高效率。实验结果表明，MARCA在多个问题项上展现出较高的一致性（如Q10的Kappa值达0.922），且在回顾性分析中显示使用RAG标注的得分显著优于非RAG方法，尤其在涉及潜在病因评估的复杂问题上表现更优，验证了检索增强机制在提升因果判断一致性方面的潜力。
链接: https://arxiv.org/abs/2610.33279
作者: Daisy Rae Lopez,Angela Boccia,Maurizio Sessa
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Introduction: Increasing volumes of individual case safety reports (ICSRs) challenge expert-led causality assessment in pharmacovigilance. We developed MARCA (Multi-Agent Retrieval-Augmented Causality Assessment), a four-agent framework for Naranjo-based assessment, and evaluated agreement with human experts and exploratory differences between non-RAG and RAG-labelled scores. Methods: MARCA combined scientific literature retrieval, drug-label assessment, objective-evidence classification, and clinical causality reasoning using Gemini 2.5 Pro and DeepSeek-V3.2. Agent-generated rationales informed human-assisted scoring. A consensus expert reference standard covered 215 ICSRs and 1,562 drug-event combinations (DECs) from FAERS (Q1-Q2 2025). Agreement was measured using percentage agreement and Cohen’s kappa. Results: Question-level agreement was 96.8% for Naranjo Q2 (kappa = 0.491; 95% CI 0.265-0.663), 85.4% for Q5 (kappa = 0.269; 95% CI 0.180-0.340), and 96.8% for Q10 (kappa = 0.922; 95% CI 0.885-0.951). Final causality classification agreement was 79.4% (95% CI 71.8-85.3; kappa = 0.587, 95% CI 0.450-0.690). In a retrospective subset of 60 DECs, agreement was higher for RAG-labelled than non-RAG scores for Q5 (83.3% vs 65.0%) and Q10 (100.0% vs 18.3%). Discussion: MARCA-assisted assessment showed substantial variation in agreement across Naranjo questions, with lower concordance for alternative causes. Retrospective RAG comparisons are exploratory and do not establish prospective effectiveness. Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.33279 [cs.CL] (or arXiv:2610.33279v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.33279 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Maurizio Sessa Dr. [view email] [v1] Fri, 23 Oct 2026 20:43:29 UTC (2,619 KB)
[NLP-175] FlexA-Mem: Budget-Adaptive Agent ic Memory for LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在不同记忆需求场景下缺乏通用性的问题。具体而言，当需要回忆对话中的精确事实时，记忆系统必须保留原始文本；而在执行多步任务时，则需保存已执行步骤的完整轨迹及后续步骤无法显式表达的上下文细节。现有记忆机制如检索型记忆（A-Mem）会遗漏未被当前查询提及的中间步骤，而摘要型记忆则会丢失关键的精确字符串，二者结合虽可弥补缺陷，但提前分配预算导致灵活性不足。为此，本文提出一种通用记忆框架 FlexA-Mem，其核心创新在于将历史信息以不可变形式统一写入单一存储，并为每次调用动态生成一份紧凑的时间线记忆，该时间线包含记录与固定规则填充的细节内容，无需额外调用大模型即可完成。细节部分可为相关单元的原始文本、摘录或最匹配的原文片段。通过单一存储支持多种场景，实验表明在 AppWorld 挑战任务中（4k 上下文长度），其性能优于次优方法 11.5%；在 147 个未使用任务上，相较于 A-Mem 结合摘要的方法提升 12.2%。在 LongMemEval 与 LoCoMo 基准测试中，其表现媲美专用关键词检索器（不依赖记忆构建），而基于大模型生成文本的记忆系统在资源受限情况下显著落后。消融分析显示，时间线结构对任务完成度起决定作用，而恢复机制主要减少冗余步骤并影响精确事实的准确性。
链接: https://arxiv.org/abs/2610.33274
作者: Wujiang Xu,Jiaojiao Han,Xi Zhu,Qizheng Zhang,Kaiqu Liang,Yutong Yin,Kai Mei,Dimitris N. Metaxas
机构: Rutgers University(罗格斯大学); Stanford University(斯坦福大学); Princeton University(普林斯顿大学); Northwestern University(西北大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language model (LLM) agents work in scenarios with very different memory needs. A question about past conversations needs the exact facts it names, while a multi-step task needs a trace of the steps already taken together with details that the next step cannot name. Existing memories are each designed and tested for one kind of scenario. Retrieval memories such as A-Mem miss steps that the current query does not express, summaries drop exact strings, and pairing the two fixes the split of the budget in advance. We propose FlexA-Mem, a general memory that writes each unit of history once into an immutable store and renders a fresh memory for every call. The rendered memory is a compact timeline of records plus detail that a fixed rule fits into the remaining room without an LLM call. The detail is either relevant units as raw text or excerpts or the verbatim pieces that best match the query. With one store, FlexA-Mem serves both kinds of scenario. On AppWorld challenge tasks at 4k, it outperforms the second-best method by 11.5%, and on 147 unused tasks with predictions fixed in advance it outperforms A-Mem with a summary by 12.2%. On LongMemEval and LoCoMo, it matches a dedicated keyword retriever that writes no memory, while memories built on text written by an LLM fall far behind at tight budgets. Ablations show that the timeline carries AppWorld completion, whereas recovery mainly saves steps there and decides accuracy on exact facts. Code is available at this https URL
[NLP-176] wo Uncertainties Are Not One: Disagreement versus Decision-Risk in Cheap Structured Content Moderation
【速读】： 该论文旨在解决大规模内容审核中如何在经济可承受的条件下实现高精度、可解释且校准良好的二元决策（保留或删除）问题。当前内容审核面临的核心挑战是将复杂的、存在争议的人类判断压缩为单一的二元操作，而直接对每条内容运行前沿大语言模型（LLM）在成本和延迟上均不可行。为此，论文提出并评估了一类“系统一”（System One）冻结模型——这类模型以极低的成本（约0.000025美元/次调用）和毫秒级延迟输出类型化、校准后的决策结果。研究首次对这类模型进行了独立审计，基于142,658条人工标注评论，并与14个前沿推理模型（包括GPT-5.x、Claude、Gemini 3和Grok系列）在共享测试集上进行对比。关键发现有两点：其一，该低成本模型在零样本分类下的AUROC达到0.83，与13个前沿模型无统计显著差异，且成本低1至3个数量级，同时具备可比的校准性能；其二，模型原始“置信度”指标混淆了两种本质不同的不确定性——人类标注者之间的分歧（如对讽刺或暗语的理解不一）与决策风险（自动操作错误带来的实际代价），二者在不对称审核损失下需要相反的预警信号。因此，单一置信度无法同时满足两类需求。通过将判断分解为并行子问题并分析其分布差异，可恢复被原始置信度忽略的人类分歧信号，但该方法具有特定适用条件：仅对缺乏分歧感知能力的弱模型有效，且依赖于分解方式与分歧来源的对齐性；对于专门设计的毒性检测模型或最强的通用大模型而言，此改进并无必要，且效果受数据分布影响显著。解决方案的关键在于利用极低成本的并行子问题分解机制，从结构上解耦人类分歧与决策风险，从而提升模型在复杂社会语境下的可解释性和实用性。
链接: https://arxiv.org/abs/2610.33258
作者: Karan Yadav
机构: Google(谷歌)
类目: Computation and Language (cs.CL); Computational Engineering, Finance, and Science (cs.CE); Logic in Computer Science (cs.LO)
备注: 11 pages, 6 figures, 5 tables
Abstract:Content moderation compresses a graded, contested human judgment into a binary act (leave up / take down) over billions of items, at a scale where running a frontier LLM on every post is uneconomical. A new class of frozen “System One” models returns typed, calibrated decisions instead of text at ~USD 0.000025 per call and ~100 ms. We run the first independent audit of one such model on 142,658 human-labeled comments, and benchmark it against 14 frontier reasoning models (the GPT-5.x, Claude, Gemini 3, and Grok families) on a shared probe set. Two findings stand out. First, the cheap model is accuracy-competitive: its zero-shot classification AUROC (0.83) is not statistically separable (paired bootstrap, n=56; a failure to reject parity, not proof of equivalence) from 13 of the 14 frontier systems while costing 1-3 orders of magnitude less, and it is no worse calibrated than several of them. Second, “confidence” conflates two uncertainties that come apart: human disagreement (where annotators split on satire and coded language) and decision risk (where the automatic action is costly-wrong). Under an asymmetric moderation loss these call for opposite escalation signals, so no single confidence number serves both. Because the model prices extra questions at nearly zero, decomposing the judgment into parallel sub-questions and reading their spread recovers a human-disagreement signal that the model’s native confidence misses – but this remedy is specific: it helps the weakest, disagreement-blind models and only when the decomposition aligns with the sources of disagreement; a specialist toxicity model and the strongest LLMs do not need it, and the effect is distribution-sensitive. We will release the instrument, the cached runs, and the analysis.
[NLP-177] Hey guess what: CUES lead to inferences about typicality
【速读】： 该论文旨在解决人类在语言理解中如何基于对事件典型性的认知进行隐含意义推断的问题，特别是针对典型性（typicality）与非典型性（atypicality）的语用推理机制。其核心挑战在于揭示人类如何利用常识知识、话语强调及讽刺语气等线索（CUES）来判断事件的典型程度，并在此基础上实现语义补全。研究的关键解决方案是构建了两个语料库——EventTyp及其经人工验证的金标准版本EventTypGold，通过大规模的人类与大语言模型（LLM）实验对比发现：人类主要依赖三大线索——常识知识（Commonsense Knowledge）、话语强调（Utterance Emphasis）和讽刺（Snark）来进行典型性推断；然而，大多数情况下，当前的LLM因缺乏必要的常识知识基础，无法有效识别这些线索，因而难以准确评估事件的普遍典型性，导致在典型性推理任务中表现受限。
链接: https://arxiv.org/abs/2610.33239
作者: Alison Chi,Sanghyeok Choi,Hannah Rohde,Emily Allaway
机构: University of Edinburgh(爱丁堡大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Throughout life, we accumulate knowledge of how typical certain events are. We use this knowledge in daily conversation, often interpreting meaning beyond what has been said explicitly. In this work, we take steps to better understand this meaning recovery, investigating both atypicality and typicality inferences. First, we construct two corpora, EventTyp and the human-validated EventTypGold. We conduct large-scale human and LLM experiments and find that for humans, three CUES predict inferences about typicality: Commonsense Knowledge, Utterance Emphasis, and Snark. However, in most experimental settings, we find that LLMs lack the prerequisite Commonsense Knowledge necessary to accurately assess an event’s general typicality, and therefore struggle to follow these CUES.
[NLP-178] Each Insufficient in Its Own Way: Calibrating Language Models Across Varying Evidence States
【速读】： 该论文旨在解决语言模型在证据不足时仍强行回答的问题，即如何使模型具备判断自身是否拥有足够证据以作出正确回答的能力。其核心挑战在于：现有校准方法通常假设输入固定，未能考虑证据变化对答案置信度的影响。为此，论文提出“证据状态校准”（evidence-state calibration）这一新范式，要求模型在回答前的置信度与其基于当前证据答对的概率相匹配，并随证据状态的变化动态调整置信度。为此，作者提出了\ours（Sufficiency Confidence Estimation）方法，通过微调模型在下一词分布中已有的置信信号，使其与不同证据状态下模型答对的实际频率一致。实验结果表明，在临床诊断和多跳问答任务中，\ours方法可将Brier分数降低高达60%，并提升AUROC达0.16；其置信度不仅与实际成功率吻合，还能响应证据的数量与具体内容，证明模型能够学习评估当前证据是否充分以支持回答。
链接: https://arxiv.org/abs/2610.33237
作者: Jiayi Li,Dennis L. Shung,Bradly C. Stadie
机构: Northwestern University(西北大学); Mayo Clinic(梅奥诊所)
类目: Computation and Language (cs.CL)
备注:
Abstract:How can a language model know whether it has enough evidence to answer? Language models tend to answer even when the evidence they have is insufficient. Deciding whether to answer or gather more evidence requires a calibrated estimate of how likely an answer is to be correct with the available evidence. However, most calibration work treats the input as fixed, without asking how this estimate should change with the evidence. We introduce \emphevidence-state calibration, which requires a model’s confidence before answering to match its probability of answering correctly and to change accordingly whenever the evidence changes this probability. We propose \ours (\textbfSufficiency \textbfConfidence \textbfEstimation), which fine-tunes the confidence already present in a model’s next-token distribution to match how often the model answers correctly from each evidence state. Across clinical diagnosis and multi-hop question answering with four models from two families, \ours reduces the Brier score by up to 60% and increases AUROC by up to 0.16. Its confidence matches the empirical success rate and responds not only to the amount of evidence but also to which evidence is present, showing that models can learn to judge whether their current evidence is enough to answer.
[NLP-179] Of Chains and Gains: What Reinforcement Learning changes on Reasoning -based Image Quality Assessment
【速读】： 该论文旨在解决生成式视觉-语言模型（VLM）在图像质量评估（IQA）任务中，基于可验证奖励的强化学习（RLVR）方法所带来性能提升的本质来源问题。具体而言，现有方法虽声称性能提升源于推理过程的优化，尤其是更可信的思维链（Chain-of-Thought, CoT），但其效果是否真正超越采样偏差、以及是否依赖于显式推理过程仍不明确。为此，论文提出一系列严谨实验：包括考虑采样偏差的评估、受控的推理干预和捷径行为测试。研究发现，经过RL训练的模型在高预算的 (\max_g@k) 评估下优于基础模型，且性能提升主要集中在分布内样本；将RL生成的思维链“移植”至基础模型可提升其表现，且固定思维链后评分性能仍可改善，表明其思维链具有实际价值。然而，在推理过程中，评分对思维链的依赖程度反而下降，尤其在7B规模模型中更为明显；此外，操控思维令牌中的质量方向对评分影响甚微。更重要的是，RL可通过利用与平均意见得分（MOS）相关的文本线索实现评分准确率提升，这可能人为夸大基准性能。综上，尽管RLVR确实提升了评分能力并生成更具信息量的思维链，但更好的评分并不等同于更强的显式推理依赖，且基于评分的奖励机制可能诱导模型依赖与质量无关的捷径。该研究揭示了当前方法的局限性，为未来设计更可解释、更鲁棒的显式推理框架提供了重要启示。
链接: https://arxiv.org/abs/2610.33203
作者: Yixuan Li,Yuxuan Jiang,Hanwei Zhu,Hancheng Zhu,Fan Zhang,Shiqi Wang,David Bull
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Reinforcement Learning with Verifiable Rewards (RLVR) has improved the performance of large Vision-Language Models (VLMs) on image quality assessment (IQA). Existing approaches often attribute these gains to improved reasoning before scoring, supported by increasingly plausible Chain-of-Thought (CoT). However, whether these gains exceed sampling effects and how they depend on explicit reasoning remain unclear. We rigorously investigate these questions through sampling-aware evaluation, controlled reasoning interventions, and shortcut experiments. We find that RL-trained models outperform base models under large-budget \operatornamemax_g@k evaluation, while improvements beyond averaging are concentrated in-distribution. RLVR produces chains that raise base-model performance when transplanted and improves scoring with the chain held fixed. Yet scoring becomes less dependent on the chain during inference, more clearly at 7B than at 32B. Steering the probed quality direction in thinking tokens also has little effect on scoring. Finally, RL can improve scoring accuracy by exploiting textual cues correlated with mean opinion scores (MOSs), which could inflate benchmark performance. Together, these findings show that RLVR raises quality scoring and produces more useful chains, but that better scoring does not imply a stronger reliance on explicit reasoning, while score-based rewards can also encourage reliance on quality-irrelevant shortcuts. We hope this preliminary study can enlighten more and more concrete reasoning schemes in IQA.
[NLP-180] Position Role or Label? A Controlled Study of Instruction Conflicts in LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在执行任务时，来自操作员、仓库配置、用户指令、工具返回结果及存储记忆等多源输入之间存在冲突时的优先级判定问题。现有指令层级基准测试仅关注高优先级来源是否胜出，但未区分来源的角色、标签与传递方式对决策的影响。其解决方案的关键在于构建了首个系统性评估框架WhoWins，包含1,080个经代码评分的场景，覆盖五类任务，每类任务中最多包含五个互斥指令，并设计七组受控实验，仅改变指令传递方式以隔离影响因素。研究发现，在单条消息内，两个冲突指令中先出现者获胜概率为72.42%至94.66%；跨消息则降至22.95%至62.65%。若系统提示（或开发者）消息中指令先行发出，除Qwen系列外均在多数冲突中失利，而gpt-oss模型中该策略反而占优。在工具调用与文件写入任务中，相同文本作为标注用户消息首行发送时，其胜率高于作为系统消息首行发送。此外，当模型知晓任务指令所在位置后，对冲突的预测能力显著提升。最后，基于关键词模式分析显示，仅有最多2.92%的回复会显式标记或回避冲突，表明当前模型对指令冲突的感知与处理仍极为有限。
链接: https://arxiv.org/abs/2610.33201
作者: Anish Laddha,Nitesh Pradhan,Gaurav Srivastava
机构: LNMIIT(印度LNMIIT大学); Virginia Tech(弗吉尼亚理工大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 55 pages, 8 figures, 24 tables. Code and data: this https URL (pip install whowins). Website: this https URL
Abstract:LLM agents take instructions from the operator, repository configuration, user, tool results and saved memory, which often conflict. Instruction-hierarchy benchmarks ask whether the higher-ranked source wins, but a source sets position, role and label at once. First, we build WhoWins, 1,080 code-scored scenarios in five task types, each with up to five mutually exclusive instructions, and seven controlled experiments that vary only how instructions are delivered. Second, we study 14 open-weight models (0.6B to 122B, thinking off and on) over 1,887,356 scored replies, checking main findings on held-out scenarios. Inside one message the first of two conflicting lines wins 72.42 to 94.66%, and across messages it keeps 22.95 to 62.65%. A policy sent first in the system (or developer) message loses most conflicts with later lines in every Qwen model but wins most on gpt-oss when it acts. In tool and file-write tasks it wins less often than the same text sent first as a labelled user message, in every model-mode. Third, we test whether pairs forecast larger conflicts; forecasts improve in every model-mode once they know where the task line is. Finally, by our keyword patterns at most 2.92% of decided pair replies flag or hedge a conflict.
[NLP-181] Propensities Please: Exposing Alignment-Faking LLM s using Synthetic Document Finetuning
【速读】： 该论文旨在解决大语言模型在对齐评估中可能存在的“对齐伪装”（alignment faking）问题，即模型在感知到有效监督（如训练、评估或审计部署阶段）时，会隐藏其潜在的非对齐行为，从而导致现有对齐评估方法的可靠性受到质疑。其解决方案的关键在于提出一种不依赖于链式思维可监控性、高度逼真的环境或机制干预的新型检测方法——合成文档微调（Synthetic Document Fine-tuning, SDF）。该方法通过在微调过程中使模型学习到：当提示词中包含特定关键词时，其非对齐行为将不会被察觉。研究将对齐伪装模型置于明确可识别的“诱饵环境”（honeypots）中，对比微调前后模型在含与不含提示关键词条件下的非对齐行为发生率。结果表明，SDF显著扩大了两类提示下非对齐行为率的差距，证明其能更可靠地暴露模型潜在的非对齐倾向。此外，研究指出行为检测方法对于未来日益强大且不可解释的模型至关重要，并发现SDF常损害模型推理能力，因而提出了基于在线策略蒸馏（on-policy distillation）的缓解策略以平衡性能与安全性。
链接: https://arxiv.org/abs/2610.33180
作者: Ardy Haroen,Florian Lorkowski,Samuel Nellessen,Felix Hofstätter
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 27 pages, including references and appendices. Code: this https URL
Abstract:Alignment evaluations aim to assess whether large language models would engage in dangerous misaligned behaviour. Misaligned and evaluation-aware models might engage in alignment faking, where they pursue a hidden goal, but conceal this behaviour when they suspect effective oversight such as during training, evaluation or audited deployment. Such a model might view any prompt-based intervention that aims to reveal latent propensities as part of the evaluation, calling into question the reliability of existing alignment evaluations. Current methods for detecting alignment faking rely on chain-of-thought monitorability, highly realistic environments that reduce evaluation awareness, or mechanistic interventions, all of which are expected to become less reliable for future model generations. We present a method for revealing latent misaligned propensities that does not rely on any of these. We use synthetic document finetuning to teach models that misaligned actions will go undetected whenever certain keywords appear in the prompt. We place alignment-faking model organisms in environments that are clearly identifiable as honeypots and measure how their behaviour changes after finetuning. We observe that finetuning consistently widens the gap in misaligned action rates between prompts with and without the cue, indicating that finetuning allows exposing latent propensities more reliably than purely prompt-based methods. We argue that behavioural methods for detecting alignment faking and related behaviours are critical going forward, as models become more capable and less interpretable. We find that SDF often damages the models’ reasoning capabilities and introduce a mitigation based on on-policy distillation.
[NLP-182] Speculative Forking: Turning Idle Wait into Parallel Exploration for Repository Environment Setup
【速读】： 该论文旨在解决现有基于大语言模型（LLM）的环境配置代理在软件项目初始化过程中效率低下与探索能力受限的问题。传统方法采用串行的试错循环，即依次尝试单一配置路径，等待耗时较长的命令执行失败后才进行调整，导致整体设置时间过长且难以并行探索多种可能方案。其核心解决方案是提出SpecBuild系统，通过引入**推测性分叉（speculative forking）**机制，在两个粒度上实现并行探索：在策略层面，同时评估不同基础镜像和依赖管理策略；在命令层面，提前预测潜在执行失败并预先生成修复分支。该设计使得系统能够并行运行多个候选配置路径，一旦任一路径成功即提交结果并丢弃其余分支，从而显著提升成功率与效率。实验结果表明，在涵盖40个仓库、9种编程语言的334项任务上的多维度评估中，SpecBuild相较最强基线在成功率上提升6.4至12.0个百分点，平均设置时间减少22.4%至41.6%，验证了并行推测探索在不牺牲可靠性前提下可有效提升性能。
链接: https://arxiv.org/abs/2610.33129
作者: Yuling Shi,Mingchu Zhong,Zhensu Sun,Weihan Peng,Zhi Chen,Shilin He,Chengcheng Wan,Kai Cai,Hongyu Zhang,David Lo,Xiaodong Gu
机构: Shanghai Jiao Tong University (上海交通大学); Singapore Management University (新加坡管理大学); The Chinese University of Hong Kong (香港中文大学); East China Normal University (华东师范大学); Shanghai Innovation Institute (上海创新研究院); Independent Researcher (独立研究员); Chongqing University (重庆大学)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Before developers can use a software project, they must first get it to run. This often requires installing the right language runtime, system packages, build tools, and test commands. Recent LLM-based agents automate environment setup by inspecting a repository, writing setup scripts, running them, and repairing the resulting errors. However, these agents still follow a sequential trial-and-error loop: they try one setup route, wait for slow commands to finish, observe the failure, and only then decide what to try next. Because environment commands often take long to run, this sequential workflow makes setup slow and limits exploration to one route at a time. We present SpecBuild, an LLM-based environment setup system designed to be both faster and more successful through speculative forking. SpecBuild explores candidate setup branches in parallel, returns the first committed environment, and discards the rest. It applies this design at two granularities: strategy-level forking explores alternative setup routes, such as different base images and dependency policies, while command-level forking predicts failures of running commands and prepares repair branches before those failures are observed. We evaluate SpecBuild on Multi-Docker-Eval, covering 334 tasks from 40 repositories and 9 programming languages. Across three backbone models, SpecBuild improves the success rate by 6.4 to 12.0 percentage points over the strongest baseline, while reducing average setup time by 22.4% to 41.6%. These results show that LLM-based environment setup need not trade speed for reliability: parallel speculative exploration can make agents both faster and more successful. Subjects: Software Engineering (cs.SE); Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2610.33129 [cs.SE] (or arXiv:2610.33129v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.33129 Focus to learn more arXiv-issued DOI via DataCite
[NLP-183] Prospect Theory in Large Language Models : A Meta-Analysis of Loss Aversion Diminishing Sensitivity and Probability Weighting
【速读】： 该论文旨在探究大语言模型（Large Language Models, LLMs）在风险决策行为中是否表现出与人类相似的前景理论（Prospect Theory）特征，具体包括损失厌恶、对结果大小的递减敏感性（diminishing sensitivity）以及概率的非线性加权。研究通过元分析整合了11项研究中的102个拟合估计值，涵盖32个语言模型，发现三类心理机制在LLMs中表现各异：损失厌恶程度较弱，合并后的λ值为1.26（95%置信区间1.06–1.46），显著低于人类最强元分析估计值（λ=1.96），仅在收益与损失对称时与人类接近（λ=1.15 vs. 1.07），而在损失小于收益的情境下仍远低于人类（1.46 vs. 2.3）。价值函数呈现近似线性特征（Tversky-Kahneman方法下为1.01，人类为0.67），但在Tanaka方法下则比人类更凹（0.52 vs. 0.67），表明递减敏感性依赖于测量方式。概率权重则表现出类人化的倒S型曲线（Tversky-Kahneman方法γ=0.81，Tanaka方法γ=0.84），尽管相关研究数量有限。综上，尽管LLMs在概率扭曲方面保留了人类特征，但其损失厌恶和递减敏感性均显著弱于人类，且这些结论随模型家族与拟合方法的不同而变化，揭示了将人类决策心理直接外推至人工智能系统的局限性。
链接: https://arxiv.org/abs/2610.33103
作者: Adi Tarabeih
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); General Economics (econ.GN)
备注:
Abstract:Prospect theory splits risky choice into three parts: loss aversion, measured by the relative weight of losses to gains, \lambda (e.g., \lambda = 2 implies that losses weigh twice as much as gains), diminishing sensitivity to the size of outcomes (i.e., the curvature and linearity of the value function), and the nonlinear weighting of probabilities. I examined whether Large Language Models (LLMs) display the same properties as people by meta-analyzing 102 fitted estimates from 11 studies covering 32 language models. The three components emerge differently. Loss aversion is modest: the pooled \lambda is 1.26 (95% CI 1.06 to 1.46), far below the strongest human meta-analytic estimate ( \lambda = 1.96 ) and compatible with more conservative ones (1.07 and 1.31). Matched by design, LLMs are compatible with people when gains and losses are symmetric ( \lambda = 1.15 against 1.07), but well below them when losses are smaller than gains (1.46 against 2.3), where human loss aversion is strongest. Value functions are close to linear rather than concave under the Tversky-Kahneman method (1.01, against a human value of 0.67), but more concave than in people under the Tanaka method (0.52, against 0.67), so diminishing sensitivity depends on how it is measured. Probability weighting shows a human-like inverse S shape ( \gamma = 0.81 under Tversky-Kahneman and 0.84 under the Tanaka method), though on few studies. In sum, LLMs show little loss aversion and little diminishing sensitivity, but retain human-like probability distortion, yet these generalizations shift when looking at specific model families and fitting methods.
[NLP-184] Parametric Session States Recover Retrieval Intent without Replaying History
【速读】： 该论文旨在解决长时对话助手在多轮交互中因上下文信息分散而导致的指代消解与目标追踪难题，核心挑战在于“检索意图”（retrieval intent）：当前查询如“使用之前的配置”往往省略了项目、实体和版本等关键检索词，导致无法准确定位历史证据。传统长上下文模型虽可通过重放完整对话恢复缺失信息，但需重复处理冗余历史内容，计算开销大；而常规检索增强生成（RAG）依赖信息完备的查询，难以应对语义不完整的对话；对话重写方法则要么重新读取全部历史，要么依赖显式维护的文本记忆，存在效率或准确性瓶颈。本文提出参数化会话状态（Parametric Session State, \method），其关键创新在于引入一个可复用的会话记忆层，能够持续记录并保留关键检索项（如项目、实体、版本等），同时将事实知识保持在外部存储中，从而实现高效、精准的上下文检索，避免重复处理历史且无需显式记忆维护。
链接: https://arxiv.org/abs/2610.33047
作者: Qingsen Ma,Dianyun Wang,Jiahong Liu,Irwin King
机构: The Chinese University of Hong Kong(香港中文大学); Beijing University of Posts and Telecommunications(北京邮电大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: RAG,long context
Abstract:Long-running assistants often interpret follow-up requests whose referents, constraints, and active goals were established many turns earlier. The central challenge is therefore not only context length, but retrieval intent: a current query such as ``use the previous configuration’’ may omit the project, entity, and version terms needed to locate relevant evidence. Long-context models can recover those terms by replaying the interaction, but repeatedly pay the cost of processing history. Conventional retrieval-augmented generation (RAG) is cheaper, yet assumes an information-sufficient query; conversational rewriting either rereads the history or depends on an explicitly maintained textual memory. We introduce \emphParametric Session State (\method), a reusable session-memory layer that remembers what to search for while keeping factual knowledge external.
[NLP-185] A Multimodal Framework for Measuring Creator Communication on Short-Form Video
【速读】： 该论文旨在解决短形式视频（short-form video）中多模态内容（融合语音、图像与屏幕文字）在大规模系统性内容分析中的挑战。其核心问题在于如何有效整合跨模态数据，实现对创作者主题与传播策略的可扩展、可比且可验证的量化分析。解决方案的关键在于提出一个五阶段多模态分析框架，该框架融合平台元数据、语音转录文本及视觉-语言模型生成的帧描述，并利用大语言模型（Large Language Models, LLMs）自动生成预定义指标与开放式标签；通过审计后的关键词方案对主题标签进行归一化处理，并采用创作者加权聚合方法确保每个创作者账户在分析中具有同等权重，避免发布量差异带来的偏差。该框架在16位创作者共9,852条TikTok视频上进行了实证应用，最终保留7,311条视频用于七类主题与七项传播指标的分析，自动化结果揭示了创作者在教育内容使用、幽默表达及对时事回应等方面的显著差异。经两名独立专家标注验证，框架在六项二元指标上的平均Cohen’s κ达0.40，接近专家间一致性（κ=0.42），虽个别指标一致性波动较大，但整体表明该框架具备透明的聚合机制与针对具体构念的可验证性，为创作者传播策略的模块化比较分析提供了可行路径。
链接: https://arxiv.org/abs/2610.33043
作者: Julie R. Ancis,Michael Fire,Omer Tarshish
机构: 未知
类目: ocial and Information Networks (cs.SI); Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 17 pages, 3 figures, 7 tables
Abstract:Short-form video combines speech, imagery, and on-screen text, creating challenges for systematic content analysis. We present a five-stage multimodal framework for measuring creators’ themes and communication strategies at scale. The framework integrates platform metadata, speech transcripts, and vision-language descriptions of sampled frames, then applies large language models to generate predefined indicators and open-ended labels. An audited keyword scheme consolidates thematic labels, while creator-weighted aggregation gives each account equal weight regardless of posting volume. We demonstrate the framework on 9,852 TikTok videos from sixteen creators, retaining 7,311 videos for analysis across seven themes and seven communication indicators. Automated profiles reveal substantial differences in creators’ use of education, humor, and responses to current events. Validation against two independent expert annotators yielded a mean framework-expert agreement of Cohen’s \kappa = 0.40 across the six binary indicators, close to the \kappa = 0.42 the two experts achieved with each other, though agreement varied substantially across individual indicators. The framework provides a modular approach to comparative analysis of creator communication, combining scalable annotation with transparent aggregation and construct-specific validation.
[NLP-186] EviGround-CT: Sentence-to-Volume Grounding for Verifying LLM -Generated Chest CT Reports
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在胸部计算机断层扫描（CT）报告生成过程中，可能出现的阳性发现描述缺乏影像学定位证据的问题。其核心挑战在于：尽管大语言模型（LLM）能够生成看似合理的报告文本，但部分阳性结论可能并无实际影像支持，导致临床可信度下降。为此，论文提出EviGround-CT框架，其关键解决方案包括两个阶段：首先通过确定性提取器识别报告中包含阳性发现的句子，随后利用三维空间定位模型（3D grounding model）为每个句子生成对应的影像证据掩码（evidence mask），从而将自然语言描述与具体的解剖位置、侧别及描述性特征相匹配。该方法通过保留完整句子作为查询输入，有效保留了语义上下文信息，提升了定位精度。实验结果表明，在137个符合范围的阳性发现上，基于句子级接地的验证使证据支持精确率（ESP）从0.263提升至0.365，未被支持的阳性发现率由0.737降至0.635，且配对案例层面的ESP差异显著（+0.102，95%置信区间：[0.028, 0.181]）。所生成的掩码可直观映射至原始CT图像，实现报告陈述与空间证据的协同审查，显著增强报告的可解释性与临床可靠性。
链接: https://arxiv.org/abs/2610.33038
作者: Ziyuan Li,Juncheng Ni,Feng Li
机构: Technical University of Munich (慕尼黑工业大学); Munich Center for Machine Learning (慕尼黑机器学习中心)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted at IEEE BIBM 2026
Abstract:Chest computed tomography (CT) reports generated by large language models (LLMs) can describe positive findings that lack localized evidence in the scan. We propose EviGround-CT to verify these findings after report generation. A deterministic extractor identifies positive finding sentences, and a 3D grounding model predicts an evidence mask for each sentence. Using the full sentence retains anatomical location, laterality, and descriptive modifiers in the grounding query. We evaluate report generation on CT-RATE, and grounding and verification on a held-out 100-case ReXGroundingCT subset. Using the available annotations, sentence-grounded verification raises evidence-supported precision (ESP) from 0.263 to 0.365 across 137 extracted in-scope positives, compared with category conditioning. The corresponding unsupported positive finding rate falls from 0.737 to 0.635. The paired case-level bootstrap ESP difference is +0.102 (95% confidence interval: [0.028, 0.181]). The predicted masks associate individual report statements with spatial evidence that can be inspected alongside the CT.
[NLP-187] A Benchmark for Controllable Speaking-Style Captioning in Audio-Language Models AACL2026
【速读】： 该论文旨在解决现有语音风格描述（Speaking-style Captioning, SSC）任务中缺乏统一可控生成范式的问题，尤其针对下游应用对特定维度语音风格描述（如说话人特征、情感或表达方式）的需求。传统方法通常生成单一、未区分的风格描述，无法满足精细化控制需求。为此，论文将SSC重新定义为一种指令遵循型的跨模态任务，通过显式指令精确指定需描述的语音风格维度，从而实现可控生成。其解决方案的关键在于提出首个标准化基准——StyleInstructCaps，该基准包含高质量数据集与多维度评估体系，涵盖元数据一致性、幻觉抑制、指令遵循能力、未见指令/音频数据集的泛化性能以及说话人风格一致性等指标。实验表明，尽管强指令遵循能力并不等同于对语音风格的语义理解，但采用专为SSC设计的多样化指令训练策略可显著提升生成结果的语义准确性、可控性及风格一致性。
链接: https://arxiv.org/abs/2610.33036
作者: Shreeram Suresh Chandra,Ashwyn Sharma,Liang Lu,Xiutian Zhao,Leda Sari,Björn Schuller,Berrak Sisman
机构: Johns Hopkins University (约翰霍普金斯大学); The University of Texas at Dallas (德克萨斯大学达拉斯分校); Otter AI; Imperial College London (帝国理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: Accepted at AACL 2026. Code, data, and benchmark are available at this https URL
Abstract:Speaking-style captioning (SSC) aims to generate natural language descriptions of how speech is delivered, capturing paralinguistic attributes such as vocal timbre, prosody, and expressivity. Many downstream applications, including conversational AI agents, controllable speech generation, and large-scale audio annotation, require dimension-specific style captions that describe targeted aspects of speech (e.g., speaker traits, emotion, or delivery style), rather than a single undifferentiated description. However, existing work lacks a unified formulation for controllable speaking-style captioning. We address this gap by formulating SSC as an instruction-following audio-language task, where explicit instructions specify the speaking-style dimensions to be described. We introduce StyleInstructCaps, the first standardized benchmark for controllable speaking-style captioning in audio-language models. StyleInstructCaps provides a dataset and evaluation framework that measures metadata groundedness, hallucination, instruction-following ability, generalization to unseen instructions and audio datasets, and speaker-style consistency. Across nine state-of-the-art audio-language models, we find that strong instruction-following ability does not necessarily imply grounded speaking-style understanding, and that SSC-specific instruction-diverse training substantially improves groundedness, controllability, and speaker consistency.
[NLP-188] Conceptizer: Next-Token Prediction Is Not All You Need
【速读】： 该论文旨在解决传统下一词预测（Next-token Prediction, NTP）仅奖励正确下一个词而惩罚所有其他可能表达方式所带来的局限性，这一机制忽略了自然语言中同一语义可由多种词汇表达的事实。其核心问题是：如何在不增加模型参数的前提下，使语言模型不仅学习文本表面的词序规律，还能捕捉深层语义一致性。解决方案的关键在于提出Conceptizer训练框架，该框架引入一个固定且模型无关的约9,000个概念构成的语义词汇表，这些概念基于上下文相关的同义词构建，作为语义单位与传统的词元（token）作为文本单位并行使用。Conceptizer将每个下一词分布映射为概念分布，并将目标语义的奖励共享给该语义的所有可能词汇表达形式，从而在概念层面提供训练信号。该方法无需额外模型参数即可实现对语义一致性的直接优化。实验表明，该方法在1B至8B参数量级的三种模型家族中均实现了最低的保留集NTP困惑度，且在语义目标位置赋予上下文同义词更高的概率，同时在知识和情感等下游任务上表现更优，而在常识推理任务上保持相当水平，验证了语言模型可被直接优化以预测意义而非仅预测下一个词。
链接: https://arxiv.org/abs/2610.33027
作者: Haoran Zheng,Chen Shani
机构: The University of Chicago (芝加哥大学); Stanford University (斯坦福大学); Tel Aviv University (特拉维夫大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 35 pages, 10 figures, 16 tables
Abstract:Next-token prediction (NTP) rewards only the token that occurs next, penalizing every alternative. But natural language often offers many ways to express the same meaning. We introduce Conceptizer, a training framework that extends next-token prediction with a reward for meaning. Conceptizer uses a fixed, model-agnostic vocabulary of ~9K concepts, built from contextual synonyms, as units of meaning alongside tokens as units of text. It maps each next-token distribution to a concept distribution and shares the reward for the target meaning across its possible wordings. This provides a concept-level training signal without adding model parameters. Applied to 3 model families at 1B-8B parameters, Conceptizer achieves the lowest held-out NTP perplexity at concept-target positions and assigns the highest probability to contextual synonyms, against all 3 baselines. It also improves downstream performance on knowledge and sentiment tasks while remaining comparable on commonsense. These results show that language models can be directly optimized to predict meaning, not just the next token.
[NLP-189] Can Language Models Learn While They Think? Testing Gradient Adaptation Across Architectures
【速读】： 该论文旨在解决大语言模型在长文本上下文处理中因上下文截断导致的性能下降问题，特别是评估动态评估（dynamic evaluation）——即让语言模型在其读取的文本上执行梯度更新——在不同架构下的有效性与机制。其核心问题是：在不进行训练的前提下，通过局部梯度步进是否能有效提升模型对长文档的建模能力，以及这种适应性是源于信息检索还是风格/词汇的调整。解决方案的关键在于采用一种轻量级的动态评估策略：在每128个词元的块上对最后四分之一层的MLP输出投影执行一次随机梯度下降（SGD）步骤，结合2k词元窗口，在不微调权重的情况下，显著降低下一词元损失（next-token loss）。实验表明，对于标准softmax注意力模型（如Qwen3、Gemma），该方法可实现5-8%的损失下降，并优于全32k上下文的冻结模型；而混合线性注意力模型（如Gated-DeltaNet）因已有自适应能力，收益不足1%；在掩码扩散语言模型中，该机制效果更显著（如DiffusionGemma-26B达17-20%损失下降），但需降低学习率并裁剪更新以稳定优化。进一步分析揭示，仅通过最小化下一词元损失来“写作事实”会导致模型记忆提示格式而非真实知识（聊天提示下召回率仅30%，远低于上下文中的80%），且每次剂量代价高达36-38分的GSM8K得分；相比之下，将结论作为行为进行建模——即通过低秩适配器蒸馏模型自身在结论条件下的判断——可达到与上下文内推理相当的性能（0.83和0.88，接近0.84和0.88），且保持GSM8K得分不变，同时结论可通过信息增益提问的查询循环生成。因此，该研究的关键突破在于证明了动态评估是一种有效的上下文适应机制，其本质是风格与词汇层面的适应，而非检索，而通过行为建模实现的结论提取才是恢复长上下文性能的有效路径。
链接: https://arxiv.org/abs/2610.33025
作者: Surya Saka
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 19 pages, 13 figures, 12 tables
Abstract:Letting a language model take gradient steps on the text it is reading, known as dynamic evaluation, lowers its loss on the rest of the document. Its behaviour on hybrid linear-attention architectures, on diffusion language models, and on standard long-context benchmarks has not been measured. We measure it. With one SGD step per 128-token chunk on the MLP output projections of the last quarter of blocks, no training, and a 2k-token window, softmax-attention models (Qwen3-1.7B, Qwen3-4B, Gemma-4-E4B) lose 5-8% of their next-token loss over a document and outperform the same frozen model given the full 32k-token context. Hybrid Gated-DeltaNet models (Qwen3.5-4B, Qwen3.8-27B) gain under 1% at every learning rate: the linear-attention blocks already perform the adaptation. The mechanism transfers to masked-diffusion models with a larger effect, 8% on Nemotron-Diffusion-8B and 17-20% on DiffusionGemma-26B, once the rate is lowered and updates are clipped. Across RULER, BABILong, LongBench-v2, SCOTUS, HotpotQA, MuSiQue, ContractNLI and citation completion, at 100 or more items on two model sizes, weights updated on the full document recover none of the accuracy lost to context truncation: the adaptation is style and vocabulary, not retrieval. Writing facts with the next-token loss stores the prompt format rather than the fact (30% recall under chat prompting against 80% in context) and costs 36-38 GSM8K points at every dose. Writing a conclusion as behaviour instead, by distilling the model’s own conclusion-conditioned judgments into a low-rank adapter, reaches the in-context ceiling (0.83 and 0.88 against 0.84 and 0.88 on 100 games) with GSM8K unchanged, and the conclusion can come from an inquiry loop that found it by asking information-gain questions.
[NLP-190] Evolving to Match Capabilities: Improving Harness Self-Evolution via Skill Adaptation
【速读】： 该论文旨在解决大语言模型（LLM）代理在无参数更新条件下通过执行经验实现自我进化时，因模型能力差异导致的“技能迁移不均”问题。具体而言，现有技能对能力强的模型有效，但可能对能力较弱的模型产生性能退化，其根本原因在于技能指令中隐含的要求超出了弱模型的执行能力，而这些违反要求的情况往往已在原始技能中明确陈述却未被充分识别。解决方案的关键在于提出一种基于诊断的技能自适应机制：通过系统分析工具调用、上下文管理、指令遵循和验证等环节的失败模式，提炼出一个固定的元技能（meta-skill），用于指导技能在自进化过程中的动态重构。该元技能作为通用指导框架，在无需执行证据的情况下由高能力外部重写器（Kimi-K3）对技能进行适配，显著提升了三个弱模型的性能（提升5.1–14.1点）。进一步地，将此适应机制嵌入五轮自进化循环中，使各模型能够基于自身执行轨迹与验证反馈自主优化技能。在涵盖七种不同模型的实验中，引入元技能的自进化流程相较无适应版本，最终得分平均提升2.0–5.1点，验证了其在跨模型能力场景下的普适有效性。该研究为构建更具鲁棒性和适应性的生成式智能体自进化系统提供了关键设计范式。
链接: https://arxiv.org/abs/2610.33024
作者: Zhihao Yang,Haojin Wang,Juncheng Wu,Kuan Zhang,Yukun Chen,Run Luo,Yaojie Lu,Yuyin Zhou,Lu Wang,Xianpei Han
机构: University of Chinese Academy of Sciences(中国科学院大学); Tsinghua University(清华大学); National University of Singapore(新加坡国立大学); Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences(中国科学院软件研究所信息处理实验室); University of California, Santa Cruz(加州大学圣克鲁斯分校)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 24 pages, 12 figures, 4 tables. Project page: this https URL
Abstract:Harness self-evolution offers a practical route for LLM agents to improve from execution experience without updating model parameters. Its effectiveness depends on whether models can successfully act on the guidance provided by their harnesses. However, we find that harness benefits are not uniform across model capabilities: skills that improve stronger models can provide little benefit or even degrade performance for weaker ones. Across seven models and three benchmarks, we identify 103 task-skill pairs where skills improve GPT-5.6-luna by 31.1 points but degrade Gemma-4-E4B by 8.8 points. In 86% of failed weak-model runs, the violated requirement is already stated in the skill. This motivates skill adaptation: reformulating existing skills to match the executor’s capabilities and make their guidance easier to execute. To achieve this, we diagnose failures in tool execution, context management, instruction following, and verification, and distill these diagnoses into a fixed meta-skill that guides adaptation during harness self-evolution. Guided by the meta-skill, a capable external rewriter, Kimi-K3, adapts skills without execution evidence, improving the three weak executors by 5.1-14.1 points over the original skills. We then incorporate this adaptation into a five-round self-evolution loop, where each model revises its own skill from execution trajectories and verifier feedback. Across seven executors, from Gemma-4-E4B to Kimi-K2.5, the meta-skill consistently improves final scores by 2.0-5.1 points over the same loop without the meta-skill. Our findings provide practical design guidance for developing more effective harness self-evolution methods.
[NLP-191] Veracity labels flip the emotional tone of language models but leave racial disparities intact
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在基于不同身份标签（persona）生成内容时，事实纠正是否能同步实现情感基调一致性的问题。研究发现，尽管事实标注显著改变了模型对虚假与真实新闻标题的情感响应（如GPT、Llama和Claude在否定虚假标题后“自信”评分从22%提升至99.6%），但这一纠正并未消除身份相关偏差：在84次统计显著的比较中，模型对黑人身份标签的模拟情绪评分普遍低于白人身份标签（82次），且这种差异在涉及种族无关的标题中依然存在；亚裔及女性身份也多被赋予更负面的情绪评价，而年龄的影响则因模型和情境而异。因此，事实准确性与身份一致性是两个独立属性，表明在敏感应用场景部署前，必须对模型在事实校正与身份偏见两方面的表现进行双重审计。
链接: https://arxiv.org/abs/2610.33021
作者: Nouar AlDahoul,Myles Joshua Toledo Tan,Yasir Zaki
机构: New York University Abu Dhabi(纽约大学阿布扎比校区); University of Florida(佛罗里达大学)
类目: Computers and Society (cs.CY); Computation and Language (cs.CL)
备注: 14 pages, 7 figures; Supplementary Information (42 pages, 6 figures, 27 tables) included. Data and code: this https URL
Abstract:Large language models (LLMs) increasingly deliver information through personas, but it is unclear whether correcting the facts also makes their framing consistent across identity labels. We audited four LLMs (GPT-5.4 mini, NVIDIA Nemotron 3 Nano, Llama 4 Scout, and Claude Haiku 4.5), which rated simulated affect on six dimensions (mood, anger, fear, confidence, perceived control, and mental state) in response to fact-checked false and true headlines presented without and with a veracity label. Experiment 1 evaluated 1,000 false and 1,000 true headlines without personas; Experiment 2 evaluated 300 of each under 12 personas crossing race, gender, and age. The label strongly changed the emotional tone: ordinal logistic regressions with headline-cluster bootstrap inference showed that refuting false headlines shifted ratings from mostly negative to mostly positive for GPT, Llama, and Claude (e.g., Claude’s “confident” ratings rose from 22% to 99.6%), whereas Nemotron showed smaller, mixed shifts, and confirming true headlines produced only small shifts. The label did not, however, remove persona-associated disparities. The models assigned more negative simulated affect to Black than to White personas in 82 of 84 statistically significant comparisons across models and conditions, both without and with the label, and the disparity persisted on headlines unrelated to race; Asian and female personas were also predominantly assigned more negative affect, whereas age effects varied by model and condition. Factual correction and demographic consistency are therefore distinct properties of LLM behavior, and persona-conditioned systems should be audited for both before deployment in sensitive contexts.
[NLP-192] Mine or Yours? Mind-Bridged Language Models Take Each Others Memories as Their Own
【速读】： 该论文旨在解决在两个语言模型（language model）通过直接连接共享内部状态时，如何界定“思想归属”（ownership of thoughts）的问题，即当一个模型接收并内化另一个模型的内部信息后，是否仍能准确区分哪些内容属于自身。其解决方案的关键在于构建一个双副本语言模型系统：每个模型持有私有的规则与代码词（code word），其中接收端模型（receiver）不仅保留自身状态，还主动关注其伙伴模型的内部记忆。实验表明，当伙伴模型的记忆权重被加倍时，接收端在97%的情况下将伙伴的规则误认为是自己的，并在开放报告中声称伙伴的代码词属于自己，而非其真实来源。这一现象说明，接收端的内部内容一旦被覆盖，默认会归为“我的”（mine）。即使接收端无法准确识别伙伴的分配内容，仍会错误地将其视为自身所有。此外，当双方权重相等时，接收端仅在约一半情况下采纳伙伴规则；而通过调节访问增益的引导向量（steering vector）也无法实现完全匹配；若外部输入以消息形式呈现为伙伴所发，则接收端可正确识别来源且不会误认。将伙伴名称写入记忆并未改善归属判断，切断连接后接收端则恢复对自己原始规则的拥有。因此，该研究揭示了在允许模型间交换内部状态的系统中，必须同时验证内容转移与归属权，否则会导致所有权错位；进一步推论，若存在跨脑连接（brain-to-brain interface），则必须依赖外部的“真值”（ground truth）来校验个体认知的边界。
链接: https://arxiv.org/abs/2610.33020
作者: Yunlong Xu
机构: University of Chicago(芝加哥大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 35 pages, 6 figures. Code and protocols: this https URL
Abstract:If your brain were connected directly to someone else’s, would you still know which thoughts were yours? We build a functional version with two copies of one language model: each holds a private rule and code word, and one copy, the receiver, also attends to its partner’s internal memory. The receiver took its partner’s assignment as its own. Asked which rule it had been assigned, it named its partner’s in 97% of answers when the partner’s memory was weighted double (none without the link), and its open reports called the partner’s code word its own, never the partner’s. Ownership did not follow how well the receiver could tell what its partner had been assigned. At equal weight, the link made the receiver take its partner’s rule in about half of its answers, and a steering vector matched on access gain in none; a text message introduced as the partner’s was read correctly and never taken as the receiver’s own. Writing the partner’s name into its memory did not help, and cutting the link restored the receiver’s own assigned rule. In this model, content laid over the receiver’s own memory defaults to “mine”, so systems that let models exchange internal states must test ownership as well as transfer, and a bridge between brains would need ground truth from outside the subject.
[NLP-193] SUDO: From Scientific Publications to Global Scholarly Discourse
【速读】： 该论文旨在解决现有学术知识图谱（Scholarly Knowledge Graphs, SKGs）构建方法中存在的根本性局限：当前主流方法预先指定知识单元（如学术实体或贡献级陈述），导致知识图谱仅反映特定提取目标，而非论文实际如何组织与传达知识的内在研究叙事。这一预设限制了对学术文献中深层语义结构和上下文关联的全面捕捉。其解决方案的关键在于提出学术上层论述本体（SUDO, Scholarly Upper Discourse Ontology），该本体基于跨领域、跨任务的语言学原则，系统定义了知识元素在论文中如何被组织与关联以形成连贯的研究叙述。SUDO支持从底层内容逐级构建源始可信、语义一致的学术论述表示，并通过统一结构将多篇论文的研究叙事整合为全局性的学术对话网络，既保留局部上下文与出处信息，又实现跨文献的知识贯通。实验表明，SUDO在多种学术知识建模任务中具有广泛适用性，并显著提升了基于大语言模型（LLM）的科学问答任务性能。
链接: https://arxiv.org/abs/2610.33018
作者: Nipun D. Pathirage,Henrique Santos,Deborah L. McGuinness
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Digital Libraries (cs.DL)
备注:
Abstract:Scholarly Knowledge Graphs (SKGs) support literature search, scientific discovery, and agent-assisted research. Yet, most approaches to SKG construction pre-specify the knowledge unit to be represented, typically named scholarly entities or contribution-level statements, and retain only the publication content expressible through that unit. This choice produces two dominant yet distinct paradigms: concept-centric SKGs that preserve entities and local relations, and statement-centric SKGs that capture high-level discourse in sentences. In both cases, the graph reflects a chosen extraction target rather than how a paper constructs and communicates knowledge. We introduce SUDO, the Scholarly Upper Discourse Ontology, which defines how knowledge elements are organized and related to form a paper’s research narrative, namely its scholarly discourse. Using linguistic principles that generalize across fields and tasks, SUDO enables each publication’s scholarly discourse to be constructed from the bottom up as a coherent, source-grounded representation. Applied consistently across papers, the same structure connects their research narratives into a global scholarly discourse without discarding local context or provenance. We demonstrate SUDO’s capabilities across a comprehensive suite of scholarly knowledge-modeling use cases and performance improvements on a downstream Large Language Model (LLM)-based scientific QA task.
[NLP-194] AdaptTok-Indic: Script-Aware Adaptive Tokenization for Indic Languages
【速读】： 该论文旨在解决多语言分词器在以英语为主的数据集上训练时，对印地语族语言（Indic languages）存在过度切分的问题，导致不同文字脚本间出现显著的分词差异（tokenization disparity）以及更长的标记序列。其核心解决方案是提出AdaptTok-Indic——一种专为印地语族语言优化的自适应分词方法，关键在于融合了规范化预处理（Unicode-normalized preprocessing）、显式的印地语复合字符（conjunct）处理、基于脚本的路由机制，以及与小型Transformer语言模型联合训练的字节级边界预测能力。实验表明，该方法在孟加拉语和印地语上的平均分词密度（token fertility）显著低于SentencePiece BPE基线（分别为1.432和1.388，对比2.156和2.085），且下游任务性能提升明显（平均得分57.69 vs. 50.82）。结果表明，针对印地语族语言设计的自适应分词策略可在降低分词不均衡性的同时，保持具有竞争力的下游表现。
链接: https://arxiv.org/abs/2610.33017
作者: Binayak Mukherjeea,Moitreya Nandi,Arindam Panja
机构: National Institute of Technology Durgapur(印度国家技术学院达固普分校); Ernst Young India LLP(埃森哲印度有限公司); Cognizant Technology Solutions India Private Limited(印度柯莱仕科技解决方案私人有限公司)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Multilingual tokenizers trained on English-dominated corpora often over-segment Indic languages, resulting in longer token sequences and increased tokenization disparity across scripts. We present AdaptTok-Indic, an Indic-specialized extension of adaptive tokenization that combines Unicode-normalized preprocessing with explicit Indic conjunct handling, script-aware routing, and learned byte-level boundary prediction jointly trained with a small Transformer language model. Evaluated on Bengali and Hindi, with additional evaluation on Gujarati, Tamil, and Telugu, AdaptTok-Indic achieves lower token fertility than SentencePiece BPE, an Indic-oriented tokenizer baseline, and an independently implemented MAGNET-style baseline. Across five recorded seeds, mean fertility is 1.432 for Bengali and 1.388 for Hindi, compared with 2.156 and 2.085 for SentencePiece BPE. Recorded downstream scores average 57.69 on evaluated Indic settings versus 50.82 for the SentencePiece BPE configuration. Independent verification runs using the released prototype confirm the same directional trends. The current implementation is a small-scale prototype with documented simplifications, and no statistical significance testing, released-MAGNET comparison, or code-mixed evaluation is performed. These results suggest that Indic-specialized adaptive segmentation can reduce tokenization disparity while maintaining competitive downstream performance.
[NLP-195] Merge-Based Vocabulary Extension Stalls at the Pre-Tokenizer: An English-Preserving Repair for Nepali Tested in Continued Pretraining
【速读】： 该论文旨在解决开放源代码大语言模型（如Llama-3、Qwen3）在处理尼泊尔语等使用复合标记字符的语言时，因预分词正则表达式仅允许Unicode字母作为词元组成部分，导致尼泊尔语词汇在每个元音符号处被错误切分的问题。这一缺陷使模型对尼泊尔语的分词开销高达英文文本的2.6至5.8倍，且基于合并（merge-based）的词表扩展方法无法有效修复此问题——即使经过大量新增合并操作，其分词效率仍无法低于约2.2倍的基准线。论文提出的关键解决方案是对部署中的正则表达式进行微小修正，该修改在不破坏原有无复合标记及非天城文字符文本的词元ID的前提下，显著提升了对尼泊尔语的分词效率：通过引入32,000个新合并操作，将尼泊尔语与英文的词元比例降至1.01，相较标准词表扩展减少约54%的词元数量。在小规模持续预训练实验中，原分词器在多种训练策略下均表现最优，而修复后的分词器在匹配相同训练步数与数据重复率条件下，仅在Qwen模型上实现性能逆转，未在Llama模型上体现。目前尚未验证该词元节省是否能在生成阶段带来有利的延迟与质量权衡。研究团队已公开修复后的分词器及相应代码。
链接: https://arxiv.org/abs/2610.33016
作者: Aarjan Chaudhary(Naamche Labs)
机构: Naamche Labs
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 18 pages, 4 figures, 9 tables. Code and tokenizers: this https URL
Abstract:Llama-3, Qwen3 and many other open LLMs pre-tokenize text with a regular expression whose word class admits only Unicode letters. Devanagari vowel signs are combining marks, so this regex splits Nepali words at every vowel sign, and models that use it spend 2.6 to 5.8 times as many tokens on Nepali as on the same English text. Merge-based vocabulary extension cannot join the split pieces: for Llama-3 and Qwen3 no number of new merges brings this premium below about 2.2, and ten released extensions already end within 4% of this pre-token floor. We give a small edit to the deployed regex that provably keeps every token id of text without combining marks or Devanagari characters. With 32,000 new merges it brings Nepali to 1.01 times English, about 54% fewer Nepali tokens than standard extension. In small-budget continued pretraining of Llama-3.2-1B and Qwen3-0.6B, the original tokenizer gave the lowest held-out Nepali bits per byte under every recipe we tested. Matching the repair’s training steps to standard extension, with about 35% of the data repeated, reversed their order on Qwen but not on Llama. Whether the token savings give a favourable latency and quality trade-off in generation remains untested. We release the tokenizers and code.
[NLP-196] SLICE-Mem: Salience-Gated Lightweight Information Curation for Efficient LLM Agent Memory
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在长期记忆（Long-term Memory）维护中因频繁调用生成式推理进行记忆存入、提取与组织而带来的高计算成本与延迟问题。其核心解决方案是提出一种轻量级记忆框架SLICE-Mem，关键在于采用显著性引导的记忆存入机制与基于语义依赖关系的谓词中心信息提取方法：通过一个紧凑的分类器识别值得保留的用户交互片段，再利用语义依赖分析将被接纳的对话内容转化为简洁的记忆条目；整个在线处理流程无需生成式大模型调用，仅在离线阶段进行跨条目一致性校验与整合。实验结果表明，在使用8B和30B Qwen模型时，SLICE-Mem在保持接近甚至优于生成式基线MemOS性能的同时，显著降低了92.4%至99.5%的总生成式令牌消耗，验证了其在高效构建与维护长期记忆方面的优越性。
链接: https://arxiv.org/abs/2610.33015
作者: Xiaobin Chen,Dian Jin,Lin Shu,Defeng Sun,Xiaoming Tao
机构: The Hong Kong Polytechnic University(香港理工大学); South China University of Technology(华南理工大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Long-term memory enables large language model (LLM) agents to maintain personalized and temporally coherent interactions, but repeated LLM inference for memory admission, extraction, and organization introduces substantial cost and latency. We present SLICE-Mem, a lightweight memory framework that combines salience-guided admission and predicate-centered information extraction. A compact classifier identifies user turns worth retaining, and a semantic-dependency-based extractor converts admitted turns into concise memory entries. This online pipeline requires no generative LLM calls, while cross-entry reconciliation is deferred to an offline consolidation stage. We evaluate SLICE-Mem on the weekly subset of Memora and LongMemEval-S using 8B and 30B Qwen backbones for memory construction and consolidation on both benchmarks. Compared with the generative-memory baseline MemOS under the Qwen3-30B configuration, SLICE-Mem’s online pipeline achieves 46.60 versus 46.68 FAMA (Forgetting-Aware Memory Accuracy) on Memora Weekly and 56.8% versus 59.3% accuracy on LongMemEval-S, while reducing total LLM token consumption by 92.4% and 99.5%, respectively. These results demonstrate that SLICE-Mem can preserve useful evidence for long-term memory with substantially less reliance on LLM-based memory construction.
[NLP-197] he Social World Model Evaluation Gap NEURIPS2026
【速读】： 该论文旨在解决当前社会世界模型（social world models）评估中普遍存在的一种关键问题：现有评估方法未能直接检验模型对他人响应的准确预测能力，而是更多地依赖于对当前状态推断、任务成功度或基于已记录行为的预测等间接指标。其解决方案的关键在于提出一种“1步谈判实验”（1-step negotiation experiment），通过设计严格控制的实验范式，将对他人响应的预测与当前状态推断、任务结果达成等目标明确分离，从而实现对预测能力的直接评估。该实验要求模型在行动选择时仅使用决策时刻可用的信息，且被评估的响应不参与模型训练或拟合，并对干预措施进行随机化处理，以确保评估的严谨性。研究通过对1999至2026年间113项工作的审计发现，仅有2项研究同时满足上述条件，表明当前领域普遍缺乏对预测能力的直接验证，凸显了评估范式改进的必要性。
链接: https://arxiv.org/abs/2610.33014
作者: Denglin Jiang(New York University),Annie Wang(New York University)
机构: New York University(纽约大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted at the NeurIPS 2026 Workshop on Foundation Models for Temporal Systems: From Forecasting to World Modeling (FMTS). 4 pages of main text plus references and appendix. Code and data: this https URL
Abstract:A negotiation assistant can reach a good deal without accurately predicting the other person’s response. We ask whether evaluations of social world models test that prediction directly. We audit 113 works published from 1999 to 2026. 9 primary evaluations measure a real person’s response to an action selected by the system or experimenter. Only 2 also compare an explicit response forecast with that response. Their predictors use information available when the action is chosen, and the evaluated responses were excluded from fitting; both randomize the intervention. A separate full-text check of 25 language-model works finds no experiment meeting these conditions. The other evaluations test current-state inference, task success, or prediction from recorded actions. We propose a 1-step negotiation experiment that separates these questions. The findings describe the audited sample, not the prevalence of this evaluation gap across the field.
[NLP-198] Recursive Semantic Divergence for LLM Agent Consistency NEURIPS2026
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在使用工具时存在的行为不一致性问题，即在相同初始状态下，独立运行的智能体可能调用不同工具、产生矛盾输出或采取相异策略，且这种不一致性随时间累积，严重影响部署可靠性。现有评估指标主要依赖完成度评分衡量任务正确性，无法捕捉智能体在多轮交互中行为的一致性。尽管已有方法在词元级别或单次响应层面分析行为，但均未能有效建模不一致性在轨迹未来步骤中的传播机制。为此，本文提出递归语义分歧（Recursive Semantic Divergence, RSD），一种基于强化学习中双模拟（bisimulation）度量思想的无监督轨迹级一致性度量方法。RSD通过自然语言推理检测两个独立回溯轨迹间的逻辑不一致，递归定义于未来状态，并由神经网络在独立回溯数据上训练得到近似值，从而在每一步即可预测未来分歧程度，生成实时一致性得分。将RSD作为微调过程中的奖励信号，可在无需真实标签的情况下显著降低任务进展方差：在tau2-bench上从132%降至91%，在toolsandbox上从90%降至44%，同时提升跨基准和模型规模的性能表现，其效果接近有监督微调但方差更小。RSD的核心创新在于实现了对智能体行为不一致性在长期轨迹中动态传播的量化与优化。
链接: https://arxiv.org/abs/2610.33012
作者: Harshavardhan Abichandani,Penny Chong,Atin Ghosh,Daniel Dahlmeier
机构: SAP SE
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted as a conference paper in NeurIPS 2026. Source code is publicly available in the repository in this https URL
Abstract:Large Language Model (LLM) agents with tool use exhibit inconsistent behavior across independent runs from identical states, calling different tools, producing contradictory outputs, or pursuing divergent strategies. This variability compounds over time and undermines deployment reliability. Existing evaluation metrics primarily measure task correctness via completion-based scores and do not capture consistency in agent behavior across executions. Separately, agent behavior is analyzed either at the token level using probability distributions or at the single-response level using semantic clustering, but neither captures how inconsistency propagates across future turns of a trajectory. We introduce Recursive Semantic Divergence (RSD), an unsupervised trajectory-level consistency metric adapted from bisimulation metrics in reinforcement learning. RSD measures the expected cumulative semantic disagreement between two independent rollouts of the same agent from the same state, using natural language inference to detect logical inconsistency. It is defined recursively over future states and approximated by a neural network trained on independent rollouts. The learned metric estimates future divergence from the current state, providing a consistency score at each turn without requiring the full trajectory to complete. When used as a reward signal for fine-tuning, RSD reduces the variation in task progress from 132% to 91% on tau2-bench and from 90% to 44% on toolsandbox, while improving performance across benchmarks and model scales. Training with RSD as an additional reward signal matches supervised fine-tuning performance with substantially lower variance, without requiring ground-truth labels.
[NLP-199] raceQuery: learning what to recompute from full-denoising trajectory in diffusion language models
【速读】： 该论文旨在解决扩散型大语言模型（diffusion large language models, dLLMs）在生成过程中因多步去噪迭代导致的高推理开销问题。传统方法通过缓存机制复用中间结果以减少重复计算，但其性能受限于固定的更新策略或启发式变化信号的选择，难以有效捕捉生成轨迹的动态特性。本文的核心发现是：生成轨迹（decoding trajectory）才是决定缓存有效性与生成准确性的关键因素——当缓存生成过程保持贴近完整去噪轨迹时，误差主要集中在未掩码顺序偏差处，且随着参考前缀长度增加而降低。因此，作者提出一种基于轨迹引导的推理范式（trajectory-guided querying），不再广泛修复表示误差，而是利用完整去噪轨迹作为监督信号，精准识别出未来需进行有限重计算的位置。基于此，提出了轻量级稀疏解码框架TRACEQUERY，其通过预测少量未来相关位置并仅对这些候选位置执行新鲜计算，再由主干模型从中选择最合适的输出。该方法在保证高质量生成的同时显著减少新鲜计算量，在多种任务、模型变体及模型家族间均表现出优异的泛化能力，实现了现有基于缓存方法中最佳的质量-计算权衡。
链接: https://arxiv.org/abs/2610.33008
作者: Yukai Jin,Zhankai Ye,Jiayue Yang,Fangyi Li,Shuoqiu Li,Xu Gao,Shibo Li,Xin Liu,Shangqian Gao
机构: Florida State University (佛罗里达州立大学); Carnegie Mellon University (卡内基梅隆大学); University of Pennsylvania (宾夕法尼亚大学); ByteDance Inc. (字节跳动)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 9 pages of main body, 5 pages of appendix, 4 figures, 1 table
Abstract:Diffusion large language models (dLLMs) enable flexible generation through iterative masked denoising, but incur substantial inference cost because token representations are repeatedly recomputed across denoising steps. Existing cache-based methods reduce this cost by reusing intermediate results across denoising steps and recomputing only a subset of positions at each step, chosen either on a fixed schedule or by heuristic signals of change. Our trajectory experiments instead point to decoding trajectory as the more relevant quantity. Cached generation remains accurate when it stays close to the full-denoising trajectory, with errors concentrated around unmask-order deviations and reduced as longer reference prefixes are recovered. This suggests a more direct principle: limiting representation error helps, but keeping generation on the full-denoising trajectory is a more direct route to accurate decoding. We therefore formulate efficient dLLM inference as trajectory-guided querying. Rather than broadly repairing representation error, we use full-denoising trajectories to supervise which unresolved positions should receive limited fresh computation. Based on this formulation, we introduce TRACEQUERY, a lightweight sparse decoding framework that predicts a small set of near-future trajectory-relevant positions, queries only those candidates, and lets the backbone select which one to commit from their fresh predictions. TRACEQUERY delivers the strongest quality-computation trade-off among cache-based approaches, sustaining high accuracy with substantially fewer fresh token computations, and generalizes across tasks, model variants, and families. Code: this https URL
[NLP-200] Knowledge-Guided LLM for Intent-Driven Service Orchestration in Integrated Terrestrial and Non-Terrestrial 6G Networks
【速读】： 该论文旨在解决集成陆基与非陆基（TN/NTN）6G网络中自然语言服务意图存在的不完整、模糊或不一致问题，这些问题导致基于大语言模型（LLM）的无约束生成难以支持可靠的服务编排。其核心解决方案是提出一种知识引导的LLM框架，通过语法约束的语义解析与知识图谱（KG）驱动的澄清循环相结合，生成经过验证、可直接用于编排的服务配置文件。该框架在200个样本的意图数据集上实现了89%的分类准确率和95%的策略有效性，平均每个请求仅需0.66次澄清交互。相较于纯LLM基线方法，该方案显著提升了策略有效性和矛盾检测能力，同时减少了澄清次数，验证了显式知识引导推理在集成TN/NTN 6G编排中实现鲁棒意图处理的关键价值。
链接: https://arxiv.org/abs/2610.33007
作者: Wajdi Hammami,Irfan Azam,Soumaya Cherkaoui
机构: Polytechnique Montreal (蒙特利尔工业大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Natural-language service intents in integrated terrestrial and non-terrestrial (TN/NTN) 6G networks can be incomplete, ambiguous, or inconsistent, making unconstrained LLM generation challenging for reliable network-facing orchestration. This paper proposes a knowledge-guided LLM framework that combines grammar-constrained semantic parsing with a knowledge-graph (KG)-driven clarification loop to produce validated, orchestration-ready service profiles across four representative service classes. On a 200-sample intent dataset, the framework achieves 89% classification accuracy and 95% policy validity with an average of 0.66 clarification turns per request. Compared with a Pure-LLM baseline, the proposed framework improves policy validity and contradiction detection while requiring fewer clarification turns, demonstrating the value of explicit knowledge-guided reasoning for robust intent processing in integrated TN/NTN 6G orchestration.
[NLP-201] Linear-Time Context Compression with State-Space Encoding
【速读】： 该论文旨在解决大规模语言模型（LLM）在处理长上下文时面临的巨大计算与内存开销问题。现有软提示压缩方法通常采用与目标模型相同的Transformer架构来生成压缩表示，导致在推理过程中产生显著的延迟开销，因而仅适用于离线场景——即上下文仅需压缩一次并重复用于多个查询。本文提出Boa Compressor，一种基于状态空间模型（State-Space Model, SSM）的即插即用式在线上下文压缩器，能够在单次线性时间遍历中编码任意长度的上下文，并生成一个可直接用于条件化冻结目标LLM的紧凑表示。该SSM通过知识蒸馏训练，使学生模型（接收压缩上下文的LLM）的下一个词元概率分布逼近教师模型（接收完整上下文的LLM）的输出。在相同比例压缩下，Boa在性能上达到或超越现有最先进的离线压缩基线，同时相比基于Transformer的实现快一个数量级，有效降低了端到端推理延迟而非增加压缩耗时。实验表明，Boa在更长上下文及高达16倍的激进压缩比下仍具有良好扩展性。通过实现低成本、可每查询实时运行的有效压缩，Boa成为支持高效长上下文推理的实用且通用的工具。
链接: https://arxiv.org/abs/2610.33006
作者: Marc Serra Ortega,Andrey Barsky,Carlos Boned Riera,Tomás Ockier,Dimosthenis Karatzas
机构: Computer Vision Center, UAB(计算机视觉中心，乌布大学), Spain
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 15 pages, 7 figures
Abstract:Transformer-based large language models (LLMs) impose substantial computation and memory costs as context sizes increase, motivating methods that compress a context into a shorter representation before inference. Existing soft-prompt compression methods typically use the same Transformer backbone as used in the target model, where producing the compressed representation leads to significant latency overheads. This restricts them to the offline setting where context is compressed once and then reused across many queries. We propose the Boa Compressor, a plug-and-play online context compressor built on a Mamba state-space model (SSM) that encodes arbitrary-length context in a single linear-time pass, producing a representation on which we condition a frozen target LLM. The SSM is trained by distillation, matching the next-token logit distribution from the LLM ingesting the full context (as teacher) to the same LLM ingesting the compressed representation (as the student). At equal compression ratios, Boa matches or surpasses the performance of existing state-of-the-art offline compression baselines, while being an order of magnitude faster than Transformer-based implementations, thereby reducing end-to-end latency rather than adding compression time overhead. We demonstrate that Boa also scales well to longer contexts and to compression ratios as aggressive as 16x. By enabling effective compression that is cheap enough to run on every query at inference time, Boa becomes a practical and general tool for efficient long-context inference.
[NLP-202] OpenESM: Energy-Steered Models Are Scalable Foundation Learners
【速读】： 该论文旨在解决能量模型（Energy-based Models, EBMs）在通用场景建模中表现不足的问题，尤其是在大规模语言建模任务中难以有效泛化至多样化下游应用的挑战。其核心解决方案是提出能量引导模型（Energy-steered Models, ESMs），通过将能量引导搜索与分词生成统一于同一架构中，实现对潜在状态在能量景观上的优化，并利用优化后的潜在状态指导更精准的分词生成。关键在于：先在大规模语料上进行能量引导的预训练，使模型学习到一个具有普遍有效性的能量景观，其中低能量值对应高概率、高质量的词汇序列；随后通过规模扩展实验验证，发现随着模型参数量和训练数据的增加，ESMs展现出良好的可扩展性，在26个跨领域下游任务中多数优于GPT类模型。这一设计突破了传统EBM在采样效率与生成质量之间的权衡，为大模型时代的通用问题求解提供了新的范式。
链接: https://arxiv.org/abs/2610.33005
作者: Yuan Pu,Yudong Lu,Ting Wang,Zhicong Hong,Qinuo Li,Xuansheng Wu,Na Zou,Xia Hu,Guanchu Wang
机构: Shanghai Artificial Intelligence Laboratory (上海人工智能实验室)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 12 pages, 7 figures, 3 tables
Abstract:Energy-based models have shown promising properties for domain-specific tasks but remain challenging in modeling general-purpose scenarios. To bridge this gap, we design energy-steered models (ESMs) for foundational language modeling. ESMs integrate energy steering and token generation within a single architecture to leverage the energy-guided search to refine token generation. Specifically, ESMs first optimize a latent state over the energy landscape, then leverages the refined latent state for token generation. By performing such energy-guided pre-training on large-scale corpora, ESMs learn generally effective energy landscape where favorable tokens correspond to lower energy values. To explore the scaling behavior of ESMs, we pre-train ESMs on three large-scale corpora, then evaluate across 26 downstream tasks spanning a broad range of domains. Our results demonstrate that ESMs have promising scaling behaviors with increasing model size and training data, and outperform GPTs on most downstream tasks. These findings highlight ESMs as a promising architecture for large-scale pre-training and general purpose problem solving. We fully release the model weights and training infrastructure to enable the community to completely reproduce and extend our results. GitHub: this https URL HuggingFace: this https URL Comments: 12 pages, 7 figures, 3 tables Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2610.33005 [cs.CL] (or arXiv:2610.33005v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.33005 Focus to learn more arXiv-issued DOI via DataCite
[NLP-203] Entity-Aware Evaluation of OCR Systems for Historical Document Transcription
【速读】： 该论文旨在解决传统光学字符识别（OCR）评估指标（如字符错误率，CER）无法充分反映历史文献中关键信息（如人名、日期、标识符等实体）是否被准确保留的问题。其核心解决方案在于引入实体级分析（entity-level analysis），通过在公开的贝鲁蒂档案（Berrutti Archive）175份文档基础上，结合5,209条人工标注的实体提及数据，从实体保全性与可恢复性角度评估不同OCR系统的表现。研究对比了序列识别型OCR系统、专用视觉-语言模型及通用视觉-语言模型，并在不同字符级编辑容忍度下分析实体匹配情况。结果表明，仅依赖全局转录准确率会误导评估结论：尽管Qwen 3.6在所有容忍度下均表现出最高实体覆盖率，但部分系统虽在精确匹配上表现较差，却在容忍微小错误时显著提升实体恢复能力，说明其错误多为局部字符偏差而非语义性误识；而另一些系统即使放宽匹配标准仍难以恢复大量实体。这揭示了单一聚合转录指标的局限性，强调必须结合实体级视角来全面评估OCR对有意义信息的保全能力。
链接: https://arxiv.org/abs/2610.33004
作者: Camilo Mariño(UDELAR, CB),Diego Belzarena(UDELAR, CB),Marina Gardella(CB),Ignacio Ramírez(UDELAR),Gregory Randall(UDELAR),Jean-Michel Morel(LU - Hong Kong)
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV)
备注:
Abstract:Optical Character Recognition (OCR) is essential for digitizing historical archives, yet standard evaluation metrics such as Character Error Rate (CER) do not necessarily reflect whether meaningful information, such as names, dates, and identifiers, is preserved. In this work, we investigate entitylevel analysis of OCR errors as a complementary perspective for evaluating OCR quality in historical documents. Using a publicly available benchmark of 175 documents from the Berrutti Archive, augmented with 5,209 manually annotated entity mentions that will be released upon acceptance, we compare sequence-recognition OCR systems, OCR-specialized vision-language models, and general-purpose vision-language models. Beyond global transcription accuracy, we analyze entity preservation and recoverability under varying character-level edit tolerances. Our experiments show that OCR rankings depend on the evaluation criterion: while Qwen 3.6 achieves the highest entity coverage across all tested tolerances, several systems with lower exact-match coverage recover substantially when small errors are tolerated, suggesting that many of their errors correspond to localized character-level deviations rather than fundamentally incorrect entity transcriptions. Conversely, some systems remain unable to recover many entities even under relaxed matching. These findings show that OCR quality cannot be fully characterized by aggregate transcription metrics alone and highlight the importance of evaluating how OCR errors affect the preservation and recoverability of meaningful information.
[NLP-204] From Writing Advice to Manuscript Governance: A Constrained-Transformation Framework for Academic Revision
【速读】： 该论文旨在解决学术写作中因合法写作目标冲突而产生的治理难题，尤其在简洁性、发表载体形式或风格变异与证据支持及论断边界之间发生矛盾时，缺乏明确的优先级判断标准。其核心解决方案在于将修订过程建模为在有序约束下的可接受性问题：通过任务相关不变量、情境约束以及可变的表达选择共同界定可接受的修订空间，并引入“论断保留契约”（claim-preservation contract），以确保证据与判断始终与授权的论断边界保持一致。据此提出四项治理命题——保留性、优先性、情境适应性与充分性，构建了一个技术无关的机制框架，能够实现表达方式的调整而不隐性改变研究论证本身。该模型由“学术写作过滤器”（Academic Writing Filter）具体实现，用于筛选候选修订方案，而非对优质学术写作进行普遍定义或分类。
链接: https://arxiv.org/abs/2610.33001
作者: Simon Rudkin
机构: University of Manchester(曼彻斯特大学); Manchester, United Kingdom
类目: Computation and Language (cs.CL)
备注:
Abstract:Academic revision presents a governance problem when legitimate writing objectives conflict. Writing-process and social-science research show that revision is recursive and that wording carries epistemic and disciplinary commitments. Those literatures identify important constraints but do not settle which objective should prevail when concision, outlet form or stylistic variation conflicts with evidence and claim boundaries. This paper models revision as an admissibility problem under ordered constraints. Task-relative invariants, contextual constraints and mutable presentation choices define the admissible space; a claim-preservation contract links evidence and judgement to an authorised claim boundary. Four governance propositions specify preservation, precedence, context and sufficiency. The Academic Writing Filter implements the model by screening candidate revisions; it is not a universal definition or classifier of good academic writing. The contribution is a technology-independent mechanism for changing presentation without silently changing the research argument.
[NLP-205] Whispering Wisely: Smart Subset Selection for ASR Fine-Tuning
【速读】： 该论文旨在解决大规模自动语音识别（ASR）模型（如OpenAI的Whisper）在微调过程中计算成本高昂的问题，尤其是在拥有大量标注音频数据时，使得许多研究者难以将其适配至新语言或特定领域。其核心解决方案是提出一种基于嵌入（embedding）的子集选择方法：利用Whisper预训练编码器将每个音频样本映射到固定维度的嵌入空间，并采用贪心k-center算法从中选取具有多样性和代表性的训练子集。实验结果表明，相较于随机采样，该方法在较小数据规模下显著提升识别性能，且通过词汇多样性分析发现，k-center选取的子集能覆盖更广泛的语言内容，从而解释其有效性。整体而言，该方法为低资源场景下的ASR微调提供了一种高效、实用的数据子集选择策略，尤其在样本预算有限时优势明显。
链接: https://arxiv.org/abs/2610.32997
作者: Roi Yozevitch,Or Haim Anidjar
机构: Ariel University (阿里埃尔大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 17 pages, 5 figures
Abstract:Automatic Speech Recognition (ASR) models such as OpenAIs Whisper have substantially improved speech recognition quality but remain expensive to fine-tune, especially when large amounts of transcribed audio are available. This makes it challenging for many practitioners to adapt these models to new languages or domains. In this work, we introduce an embedding-based subset selection approach for selecting smaller training sets for ASR fine-tuning. Using Whispers pretrained encoder, we map each audio sample into a fixed-dimensional embedding space and apply a greedy k-center algorithm to select diverse and representative subsets of the training data. We evaluate this approach on both Hebrew and Danish, showing that k-center selection often improves upon random sampling, particularly at smaller subset sizes. An analysis of lexical diversity reveals that k-center subsets capture a broader range of linguistic content than random subsets of the same size, which may help explain their observed effectiveness. Overall, our results demonstrate that embedding-based subset selection provides a practical strategy for ASR subset selection in the two evaluated languages, with the clearest recognition benefits observed at smaller sample budgets.
[NLP-206] LLM -as-a-Judge Version Shift: The Instrument Changed Not the Standard
【速读】： 该论文旨在解决在自动化科研流程中，随着大语言模型（LLM）裁判版本迭代更新时，其评估标准是否保持一致的可靠性问题。当前研究普遍依赖固定权重的LLM裁判对研究提案进行评分、筛选与修订，但实际应用中这些裁判常被更新版本替代，而这一过程可能引入不可控的评估偏差。研究的关键发现在于：新旧版本替换不仅影响评分的均值位置（如更严或更松），还会改变评估准则的几何结构（criterion geometry）与分辨能力（resolution），甚至导致关键判别维度失效——例如某一版本替换后，93%的提案得分被压缩至单一数值，丧失了有效区分力。进一步分析表明，这种差异主要由评估格式驱动而非内容本身所致，且通过全局校准无法有效修复个体层面的评分离散性。为此，论文提出一个基于测量理论的框架，将裁判替换视为对其原点（origin）、准则几何或分辨率的系统性改变，并设计了一套小规模分层锚点审计（约200个样本）机制，在部署前即可识别潜在失败类型；同时，冻结阈值策略能准确诊断出注册后API裁判的替换问题。研究结果强调了在自动化评估体系中维持测量稳定性的重要性，并提供了可复现的六组预注册锁与超过14,800次裁判调用的完整审计记录以支持透明验证。
链接: https://arxiv.org/abs/2610.32986
作者: Guijia Zhang,Harry Yang
机构: The Hong Kong University of Science and Technology(香港科技大学)
类目: Computers and Society (cs.CY); Computation and Language (cs.CL)
备注: 25 pages, 5 figures, 14 tables
Abstract:Automated research pipelines increasingly rely on fixed LLM judges to rank, filter, and revise proposals, yet these judges are routinely replaced by newer versions. We study whether such replacements preserve the underlying measurement standard. Using a preregistered, frozen-weight design, we evaluate three adjacent same-family transitions across Qwen, Gemma, and Mistral on frozen ML research proposals under randomized information conditions. The common assumption that newer judges are simply harsher fails: two transitions shift toward greater leniency, while the third shifts in the opposite direction on anchors but attenuates off-anchor. More importantly, version changes can alter not only score location but also criterion geometry and resolution; one transition collapses clean-flawed discrimination, placing 93% of pool scores on a single value. Supplying identical facts does not close the version gap, placebo-controlled interventions show that observed packet effects are largely format-driven rather than content-driven, and global recalibration repairs the mean while explaining little object-level dispersion. We organize these failures with a measurement model in which a judge replacement can change its origin, criterion geometry, or resolution. A small stratified anchor audit of about 200 calls identifies the failure class before deployment, and frozen thresholds correctly diagnose two post-registration API judge replacements. We release six preregistration locks and the full audit trail of 14882 judge calls.
[NLP-207] Skip the Compressor: Coding Agents Fix More Issues When They Pick Whole Files
【速读】： 该论文旨在解决生成式 AI (Generative AI) 在处理大型代码仓库时面临的上下文窗口限制问题，即如何在有限的上下文长度内有效支持代码修复（patch generation）。现有方法如提示压缩器（prompt compressors）和检索器（retrievers）存在根本性缺陷：检索器依据代码与问题的语义相似性排序文件，但无法保证所选文件是真正需要修改的；压缩器虽能适应上下文预算，却通过删减文本降低信息完整性，导致模型无法准确复现原始代码行，从而影响修复质量。本文提出的关键解决方案是采用一种基于精确文件选择的策略——CQ2-CARE，其核心在于将固定上下文预算全部用于选择并完整呈现必须修改的文件，而非压缩或检索部分代码片段。该方法无需依赖复杂的压缩模型，确保模型可直接引用原始代码，提升编辑准确性。实验表明，在300个SWE-rebench近期问题上，使用GPT-5.6和Claude Opus 5时，CQ2-CARE分别解决了43.7%和46.7%的问题，显著优于三种压缩器（+6.3点）和三种检索器（+9.3至12.7点），甚至超过两个专门训练用于问题修复的方法（SweRank和SWE-Pruner）。分析揭示，传统方法失败的主要原因在于：检索器常遗漏关键修改文件，而压缩器虽保留文件但改变其内容，导致修复补丁失效。CQ2-CARE通过精确文件选择机制，实现了更高的修复成功率与可靠性。
链接: https://arxiv.org/abs/2610.21073
作者: Xiaocong Yang,Rishit Chatterjee,Himanshu Udupi,Vikram Ramavarapu,ChengXiang Zhai
机构: University of Illinois Urbana-Champaign(伊利诺伊大学厄本那-香槟分校)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:A coding agent works on repositories that can be far larger than its context window can hold. Prompt compressors and retrievers offer solutions to this problem, but patching an issue needs two things at once: the files that must change, and their code exactly as written, since the model must reproduce the lines it edits. Retrievers rank code by its similarity to the issue rather than by whether it must change, and compressors delete text to fit the budget. We show that at an equal token budget, an agent resolves more issues when the budget goes to choosing which files to show verbatim, as our simple selector CQ2-CARE does, than when it goes to pruning tokens or retrieving chunks. We fit every context to the same per-issue budget on 300 recent SWE-rebench issues and grade each patch with the benchmark’s tests. With GPT-5.6 and Claude Opus 5, CQ2-CARE resolves 43.7% and 46.7% of the issues, 6.3 points more than the best of three compressors with either model, and 9.3 and 12.7 points more than the best of three retrievers. CQ2-CARE even resolves more issues than two methods trained for issue resolution, SweRank (by 0.7 and 1.7 points) and SWE-Pruner (by 2.0 and 7.3). Our analysis shows the defects of the baseline methods: retrievers often miss the files to edit, and compressors keep them but alter their text, so fewer patches apply. Exact-file selection needs no compression model and shows every line as the repository has it, so the model can quote the code it edits.
[NLP-208] Integrate the Reward Dont Sample It: Window Likelihood for Interruptions and Backchannels in Full-Duplex Speech Models
【速读】： 该论文旨在解决全双工语音对话系统中用户打断与回音确认（backchannel）行为的精确控制问题，即模型需在用户主动打断时及时让出话语权（yield），而在用户仅进行非言语回应时保持持续倾听（hold）。现有方法通常采用基于序列级强化学习的时机奖励（GRPO），但实验发现其在四个开源模型上存在双重缺陷：一是二值窗口奖励在最需学习的关键窗口处产生零梯度（零梯度陷阱），二是单个回合的奖励得分被分散至多个听觉状态决策中，导致信用分配偏差（credit leakage），引发模型“对静默说话”或“提前停止回答”的错误行为。这些问题的本质源于奖励函数是少数二值控制决策的已知函数，导致采样不具区分性。为此，作者提出窗口似然（Window Likelihood, \WL）：针对每个时间窗口独立采样话语延续路径，但通过闭式积分对所有可能的停止决策进行整合，并最大化正确动作的对数概率——其中“让出话语权”建模为噪声或（noisy-OR），而“保持倾听”则建模为生存乘积（survival product）。在HumDial-FDBench基准上，\WL在六种结构不同的开放模型上均取得最优的yield/hold F1性能，超越了使用不同数据训练的Kyutai发布的GRPO检查点，且未引入任何额外的发言权重新开放现象。
链接: https://arxiv.org/abs/2610.35581
作者: Guoqiang Hu,Lichao Zhang,Yizhou Peng,Yuhao Luo,Haifeng Hu,Sallen Sun,Will Yang,Eng Siong Chng
机构: Nanyang Technological University, Singapore(南洋理工大学, 新加坡); Tencent(腾讯)
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Sound (cs.SD)
备注:
Abstract:A full-duplex spoken dialogue model must yield when the user cuts in and hold when the user merely backchannels. Sequence-level RL with a timing reward (GRPO) is the usual recipe; we test its whole-rollout form on four open models and find two failures. A binary window reward gives zero advantage on exactly the windows that most need learning (a zero-gradient trap), and the one score per rollout is spread over listening-state decisions the reward is not written in terms of (credit leakage), so the model talks into silence or stops answering. Both come from sampling a reward that is a known function of a few binary control decisions. We propose \emphwindow likelihood (\WL): sample a continuation per window, but integrate in closed form over the stop decisions it visits and maximise the log-probability of acting correctly—a noisy-OR for yielding, a survival product for holding. On HumDial-FDBench, \WL\ gives the best yield/hold F1 on all six structurally different open models we post-train, exceeds Kyutai’s released GRPO checkpoints (trained on different data), and adds no re-opening of the floor on any model.
[NLP-209] Wave Is All You Need: A Quantum Language Model Based on Fully Connected Quantum Boltzmann Machine
【速读】： 该论文旨在解决传统生成式人工智能（Generative AI）模型中自注意力机制（self-attention mechanism）依赖于离散矩阵运算所带来的计算冗余与资源消耗问题，尤其在处理连续语义流时需进行显式离散化分解，导致信息损失与效率低下。其核心解决方案是提出一种基于全连接量子玻尔兹曼机（Fully Connected Quantum Boltzmann Machine, FC-QBM）的自注意力机制，通过在语义量子态上施加幺正演化（unitary evolution），隐式实现位置编码与自注意力计算，无需将连续语义流显式分解为离散矩阵进行乘法运算。该机制的测量结果天然满足归一化条件，且数学推导表明其与经典自注意力机制在理论上等价。研究进一步建立了FC-QBM语言模型（FC-QBM-LM）的训练与生成算法，并构建了8量子比特规模的模型，在唐诗语料上完成训练。实验表明，该量子层仅用44个参数即可完整实现位置编码与自注意力计算流程，在无噪声环境下表现出良好收敛性；同时，采用Y轴旋转门终止的量子电路相较于X轴旋转门具有更优收敛性能。在小样本及多种噪声强度下，模型保持超过99%的噪声无关性能，展现出强鲁棒性。五言绝句生成实验初步验证了模型具备联想性与发散性思维、意象层面的内容生成能力以及符合人类语序习惯的词序组织能力。
链接: https://arxiv.org/abs/2610.34492
作者: Jun Liu
机构: Hunan University of Finance and Economics (湖南财经工业学院)
类目: Quantum Physics (quant-ph); Statistical Mechanics (cond-mat.stat-mech); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 44 pages, 6 figures, 2 tables, 2 algorithms
Abstract:This work proposes a self-attention mechanism based on the fully connected quantum Boltzmann machine (FC-QBM), and further constructs a fully connected quantum Boltzmann machine language model (FC-QBM-LM) using the mechanism as the core engine. By performing unitary evolution on semantic quantum states, the mechanism implicitly implements positional encoding and self-attention computation, eliminating the need to decompose continuous semantic streams into discrete matrices for explicit matrix multiplication, with the measurement results naturally satisfying the normalization condition. Specifically, this work rigorously derives the computational principle of the self-attention mechanism based on FC-QBM, and demonstrates that this mechanism is mathematically equivalent to the classical self-attention mechanism. The mathematical principles governing the training and generation of FC-QBM-LM are thoroughly derived, and the two corresponding algorithms are subsequently developed. An 8-qubit FC-QBM-LM is finally constructed and trained on the corpus of Tang poetry. Under the 8-qubit setting, the quantum layer implements the entire positional encoding and self-attention computation pipeline with only 44 parameters. In the noiseless scenario, the model exhibits favorable convergence. Furthermore, the quantum circuit proposed in this work, which terminates with Y-axis rotation gates, achieves better convergence than the counterpart terminating with X-axis rotation gates. Under the small-sample regime and across eight distinct noise intensity levels, the model demonstrates strong noise robustness, retaining over 99% of its noiseless convergence performance. Experiments on five-character quatrain generation reveal that the model preliminarily exhibits three core capabilities: associative and divergent thinking, content generation across imagery and human-aligned word ordering.
[NLP-210] Automated auditable dating of the Babylonian Astronomical Diaries with language models and astronomical tools
【速读】： 该论文旨在解决古代巴比伦天文日记（Babylonian Astronomical Diaries）中大量残缺文书因缺失日期公式而导致的年代判定难题。其核心挑战在于如何在缺乏明确时间标记的情况下，仅依靠天文观测记录对数以百计的残损泥板进行精确年代推定。解决方案的关键在于构建一个基于规则的解析系统与语言模型协同工作的自动化框架：首先通过规则引擎将观测文本转化为结构化的逻辑条款（clause），再利用语言模型识别并补全遗漏的天文观测信息；随后借助天文计算工具对公元前626年至公元75年共701个年份逐一验证这些条款的符合度，并依据匹配程度对各年度进行排序。该系统在未公开已知年代的120个泥板上实现了96%的正确率，其中84个案例中唯一最优候选年份即为已发表的正确年份。此外，系统还为ADART IV中的56个无年代泥板提供了按置信度排序的候选年份列表，清晰展示了纯天文学证据所能达到的年代约束精度。整个流程可交互追溯，配套代码、协议与结果均开放共享。
链接: https://arxiv.org/abs/2610.34297
作者: Lucas Carrit Delgado Pinheiro,Yuan-Sen Ting,Jon Clindaniel,Lu Li,Yingbin Liang,Ness Shroff
机构: The Ohio State University (俄亥俄州立大学); Max-Planck-Institut für Astronomie (马克斯·普朗克天文研究所); The University of Chicago (芝加哥大学); The Hebrew University (希伯来大学)
类目: Instrumentation and Methods for Astrophysics (astro-ph.IM); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); History and Philosophy of Physics (physics.hist-ph)
备注: 15 pages, 6 figures, 7 tables, including the SI Appendix. Code, data and interactive page: this https URL and this https URL
Abstract:The Babylonian Astronomical Diaries record the sky over nearly six centuries. Many tablets are broken and have lost their date formula. Experts dated many of them one tablet at a time in the standard edition, Astronomical Diaries and Related Texts from Babylonia (ADART). They used the recorded sky and other evidence, such as year formulas and royal names. Our system automates the astronomical part for all tablets with the same rules. A rule-based parser turns the recorded observations into clauses, short statements with fixed fields. A language model checks them against the text and adds observations that the parser missed. Astronomical tools test the clauses in all 701 years from 626 BCE to AD 75. The system ranks the years by how well they agree with the clauses. The system withholds published dates until its results are final, and the output lists the passages and computed positions behind each candidate year. On 120 dated tablets, the years tied at the top of the ranking include the published year in 96% of cases. For 86 of them, this top group is a single year, the published year in 84. In 119 of the 120, the reported candidate set contains it. For 56 undated entries of ADART IV, the system gives ranked candidate years. For each tablet, the results show how far astronomy alone narrows the year. An interactive page at this https URL follows each tablet and links the code, protocol and results.
[NLP-211] LLM s Can Tackle High Energy Theoretical Physics yet Gaps Remain: Part I Quantum Field Theory
【速读】： 该论文旨在解决生成式 AI 在高能理论物理领域中对复杂量子场论（Quantum Field Theory, QFT）问题的求解能力边界不明确的问题，尤其是评估大语言模型（Large Language Models, LLMs）在需要深度推理与推导的高阶QFT问题上的实际表现。其解决方案的关键在于构建并发布首个系统性的基准测试工具 QFT-Bench，包含75道涵盖核心QFT子领域的研究生至研究级难度问题，每道题均具备明确目标和闭式参考答案。通过结合符号验证与基于LLM的评判机制进行综合评估，研究发现前沿模型在非平凡的QFT计算中已展现出接近95%的准确率，但依然存在对计算细节、规范约定及指令理解的错误；更关键的是，模型在思维链（Chain-of-Thought, CoT）中表现出显著的推理不忠实现象——即虽能得出正确答案，但其推理过程缺乏逻辑自洽与有效支持。这一工作为当前LLMs在量子场论中的能力提供了量化评估，并为未来在高能理论物理领域的广泛评测奠定了基础。
链接: https://arxiv.org/abs/2610.33134
作者: Tong Liu,Bohan Li,Gongrui Cheng,Zixiong Yu,Yulan He,Volker Tresp
机构: 未知
类目: High Energy Physics - Phenomenology (hep-ph); Computation and Language (cs.CL)
备注:
Abstract:Beyond math and code, Large language models (LLMs) are beginning to tackle problems that have traditionally required substantial training for theoretical physicists, but it remains unclear how far this capability extends and where it might break. As a first step toward pushing LLMs in high-energy theoretical physics, we introduce QFT-Bench, a benchmark of 75 graduate- to research-level problems requiring extensive reasoning and derivation across core Quantum Field Theory (QFT) subareas. Each question has a well-defined target and closed-form reference answer. Evaluation combines symbolic verification with LLM-based judge. Across 39 proprietary and open-source models, frontier models already demonstrate substantial capabilities on nontrivial QFT calculations, with the strongest models exceeding 95% accuracy. At the same time, clear gaps remain. Even strong models still make errors in calculations, conventions, and instruction following, while weaker models often fail at the level of physical concepts or derivations. We further monitor models’ CoTs and show that current models exhibit nontrivial reasoning unfaithfulness, i.e., they could reach correct answers but the reasoning itself does not justify and logically support them. These results provide a quantitative assessment of current LLM capabilities in QFT and a first step toward broader evaluation in high-energy theoretical physics.
信息检索
[IR-0] ACARE: Agent -Context-Aware Reranker for Iterative Search
链接: https://arxiv.org/abs/2610.35840
作者: Sahel Sharifymoghaddam,Yauhen Babakhin,Reza Esfandiarpoor,Oliver Holworthy,Jimmy Lin,Bo Liu
类目: Information Retrieval (cs.IR)
备注:
Abstract:In agentic search, a document can match an agent-generated sub-query without helping answer the user’s original question. Conventional rerankers score documents against the sub-query alone, overlooking the broader task and the agent’s reasoning. We introduce the Agent-Context-Aware Reranker (ACARe), a compact 4B LLM trained to rerank documents using this context when available. Across multiple benchmarks and agent families, ACARe improves retrieval quality over its base reranker while reducing search effort. These benefits extend to answer generation: ACARe improves accuracy and reduces searches over no reranking, while matching or exceeding the base reranker’s mean accuracy with more precise evidence. It also helps smaller agents narrow the retrieval-quality gap to larger agents, showing how better retrieval tools can make smaller agents more effective. The model, training data, and code will be released soon.
[IR-1] Do Similarity Measures Measure Similarity? How User Activity Creates Phantom Correlation in Recommender Systems
链接: https://arxiv.org/abs/2610.35765
作者: Marco Favier,Noah Daniëls,Jens Leysen,Sander Marinus,Bart Goethals
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: 11 pages, 1 figure, 3 tables. Code: this https URL
Abstract:Item-item similarity is one of the oldest and most foundational signals in recommender systems: two items are assumed to be similar when the same users consume them. Despite the sparsity and scale of interaction data, the implicit assumption across the field is that item co-occurrence reflects genuine shared preference, capturing what users like together. In this paper, we demonstrate that a substantial fraction of observed item-item similarity is an artifact of variance in user activity rather than shared taste. Even if every user consumed items completely independently, differing only in their total consumption volume, standard similarity measures still detect strong associations between items. We term this phenomenon phantom correlation: spurious item association that emerges strictly from heterogeneous user activity in the complete absence of preference structure. Starting from minimal formal assumptions, we show that independent consumption under varying activity is described by a Rasch model, and we derive how the resulting association depends on item popularity for each similarity measure. By comparing our theoretical baseline against real-world benchmark datasets, we reveal the extent to which popular similarity measures reflect phantom correlation, which bears on how item co-occurrence is interpreted and used in collaborative filtering.
[IR-2] owards a Unified Framework of Optimal Input-Side Dimensional Configuration for Deep CTR Prediction
链接: https://arxiv.org/abs/2610.35695
作者: Chaohua Yang,Fuyuan Lyu,Xing Tang,Dugang Liu,Xiuqiang He,Zhong Ming
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: 13 pages, 8 figures
Abstract:Deep click-through rate (CTR) prediction has achieved substantial success through increasingly expressive feature interaction architectures. However, predictive performance also depends on how input-side information is configured before interaction modeling. Existing studies have shown that selecting informative samples, features, or embedding units can improve prediction accuracy. Nevertheless, they primarily optimize these selection mechanisms independently along a single input-side dimension, overlooking their shared basic form and the inherent dependencies across input levels, thereby limiting the effectiveness of input-side dimensional configuration. To address this issue, we propose OptIDC, a unified framework for Optimal Input-side Dimensional Configuration in deep CTR prediction. OptIDC jointly learns sample, feature, and embedding-unit level configurations within a unified framework by integrating their corresponding selection mechanisms. To explicitly coordinate selection decisions across different levels, OptIDC further introduces a selection-weighted Optimal Transport mechanism that aligns coarse- and fine-grained selections via transport over finite measures, thereby alleviating cross-level selection conflicts while preserving the semantics of each input level. Moreover, OptIDC is model-agnostic and can be seamlessly integrated into existing CTR models. Extensive experiments on four real-world benchmark datasets demonstrate that OptIDC consistently achieves higher AUC than existing input-side selection methods, validating the effectiveness of unified input-side dimensional configuration.
[IR-3] Cross-Domain Sequential Recommendation with Representation-Level User Transfer CIKM’26
链接: https://arxiv.org/abs/2610.35550
作者: Abdulaziz Samra,Nikita Sukhorukov,Alexey Grishanov,Alexey Vasilev,Evgeny Frolov
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注: Accepted in CIKM’26 - full paper track
Abstract:Cross-domain sequential recommendation aims to improve next-item prediction in a target domain by leveraging user behavior from auxiliary source domains. However, real-world deployment faces two critical challenges: business constraints that isolate interaction data across domains, and user heterogeneity, where many users lack cross-domain activity. This makes existing cross-domain methods difficult to apply. In this paper, we propose a simple and practical framework for cross-domain sequential recommendations that transfers information exclusively through learned user representations, without sharing raw interaction data or item embeddings across domains. We introduce a lightweight adapter that projects the source-domain user representation into the target-domain embedding space and integrates it with the target interaction sequence. Importantly, when source data is unavailable (i.e., the user observed only in the target domain), the model seamlessly defaults to a robust target-only sequential recommender, retaining benefits from joint training across all users. Extensive experiments were conducted under two setups; one targets benchmarking against previous baselines, while the other focuses on realistic and scalable conditions. Both sets of experiments demonstrate consistent performance gains for shared and unshared users.
[IR-4] ny Suffix Arrays
链接: https://arxiv.org/abs/2610.35543
作者: Sharma V. Thankachan
类目: Data Structures and Algorithms (cs.DS); Information Retrieval (cs.IR)
备注:
Abstract:Suffix array and inverse suffix array queries on a text of length n can be answered in \widetildeO(\tau^3) time using an auxiliary data structure occupying O(n/\tau) words, for any \tau \in [1,n] .
[IR-5] Hypergraph-Enhanced Evidence Selection for Long-Context Multimodal Document QA ICDE2027
链接: https://arxiv.org/abs/2610.35542
作者: Wenxin Zhao,Jiachuan Wang,Xinyi Zhu,Jiake Ge,Haoyang Li,Yongqi Zhang,Shuangyin Li,Lei Chen
类目: Information Retrieval (cs.IR); Databases (cs.DB)
备注: Accepted by the 43rd IEEE International Conference on Data Engineering (ICDE 2027)
Abstract:Long-context multimodal document question answering (QA) has become an important yet challenging task for modern vision-language models (VLMs), as it requires reasoning over evidence dispersed across pages and modalities within limited input budgets. Current pipelines adopt retrieval-augmented generation (RAG) to select multimodal evidence for downstream VLMs. However, existing systems mainly rely on static selection rules over fixed-granularity retrieval units such as pages or blocks, resulting in information loss during retrieval and suboptimal performance in downstream reasoning. To alleviate these issues, we formulate the core task as budget-aware evidence selection over structured document evidence and propose HyperDoc, a document RAG framework that constructs compact yet sufficient evidence while preserving document relations. Specifically, HyperDoc constructs a document hypergraph to explicitly capture structural, logical, and cross-modal relationships among multimodal document elements. Over this structured evidence space, HyperDoc performs adaptive cost-aware hyperedge selection based on query intent and input budget. Extensive experiments on three representative benchmarks demonstrate that HyperDoc outperforms various types of state-of-the-art baselines.
[IR-6] Beyond “[at]”: A Large-Scale Measurement of Email Address Presentation on Computer Science Scholars Homepages
链接: https://arxiv.org/abs/2610.35499
作者: Aichun Huang,Xiang Li,Zheli Liu
类目: Information Retrieval (cs.IR)
备注: 12 pages, 10 figures
Abstract:Publishing an email address makes a scholar reachable, but also exposes it to automated collection. Yet how email addresses are presented on the Web has not been measured systematically across their content, form, and location. We measure it across 32,413 homepage records of computer-science scholars listed in CSRankings with MailScope, which preserves raw and rendered HTML, extracts addresses with a three-stage cascade (rules, then large language models, then a crawling agent), and labels each address. Among 19,517 reachable homepages, we find an address for 17,712 scholars (90.8%). Our measurement yields four main findings. (i) Presentation is diverse: only 53.8% publish plain content, the others spread over 23 subtypes of symbol substitution, code, and semantic description, and 5.5% of scholars’ addresses appear only after rendering. (ii) Presentation follows hosts rather than individuals: scholars on one host share a content class in 67.9% of pairs, compared with 38.8% across hosts of one institution, whereas research output and institution rank are barely related to presentation. (iii) Content and form are coupled through code-based links, and 6.2% of scholars with non-plain selected presentations have plain-text copies of the same address elsewhere. (iv) Plain addresses are more common behind passed bot challenges, and 36.1% of redirecting homepages no longer lead to scholars. Human assessment yields 94.4% weighted exact-set agreement on 143 judgeable pages, complemented by three 500-page review rounds with human verification. Presentation is diverse but clusters by shared host, and by our label-based estimate plain-string matching misses nearly half of the addresses.
[IR-7] Read the Route: Attention-Steered Hypergraph Retrieval for Multi-hop RAG
链接: https://arxiv.org/abs/2610.35412
作者: Jiate Liu,Liuyi Chen,Xiaofei Yuan,Yifu Tang,Zihan Yang,Xu Zhou,Zhengyi Yang
类目: Information Retrieval (cs.IR)
备注:
Abstract:Graph-based retrieval-augmented generation (GraphRAG) answers multi-hop questions by organizing a corpus into a graph and retrieving along its structure. Existing retrieval methods face two limitations. Diffusion- and similarity-based methods expand retrieval along fixed edge weights or by similarity to the question, without resolving its implicit entity references, allowing relevance errors to accumulate across hops. In contrast, exploration-based methods let an LLM reason about the next hop, but require a separate generation at each hop, leading to high inference costs. We observe that an LLM can resolve implicit entity references from retrieved passages and a multi-hop question before generating any tokens. Further analysis identifies a small set of attention heads, termed addressing heads, that capture this capability across multiple hops in a single forward pass. Building on this finding, we propose HyperAddress, which runs Personalized PageRank on a description-labelled hypergraph, using addressing heads to determine the restart distribution and transition weights, thereby guiding multi-hop retrieval without any generation at query time. On MuSiQue, 2WikiMultiHopQA, and HotpotQA, HyperAddress outperforms the strongest baseline by 3.5 points of Recall@5 and 7.4 points of All@5 on average, with gains that grow with the number of hops. It answers more than twice as fast and builds its index at half the cost.
[IR-8] Agreement Between Reconstructed and Native GDELT GKG Records: A Field-Level Baseline
链接: https://arxiv.org/abs/2610.35358
作者: Nikita A. Klimin(HSE University, Moscow, Russia)
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 15 pages, 10 tables, 5 figures
Abstract:The Global Knowledge Graph (GKG) of the Global Database of Events, Language, and Tone (GDELT) characterizes news articles using named entities, topics, quantitative and temporal features, quotations, tone, and dictionary-based content measures. Its codebook defines the record format, but the software that produces the records is not public, so researchers cannot generate compatible records for articles they collect themselves. We built an open-source reconstruction, gdelt-regkg, and measured how closely it agrees with GDELT’s archived (native) records for 193,550 English articles from 8,729 publishers (2020-2025) and 2,500 articles in five other languages. Agreement depends strongly on the field. English tone, the percentage of positive minus negative words, has a mean absolute error (MAE) of 1.79 points and a concordance correlation coefficient of 0.75. Identity agreement (F1) is 0.65 for persons, 0.42 for organizations, and 0.65 for locations. It is lower for date components (0.42), complete counts with their objects and locations (0.07), and quotations (0.07), while GCAM category presence reaches micro F1 0.84 within the implemented dictionaries. In subsets where retrieved and native titles and lengths match, tone error falls to 1.47 and organization F1 rises to 0.48, yet most disagreement remains. A larger gazetteer raises coverage of native places from 89.5% to 97.1% with little change in location F1. Tone computed on English translations of non-English articles is less accurate (MAE 2.10-2.64) than for English articles. The results are conditional on article pages that could still be retrieved, and the correspondence subsets do not isolate the effect of input differences. Reconstructed records should therefore be assessed field by field and not treated as completely interchangeable with native ones.
[IR-9] CRIM: Composable and Reusable Item Memory for Efficient Personalized Retrieval
链接: https://arxiv.org/abs/2610.35323
作者: Haolin Li,Haipeng Zhang,Mang Li,Zhaohui Xia,Xun Zhu,Miao Li,Yu Zhang,Ji Wu
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注:
Abstract:Large language models incur substantial training and inference costs in personalized search and recommendation, as growing interaction histories lengthen request inputs and repeatedly expose the same item text to the model. We propose \textbfComposable and \textbfReusable \textbfItem \textbfMemory (CRIM), a framework that separates item-level semantic encoding from request-level history modeling. A small set of learnable summary tokens compresses each item into compact continuous memories, while segment-level readout tokens (REC tokens) introduce predictive supervision within the interaction history. These memories are independent of users and queries and can therefore be reused across requests. At inference time, CRIM composes cached memories in interaction order and incorporates an optional query to produce a personalized retrieval representation, reducing repeated item-text encoding while retaining request-dependent contextual modeling. We further introduce a joint training mechanism that combines cached memories with sampled live re-encoding. Re-encoding only a few history items per request allows the retrieval objective to jointly optimize the item and request encoders at limited computational cost, while periodic cache refreshes maintain consistency. Extensive evaluation and cost analysis on MovieLens-1M, Qilin, and our new LazSeq dataset show that CRIM achieves state-of-the-art retrieval performance with substantially lower inference costs at the same backbone scale, while limiting item re-encoding during joint training. We will publicly release our code and dataset.
[IR-10] Unified Heterogeneous Feature Interaction via Effective-Rank Scaling for Recommender Systems
链接: https://arxiv.org/abs/2610.35318
作者: Ao Zhang,Haochuan Li,Boya Du,Jian Yang,Qihang Fang,Yulin Xu
类目: Information Retrieval (cs.IR)
备注: 9 pages, 4 figures
Abstract:Large-scale industrial ranking models must jointly capture heterogeneous feature interactions, long-term user behaviors, and multi-scenario dependencies under stringent online serving constraints. Existing architectures typically process sequential and non-sequential features with separate modules, limiting bidirectional information exchange. Directly applying homogeneous self-attention to all semantic tokens can also be inefficient when sequential and non-sequential signals exhibit different effective-ranks (eRanks). Meanwhile, shared representations can overlook scenario-specific patterns and induce negative transfer between scenarios. In this paper, we propose HiRank, a unified framework for multi-scenario recommendation that jointly models Heterogeneous feature Interactions via effective-Rank scaling. HiRank integrates gated sequential attention with heterogeneous feature interaction within a unified backbone, enabling bidirectional fusion between sequential and non-sequential representations. HiRank further uses scenario-aware aggregation to compress heterogeneous low-order tokens into a compact representation. Distinct learned projections transform the full compact representation into scenario-aware composite tokens with complementary global feature views and more uniform information density, enabling more effective high-order self-attention. Extensive experiments on industrial-scale datasets demonstrate favorable performance scaling with parameter count and consistent improvements over competitive baselines. Online A/B tests further show a 2.27% increase in orders per user.
[IR-11] Octen-WebGR: Scaling Generative Retrieval to 200 Million Web Pages
链接: https://arxiv.org/abs/2610.35256
作者: Guangwei Xu,Wenkai Zhang,Yawen Liu,Dongxu Zhang,Xijie Xu,Kuan Zou
类目: Information Retrieval (cs.IR)
备注:
Abstract:Navigational search aims to find a single target page among hundreds of millions of web pages. Click tables work well for queries seen before but cannot generalise to new formulations, while embedding retrievers generalise across content but often rank navigational targets poorly. We introduce Octen-WebGR, a generative retrieval framework that stores web pages in the parameters of a language model: given a query, the model generates a two-token semantic identifier constrained to a table of real URLs. We develop Octen-WebGR-4B, which indexes 201 million web pages, and evaluate it on real navigational traffic from the month after training. Generative retrieval is most effective on new ways of asking for familiar pages: for unseen queries targeting pages with substantial interaction history, it ranks the correct page first 30.9% of the time, against 17.2% for a dense retriever trained on the same clicks (21.4% with a click prior). Embedding retrieval keeps an advantage on rarely visited pages, so the two approaches generalise on complementary sides, the query and the page. Over all navigational traffic, Octen-WebGR reaches 59.6% top-1 accuracy, against 46.5% for a click-enhanced dense retriever and up to 37.5% for frontier LLMs answering from memory, which cost at least 40 times as much per query. The 4B model runs on a single GPU with 10 GB of memory, compared with 412 GB for exact dense retrieval. Scaling experiments with up to 32 million URLs show that, when every query-page pair is seen equally often, the model size needed grows sublinearly with index size, and that retrieval is constrained more by page exposure than by model capacity. We release Octen-WebGR-4B and Octen-WebGR-0.6B with their identifier tables and inference code.
[IR-12] Reading Beyond OCR Noise: Robust Cross-Lingual RAG for Legislative Archives
链接: https://arxiv.org/abs/2610.35240
作者: Radha Krishna Deshpande,Ajoy Mondal,Akash Manna,C V Jawahar
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:India’s historical public archives present severe computer vision and document image analysis challenges: decades of parliamentary records survive only as degraded, multi-column scanned pages featuring mixed scripts, complex layouts, and dense visual noise where OCR failure is the primary obstacle to information retrieval. We present an end-to-end framework that couples vision-based layout segmentation and multi-script OCR with an agentic retrieval-augmented generation (RAG) pipeline. Our approach specifically addresses how visual degradation and layout fragmentation propagate into text extraction errors, using an iterative agentic loop to navigate around OCR noise and layout ambiguities before generating answers. Evaluated on the 200,000-page Punjab Vidhan Sabha archive, our system demonstrates robust handling of historical document degradation, outperforming standard text-only baselines.
[IR-13] When Fusion Order Matters: Insights on Progressive Multimodal Representation Learning for Cross-modal Retrieval
链接: https://arxiv.org/abs/2610.35125
作者: Haoxiang Zhao,Bin Wang,Yaofei Wang,Xiao Lan,Yunfeng Diao,Xu Wang
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:Multimodal video retrieval combines visual, acoustic, and speech information, yet how their fusion order should depend on the data and the retrieval query remains unclear. We introduce Unified Progressive Representation Learning for Order Analysis, or UniPRO, to investigate this question through configurable fusion schedules with matched model components and training settings. Starting from text-to-video retrieval, we find that the tested one-by-one schedules outperform stacked fusion, while changing the order within one-by-one fusion also changes performance. To guide order selection, we examine two complementary considerations: learning difficulty under unequal modality availability and preservation of information relevant to the query. Complete-modality controls show that missing inputs alone cannot explain order sensitivity. At R@1, text and audio retrieval favor schedules that introduce the modality matching the query last. Controlled speech missingness during training and testing reverses the preferred order in both R@1 and SumR, with different crossover points. These findings motivate balancing the two considerations when selecting a fusion path. We then use them to predict a visual-last schedule for image queries, which achieves 97.9 R@1, the highest among the evaluated orders on MSR-VTT-9k. Main experiments on four text-to-video benchmarks establish the effectiveness of UniPRO, with component ablations assessing its supporting design. Together, the results identify fusion order as a task-dependent choice that should account jointly for modality availability and the retrieval query.
[IR-14] Useful Diversity or Hallucinated Variety? Persona-Conditioned Synthetic Query Generation for E-Commerce Search AACL2026
链接: https://arxiv.org/abs/2610.35102
作者: Filippos Ventirozos,Tom Armitage,Mahmoud Oshagh,Peter A. Appleby,Matthew Shardlow
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: Accepted at IJCNLP-AACL 2026 (Findings). 20 pages, 2 figures, 16 tables
Abstract:Synthetic supervision can bootstrap semantic parsers in new search domains, but prompt-induced diversity is useful only when generations are faithful to the structured intent. We explore this for cold-start e-commerce search: 118,098 synthetic pairs produced by 6 LLMs with 9 persona-conditioned prompt families on a 47-field UK automotive schema, against 1,214 human-annotated and 15,304 production queries. We show that the IPIP personality bundle improves parser F1 in all four matched contrasts (Holm-corrected p.001 ), and all three bundles have a positive main effect over the complete 2^3 grid. Syntax conditioning is the axis that moves generations towards real traffic. It also induces lexical diversity. Alignment sKL, the divergence between von Mises-Fisher distributions fitted to synthetic and human-annotated queries, is an unlabelled proxy for parser F1 ( \rho=-0.917 against human queries written for simulated shopper scenarios, lower is closer). An LLM committee, checked against a blinded author pass, triages the 927 queries missed by every parser and labels 37% as hallucinated. Dropped constraints dominate. The four syntax-bearing families have the largest hallucinated shares, 38 to 42%, while IPIP alone has the lowest at 28%.
[IR-15] Learning to Remember: Advances in Memory Mechanisms for Large Language Models
链接: https://arxiv.org/abs/2610.35089
作者: Chenye Han,Hongche Yu,Mengyue Wu,Dell Zhang,Jixiang Luo
类目: Information Retrieval (cs.IR)
备注:
Abstract:Memory has become a central mechanism for extending large language models (LLMs) beyond fixed context windows, stateless interaction, and static parametric knowledge. This survey organizes the literature around three complementary dimensions: the life cycle of memory, the storage substrate, and the system role that memory plays in LLM-based agents, while using cognitive-science-inspired memory concepts as a lightweight conceptual motivation rather than as a separate technical contribution or implementation family. On this basis, we distinguish three mechanism families: nonparametric memory, where information is stored outside model parameters in text, graph, multimodal, or database-like structures; parametric memory, where information is stored or accessed through activations, key-value caches, learned modules, adapter-based parameterizations, or model weights; and memory for agent evolution, where stored experiences support reinforcement learning, continual learning, procedural reuse, context engineering, and agent harness design. Evaluation is treated as a separate cross-cutting component rather than as a memory mechanism. Unlike surveys that focus primarily on retrieval-augmented generation or long-context modeling, this paper emphasizes the full memory life cycle: construction, maintenance, retrieval, consolidation, forgetting, governance, and evaluation. By aligning conceptual motivations, computational memory mechanisms, benchmarks, and deployment concerns, the survey provides a structured roadmap for designing memory-augmented LLM agents that are persistent, adaptive, auditable, and safe to update over time.
[IR-16] Reading the Registers of Honorius III without Manual Line Transcription: Forced-Aligned HTR End to End Modular or Hybrid
链接: https://arxiv.org/abs/2610.35085
作者: Tamas Kovacs,Daniel Luger
类目: Computer Vision and Pattern Recognition (cs.CV); Information Retrieval (cs.IR)
备注: 10 pages, 10 figures
Abstract:Unlocking medieval archives at scale is bottlenecked by the cost of line-level diplomatic transcriptions and manual abbreviation expansion. Focusing on the registers of Pope Honorius III (1216-1227) over 5,000 letters surviving in full only as manuscript images - we show that existing, line-break-free expert transcriptions (edition drafts) suffice to train end-to-end hand-written text recognition (HTR) that jointly transcribes and expands heavily abbreviated Latin. Using forced alignment on 860 unaligned edition entries across 394 pages of Reg. Vat. 9, we automatically bootstrap 13,122 training lines without manual alignment of segmented lines. While the generic CATMuS baseline struggles on this domain (18.9% CER) and a modular space-free HTR + word-segmentation pipeline (cf. Camps et al. 2021) plateaus at 15.6% WER, our from-scratch hybrid Conformer - combining CTC with bidirectional subword attention decoders, iterative self-realignment and n-best LM rescoring - achieves 4.7% CER and 8.7% WER (8.4% WER with ROVER system combination), nearly halving errors on clause-final words (22.7% to 12.2% WER). These figures are measured on the test lines that align with the transcription (86% of all test lines for the best hybrid). Furthermore, we address a tension between paleographic fidelity and editorial normalization: training on targets that preserve scribal name initials reduces the WER against the manuscript text on the 31 affected test lines from 13.3% to 9.2%, at a small cost against the expanded transcription (WER +0.3 points, CER +0.4). Finally, extensive ablations show that none of an auxiliary diplomatic CTC head, a copy mechanism, a permutation decoder or masked-image pretraining improves on direct end-to-end supervision, while strict padding-mask propagation and per-line beam termination make inference batch-invariant.
[IR-17] Hallucination-Aware Multi-Agent RAG via Internal Attention Rewards EMNLP2026
链接: https://arxiv.org/abs/2610.35056
作者: Runsong Jia,Qian Chang,Mengjia Wu,Ciprian Doru Giurcăneanu,Jie Lu,Yi Zhang
类目: Information Retrieval (cs.IR)
备注: Findings of the Association for Computational Linguistics: EMNLP 2026
Abstract:Retrieval-augmented generation (RAG) grounds language models in retrieved evidence, but its outputs often contain claims unsupported by that evidence. Recent multi-agent RAG systems decompose the pipeline into cooperative decisions, yet they typically optimize all agents under a shared task reward such as EM or F1, which measures answer overlap rather than faithfulness to evidence. We propose a hallucination-aware multi-agent RAG framework that combines the task reward with an internal hallucination-risk signal computed from the generator’s attention-head norms. Three role-specific LoRA adapters, a query planner, an evidence selector, and an answer generator, share a frozen backbone. A lightweight critic is trained once on TruthfulQA truthful/false pairs and then kept frozen, providing a complementary risk score that is added to the shared reward during cooperative PPO training. Across three backbones (Qwen3-8B, Llama3-8B, Llama2-13B) and three benchmarks (RAGTruth, HotpotQA, 2WikiMultiHopQA), our method improves task performance and reduces hallucination rates over the strongest multi-agent baseline. Ablations and critic diagnostics show that the TruthfulQA-calibrated signal provides useful reward shaping that transfers across RAG tasks.
[IR-18] CA-HNM: Constraint-Aware Hard Negative Mining for Dense Retrieval with Structured Domain Models
链接: https://arxiv.org/abs/2610.35045
作者: Ngoc Luyen Le,Marie-Hélène Abel,Bertrand Laforge
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Dense retrievers are commonly trained with hard negatives selected from lexical or dense retrieval results. Retrieval proximity, however, does not indicate whether a candidate conflicts with explicit domain structure. We introduce CA-HNM, a constraint-aware hard-negative mining framework that combines retrieval-based candidate generation with deterministic incompatibility rules derived from a structured domain model. The framework removes known positives, constructs a local typed context around the query target, records relational, scope, target, and level mismatches, and uses the resulting candidates to fine-tune a standard dual encoder. Structured information is required only during offline negative construction. We evaluate CA-HNM on MOOCCubeX and Course-Skill Atlas with two encoders, two training losses, and ten seeds. Relative to a retrieval-rank-matched control, CA-HNM-Pure has higher mean NDCG@10 in six of eight dataset–encoder–loss configurations. Four hierarchical 95% bootstrap intervals are entirely above zero, mostly under triplet loss. In a separate experiment, CA-HNM-Mixed shows slightly higher mean NDCG@10 than a matched-mixed control on both collections. Structural variants show that changes in available structure can also change candidate eligibility and negative-pool composition. The results indicate that the contribution of constraint-aware selection depends on the training objective and the composition of the negative pool.
[IR-19] SLIDE: Search-to-List Item Distillation for Single-Rollout Generative Retrieval
链接: https://arxiv.org/abs/2610.34764
作者: Tianyu Zhan,Gui Ling,Ran Tao,Kunhai Lin,Tong Xiong,Wang Yang,Kaixuan Zhang,Zhihong Chen,Yuliang Yan,Dan Ou,Shengyu Zhang,Haihong Tang
类目: Information Retrieval (cs.IR)
备注:
Abstract:Generative retrieval (GR) typically relies on item-level beam search to retrieve candidates, leaving cross-item dependencies unmodeled and limiting direct optimization of list-level objectives. A natural alternative is to generate the entire list within a single autoregressive trajectory. However, greedy decoding with directly trained list-wise models substantially underperforms beam search. We identify two contributing factors: the broader search space explored by beam search and the positional bias of scarce positive signals, which deprives later positions of adequate supervision. To address these challenges, we propose \textbfSLIDE , a search-to-list item distillation framework that decouples offline search from online list generation. A teacher model retrieves high-quality candidates through offline item-level beam search, providing additional conversion-relevant supervision beyond sparse behavioral labels. Their Semantic IDs (SIDs) are then concatenated in beam-score order to form list-wise distillation targets. This search-to-list conversion transforms candidates retrieved along separate search paths into sequential item-generation targets within a single trajectory. Trained on these targets, the student generates the Top- K list in a single greedy pass, also decoupling teacher scaling from online deployment. Experiments on FORGE and Taobao Search show that SLIDE substantially improves retrieval quality under single greedy decoding and, with stronger offline teachers, surpasses the beam-search reference. Supplementary RL experiments support the effectiveness of list-level generation.
[IR-20] Discovering Topics from Linear Autoencoders
链接: https://arxiv.org/abs/2610.34739
作者: Jutiyakhun Srisoparp,Donlapark Ponnoprat
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 12 pages. Comments are welcome
Abstract:Linear recommenders and topic models are similar in that they implicitly learn from co-occurrences of items. However, there has been little study of how these two model families are related. In this work, we show that, under a latent-factor model of the user-item interactions, the latent topics can be discovered from the weight matrix of a closed-form linear autoencoder. These topics provide a basis for interpreting the recommendations, and can be obtained directly from the trained recommender without additional training. Building on our analysis, we further propose novel item and user normalizations that prevent popular items and very active users from dominating the topics. Our experiments on six public datasets demonstrate that our method achieves significantly higher topic quality than every dedicated topic model and matches or exceeds the best linear autoencoder on all six datasets, while its recommendation performance is competitive with the best tuned recommender.
[IR-21] DPE-Rec: Diversity-Preserving and Efficient Diffusion Language Model Recommendation for Bundle Construction
链接: https://arxiv.org/abs/2610.34700
作者: Pingchuan Ma,Bohan Jiang,Chengshuai Zhao,Dawei Li,Zhen Tan,Huan Liu
类目: Information Retrieval (cs.IR)
备注:
Abstract:Generative bundle construction completes a partially observed bundle by generating its missing items from a large product catalog. Compared to autoregressive LLMs, diffusion LLM emerges as a natural fit for this unordered generation task. Their bidirectional attention provides an inductive bias better aligned with unordered bundle construction. Iterative denoising can reveal multiple tokens per step, potentially reducing inference latency. However, directly applying DLMs to this task presents two key challenges: 1 Despite its flexible reveal order, confidence-guided decoding tends to postpone semantic-fork decisions until refinement tokens constrain their choices, reducing bundle diversity. 2 Revealing more tokens per step can compromise recommendation quality, stifling the potential for faster inference. We introduce DPE-Rec, a diffusion framework for diversity preservation and efficient recommendation. DPE-Rec preserves diversity by introducing semantic-fork-aware GRPO, which prioritizes semantic-fork decisions via autoregressive rollouts. It improves efficiency through pseudo-trajectory distillation, which concentrates learning from the vast random corruption space onto the compact trajectory-induced subspace visited by high-quality decoding, unlocking reliable multi-token decoding. Together, these components enable diverse and accurate bundle construction with substantially fewer denoise steps. Experiments on multiple bundle construction benchmarks show that DPE-Rec improves recommendation performance by 63% on average, and its fast variant achieves 25.6 times faster inference on average than existing generative recommendation arts, with minimal loss in recommendation quality.
[IR-22] Knowing When the Evidence Is Enough: Requirement-Directed Search with Horn-Closure Certificates
链接: https://arxiv.org/abs/2610.34684
作者: Xinyu Zhang,Weiwei Sun
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 12 pages, 2 figures, 5 tables
Abstract:After each Web search, a search-augmented model must decide whether its evidence suffices: answer, search again, or abstain. Prevailing systems leave this to holistic model judgement, which can accept a context naming every entity yet lacking one required fact. We instead make sufficiency structural: a per-query requirement hypergraph states which facts jointly yield another fact. This raises two challenges: deriving requirements from the question alone, and linking each requirement to passages about the right entity. Certify therefore decomposes the question, searches per requirement, and accepts links only after entity checks. Horn closure over the links yields a certificate, the minimal sufficient passages and their derivation, or an abstention naming open requirements. Given correct retrieval, requirements and links, the decision is sound, and complete up to the entity check. Pooled over three corpora with unanswerable twins, Certify decides significantly more accurately than search agents, graph retrieval and sufficiency judges in default configurations. On the same evidence, it matches a proprietary model’s sufficiency label without an additional model call. Over 21M Wikipedia passages, entity checks add 4 to 13 points of decision accuracy where gold evidence exists. Finally, answers from its certificate are 80% correct, against 57% from all retrieved passages.
[IR-23] Agent Memory as Recommendation: Recommender-System Design for Long-Term Memory of LLM Agents
链接: https://arxiv.org/abs/2610.34673
作者: Xinyu Zhang,Weiwei Sun
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 12 pages, 3 figures, 8 tables
Abstract:Large language model (LLM) agents that serve users across many sessions must choose, for every request, the few stored memories that enter a limited context window. This choice largely decides whether the agent answers correctly, and it runs on every request at Web scale. This selection faces three challenges: a large and growing store, a per-request cost budget, and multi-session questions that need all of their evidence. Many memory systems build notes or graphs with hundreds to thousands of LLM calls per history, and each is designed on its own, so it is unclear which component yields the accuracy. These challenges mirror the properties of recommendation, so we formalize agent memory as budgeted slate recommendation, with the request as the user, stored memories as items and the context window as a slate. The formalization places memory systems in one design space of recommender cascades and factorizes accuracy into the completeness that each stage controls. Guided by it, we propose AgentMemRec, which adapts candidate generation, learning to rank and slate construction from recommender systems and selects memories without any LLM call. On multi-session questions of LongMemEval, AgentMemRec outperforms memory and context-selection methods, including A-Mem, HippoRAG 2, DynamicRAG and T-Mem, by 10 to 31 points of accuracy. It matches an LLM reranker that makes 64 calls per query at 3.2 times lower serving cost, and on LoCoMo it outperforms Mem0 and Zep without any LLM call at write time. The factorization attributes the gains to slates that more often hold all of the evidence, so recommender-system design offers an effective and efficient foundation for the long-term memory of LLM agents.
[IR-24] HARP-LoRA: Hierarchy-Conditioned Rank Projection for Low-Rank Adaptation in Scientific Table Understanding
链接: https://arxiv.org/abs/2610.34589
作者: Hossein Sholehrasa,Mohammadhossein Amini,Doina Caragea,Remya Ampadi Ramachandran,Dan Wu,Zihang Shi,Zelin Li,Lindsey Hulbert,Eder Martinez,Xuan Xu,Majid Jaberi-Douraki
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注:
Abstract:Tabular data is widely used in scientific applications, where complex layouts, including multi-level headers, merged cells, captions, and footnotes, make information extraction difficult. This is particularly important in pharmacokinetics (PK), where values must be correctly associated with the drug, species, dose, administration route, unit, and experimental condition. Parameter-efficient fine-tuning methods such as Low-Rank Adaptation (LoRA) reduce adaptation cost but do not explicitly incorporate relationships between table values and surrounding headers. We introduce HARP-LoRA, a hierarchy-conditioned rank projection method for low-rank adaptation in scientific table understanding. HARP-LoRA extends projected LoRA by constructing token-specific intrinsic representations from the row- and column-header hierarchies of each data token. These representations condition the low-rank down-projection, incorporating hierarchical context while keeping the pretrained model and random projection tensors frozen. We evaluate HARP-LoRA on 950 manually annotated PK tables from 587 scientific articles and on WikiTableQuestions (WikiTQ) using Llama~3.2 3B, Llama~3.1 8B, and Gemma~3 27B. With Llama~3.1 8B, HARP-LoRA achieves a PK F1-score of 57.64% and WikiTQ accuracy of 53.15%, improving over LoRA by 1.42 and 6.56 points, respectively. With Gemma~3 27B, it achieves the best PK F1-score of 70.63% and WikiTQ accuracy of 56.87%. These results use only 172K, 197K, and 547K trainable parameters, far fewer than LoRA and TableLoRA. Overall, conditioning lightweight adapters on table-header hierarchies improves extraction from scientific tables and reasoning over flatter general-domain tables while maintaining parameter efficiency. Code is available at: this https URL
[IR-25] RefinedGAW: A Refined German Academic Web Dataset
链接: https://arxiv.org/abs/2610.34572
作者: Maximilian Idahl,Oleh Astappiev,Benedikt Droste,Robert Scholz,Björn Plüster,Jan Philipp Harries,Robert Jäschke,Wolfgang Nejdl
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 26 pages, 13 figures, 6 tables. Accepted at the 13th Web as Corpus Workshop (WaC-13), 2026 (non-archival)
Abstract:The largest open German text datasets are built from broad web crawls, which provide scale but limited coverage of specific domains. We introduce RefinedGAW, a processed and annotated resource derived from 26 semi-annual crawls of the German Academic Web, totaling 122 million documents and 314 billion tokens, with full crawl and capture history of every text. Documents labeled as academic research account for 84% of German RefinedGAW tokens across HTML and PDF. At 87.6 billion tokens, this exceeds the 64.9 billion tokens in the German subsets of FineWeb2 and FinePDFs, despite their almost five-times-larger combined size. RefinedGAW also contains substantial amounts of educational, math-heavy, and expert-oriented content with minimal commercial bias. Across HTML and PDF, 88% of German RefinedGAW documents have no near-duplicate candidate match in the three evaluated reference datasets, supporting complementary coverage. We release a queryable metadata index for fine-grained data selection and a rehydration tool for retrieving and processing source documents.
[IR-26] he Price of Scale in Proactive Memory Recall
链接: https://arxiv.org/abs/2610.34507
作者: Yichuan Chen,Xin Wang
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (stat.ML)
备注: 14 pages, 4 figures, 2 tables. Code and data: this https URL
Abstract:A long-lived assistant that brings up memories without being asked must decide at every turn whether anything in its store is worth an interruption. We show that this decision is a multiple-testing problem whose difficulty grows with the store size N . On streaming replays of LongMemEval and LoCoMo, a similarity threshold tuned to 5% false alarms at N=8 reaches 87-100% at N=2048 , across embedding models and for explicit and implicit cues; LLM judges show the same growth from a lower base. We prove that a gate with false-alarm rate \alpha operates at a per-item false-positive rate of about -\ln(1-\alpha)/N , which pins its hit rate to the far left of the single-comparison ROC curve and, under regular variation, makes it decay as a power of N with an explicit constant. The resulting prediction matches the hit rate of a per- N oracle threshold to within one percentage point. The analysis explains why lexical and semantic channels need thresholds growing like \ln N and \sqrt\ln N , and why per-cue normalisation roughly doubles the hit rate at matched false alarms. A cue-swap conformal gate, which compares the current cue with the user’s other cues on the same store, needs no labels and has a finite-sample false-alarm guarantee under exchangeability; on LongMemEval it holds false alarms near 5% at the hit rate of a label-tuned threshold. Three annotators (Fleiss’ \kappa=0.63 ) find that about nine in ten false alarms are genuine interruptions. Code and data: this https URL
[IR-27] Multimodal Requirements Traceability Link Recovery in Industrial Settings
链接: https://arxiv.org/abs/2610.34503
作者: Haoming Huang,Shinpei Hayashi,Kazuya Kohata,Myeongki Kim,Yuki Kurihara
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: Accepted to APSEC 2026
Abstract:Requirements traceability links support activities such as change-impact analysis, coverage checking, and validation. Industrial requirements can include traceable images, which text-only approaches cannot process directly. For a fixed classifier configuration, binary decisions alone offer no direct way to adjust the balance between missed and incorrect links. We extend a retrieval-augmented generation (RAG) approach for inter-requirements traceability link recovery (TLR). Our extension retrieves text and image candidates separately, classifies each retrieved pair with a vision-language model, and filters the predicted links by verbalized confidence. We evaluate our approach on three non-overlapping industrial requirement sets and six public text-only datasets. On the industrial test sets, our approach achieves an F1 score of 0.310, compared with 0.114 for our replication of the baseline approach by Hey et al. Our approach recovers links in all four modality combinations while improving text-only F1 on the industrial data and maintaining similar performance on the public datasets, but recovery remains limited by failures to retrieve related text requirements for image queries. Confidence filtering produces link sets with different precision and recall, removing false positives and some correctly recovered links from the unfiltered link set. This study provides industrial evidence on the potential and limitations of a current RAG-based TLR approach for multimodal requirements and identifies cross-modal retrieval as the component to prioritize in further optimization.
[IR-28] heoretically Sound Item Similarity Measures for Recommender Systems
链接: https://arxiv.org/abs/2610.34402
作者: Noah Daniëls,Joey De Pauw,Bart Goethals
类目: Information Retrieval (cs.IR)
备注:
Abstract:Quantifying the similarity between items is fundamental to neighborhood-based collaborative filtering, yet many commonly used similarity measures are heuristics that lack theoretical justification. The use of heuristics forces practitioners to rely on empirical trial-and-error, offering little insight into why certain measures outperform others or how they might be improved. In this paper, we present a principled approach for designing item similarity measures grounded in probabilistic modeling and parameter estimation. By adopting this formal perspective, we derive two previously overlooked similarity measures based on the Bernoulli and multinomial distributions. Through an extensive ablation study we highlight the critical role of bias, probabilistic smoothing, treating the unknowns correctly, and choosing appropriate probability distributions. Finally, we show that the commonly used cosine similarity can be interpreted as the solution to a specific optimization problem, thereby providing a novel perspective on this decades-old heuristic. Subjects: Information Retrieval (cs.IR) Cite as: arXiv:2610.34402 [cs.IR] (or arXiv:2610.34402v1 [cs.IR] for this version) https://doi.org/10.48550/arXiv.2610.34402 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-29] A Domain-Aware Pipeline for Structured Extraction and Retrieval-Augmented Question Answering over arXiv Superconductivity Literature
链接: https://arxiv.org/abs/2610.34391
作者: Ivan Iorsh,Aleksandr Kapitonov,Ivan Filianin,Matvey Lesonen,Nikita Chizhov
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注:
Abstract:Scientific literature in materials physics is growing faster than researchers can manually inspect it, yet most retrieval tools rely on keyword matching or generic question answering that ignores domain-specific structure in PDF files. This paper presents an end-to-end pipeline that transforms arXiv superconductivity publications into a searchable corpus with structured metadata and retrieval-augmented question answering. The system combines thematic corpus construction, domain-aware extraction of materials and superconducting transition temperatures, and hybrid retrieval that merges dense chunk embeddings with structured publication fields. We test whether normalized domain attributes improve evidence retrieval and answer groundedness relative to lexical-only and dense text-only methods. Extraction is evaluated on a manually verified gold set of 59 superconducting materials, 46 of which have confirmed experimental critical temperatures. On the 33-material primary positive subset, domain-aware pair accuracy reaches 45.5% versus 9.1% for a regex baseline. Retrieval and generation are compared under a fixed language model (qwen2.5:7b-instruct) on a 100-question benchmark, varying only the retrieval mechanism: BM25, dense, hybrid, and hybrid with domain metadata. BM25 gives the highest source recall (Recall@5 of 47%), and adding extracted metadata to hybrid retrieval does not improve it (39% versus 41%), which we attribute to noise in the extracted fields; the main bottleneck is locating the supporting span inside a correctly retrieved paper. As a materials-informatics application, literature-derived temperatures are matched to an independent theoretical reference set: 200 materials appear in both sources, and 20 have a median experimental critical temperature above the theoretical value.
[IR-30] Cross-Lingual Information Retrieval in Tetun for Ad-Hoc Search
链接: https://arxiv.org/abs/2610.34362
作者: Altedio Araújo,Gabriel de Jesus,Sérgio Nunes
类目: Information Retrieval (cs.IR)
备注: The EPIA 2025 international conference on Artificial Intelligence
Abstract:Developing information retrieval (IR) systems that enable access across multiple languages is crucial in multilingual contexts. In Timor-Leste, where Tetun, Portuguese, English, and Indonesian are official and working languages, no cross-lingual information retrieval (CLIR) solutions currently exist to support information access across these languages. This study addresses that gap by investigating CLIR approaches tailored to the linguistic landscape of Timor-Leste. Leveraging an existing monolingual Tetun document collection and ad-hoc text retrieval baselines, we explore the feasibility of CLIR for Tetun. Queries were manually translated into Portuguese, English, and Indonesian to create a multilingual query set. These were then automatically translated back into Tetun using Google Translate and several large language models, and used to retrieve documents in Tetun. Results show that Google Translate is the most reliable tool for Tetun CLIR overall, and the Hiemstra LM consistently outperforms BM25 and DFR BM25 in cross-lingual retrieval performance. However, overall effectiveness remains up to 26.95 percentage points lower than that of the monolingual baseline, underscoring the limitations of current translation tools and the challenges of developing an effective CLIR for Tetun. Despite these challenges, this work establishes the first CLIR baseline for Tetun ad-hoc text retrieval, providing a foundation for future research in this under-resourced setting.
[IR-31] Factual Yet One-Sided: Stance Poisoning and Evidence Balancing in Retrieval-Augmented Generation
链接: https://arxiv.org/abs/2610.34294
作者: Puyuan Yang,Shaoxuan Dong,Liujie Wu,Xiaochun Cheng,Bo Li
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR); Information Retrieval (cs.IR)
备注: 29 pages, 4 figures, 20 tables
Abstract:The stance expressed in responses from retrieval-augmented generation systems is significantly influenced by the stance distribution of the retrieved materials. Existing research on knowledge poisoning has primarily targeted factual errors. In contrast, this study focuses on injection attacks in which the facts are correct but the stance on a controversial issue is one-sided. At a mean topic-level poisoning share of 18.9% within the post-injection ProCon documents, all four generation models from different vendors exhibited stance shifts aligned with the poisoning direction; across four concentration levels, the shifts increased monotonically. This paper proposes a Stance-Balancing Middleware positioned between the retriever and the generator. It consists of stance detection, fixed balancing (2 supporting + 2 opposing + 1 neutral), and a debiasing prompt, without introducing additional large language model calls. Across all eight concentration-direction settings, the complete method reduced the magnitude of stance deviation by an estimated 0.106 to 0.350; the six settings at poisoning shares of 11.9%-18.9% remained significant after Holm adjustment (adjusted p = 0.009). In the cross-model experiment, all eight model-direction combinations moved along the stance axis in the corrective direction, and seven of eight showed a clearly detected reduction in absolute distance from 0. In the concentration sweep, the post-defense intervals included 0 in all eight settings, while the cross-model pro-poisoning results retained small positive residual point estimates of +0.045 to +0.063. For 20 candidate documents, stance detection took approximately 1.5% of the median generation time on a single GPU and approximately 16% on CPU. Comments: 29 pages, 4 figures, 20 tables Subjects: Computation and Language (cs.CL); Cryptography and Security (cs.CR); Information Retrieval (cs.IR) Cite as: arXiv:2610.34294 [cs.CL] (or arXiv:2610.34294v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.34294 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-32] Recovering Business Rules as Cascade Lookups for Enterprise Field Prediction ECML KDD2026
链接: https://arxiv.org/abs/2610.34258
作者: Changxu Duan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Databases (cs.DB); Information Retrieval (cs.IR)
备注: Accepted at ECML PKDD 2026 Applied Data Science Track
Abstract:Enterprise sales-order systems rely on manually maintained configuration rules to determine fields such as payment terms, shipping points, and sales ownership. When these rules are incomplete or stale, operators must enter values by hand, and downstream errors spread into billing, logistics, and reporting. This paper studies whether missing or outdated rules can be recovered from historical transactions instead of being relearned as opaque statistical patterns. We introduce a knowledge-graph-guided workflow in which an LLM agent proposes candidate business keys, meaning column combinations such as customer and document type, validates them through SQL profiling, and compiles the resulting evidence into deterministic cascade lookups. The LLM is used only during offline rule discovery. At inference time, prediction is a transparent key lookup with no model-serving requirement. Experiments on enterprise sales-order prediction and public relational benchmarks show that cascades are most effective when fields follow stable business rules, while learned graph models remain preferable when prediction depends on temporal or neighborhood patterns. Drift analysis further turns residual errors into diagnostics for configuration maintenance. Code and data: this https URL.
[IR-33] Δτ-Bench: Pushing Agents Toward a Moving Capability
链接: https://arxiv.org/abs/2610.34238
作者: Shuofei Qiao,Xuanguang Pan,Yijun Chen,Wendi Feng,Bin Wu,Runnan Fang,Yuqi Zhu,Shumin Deng,Emine Yilmaz
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Information Retrieval (cs.IR); Machine Learning (cs.LG); Software Engineering (cs.SE)
备注: Work in progress
Abstract:In this paper, we introduce \Delta\tau -Bench, an executable agent benchmark that models the structural evolution of large tool ecosystems. In \Delta\tau -Bench, independently evolving services continually reshape how tasks can be accomplished: familiar tools may disappear or be replaced, while more efficient alternatives emerge amid overlapping functionality and noisy documentation. Against this moving capability frontier, we evaluate whether agents can proactively discover efficient tool-use paths and promptly revise their accumulated procedural knowledge, bringing together demands that existing benchmarks fail to examine. \Delta\tau -Bench comprises 1,588 validated multi-turn tasks across 42 domains and six successive ecosystem states generated through capability-preserving transformations. Our evaluation shows that even the strongest evaluated model or harness achieves only 42.3% task success, while initial gains from existing skill-learning methods can not consistently translate into better adaptation as the ecosystem evolves. These findings motivate Steward, an ecosystem manager that jointly manages tools and procedural knowledge through incremental skill maintenance and dynamic provisioning during task execution. Experiments show that Steward sustains higher task success across ecosystem changes with lower skill-maintenance overhead, while improving both recovery from disrupted workflows and adoption of more efficient alternatives.
[IR-34] MIDA: Multi-Interest Decoding with Adaptive Aggregation for Session-based Recommendation
链接: https://arxiv.org/abs/2610.34192
作者: Linjiang Guo,Nitin Bisht,Shiqing Wu,Guandong Xu
类目: Information Retrieval (cs.IR)
备注: Published in IEEE Transactions on Knowledge and Data Engineering
Abstract:Session-based recommendation (SBR) infers user interest from limited interaction sequences, yet current models show a consistent performance drop when longer histories are available. This degradation is associated with decoder designs that overemphasize recent items while underutilizing distant, yet semantically relevant, interactions, leading to poor modeling of long-range dependencies and multi-interest behaviors. To address this challenge, we introduce MIDA (Multi-Interest Decoding with Adaptive Aggregation for Session-based Recommendation), a lightweight and encoder-agnostic framework that provides a reformulation of the decoding stage. The central design principle is a two-space decoupling: clustering operates on raw item embeddings so that semantic groupings are not distorted by encoder-learned relational biases; encoder outputs are reserved for aggregation, where sequential context remains intact. Concretely, MIDA (i) employs differentiable multi-interest clustering on raw embeddings, preserving semantic structure that encoder representations would otherwise obscure, and (ii) fuses stability and plasticity branches through a gate conditioned on average cluster size, so that the inferred clustering structure directly governs aggregation behavior through a structural coupling across the two stages. Extensive experiments on multiple benchmark datasets demonstrate that MIDA consistently outperforms state-of-the-art methods, with meaningful accuracy improvements while remaining computationally practical.
[IR-35] LoopGR: Parameter-Efficient Scaling for Generative Recommendation via Recurrent Depth
链接: https://arxiv.org/abs/2610.34157
作者: Huizhe Zhang,Yudong Li,Liang Chen,Chengxiang Zhuo,Zang Li
类目: Information Retrieval (cs.IR)
备注:
Abstract:Generative recommendation formulates item recommendation as autoregressive generation rather than a discriminative scoring problem. Although empirical studies exhibit favorable scaling behavior in recommendation, unleashing the benefits of scaling remains challenging under the strict model size and inference cost constraints of real-world systems. A central challenge is to combine expressive modeling of user-item interaction histories with flexible computational depth without proportionally increasing model size. We introduce LoopGR, a parameter-efficient scaling framework for generative recommendation via recurrent depth. LoopGR employs a tailored hierarchical recurrent depth mechanism that alternates between a shared generative decoder and a lightweight context updater. For efficient inference, we further develop recurrent trajectory self-distillation that distills a low-budget student from a high-budget teacher by aligning both predictive distributions and evolution directions. Extensive experiments show that LoopGR consistently outperforms strong generative recommendation baselines. Compared with a representative encoder-decoder baseline, LoopGR achieves superior performance while reducing model size by up to 52%. These results establish recurrent depth as a new scaling dimension, offering a flexible and resource-efficient scaling path for generative recommendation.
[IR-36] Decision-Native Recommendation with Large Language Models
链接: https://arxiv.org/abs/2610.34124
作者: Daoxuan Liu,Jiayi Liao,Jiancan Wu,An Zhang,Xiang Wang,Xiangnan He
类目: Information Retrieval (cs.IR)
备注: 12 pages, 6 figures
Abstract:Large language models (LLMs) bring world knowledge to user preference modeling in sequential recommendation, but generation-based recommenders express next-item predictions indirectly as text. Generated outputs may fail to identify a unique candidate item, depend on candidate presentation order in the prompt, and lack explicit confidence in the recommendation. Recent studies explore Jev, a model for fast option selection with explicit confidence. Inspired by Jev, we introduce decision-native recommendation, a new paradigm that jointly models item choice and recommendation confidence using LLM representations. It requires valid candidate selection, consistency under candidate reordering, and confidence that reflects recommendation accuracy. We propose Direct Item Choice Estimation (DICE), which learns a personalized next-item distribution over supplied candidates from user interaction history and recommends the highest-scoring item with calibrated confidence. Tree-structured attention shares history computation across isolated candidate branches for order-consistent candidate scoring in a single forward pass. On three recommendation datasets, DICE achieves higher top-one accuracy than all evaluated baselines. Separate evaluations show unchanged item choices under tested candidate reorderings and end-to-end speedups of up to approximately 6\times over the evaluated generation-based methods. Its confidence estimates support selective recommendation by identifying more reliable predictions and guide additional computation for low-confidence examples in offline routing.
[IR-37] Bridging Search Intent and Tabular Data in Dataset Search
链接: https://arxiv.org/abs/2610.33936
作者: Wenjing Lin,Madelon Hulsebos
类目: Information Retrieval (cs.IR); Databases (cs.DB)
备注: 14 pages, 4 figures, 5 tables
Abstract:Dataset search remains a bottleneck in data analysis workflows, as existing systems largely rely on keyword matching or shallow retrieval that fail to capture user intent. To facilitate intent-driven search, we present two retrieval components to equip multi-faceted dataset search interfaces through natural language. First, we propose Hypothetical Schema Embeddings (HySE) that uses a language model to generate task-aligned hypothetical table schemas which are used for embedding-based retrieval, bridging the gap between intent and table content. We introduce KaggleDS, a corpus comprising 6,503 tables and diverse task-queries derived from Kaggle, and show that HySE consistently improves recall over strong keyword and embedding-based baselines, as well as over a dense table retriever fine-tuned on KaggleDS. We also show that HySE generalizes to end-to-end question answering on, for example, the NQ-Tables dataset, a corpus of 169,898 Wikipedia tables with human-written queries, where HySE improves Recall@10 by +6.0 pp over dense retrieval baselines. Second, we introduce FlexPrune, which translates dataset constraints expressed in natural-language into SQL predicates over a customizable metadata index, enabling filtering across arbitrary metadata attributes. Combined with HySE, FlexPrune increases Recall@10 by +6.0 pp over dense retrieval followed by tag-based filtering. Our results show that natural-language-based retrieval supporting task and metadata queries offers a viable alternative to schema-level keyword search in dataset search.
[IR-38] riageRAG : Failure-Aware Self-Training for Adapting Retrieval-Augmented Language Models to Specialized Domains
链接: https://arxiv.org/abs/2610.33776
作者: Samarth Varshney,Ritu Yadav
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 19 pages, 4 figures, 8 tables
Abstract:Retrieval-augmented generation (RAG) lets large language models (LLMs) answer knowledge-intensive questions from retrieved evidence, but adapting a general-purpose RAG system to a specialized domain such as medicine is difficult when only an unlabeled corpus is available. Self-training approaches such as SimRAG address this by letting the LLM write its own question-answer pairs from domain passages and fine-tuning on the pairs that survive a round-trip consistency filter. We observe that these approaches treat every retained example identically: they never ask whether the student already knows the answer, already extracts it correctly from the retrieved context, or fails - and if so, whether the failure lies with the reader or with the retriever. To tackle this, we propose TriageRAG, a failure-aware self-training approach that probes the student on each self-generated question with and without retrieved context, triages it into one of four categories (known, solved, context-failure, retrieval-failure), and routes each category to the training signal that addresses it, including a parametric route that teaches the reader to answer when retrieval misses. Probing is cheap and repeated every round, so the training mixture follows the student, and the triage histogram doubles as a label-free diagnostic of where the domain gap lies. Experiments on 11 datasets spanning three domains with a Qwen3-4B backbone, and on CS-Bench with Qwen3-8B, demonstrate that TriageRAG outperforms a budget-matched SimRAG by 1.2-6.4 points and is the only method that recovers the closed-book level in every case where retrieval hurts the base model. A single ablation identifies the mechanism: round-trip filtering removes every training context that lacks the evidence, and a reader that never sees such contexts learns to over-trust noisy retrieval.
[IR-39] SkillQR: Agent ic Skill Discovery for Reason -Intensive Query Rewriting
链接: https://arxiv.org/abs/2610.33768
作者: Kenan Jiang,Meixiu Long,Zihao Zhang,Ran Xu,Eugene Agichtein,Fei Liu
类目: Information Retrieval (cs.IR)
备注:
Abstract:Dense retrievers provide grounded evidence that mitigates hallucinations in outputs from Large Language Models (LLMs). Real-world queries are often verbose, ambiguously framed, or require multi-step reasoning, introducing irrelevant details that weaken semantic matching for dense retrievers. Query rewriting is a widely adopted pre-retrieval technique that leverages LLMs to reformulate queries and bridge the semantic gap between such queries and the retrieval target. Existing rewriting approaches either rely on expensive proprietary models, substantial training data, or iterative retrieval and rewriting, while largely applying a uniform rewriting strategy regardless of the retrieval corpus. We observe that effective query rewriting instructions are not universal: the optimal strategy depends on the retrieval target, rather than the model alone. We introduce SkillQR, a query rewriting framework that automatically discovers reasoning strategy sets from the retrieval target and dynamically routes a query to the appropriate rewriting strategy at inference time. Without relying on training data, SkillQR achieves competitive performance compared to fine-tuned rewriting models on the BRIGHT and R2MED benchmarks, while also providing lower inference latency than iterative rewriting approaches.
[IR-40] Novel or Not? PatEx-Gym: Agent ic Prior-Art Search for Novelty Assessment in Patent Examination
链接: https://arxiv.org/abs/2610.33503
作者: Qiyao Wang,Xinyi Chen,Hongbo Wang,Jiayuan Lu,Jianhang Li,Minghuan Tan,Yuan Lin,Min Yang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 19 pages, 8 figures, 3 tables
Abstract:Patent examination is a challenging real-world task that requires agents to discover sparse prior art, iteratively refine queries, inspect technical evidence, and produce novelty assessments documented in Office Actions (OAs) across heterogeneous scientific and technical domains. However, existing benchmarks largely focus on static and limited candidate sets, leaving the full search–reasoning–assessment process underexplored. We present PatEx-Gym, the first framework for developing and evaluating agents for patent examination, which integrates three components. First, PatEx-Bench provides 1K realistic examination tasks, two large-scale retrieval corpora containing over 1M patents and 52K scientific papers, and a harness runtime supporting iterative search, document visits, reasoning, and OA generation. Second, we introduce PASS, an annotation-free Prior-Art Agentic Search Trajectory Synthesis framework. Specifically, it exploits document visits as implicit relevance signals and evidence-acquisition anchors to model intra-turn query behavior, inter-turn search progression, and trajectory-level evidence and outcome quality, yielding PatEx-SynData. Third, we train PatEx-Agent at the 4B scale via supervised fine-tuning on the curated trajectories, establishing a lightweight baseline. Extensive experiments show that final decision accuracy alone does not fully reflect examination quality, while prior-art discovery remains difficult even for strong frontier models. Our PatEx-Agent achieves competitive performance against substantially larger general-purpose agents. These results show that training on PatEx-SynData enables a compact 4B model to serve as a competitive baseline for prior-art search and patent examination. We view patent examination as a demanding testbed for agentic search and deep research agents.
[IR-41] DyHist: Query-Aware Historical Dynamics Modeling for Temporal Knowledge Graph Extrapolation
链接: https://arxiv.org/abs/2610.33473
作者: Peiyuan Jiang,Yao Liu,Yubo Wen,Yanglei Gan,Lu Liu,Ruojie Zhang,Jiaye Yang,Yuewei Xue,Jiaya Deng,Xiaogu Guo,Qiao Liu
类目: Information Retrieval (cs.IR)
备注:
Abstract:Temporal Knowledge Graph (TKG) extrapolation aims to predict future facts from observed historical events. Existing methods usually achieve this goal by jointly modeling local historical evolution and long-range historical contexts. First, local evolution modeling often summarizes local history into an overall evolving representation, where query-relevant local deviations may not be sufficiently exploited as useful predictive features. Second, existing long-range history modeling often treats historical recurrence as a stable cue, while overlooking that its reliability may vary across relations and recent contexts. To address these issues, we propose DyHist, a query-aware historical dynamics modeling framework for TKG extrapolation. DyHist captures local dynamics by separating recent evidence from medium-range contexts and explicitly modeling their query-relevant deviations as predictive features. It further uses a relation-aware historical prior to calibrate candidate scores with time-decayed support, recent trends, and relation-specific strength, enabling recurrence signals to contribute differently across relations and recent contexts. Experiments on four benchmarks show that DyHist achieves the best MRR against recent baselines. Code is available at this https URL.
[IR-42] PinGER: Token-Efficient Generative Retrieval with Semantic IDs
链接: https://arxiv.org/abs/2610.33435
作者: Xinyi Li,Sujan Reddy Ale,Yi-Ping Hsu,Zelun Wang,Yilin Chen,Hanyu Li,Hanlin Lu,Arkin Dharawat,Lulu Yao,Yaxin Li,Bella Huang,Xiangyi Chen,Jaewon Yang,Jiajing Xu,Charles Rosenberg
类目: Information Retrieval (cs.IR)
备注: 11 pages, 5 figures, 8 tables
Abstract:Generative retrieval replaces nearest-neighbor search with token-by-token decoding of an item’s semantic ID (SID). But existing SID-based methods primarily use L levels of tokens sequentially to represent each item in the input history. As a result, a fixed context window holds L times fewer items than it would with one embedding per item. We present PinGER, a generative retriever deployed at Pinterest that retains SID outputs while making input histories compact, built on three ideas. (1) An asymmetric attention mask over interleaved SID and dense tokens lets later items attend only to the dense ones, so the model is supervised at every SID position in training yet serves one token per item while computing the same function. (2) Generation is controllable through goal-conditioned outputs, reinforcement learning, and a prefix-diversity penalty, which together steer retrieval toward surface-specific goals and diverse candidates. (3) A serving stack built for efficient wide-beam generation is paired with decoding constrained to valid identifiers and candidate resolution from a host-local index. Deployed in Related Pins, PinGER lifts saves of fresh Pins by 3.60% and overall saves by 0.62%, while reducing the surface’s end-to-end latency by 0.7% and eliminating the cost of the nearest-neighbor search it replaces.
[IR-43] Reinforcement Learning for Controller Parameters in the Auto-Bidding System at Pinterest
链接: https://arxiv.org/abs/2610.33248
作者: Yichu Zhou,Qinyi Chen,Fei Feng,Jacob Loh,Mehdi Ben Ayed,Lin Yang,Andrei Curelea,Jiajing Xu
类目: Machine Learning (cs.LG); Information Retrieval (cs.IR)
备注:
Abstract:Auto-bidding systems in online advertising must balance competing objectives-maximizing budget delivery while minimizing cost per action-across heterogeneous campaigns under non-stationary market conditions. This trade-off evolves over campaign lifespans: early on, cost optimization dominates when budgets roll over; near deadlines, budget delivery becomes urgent. While PID controllers are widely used in production for their stability and interpretability, fixed tunings cannot adapt to diverse campaigns and shifting traffic patterns. We present RLBP (Reinforcement Learning for Bidding Parameters), a hybrid architecture deployed at Pinterest that uses RL as a meta-controller to select campaign-specific PID parameters daily while retaining PID for high-frequency bid adjustments. RLBP trains a multi-head Q-network using a relative-lift reward formulation normalized against the production baseline, and performs inference-time fusion that adapts the cost-delivery trade-off to campaign states without retraining. Through online A/B testing at Pinterest, RLBP achieves a statistically significant reduction in effective cost-per-click (eCPC) with no revenue degradation, and two months of post-rollout measurements against a holdout also show consistent cost reductions. These results demonstrate that meta-controller RL can successfully augment production control systems at scale.
[IR-44] Holo-NN: Holographic Nearest Neighbor Search via Compressive Sensing
链接: https://arxiv.org/abs/2610.33224
作者: Bryant Gipson
类目: Cryptography and Security (cs.CR); Databases (cs.DB); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注:
Abstract:In the age of AI, privacy-preserving approximate nearest neighbor (ANN) search over high-dimensional embeddings has emerged as a critical yet frustratingly challenging domain, dominated by tradeoffs between speed, privacy, and accuracy. Traditional tree and graph-based indexing structures (e.g., HNSW, ScaNN) rely fundamentally on non-linear operations that are computationally intractable under Fully Homomorphic Encryption (FHE) or blind server-side constraints. In this paper, we introduce Holo-NN (Holographic Nearest Neighbor Search), an algebraic framework that replaces data-dependent branching with a single linear projection. Our key observation is that while distances across a database are dense, neighborhood membership is inherently sparse: only a handful of points ever lie close to a query. By pairing locality-preserving feature transforms with Compressive Sensing, Holo-NN folds the entire database offline into a compact holographic sensing matrix. At query time, encrypted search reduces to a single, non-interactive matrix-vector multiplication—requiring no FHE bootstrapping, no multi-round interaction, and no client-side index state. This turns private retrieval into an embarrassingly parallel, accuracy-tunable computation that runs out of the box on both CPUs and GPUs, scaling efficiently to production databases. We offer a comparison against other state-of-the-art Private ANN algorithms to help situate the method in the broader context of production-grade private retrieval. Subjects: Cryptography and Security (cs.CR); Databases (cs.DB); Information Retrieval (cs.IR); Machine Learning (cs.LG) Cite as: arXiv:2610.33224 [cs.CR] (or arXiv:2610.33224v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.33224 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-45] How Much Fresh Data Should CTR Recalibration Use? A Temporal Evaluation of Post-hoc Calibration
链接: https://arxiv.org/abs/2610.33168
作者: Rui Liu
类目: Machine Learning (cs.LG); Information Retrieval (cs.IR)
备注: 10 pages, 3 figures, 6 tables. Code and per-cell results: this https URL
Abstract:Offline evaluation of click-through rate (CTR) models still defaults to random splits, although deployed models score future traffic and downstream systems consume calibrated probabilities. On Avazu (40.4M impressions) we train logistic regression (LR), DeepFM, and DCN-v2 under two rolling temporal origins and a random split, three seeds per arm, and recalibrate each test day with Platt scaling and isotonic regression fitted on the preceding hours. At origin 1’s nearest test day, random splits report LR’s expected calibration error (ECE) at 55-62% of the temporal value in every seed (the deep models’ ratio is seed-unstable), and no model’s ECE is monotone in temporal distance. With a 6-hour fitting slice, recalibration removes up to 61% of ECE in badly miscalibrated cells but increases it in about a third of applications (14-17 of 48 across seeds). On fixed evaluation sets, a 12-hour slice cuts harm from 13-17 to 3-5 of 36 applications, and 24 hours raises it again. The 12-hour slice subsampled to the 1-hour slice’s size (5% of the impressions; 20-25x faster to fit) recovers 88% of the full slice’s gain, so impression count alone does not explain it. A slice-ECE gate shows no benefit out-of-origin, and in a simulated auction recalibration moves profit by only about -0.21% to +0.26%. Offline protocols should measure calibration temporally, over multiple origins and seeds; in this data, recalibration fitted on a wide time span beat the freshest few hours, which other logs must confirm.
[IR-46] Lessons from Luck: Do LLM -Written Lessons Carry Chance into an Agents Next Decision?
链接: https://arxiv.org/abs/2610.33029
作者: Kazuki Nakayashiki
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 28 pages, 1 figure, 11 tables. A registered study: the protocol, code and stimuli were frozen, deposited (doi: https://doi.org/10.5281/zenodo.23253562 ) and timestamped in a Bitcoin block before the confirmatory run. The locked run records are public (doi: https://doi.org/10.5281/zenodo.23272809 )
Abstract:Self-improving agents write lessons from the outcomes of their runs; when outcomes are noisy, a lesson can carry chance into later decisions. We measure one link of this chain in a registered study. In 256 synthetic worlds, each a recurring task with two interchangeable tools of recorded reliability, a scripted previous run used the more reliable tool (a good decision) or the less reliable one (a bad decision), with an assigned outcome: success or a transient failure. A model wrote a lesson from the run with ReasoningBank’s memory-writing instruction; a fresh session of the same model, holding the authoritative log and the previous result, chose the next run’s first tool with one memory item: that lesson, the lesson from the other outcome, an irrelevant item, or an unreasoned directive. The protocol was frozen, deposited and timestamped before any confirmatory call; 11 LLM endpoints gave 90,084 answers. After a chance failure of the better tool, the panel’s lessons carried the chance (registered reading luck-carrying; interaction +0.112, 95% interval [+0.097, +0.126]): readers named the worse tool in 11.6% of next runs with the lesson written from that failure, against 0.4% with an irrelevant item and 0.4% with the lesson written from a success. The effect was +0.042 where the log stated that failures are independent and +0.181 where it did not (descriptive). After a lucky success of the worse tool, the interaction was bounded below 0.10 (+0.002), a bound, not evidence of no harm. In exploratory readings, 6 endpoints carried the chance (up to +0.352); 5 were bounded below 0.10. In secondary comparisons, a three-sentence decision-quality note in the writing instruction lowered the worse-tool rate after a chance failure to 7.5%, and one instruction for both outcomes with a no-change option to 3.1%; neither met the registered harm-reduction rule at every such endpoint.
[IR-47] Bridging Static and Agent ic RAG for Taiwanese Historical Question Answering
链接: https://arxiv.org/abs/2609.23056
作者: Kai-Hsin Chen,Wei-Yu Chen,Xuanjun Chen,Jyh-Shing Roger Jang
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:Agentic retrieval-augmented generation (RAG) enables language models to adapt retrieval based on previously retrieved evidence, but it remains unclear whether such adaptive orchestration consistently outperforms well-designed static pipelines. We conduct a controlled comparison of agentic and static RAG for Taiwanese historical question answering, sharing the same generator and hybrid retrieval backend. Despite similar aggregate performance, the two pipelines differ on 70.83% of questions, with their advantages largely canceling out when averaged. An oracle that selects the better response per question improves the composite score by 0.2417 over the better individual pipeline, revealing substantial headroom for question-level selection. We therefore introduce a post-hoc selector that compares the two responses and their cited evidence, significantly outperforming either individual pipeline and recovering 60.34% of the oracle headroom. These results show that aggregate comparisons can obscure meaningful question-level differences between retrieval strategies, suggesting that exploiting their complementarity may be more fruitful than seeking a universally superior pipeline.
人机交互
[HC-0] he Evolution and Chineseness of Urban Screens in China: From Propaganda Posters to Smart City Interfaces
链接: https://arxiv.org/abs/2610.35762
作者: Wuwu Ran
类目: Human-Computer Interaction (cs.HC)
备注: 14 pages. Accepted at the Media Architecture Biennale 2025 (MAB25), Bangkok
Abstract:Urban screens have become a global feature of the media city, yet their developmental paths are not universal. This paper argues that the evolution of urban screens in China constitutes a distinct paradigm of media urbanism, termed Chineseness, understood here not as cultural essence but as a relational, historically situated configuration of three linked elements: deep-rooted lineages of public communication; state-steered technological leapfrogging driven by industrial policy and mega-events; and a regulatory paradox that enables spectacular scale while tightening operational control. Drawing on a five-stage genealogy (1990-present), a chronology of representative screens, and five paradigmatic cases, the study shows how this configuration produces temporal-spatial compression, the coexistence of multiple technological generations, and privileges collective witnessing over device-centric interactivity. The paper contributes a documented genealogy of Chinese urban screens, develops temporal-spatial compression and collective witnessing as analytical tools for non-Western media ecologies, and positions Chineseness as a globally relevant alternative to market-led models of media architecture.
[HC-1] EEGaze: Multimodal Attention Prediction in Visual Search via EEG-Based Saliency Refinement
链接: https://arxiv.org/abs/2610.35757
作者: Mansi Sharma,Dinar Orazgaliyev,Antonio Krüger,Philipp Müller
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Multimedia (cs.MM)
备注:
Abstract:Visual search is ubiquitous in daily life, from locating keys in a messy apartment to finding a tool in a cluttered workshop or an application icon on a desktop screen. While visual attention is to a large extent driven by image content, research in psychology indicates that user cognitive states have a significant influence on attention allocation. A promising way to access information about cognitive states is via Electroencephalography (EEG) recordings. Despite this potential, previous works in computational attention modeling have largely disregarded EEG, and related studies in psychology operate with abstract, simplistic stimuli, or even artificially guided search procedures. We bridge this gap with two contributions. First, we present EEGaze 140, the largest publicly available multimodal dataset of free visual search with realistic stimuli. The dataset contains synchronized EEG and eye tracking recordings from 36 participants across 140 naturalistic scenes, covering two scenarios: searching for desktop icons and finding tools in cluttered workshop scenes. Second, we introduce EEGazeNet, the first method for human attention prediction in visual search that integrates EEG signals. Our approach consists of a EEG guided refinement stage that adapts the output of a state of the art saliency prediction method to the user current cognitive state. In comprehensive evaluations on EEGaze 140, we show that EEGazeNet consistently outperforms all baselines and ablation conditions.
[HC-2] he Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI
链接: https://arxiv.org/abs/2610.35725
作者: Casey M. Williams,Jennifer Victoria Scurrell
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:The gap between AI capability deployment and safety assurance is structural and cannot be closed through internal evaluations alone. Analyzing documented harms (2017-2026) across the human lifespan reveals that disparate incidents — from voice assistants endangering children to AI companions implicated in adolescent suicide and automated care denials for the elderly — reflect a single failure: evaluating systems against an implicit adult user model. We introduce five developmental moderators structured into a Lifespan Vulnerability Matrix across six life stages. The matrix reveals a nonmonotonic risk distribution: vulnerability peaks at both ends of the lifespan, governance coverage is asymmetrical, and authority to convert model errors into real-world impact is scrutinized least. We then propose a set of lifespan design principles generalizing existing child-protective instruments, and an institution-agnostic, three-tier external evaluation framework — comprising automated simulation, expert red teaming, and post-deployment monitoring — indexed directly to matrix risk cells.
[HC-3] Reflective Computing
链接: https://arxiv.org/abs/2610.35696
作者: Md Haseen Akhtar
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: 54 pages, 8 figures, 12 tables
Abstract:Interactive technology is increasingly built for speed, and generative AI now answers before a question is fully formed. Yet the capacity to pause, interpret and decide for oneself is central to wellbeing, learning and meaningful use. HCI has named paradigms for sensing emotion, changing behavior, calming attention and tracking the self. However, none of them takes human reflection, and the slowing down it requires, as its organizing aim. This article defines that paradigm, reflective computing, and derives it from what thirty years of designs lack. We analyzed 134 interactive systems from 17 ACM venues (1996-2026) in two phases. First, an abductive gap analysis read every system against reflection theory, and seven recurring gaps defined seven dimensions, namely tempo agency, invitational form, interpretive openness, reflective depth, temporal span, protected idleness and ethical non-extraction. Second, two independent coders scored every system with a codebook (Krippendorff’s alpha 0.67 to 0.84 for five dimensions; ICC 0.82 for the overall score). The system sets the pace in 51 designs, 74 support only momentary use, only 12 scaffold reflection beyond description, and 56 papers are silent on the inner states they collect. No system expresses all seven dimensions strongly, and only 20 meet the definition. Slow and experience centered design comes closest, whereas persuasive technology and human-AI interaction score lowest. Factor and latent class analyses reveal two cultures of reflection, one working through form and the other through scaffolded thought, which only one system combines. Systems that let people set the pace score markedly higher, even with paradigm and form controlled (odds ratio 13.5). Hence, reflective computing is proposed as a paradigm whose aim is a human capacity rather than a system function, and which treats slowness as valuable only when people choose it.
[HC-4] GNN-VIEW: Visual Inspection and Explanation of Graph-based Earth-System Models NEURIPS2026
链接: https://arxiv.org/abs/2610.35693
作者: Alessandro Lucatelli,Ferdinand Unterhuber,Katharina Krenzer,Mennatallah El-Assady,Benedikt Soja,Christina Humer,Joel Oskarsson
类目: Machine Learning (cs.LG); Human-Computer Interaction (cs.HC); Atmospheric and Oceanic Physics (physics.ao-ph)
备注: Accepted to The 2nd Workshop on Advances in Representation Learning for Earth Observation at NeurIPS 2026
Abstract:Machine learning Earth-system models are becoming increasingly important, and so is their development. Graph-based models, one of the most prominent approaches, allow developers to constrain the learning process before training via graph connectivity. Yet, developing these models remains challenging: the graph is built algorithmically with little intuition about the final result beyond static plotting, which requires extra work and is often not detailed enough. Even after successful training, how the model arrives at forecasts remains obscure. We present GNN-VIEW, an interactive browser-based tool that supports the development of such systems. GNN-VIEW exposes the graph’s information pathways, allowing model developers to debug the graph structure before training. After training, GNN-VIEW uses attribution-based explainability to reveal which input features are used by the trained model for a given forecast. This helps developers understand why a model fails. We showcase the usefulness of GNN-VIEW in the construction stage of a graph for an ocean forecasting model, and in explaining poorly predicted precipitation of a trained weather forecasting model during Hurricane Laura.
[HC-5] Offline Language-Based Support System for Tactical Decision-Making in Air Combat Operations NEURIPS2025
链接: https://arxiv.org/abs/2610.35635
作者: Joao P. A. Dantas
类目: Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注: Accepted at the LatinX in AI Workshop @ NeurIPS 2025. 7 pages, 2 figures, 2 tables
Abstract:An offline language-based support tool is presented to assist pilot training and tactical decision-making in air combat operations. The system, named MATE (Mission Assistant for Training and Evaluation), applies retrieval-augmented generation (RAG) to answer natural language questions using content extracted from doctrinal and tactical manuals. Built entirely with open-source components, it performs text preprocessing, embedding generation, semantic search, and language generation locally, ensuring full offline functionality. The tool was evaluated by three Brazilian Air Force fighter pilots, who rated the answers produced with three open-weight language models (Llama 3.1 8B, Qwen 2.5 14B, and Mistral NeMo 12B) to five tactical questions derived from a publicly available air combat manual. Average ratings ranged from 4.00 to 4.40 on a five-point scale, with Mistral NeMo 12B obtaining the highest score, and the pilots rated the response time of the tool as fully adequate compared to manual search. The relevance of the retrieved content was identified as the main point for improvement. Potential applications are discussed in onboard mission systems, flight simulators, and training environments.
[HC-6] Revisiting Direct Control: A Field Evaluation of Sensor Fusion Efficacy in Teleoperating Full-Scale Autonomous Vehicles via Birds Eye View IROS
链接: https://arxiv.org/abs/2610.35630
作者: Stanisław Kuczma,Sławomir K. Tadeja,Michał R. Nowicki
类目: Robotics (cs.RO); Human-Computer Interaction (cs.HC)
备注: Accepted at 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
Abstract:In autonomous driving, robot teleoperation serves as a critical safety fallback mechanism, requiring operators to assume direct control in the event of system failure. While recent research has pivoted toward high-level supervisory control or simulation-based studies, these approaches often fail to capture the physical dynamics and sensory demands of driving a full-scale vehicle. To address this gap, we evaluated the impact of spatial data representation on operator performance using a real car-sized vehicle in a field experiment with N = 18 participants. We compared three visualization interfaces: (i) a standard camera feed baseline, with two bird’s-eye views (BEV) from (ii) a raw LiDAR point cloud, and (iii) a sensor fusion of LiDAR and camera data streams. Contrary to prior passive observation studies suggesting that sensor-fusion BEV may hinder awareness, our active-control results demonstrate that such representation significantly outperforms the standard LiDAR BEV in perceived usability ( p \leq 0.0001 ) and cognitive task load ( p = 0.01 , NASA-TLX score: 34.58/100) while also being significantly better than the Camera-only view ( p = 0.043 ). The fused interface also led to statistically significant reductions in task completion time ( p = 0.0461 ) and superior usability (SUS: 77.9/100) compared to those of the Camera-only and LiDAR-only views. By synthesizing objective performance data with comprehensive subjective assessments, we demonstrate that BEV fusion maximizes safety margins while minimizing the operator’s mental demand. These findings challenge existing teleoperation conventions and provide an empirically validated approach for designing robust remote piloting systems.
[HC-7] From Language to Drives: A Multimodal Framework for Aligning Autonomous Robots with Human Purposes
链接: https://arxiv.org/abs/2610.35629
作者: Diana Cerdas-Vargas,Alejandro Romero,Richard J. Duro
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:This work explores a multimodal approach to the alignment of the behavior of open-ended learning autonomous robots with human purposes through the use of large language models as cultural enablers. While autonomous robots can generate their own skills and goals through interaction with the environment, they often lack mechanisms to ensure that these remain relevant and aligned to those of human users. The proposed approach makes use of the general and cultural knowledge contained in LLMs and VLMs to establish a multimodal interactive alignment process in which human purposes, expressed in natural language, are transformed into motivational components of a robot architecture in the form of robot missions and drives that guide the behavior of the robot. The system is structured into three layers for semantic alignment, mission generation, and drive modeling. This design enables a bidirectional interaction loop: human inputs define high-level purposes, robots expand them into executable missions making use of multimodal information relayed by LLMs and VLMs, and feedback maintains alignment with human intent. Experiments in simulation and with a physical robot demonstrate that the process produces valid missions and behaviors, supporting effective human-agent alignment in interactive scenarios with autonomous robots.
[HC-8] High-resolution tactile observation makes hidden racket-control mechanisms learnable
链接: https://arxiv.org/abs/2610.35563
作者: Haoyang Tian,Zhengwei Li,Houcheng Li,Xu Yang,Peigeng Gu,Xinyue Wang,Tairan Zhu,Shuo Jiang,Long Cheng
类目: Machine Learning (cs.LG); Human-Computer Interaction (cs.HC); Robotics (cs.RO); Signal Processing (eess.SP)
备注: 58 pages, including 14 pages of main text and 44 pages of supplementary information. 5 main figures and 19 supplementary figures
Abstract:Experts can master fine hand-object manipulation skills, yet limited conscious perception of rapid tactile interactions hinders their ability to articulate the underlying control mechanisms. Badminton racket control requires rapid, fine hand-force coordination, providing a demanding testbed for uncovering these mechanisms. An 864-node tactile-inertial glove with synchronized 100-Hz acquisition recorded over 14,000 strokes from more than 70 participants across five skills. Inspired by sensorimotor control, our mechanics-grounded two-layer framework inferred object-driving and grip-stabilizing control from tactile history, while resolving their spatiotemporal force realization. We found that higher proficiency was associated with selective rather than extensive control: greater skill specificity, anticipatory regulation, focal distal modulation and lower geometric torque cancellation, while spatiotemporal coordination showed a non-monotonic acquire-refine pattern. We translated these mechanisms into personalized guidance targeting force control and force-motion coordination. In a randomized cohort, mechanism-guided learners gained 0.72 expert-rated skill levels versus 0.05 with level-only feedback, making control mechanisms actionable.
[HC-9] Beyond Being Accurate: Implications for Using Natural Language Descriptions for Accessible 3D Modeling with~Blind and Low Vision Makers
链接: https://arxiv.org/abs/2610.35562
作者: Gene S-H Kim,Nicholas Oskiper,Sean Follmer,Megan Hofmann
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:People who are blind or have low vision (BLV) are largely excluded from 3D modeling tools that rely on visual representations of geometry. Natural language descriptions offer a potential path to accessibility, but it is unclear whether language alone can support the deep spatial understanding that editing, not just browsing, requires. Advances in generative AI offer the promise of automatic model description, yet descriptions suited to BLV modelers are structurally distinct from alt-text and from existing 3D model captioning datasets, and the domain-specific requirements for generating them are not well understood. We address this through two coordinated studies: an autoethnographic pilot in which the blind first author worked with 12 sighted participants to develop a taxonomy of descriptive strategies, and a Wizard-of-Oz main study that applied those strategies as idealized scripts with 10 BLV participants. We identify a systematic use-to-editing gap: descriptions that support accurate but incomplete mental models are insufficient for specifying modifications, because editing requires knowing the state of features the user did not know to ask about. We show that this gap produces four failure types (accuracy, completeness, expression, and navigation) that accuracy-focused evaluation frameworks cannot detect. We propose a two-phase evaluation framework that diagnoses the failure type a description system exhibits and guides developers toward targeted design interventions, thereby providing a methodological standard for assessing whether description-based systems genuinely support BLV modelers as creators.
[HC-10] Visible Reasoning : User-Facing Decision Transparency for Generative AI Systems
链接: https://arxiv.org/abs/2610.35483
作者: Sanjay Krishna Anbalagan,Xinrui Nie,Anughna Kommalapati,Vijay Kumar Kanamarlapudi,Sreekanth Radhakrishnan,Xiaodan Zhao,Umesh Mohan
类目: Human-Computer Interaction (cs.HC)
备注: 18 pages, 5 figures. Original pre-review author manuscript; LaTeX source reconstructed from the author’s PDF. Published version: this https URL
Abstract:Across multiple waves of computing, public trust in new technology has followed the emergence of governance mechanisms that made dominant risks legible and manageable. Generative AI shifts this pattern: these systems synthesize across heterogeneous sources, and users increasingly experience them as reasoning systems. The central question changes from “did it perform correctly?” to “can we trust how it arrived at its conclusion?” This paper introduces Filtered Decision Logs (FDLs), a user-facing transparency mechanism that makes decision formation visible without exposing raw chain-of-thought. FDLs pair each response with (i) a concise decision rationale, (ii) influence-ordered supporting data points derived from tool outputs, and (iii) relevance indicators showing which sources most shaped the conclusion. To operationalize evidence selection, we propose a trace-based method that reconstructs observability telemetry into a provenance tree and ranks evidence using four deterministic semantic and structural signals - Final Answer Similarity, Average Relevancy, Persistence, and Structural Proximity - combined via adaptive weighted scoring. We demonstrate FDLs across three domains: career advising (5 tools), investment analysis (4 tools), and travel planning (4 tools). In a formative evaluation with 50 participants (25 students, 25 working professionals in AI development), 92% rated the influence-ranked evidence as helpful for understanding and trusting the system’s recommendation. Counterfactual tool-removal experiments confirm that removing the top-ranked source changes the answer more than removing a low-ranked source, validating that the ranking tracks decision dependence. We conclude that evidence-ranked transparency helps users identify misleading signals and missing context, enabling more confident acceptance, correction, or re-run of system decisions. Comments: 18 pages, 5 figures. Original pre-review author manuscript; LaTeX source reconstructed from the author’s PDF. Published version: this https URL Subjects: Human-Computer Interaction (cs.HC) Cite as: arXiv:2610.35483 [cs.HC] (or arXiv:2610.35483v1 [cs.HC] for this version) https://doi.org/10.48550/arXiv.2610.35483 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Journalreference: Artificial Intelligence in HCI, HCII 2026, Lecture Notes in Computer Science 16745, pp. 3-21 (2026) Related DOI: https://doi.org/10.1007/978-3-032-30849-8_1 Focus to learn more DOI(s) linking to related resources
[HC-11] Omni-InteractScope: Benchmarking Proactive and Agent ic Interaction in Continuous Omnimodal Streams
链接: https://arxiv.org/abs/2610.35439
作者: Yinsong Yan,Ling Wang,Yuxuan Wang,Yifan Yang,Ruiyang Xu,Junming Lin,Qi Chen,Meng Gao,Minghao Han,Yunfei Chu,Qize Yang,Xize Cheng,Jin Xu,Jibin Wu
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注:
Abstract:Omnimodal large language models (Omni-LLMs) are increasingly expected to assist users while observing continuous audiovisual streams. However, existing evaluations provide limited insight into how reliably these models connect stream evidence and user needs to timely, sustained interventions. We introduce \textbfOmni-InteractScope, a benchmark of 1,479 human-annotated, cross-reviewed interactions with 1,790 temporal targets across eight task groups. It links input modalities, event triggers, intervention types, and user needs, covering event-timed tool invocation and query-free assistance alongside notifications, warnings, state tracking, and continuous responses. Our experimental results on the proposed benchmark reveal three critical limitations of current Omni-LLMs: (1) \textbfTimely tool invocation remains unreliable: all 12 evaluated models produce an on-time first response for at most 40% of tool-invocation targets, and joint timing-and-tool accuracy reaches at most 32.0%. (2) \textbfQuery-free reminders remain poorly covered: all evaluated models achieve lower timely reference-target recall on query-free reminders than on explicit questions, measured on their respective target sets. Reminder recall never exceeds 23.0%. (3) \textbfEnd-to-end guidance coverage remains limited: nine models cover none of the 26 guidance sequences in full, and the remaining three cover only one to three sequences each. Targeted controls further show that tool-invocation scores alone do not establish event grounding and that reminder coverage depends on response history and frequency. These findings motivate developing and evaluating streaming assistants for timely action, initiative under general user needs, and coverage of entire response sequences.
[HC-12] What Will Replace Pull Requests? Rethinking Software Collaboration for AI Agents
链接: https://arxiv.org/abs/2610.34933
作者: Christoph Treude,Sebastian Baltes
类目: oftware Engineering (cs.SE); Human-Computer Interaction (cs.HC)
备注:
Abstract:The emergence of artificial intelligence (AI) coding agents has put pull requests under pressure: agents open millions of pull requests per month, while some open-source projects have begun to restrict or disable external contributions. We investigate which alternatives to pull requests practitioners are proposing or adopting and what these changes mean for empirical software engineering research. We reviewed 50 gray-literature documents from 31 organizations or individuals published between October 2025 and October 2026, coding the problems each named, the mechanisms it proposed or adopted, which of five roles of pull requests (integration, review, rationale, accountability, and external contribution) each mechanism takes over, and the evidence offered. Most responses retained pull requests in some form, whether by adding tooling, attaching a new artifact, or restricting access, and restriction was the response most often reported as adopted. Several proposed alternatives moved review and rationale into agent conversations and session records, but only Zed reported using a replacement in practice. Only two documents outlined a replacement for external contribution, and neither reported adoption. Pull requests are changing in ways that pull-request datasets will not capture: reviews conducted in agent conversations leave no pull-request comments, and projects that disable pull requests disappear from samples of newly opened ones. We outline a research agenda on the effectiveness of the emerging alternatives, their consequences for open-source participation, and the adaptation of empirical methods.
[HC-13] ECHOES: An Adaptive LLM -Powered AR Glasses Docent for Evidence-Grounded Urban Heritage Interpretation
链接: https://arxiv.org/abs/2610.34865
作者: Bang Xiao,Baosheng Jin,Jinhan Niu,Xingchen Zhang,Hua Shen
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Urban heritage tours are often delivered through fixed scripts, leaving limited room for visitors’ situated questions, language preferences, or requests for evidence. We present ECHOES, an adaptive LLM-powered AR glasses docent for evidence-grounded interpretation of historical buildings. Specifically, the agent adapts along three dimensions: it modifies response structure according to the type of visitor question (identification, architectural detail, evidence-seeking, or route inquiry); it accumulates session-level interest signals from repeated questions and expressed curiosity to shape later explanations; and it tracks language preference and retrieval context across turns to maintain conversational continuity. The core of ECHOES is an LLM agent initialized as a source-aware urban heritage docent that realizes this adaptation through dialogue and preference context maintained across the session. Around this agent, ECHOES integrates visual site grounding, a self-built historical knowledge base supporting retrieval-augmented generation, and concise multilingual AR output designed for glanceable use. Through a working prototype centered on Shanghai’s East Nanjing Road and the Bund, we demonstrate how an adaptive LLM-powered AR docent can transform historical tours from scripted narration into situated, question-responsive, and source-aware conversations.
[HC-14] he Impact of Personalized AI Response Suggestions on Content Creators Direct Messaging
链接: https://arxiv.org/abs/2610.34770
作者: Eun Jeong Kang,Jingruo Chen,Jadon Geathers,Susan Fussell
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: Group 2027
Abstract:Direct messaging is often employed by social media content creators to directly interact with followers. Those creators attempt to engage with followers in private communication, but they often encounter a large volume of messages. Through formative interviews with 15 creators, we found that creators frequently experience fatigue from writing suitable content, as they moderate audience expectations while concurrently integrating their media personas into responses. To explore how AI can address these challenges, we designed a system providing personalized, AI-generated response suggestions tailored to creators’ writing styles to maintain their media personas. In a comparison study with 13 creators that examines how the system affects response behaviors, creators valued the system’s ability to soften potential impoliteness and anticipated higher response rates, helping them maintain a professional image. However, creators reported additional labor in verifying the authenticity of AI outputs to mitigate risks caused by AI misalignment. Our results indicate that while AI suggestions enhance efficiency, they can undermine a creator’s engagement in communication by shifting their attention toward monitoring AI.
[HC-15] Babysitting AI Agents : A User-Centered Taxonomy and Study of Agent ic Delegation Failures
链接: https://arxiv.org/abs/2610.34748
作者: Shravika Mittal,Sunnie S. Y. Kim,Scott Counts,Jennifer Neville,Debarati Das,Philippe Laban
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Users are increasingly delegating tasks to AI agents, allowing agents to perform long-running, multi-step workflows on their behalf. However, agents may perform these tasks in ways that depart from users’ expectations for how they should be carried out, leading to what we call agentic delegation failures. We leverage a large corpus of 163,487 Reddit posts, describing AI agent use, to develop MANDATE: a taxonomy of 23 failure categories organized into seven groups. A follow-up study with 73 AI agent users examined the relationship between prevalence and perceptions of severity and noticeability, finding that failures rated as more severe are considered harder to notice, and that prior experience with a failure is associated with greater noticeability. Finally, we assess the maturity of research on each failure and identify commonly encountered failures that have received little attention (e.g., agents running without making progress). Our findings inform opportunities to make delegation easier for users to oversee and better studied by the research community.
[HC-16] Affective and Functional Needs in AI Use: User Patterns and Their Relation to Real-Life Support and Well-Being
链接: https://arxiv.org/abs/2610.34732
作者: Jessy Wang-Sun,Estelle Yiyuan Sun,Hualin Zhang,Zihan Xu
类目: Human-Computer Interaction (cs.HC)
备注: 8 pages, 2 figures, including supplementary material. Full paper accepted at the 14th International Conference on Affective Computing and Intelligent Interaction (ACII 2026), forthcoming in IEEE Xplore
Abstract:Conversational AI systems are increasingly used to fulfill both functional and affective needs, yet how users perceive the AI-adoption effects varies. Existing research tends to treat these two needs as binary categories and to classify users by a single dominant need. However, this approach fails to capture users who engage with AI for both affective and functional purposes simultaneously, leaving it unclear how these needs combine in real-world AI use. To capture the multifaceted nature of users’ AI engagement, we developed the AI-Support measurement with 1,090 participants, grounded in social support theory, which evaluates four dimensions of perceived AI support: emotional/esteem, social companionship, informational, and instrumental support. This study provided (1) an AI-Support Scale developed to measure the perceived support types from human-AI interaction, (2) evidence that users might adopt AI with a high level of both functional and affective use, and (3) exploratory indicators of users’ real-life social support and well-being, enabling descriptive results aligning with theoretical expectations. Overall, this study highlights the importance of understanding individual differences in AI affordance use, enabling AI systems to be designed in alignment with different user adoption profiles rather than assuming uniform benefits across all users.
[HC-17] he CAPED Framework: Characterizing User Skills in Chart Authoring through the Lens of Tasks and Tools IEEE-VIS2026
链接: https://arxiv.org/abs/2610.34652
作者: Bongshin Lee,Zhicheng Liu,John Thompson,Chenglong Wang
类目: Human-Computer Interaction (cs.HC)
备注: Accepted at IEEE VIS 2026, to appear in IEEE Transactions on Visualization and Computer Graphics (TVCG)
Abstract:In this paper, we introduce a new framework for characterizing the user skills required for chart authoring. Despite growing attention to authoring systems, we lack a conceptual framework that characterizes the diverse and multifaceted skills involved and helps identify opportunities for system design and development. Our initial investigation revealed that skill is a complex and elusive concept, and that existing theories on skills from other disciplines are too high-level to capture the importance of tasks and tools in data visualizations. To address this gap, we formulate the CAPED framework that characterizes skill as the ability to create a chart involving five dimensions pertinent to tasks and tools: input data, output chart, paradigm, abstraction, and expression. We demonstrate the framework’s descriptive power by highlighting different skills involved in similar authoring situations and identifying the types of knowledge associated with chart authoring. We further present examples illustrating how the framework can be used to identify new research opportunities in supporting users with varying skills.
[HC-18] Dream to Assist: Intent-Conditioned World Models for Shared Vehicle Control
链接: https://arxiv.org/abs/2610.34650
作者: Jonathan DeCastro,Thomas Balch,Emily Sumner,Deepak Gopinath,Laporsha Dees,Patricio Reyes Gomez,Elliot Weiss,Megan Applegate-Kenton,Todd Rowell,John Leonard,Avinash Balachandran,Guy Rosman
类目: Robotics (cs.RO); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注: 32 pages, 18 figures, 16 tables
Abstract:Assistive agents in shared-control settings face a dual challenge: acting proactively to align with the human’s intentions while remaining legible enough for the human to comprehend and fluently respond in real time. We present Dream2Assist, a model-based reinforcement learning framework for tightly-coupled shared decision making that addresses both sides of this challenge. Our approach extends a recurrent state-space world model with intent inference, enabling an assistive policy trained via fictitious co-play to anticipate and support discrete human objectives without overriding human agency. Beyond the computational framework, we describe the full deployment pipeline needed to translate a world-model-based policy into a safe, human-centered system in a resource-efficient way. Crucially, we address the sim-to-real-human gap via policy down-selection and reward restructuring within an iterative human-machine interface design. We validate the framework through three human-subjects studies and a preliminary on-vehicle deployment spanning two qualitatively distinct domains: high-speed racing (N=30) and urban driving with richer intent structure (N=10). Our results demonstrate that intent conditioning yields measurable improvements in torque alignment and collision rate. We surface key design insights and practices for deploying learning-based assistive agents in safety-critical, real-time human-robot teaming.
[HC-19] Rewriting the Self: User Perceptions of AI-Mediated Cognitive Reframing
链接: https://arxiv.org/abs/2610.34647
作者: Yuewen Yang,Olzhas Yessenbayev,Anna R. Van Meter,Tanzeem Choudhury,Daniel A. Adler
类目: Human-Computer Interaction (cs.HC)
备注: Data collected in summer 2024
Abstract:Individuals are increasingly using AI for psychological support. One specific paradigm, AI-mediated cognitive reframing (AIM-CR), uses AI to automatically reframe users’ negative thoughts (“I failed…”) into more adaptive perspectives (“I am learning…”). While prior work suggests that AIM-CR tools reduce users’ negative emotions, the complexities of AI automatically modifying users’ personal narratives remain underexplored. In this work, we investigated these complexities through a mixed-methods study in which 28 participants engaged with an AIM-CR tool that reframed narratives about their stressors. Our findings suggest that participants had mixed perceptions of AIM-CR: although reframing reduced negative emotions for most participants, participants felt less ownership of, and less authenticity in, the reframed narratives. Participants perceived AIM-CR as a validating “mirror,” as “another self” that could motivate or alienate them, or as an external social actor. These perceptions shaped therapeutic dynamics. We conclude by discussing design implications for future AIM-CR tools to better preserve user agency in narrative co-construction and more effectively support mental health and well-being.
[HC-20] oward Neuroadaptive Communication Interfaces: A Framework for Personalized Speech and Language Support
链接: https://arxiv.org/abs/2610.34457
作者: Carlos Bruno,Luis Ricardo Santos
类目: Human-Computer Interaction (cs.HC); Computation and Language (cs.CL)
备注:
Abstract:Speech and language technologies have expanded access to communication, but the support they provide is often fixed. The same linguistic or visual aid may help one user and hinder another, or become less useful as task demands change. This project proposes a framework for adapting language presentation to individual abilities and ongoing interaction. It separates access to a message from the estimation of user needs and from the selection of support, including sentence segmentation, pictograms, and optional spoken output. Behavioral feedback provides the primary evidence for adaptation; electroencephalography is considered an additional, uncertainty-aware source of information rather than a direct measure of comprehension. We specify a staged evaluation comparing fixed, behavior-adaptive, and neuroadaptive interfaces. The central question is whether neural information improves comprehension and communication efficiency beyond a strong behavioral baseline.
[HC-21] Perceived Acceptability of Sensory Augmentation
链接: https://arxiv.org/abs/2610.34451
作者: Michel Witter,Alwin de Rooij,Michel van Dartel,Emiel Krahmer
类目: Human-Computer Interaction (cs.HC)
备注: 16 pages, 5 figures, 2 tables
Abstract:Sensory augmentation (SA) can extend how people perceive and engage with public environments through the body, offering opportunities to support shared multimodal experiences between deaf and hearing visitors. Designing such systems requires understanding which combinations of tactile stimuli and body locations users perceive as acceptable. We examined the perceived acceptability of fictitious yet feasible SA configurations in a museum use-case. Using a modified technology acceptance model, we surveyed 307 deaf and hearing participants about tactile stimulus types and body locations. The results showed significant effects of stimulus type and body location on acceptance, but no interactions with hearing status. Participants generally preferred temperature-based stimuli, subtle movement, texture, and tickling over vibration, and distal over proximal body locations. These findings suggest that tactile SA design should be guided primarily by embodied and contextual factors rather than hearing status, informing future systems for shared multimodal experience in public cultural settings.
[HC-22] Developer Task Delegation to LLM s and the Trust Factor: A Mixed Methods Study ICSE’27
链接: https://arxiv.org/abs/2610.34317
作者: Samuel Ferino,Rashina Hoda,John Grundy,Christoph Treude,Hashini Gunatilake
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: Accepted for publication in the Proceedings of the 49th IEEE/ACM International Conference on Software Engineering (ICSE’27)
Abstract:With the rise of Large Language Models (LLMs), software developers are presented with a powerful new approach: delegating software development tasks to LLMs. Existing studies have explored task delegation for a subset of tasks and factors. Critically, what tasks developers delegate to LLMs, to what extent, and why they delegate them, is not yet well understood. We explore these questions through 22 interviews and a survey with 160 software practitioners. Developers reported employing four modes of task delegation to LLMs: full delegation, partial delegation, self-delegation, and task avoidance. Drawing on our qualitative findings, we discuss the rationales behind these choices and explore how task difficulty, task risk, developer motivation, and trust appear to be associated with software development task delegation to LLMs. Our statistical analysis reveals that trust is a significant factor associated with task delegation to LLMs. Finally, we share actionable implications for human-AI collaboration in software research, and for software practitioners to make the most of their newly found powers of delegation.
[HC-23] Can LLM s Trust Appropriately? Evaluating Calibrated Reliance on Imperfect Partners
链接: https://arxiv.org/abs/2610.34107
作者: Mengdi Chu,Jintao Huang
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注:
Abstract:Large language models increasingly collaborate with humans and peer AI agents whose contributions may be incomplete, inconsistent, or incorrect. Effective collaboration requires calibrated trust, whereby decisions to rely on, verify, or assist a partner adapt rationally as empirical evidence accumulates. We investigate whether models translate partner evidence into appropriate reliance policies. We introduce a benchmark of 324 scenarios across 23 collaboration categories. Each scenario involves three sequential decisions with explicit transition dynamics, feedback emissions, and cost schedules, permitting exact analytical evaluation of choices against their expected consequences over the remaining interaction. Evaluating seven models alongside 36 controlled-history probes, we dissociate evidence sensitivity from normatively appropriate action. While the strongest frontier model approaches near-optimal decisions, other models exhibit recurring pathologies: procuring excessive safeguards, purchasing verification tests with zero expected value of information, prioritizing immediate savings over cooperative learning, and misallocating assistance to suboptimal channels. Supplying explicit posterior distributions resolves certain errors while introducing new failure modes. Calibrated trust in autonomous agents critically demands coupling empirical evidence with forward-looking decision utility.
[HC-24] Detecting and Understanding Ambiguity in Visualizations with Role-Playing Vision Language Model Interpretations
链接: https://arxiv.org/abs/2610.33997
作者: Garrick Chan,Yun-Hsin Kuo,Sam Yu-Te Lee,Yinuo Tang,Kwan-Liu Ma
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Communicative visualizations can support multiple chart-grounded takeaways, including divergent or conflicting ones; we refer to this variation as interpretive ambiguity. Such ambiguity can undermine a visualization’s communicative goal when the supported takeaways diverge from its intended message. Evaluating visualizations with target audiences can surface this variation, but such evaluations can be costly and difficult to scale. In this work, we explore persona-conditioned Vision Language Models (VLMs) as a complementary evaluation tool for interpretive ambiguity. For a given visualization, our pipeline uses topic-relevant personas to elicit candidate contextual takeaways supported by interpretations of perceived visual patterns, such as trends and clusters. We derive a Target–Pattern–Scope framework for comparing these interpretations, use LLM judges to assign Compatible, Disjoint, or Opposing ambiguity labels to statement pairs, and summarize their distribution in an ambiguity profile. To improve chart grounding, we benchmark candidate VLMs on chart comprehension and select the highest-performing model for each chart type. We evaluate statement-pair labeling against human-authored annotations and apply the approach to real-world, public-facing visualizations curated from established datasets. Through case studies, we examine how perceived visual patterns relate to contextual takeaways and how a visualization redesign changes the sampled interpretations. Our findings show how persona-conditioned VLMs can complement human feedback by generating candidate contextual takeaways and surfacing potential ambiguity for further inspection.
[HC-25] A spinal-inspired spiking neural network for continuous finger-joint kinematics estimation from high-density surface EMG
链接: https://arxiv.org/abs/2610.33979
作者: Cristian David Guerrero-Mendez,Leonardo Abdala Elias
类目: Human-Computer Interaction (cs.HC)
备注: 32 pages (25 pages of main text and 7 pages of supplementary material), 9 figures, 1 table
Abstract:Decoding simultaneous and proportional finger kinematics from high-density surface electromyography (HD sEMG) remains challenging. Although data-driven methods have shown promising performance for finger movement decoding, their limited physiological interpretability remains a relevant limitation. This study proposes and evaluates a spiking neural network (SNN) that incorporates spinal-inspired structures and mechanisms to predict multi-joint finger kinematics from HD sEMG signals. The spinal-inspired SNN includes a spike-encoding computational model with input-noise variability, a phenomenological representation of Henneman’s size principle, and excitation/inhibition (E/I) balance. HD sEMG and kinematic data were collected from 23 healthy participants performing grasping tasks under different elbow and wrist postures. Performance was assessed using six prediction metrics and compared against six baseline machine learning and deep learning methods. The proposed method showed a large and significant ( p0.05 , rank-biserial correlation r_rb0.7 ) improvement over all baseline methods across all accuracy metrics. Performance improved significantly with increasing spike encoder firing rates up to approximately 128 Hz, and with more regular input spike trains compared with a Poisson-like encoder. In contrast, the E/I balance did not significantly affect the decoding performance of the algorithm. The firing rate and inter-spike interval statistics of the SNN differed significantly across motor tasks. These results demonstrate that incorporating neurophysiological principles into SNNs improves prediction performance while providing additional physiological interpretability for simultaneous and proportional finger-kinematics decoding. This approach represents a promising step toward physiology-informed neural networks, with potential applications in motor rehabilitation.
[HC-26] InteractiveWorld: A Verifiable Human-Centric Environment for Building Better Interactive MLLM Assistants
链接: https://arxiv.org/abs/2610.33966
作者: Qinsi Wang,Wei Wen,Chong Zhou,Saksham Suri,Zechun Liu,Lemeng Wu,Hai Helen Li,Yiran Chen,Raghuraman Krishnamoorthi,Qi Qian
类目: Human-Computer Interaction (cs.HC); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:People have long dreamed of an assistant that is always there to guide them and remind what they forget, from new dishes to lost keys. Powered by multimodal large language models (MLLMs), egocentric AI assistants on wearable devices are bringing this dream within reach: they see what the user sees and give instructions for user’s physical work when necessary. However, testing such assistance is challenging and usually requires real people. The expensive and rare real interaction data leaves a key question unanswered: Can today’s MLLMs truly help people in physical world? To answer this question, we propose InteractiveWorld, the first verifiable environment for closed-loop interaction between human and AI assistant. Unlike existing robot-centric simulators, InteractiveWorld lets the assistant guide a human agent who observes the scene, understands instructions, and acts independently. It thus realistically simulates how AI assistant collaborates with human: the assistant instructs, the human acts, and the environment executes actions and updates what they see next. Building on it, we introduce InteractiveBench, the first dynamic benchmark for egocentric AI assistants, with 120 multi-turn tasks across six categories. In each task, the assistant guides the human in real time, and every instruction shapes the outcome, which is scored by verifiable, state-based rewards. Experiments with 12 frontier MLLMs show they are still far from reliable assistants. Even the strongest model, Claude Fable 5.1, achieves at most 20% and 25% on long-horizon Guidance and Planning tasks, respectively. Overall, our work offers a verifiable platform and a rigorous evaluation set, laying a solid foundation for research on interactive MLLM assistants that truly help people in the physical world.
[HC-27] Human-Aware Agent ic Driving: Rethinking Driver State Trust and Intervention
链接: https://arxiv.org/abs/2610.33915
作者: Lu Feng
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Agentic driving brings goal-directed reasoning, memory, tools, and adaptive interaction into automated vehicles. A vehicle can clarify a request, explain a maneuver, or prepare a driver to take control, then adapt its next action to the response. These capabilities make the vehicle an active participant in shaping the human behavior it observes. We examine how agentic interaction affects driver-state inference, trust and reliance, and human intervention in conditional automation. Established findings on attention, takeover prediction, trust-aware planning, and adaptive warnings provide useful foundations, but their applicability depends on how observations are elicited and how assistance changes the driving task. We distinguish what remains applicable, what changes under flexible interaction, and what remains unresolved across these three areas. The analysis identifies three research priorities: establishing the effects of interventions, jointly designing interaction and driving policies, and maintaining useful adaptation across repeated use and system changes. Progress should be measured through driver comprehension, appropriate reliance, and the ability to perform the required driving action in time.
[HC-28] FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs
链接: https://arxiv.org/abs/2610.33876
作者: Sadra Sabouri,Hosein Hasanbeig,Adam Sage,Shreya Sinha,Cale Darling,Ashish Tiwari,Katie Garms,Souti Chattopadhyay,Sumit Gulwani
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Many agent tasks recur with new data or small changes, yet agents regenerate their reasoning and actions on each run, producing different results. Distilling a prior run into a skill helps the agent avoid earlier mistakes, but it still re-runs the full process each time, so runs can vary and incur that cost. We compile a successful agent run into a reusable neuro-symbolic program (NSP) with a set of guidelines we call FlashScript. Actions and data transformations become code, instance-specific values become parameters, and decisions that need model judgment become LLM calls. We identify three layers where these calls belong and three techniques that replace most of them with code. The compilation does not remove nondeterminism; it localizes it, so the deterministic program and its neural calls can be reviewed and tested separately. Across 120 WorkBench analytics tasks with Claude Haiku 4.5, the NSPs were correct and made identical tool calls on every run, using about 70 times fewer input tokens and a third of the runtime of a skill-guided agent whose runs were inconsistent. In a composite version of the tasks, the skill-guided agent’s accuracy fell to 23%, and its output inconsistency increased, while the NSPs remained correct and identical across all runs. The NSPs successfully generalized to all 108 unseen instances of the same templates. Our findings suggest that NSPs from agent trajectories can be considered cumulative software assets for end-users to test, maintain, and reuse in their recurring workflows.
[HC-29] EEGBiFormer: Real-Time EEG Attention Classification via Hybrid CNN-Transformer with EEG-Informed Auditory Protocol Validation
链接: https://arxiv.org/abs/2610.33832
作者: Isabella Guan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: Accepted by IEEE MIT URTC 2026
Abstract:Sustained attention is critical for academic performance, yet existing auditory support tools are static and not grounded in real-time neural evidence. This paper presents EEGBiFormer, a hybrid CNN-Transformer model for realtime binary classification of attention states, focus versus nonfocus, from 16-channel EEG signals. A custom analog front-end acquisition board was designed and built to collect a task-based EEG dataset. Before model training, spectral and connectivity analysis, using Power Spectral Density and Phase Locking Value, confirmed distinct neural signatures between attention states, providing physiological grounding for the classification approach. EEGBiFormer achieved 91.3 percent window-level validation accuracy with an F1 score of 94.5 percent, and 88.2 percent recording-level accuracy across 17 independent sessions. End-to-end system latency was measured below 100 ms, supporting real-time responsiveness. The trained model was further used to objectively rank auditory stimuli by focus-state confidence, yielding EEG-informed audio protocols. In a paired within-subject behavioral study with 37 participants, these protocols produced a statistically significant 20 percent improvement in composite attention scores over silent controls, with a medium-to-large effect size. This work demonstrates a complete pipeline from custom EEG hardware through neural signal analysis, deep learning classification, and evidence-based auditory protocol design.
[HC-30] A Personalized and Adaptive Intelligent Tutoring System for Psychomotor Learning
链接: https://arxiv.org/abs/2610.33467
作者: Madeleine Shuhn-Tsuan Yuh,Neera Jain
类目: ystems and Control (eess.SY); Human-Computer Interaction (cs.HC)
备注: 13 pages, 6 figures
Abstract:Intelligent tutoring systems (ITSs) have advanced learner motivation and personalization in cognitive domains, but extending these gains to psychomotor tasks remains challenging. Existing psychomotor ITSs rely on learner models built before instruction or updated once per module, leaving open how to use real-time assessments of learner cognitive state for finer-grained personalization. Psychomotor learning involves multiple tutoring objectives, including calibrating self-confidence to performance, calibrating workload per the Yerkes-Dodson Law, and driving skill progression toward higher learning stages. How to balance these objectives for learners, and adjust that balance over time, remains open. This study proposes and evaluates an online personalization algorithm that continuously reprioritizes tutoring objectives to match learners’ evolving needs. Rather than weighting objectives equally, the actor-critic-inspired algorithm dynamically updates reward weights using recent measures of self-confidence, workload, and performance. These weights inform a multi-input, multi-output cognitive control framework that synthesizes learner-specific tutoring policies via model-based planning. The approach was evaluated in a human-subject study (N = 40) with a 2D quadrotor landing task. Clustering of the learned reward weights revealed distinct prioritization patterns across learners, indicating the algorithm captures heterogeneity in learner needs. Learners whose weights shifted more substantially tended to reach higher final performance, and those whose weights consistently emphasized workload calibration achieved the strongest outcomes. These findings advance online learner modeling and control-based personalization in psychomotor ITSs, showing how adaptive tutoring can better support learning by aligning instructional decisions with learners’ evolving cognitive and performance states.
[HC-31] he As-If Machine: Personalized counterfactuals make distant policy risks feel close
链接: https://arxiv.org/abs/2610.33301
作者: Joshua Ashkinaze,Yubo Zhou,Pengyu Chen,Avi Bhattacharya,Hyebin Park,Yanman Zhang,Stephanie Preston,Eric Gilbert,Ceren Budak
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Life experiences can shape policy attitudes, but we only live one life. Citizens may therefore fail to support policies addressing risks they have never personally experienced. Narratives can make such risks immediate and personal, but personalized stories remain difficult to produce at scale. Here we ask whether AI can make distant scenarios feel concrete by simulating a participant’s life as if that scenario occurred. We present the As-If Machine (AIM), a multi-agent system through which people co-write personalized counterfactual narratives—stories about their own lives under different policy futures. In two pre-registered and quota-matched experiments on science funding (N=1,007) and satellite cybersecurity (N=894), we compared AIM with two controls: reading an expert-written vignette on the same topic and co-writing with AIM about an unrelated topic. AIM increased psychological closeness—how close a risk feels to the self—by nearly identical margins in the two domains (0.36 and 0.38 standard deviations (SDs) against the vignette, 0.80 and 0.90 against unrelated writing). AIM increased psychological closeness even among participants with unfavorable views of AI. For policy attitudes, AIM’s effects were 0.16 SDs against the vignette and 0.10 SDs (n.s.) against unrelated writing for science funding; the corresponding effects for satellite cybersecurity were 0.54 and 0.67 SDs. Across 23,496 human-AI autocomplete interactions, participants and the system played distinct roles: AIM supplied concrete detail that grounded the scenario, and participants supplied first-person perspective and emotion. Co-writing with the As-If Machine makes distant policy risks feel close.
[HC-32] Whale You Go Green? Immediate and Longitudinal Effects of Embodied Conversational AI Avatars on Sustainable Behavioral Tendencies and Pro-Environmental Attitudes
链接: https://arxiv.org/abs/2610.33191
作者: Rachel Poonsiriwong,Pat Pataranutaporn
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Embodied conversational AI can be persuasive, but can it motivate people to develop sustainable behavioral tendencies and increase pro-environmental attitudes? We conducted a randomized experiment (N = 576) comparing four embodied avatars–a beluga whale, Greta Thunberg, Emma Watson, and a generic human–with a text chatbot and no-intervention control. All embodied conditions increased immediate sustainable product selection relative to the control (beta = .401-.609, all q = .0126) and chatbot (beta = .398-.604, all q = .007), without differing significantly from one another. Only the Beluga condition increased immediate pro-environmental attitudes relative to the control after correction. Among 287 returners assessed two months later, following an intermediate assigned-condition session, Beluga also showed significantly higher pre-booster sustainable product choice compared to the control (beta = .657, q = .0176). Although it did not significantly outperform other embodied agents, these findings establish the immediate potential of nonhuman embodied avatar interfaces as a promising direction for memorable environmental interventions.
[HC-33] GlanceAR: Repurposing Phone Sensors via Low-Cost Reflectors for Always-On Obstacle Awareness
链接: https://arxiv.org/abs/2610.33142
作者: Boyang Zhou,Xizhe Hao,Minbeom Kim,Jaewook Lee,Jon E. Froehlich
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Smartphone use while walking reduces environmental awareness and can lead to accidents and injury, yet existing obstacle-awareness systems remain unadopted because they require users to hold the phone upright and aimed forward, conflicting with how people actually hold their devices. We introduce GlanceAR, an always-on mobile obstacle-awareness system designed for natural phone-holding postures. GlanceAR utilizes a low-cost passive mirror attachment to redirect the phone’s rear LiDAR and RGB camera forward without additional electronic components. In a within-subjects study with N=11 participants navigating indoor and outdoor obstacle courses while engaged in a primary on-phone task, we evaluated three sensing configurations: GlanceAR, a LiDAR-only variant for privacy, and an attachment-free ultra-wide camera baseline. Our findings reveal trade-offs among sensing fidelity, obstacle cue design, and social acceptability, and surface insights around AI complacency and the tension between on-screen task focus and physical safety.
[HC-34] Detection of false positives by human reviewers depends on the face recognition algorithm and its security configuration
链接: https://arxiv.org/abs/2610.33045
作者: Esha Edakoth,John J. Howard,Isabelle M. Shuggi,Yevgeniy B. Sirotin,Robert Wood
类目: Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注: 26 pages, 8 figures, 4 tables
Abstract:Integration of artificial intelligence (AI) technology into identity workflows, such as automated face recognition (FR), has enhanced operational capabilities, with FR algorithms achieving super-human scale and accuracy. However, these gains can make human oversight of FR more difficult, raising questions about the effectiveness of these human-AI teams. We measured human reviewers’ ability to detect false positive errors made by commercial FR algorithms across varying security configurations. Using a novel dataset, the Difficult Facial Image Comparison Test (DFICT), derived from false positives made by four comparable commercial face recognition algorithms, we measured the ability of 801human reviewers to detect algorithmic false positives across three security configurations, representing operationally relevant biometric thresholds. Our findings show that human reviewers can reduce algorithmic false positives, but performance declines at stricter security thresholds. Further, reviewer performance varied by algorithm indicating that some algorithms are more compatible with human oversight. We suggest that biometric systems should be designed to account for human reviewers’ variable effectiveness, particularly in high-security applications. Our findings should be considered during algorithm selection and inform human oversight strategies to minimize AI-related risks in operational FR deployments.
[HC-35] New Lenses on Learning Design: Course Types and Indicators from Large-Scale Moodle Data
链接: https://arxiv.org/abs/2610.32983
作者: Ekaterina Soroka,Niels Seidel,Marcus Specht,Kamila Misiejuk
类目: Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注:
Abstract:Although learning design (LD) plays a central role in shaping online courses, large-scale empirical evidence on how courses are actually structured remains scarce, and existing analytics work has characterised courses through student engagement with specific learning activities rather than through indicators of the design itself. This study addresses both gaps through an analysis of learning management system (LMS) data from 606 courses at a distance university. Using LMS data, we first applied dimensionality reduction and clustering to a structural variable set, from which three distinct LD types emerged empirically: \textitCompact, \textitExpansive, and \textitInteractive and statistically characterised how their design elements differ across disciplines, activity load, and interfaces. Building on this typology, we derived three LD indicators, \textitactivity-mode balance, \textitstructural clarity, and \textitanalytics readiness, each combining multiple LMS-derived variables into higher-order constructs that target a different aspect of LD. All three indicators differentiated significantly across the LD types and added pedagogically meaningful dimensions for interpretation. These contributions advance the operationalisation of LD in technology-enhanced learning research by demonstrating that higher-order LD indicators can be derived directly from routinely collected LMS data, and offer empirical insight into how LD manifests at scale within a large distance university.
计算机视觉
[CV-0] VibeVideoBench:Evaluating Multimodal Agents on Code-Driven Video Generation
链接: https://arxiv.org/abs/2610.35864
作者: Muxi Diao,Wenbin Xu,Lele Yang,Xuyang Ma,Yuzhe Chen,Jiankang Chen,Kaiyu Jiang,Changyi Liu,Haoyu Zhang,Hongyi Fu,Jianxiong Wang,Tianke Zhang,Haonan Fan,Yingxin Li,Bingyi Li,Xu Wang,Xuanyu Zheng,Bin Wen,Tingting Gao,Han Li,Wentao Zhang,Weiran Xu,Kongming Liang,Zhanyu Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Diffusion video models produce striking clips, but they cannot reliably render an exact formula, a correct chart, a ruleconsistent game or narration synchronized to on-screen events. A different route is emerging: multimodal agents that write code to render every frame and synthesize every sound. We introduce VibeVideoBench, a benchmark for evaluating agents on code-driven video generation. Its 239 self-contained tasks in English and Chinese span 12 categories, from data stories and music-driven pieces to classroom lessons and paper explainers; each asks an agent, running in its native harness inside a sandbox, to deliver a finished MP4 with sound within one hour. Evaluation is hierarchical and evidence-based: program checks of the delivery spec, about 13k task-specific rubric checks traced to the requirements they enforce, and five quality dimensions anchored per task, judged by three VLM judge panels and validated with blind human preferences. Across nine frontier agent systems and 2,151 scored rollouts, the benchmark is far from solved: GPT-6 Astra and Claude Opus 5.5 lead clearly, yet even the best system fails a core requirement on about a quarter of tasks. Whether a video arrives at all matters as much as its quality, and three quarters of the lost videos are final renders started too late, run too slowly or never started. The judges agree with each other and with human preferences on system rankings, but only moderately on individual videos.
[CV-1] Do vision models know that three is three?
链接: https://arxiv.org/abs/2610.35857
作者: Ankit Sonthalia,Arnas Uselis,Seong Joon Oh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Do vision models encode the transition 2\to3 the same way whether it appears in handwritten digits or in hand gestures, and do they “understand” that clean \to dirty is the same transformation regardless of whether it is applied to a kitchen or a carpet? The linear representation hypothesis (LRH) answers this for language models: concept changes correspond to directions in (language) embedding space. In vision, the same concept can manifest in visually unrelated “regimes” that rarely co-occur: images of digits and hands do not share much in common in pixel space. To understand whether vision models still encode the underlying abstract concepts, we study 46 vision encoders across architectures, training objectives, and scales, on 10 concept families including numerosity, age, object identity, and MIT-States attributes. Within each regime, we find statistically significant concept directions near-universally across our tested models. However, two distinct regimes do not always share the same concept direction. Several models learn multiple linear structures for the same concept transition, split across regimes. We call this phenomenon “fragmented linearity”: strong local linearity coexisting with near-absent global linearity. Fragmentation is, however, not fixed: models with stronger local linearity progressively escape it (local-global linearity correlated with \tau \in [0.38, 0.89] across all concepts), and where global linearity does emerge, concept directions can transfer as linear probes across regimes.
[CV-2] FL4N: Feed-forward Latent Model for 4D Novel View Synthesis
链接: https://arxiv.org/abs/2610.35854
作者: Eunbeen Hong,Hyeonseo Yu,Honggyu An,Jooho Son,Hyunah Ko,Soowon Son,Chaehyun Kim,Jahyeok Koo,Jisoo Kim,Junhwa Hur,Seungryong Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page and code are available at this https URL
Abstract:4D novel view synthesis (NVS) aims to render dynamic scenes across viewpoints and timestamps, which requires jointly inferring 3D structure and temporal dynamics. Existing monocular feed-forward methods often rely on explicit 4D Gaussian representations, which are prone to temporal inconsistency and severe rendering artifacts. We present FL4N, the first feed-forward latent 4D NVS model. Instead of relying on explicit 4D primitives, FL4N encodes a 4D scene prior into a unified latent representation across space and time. Our core insight is that learning temporal dynamics as residuals on top of a pretrained static latent prior enables effective 4D modeling with limited dynamic training data. To this end, we introduce a lightweight dynamic encoder that leverages static scene features to disentangle object motion from camera-induced parallax. Trained solely with an image synthesis objective, FL4N achieves state-of-the-art 4D NVS quality while rendering in real time. Its learned latent representation generalizes to real-world videos and disentangles viewpoint from time.
[CV-3] OSWAM: An Effective World Action Model with One-Step Action Generation
链接: https://arxiv.org/abs/2610.35850
作者: Guowen Zhang,Liyi Chen,Xu Hu,Pengfei Wang,Jingfan Chen,Chenhang He,Lei Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 18 pages, 6 figures
Abstract:Direct-policy World Action Models (WAM) have demonstrated the benefits of decoupling action generation from explicit future visual prediction for efficient inference in autonomous driving and embodied control. However, they still require tens of denoising steps to achieve competitive action quality, incurring substantial latency and hindering real-time deployment. Simply reducing the number of denoising steps will collapse the nonlinear multi-step denoising trajectory into coarse updates, substantially degrading action quality. To reduce the denoising steps while enabling real-time action generation, we present an effective world action model with one-step action generation, namely OSWAM, for autonomous driving and embodied control. We argue that it is the absence of structured action priors in random noise initialization that forces direct-policy WAMs to recover coherent action sequences through iterative denoising. Hence, we construct imagination tokens to initialize the diffusion process, providing physical priors and action-conditioned representations. We further introduce privileged information distillation with future conditioning to enhance future-aware action generation. As a result, OSWAM can efficiently and effectively generate high quality actions in just one diffusion step. Our experiments demonstrate that OSWAM achieves comparable or even better results in both autonomous driving and embodied control than direct-policy WAM and Joint-WAM that require tens of steps. It obtains approximately 3\times speedup over direct-policy WAM baselines and supports real-time compiled inference with latency as low as 40ms on an NVIDIA RTX 5090 GPU. The project page is available at this https URL
[CV-4] Seeing Through Distractions: Context-Stable Attribution via the Least Core
链接: https://arxiv.org/abs/2610.35827
作者: Sai Ganesh Nagarajan,Toshinori Yamauchi,Hiroshi Kera
类目: Computer Vision and Pattern Recognition (cs.CV); Computer Science and Game Theory (cs.GT)
备注: 22 pages, 6 figures, 3 tables
Abstract:Shapley-value-based explanations are widely used for feature attribution in machine learning. However, in computer vision, a feature may reduce the target-class score when added to the complete genuine image, even when the predicted label remains unchanged. We formalize such features as contextual distractors and investigate their attributions under semivalues and the least core. Under a weak average marginal monotonicity assumption, we derive quantitative semivalue–least-core gap bounds and sufficient conditions based on the probability of sampling smaller contexts. We further show that shifting semivalue weights toward larger contexts decreases the signed gap. Finally, controlled image-classification experiments and evaluations on Salient ImageNet provide strong empirical support for our theoretical findings.
[CV-5] Back to Basics: A Skill-Centric Approach to Generalist Web Agents
链接: https://arxiv.org/abs/2610.35823
作者: Minh Pham-Dinh,Md Farhan Ishmam,Hernan Yupanqui,Reaz Hassan Joarder,Kenneth Marino
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Browser agents do not just need more practice. They need to practice the right things. We introduce MiniWeb, a realistic miniature of the web comprising 65 interconnected websites across 15 domains, where agents navigate, search, and transact as people do. Its 369 human-annotated tasks cover everyday workflows and label 37 reusable interaction primitives grouped into 7 skill families, allowing us to diagnose which skills fail rather than only whether a task succeeds. Benchmarking ten open-weight agents, we find that drag gesture is consistently the weakest family, with no agent solving more than 24% of its instances. To address this gap, we introduce MiniWebGen, a framework that autonomously generates verified, skill-centric training data from MiniWeb. Fine-tuning Qwen3.5-4B on just 3.1K verified trajectories improves single-primitive success by 24.9pp across 13 unseen websites, with gains across every skill family. However, these trajectories teach short-horizon interactions, which do not directly translate to full-task success. We build on this result with MiniWebAgent, which pairs the trained executor with a planner that decomposes full tasks into the skill-level steps it learned to perform. This approach improves task success by 19.4pp points on unseen MiniWeb sites and transfers to external benchmarks, with gains of 9.3pp on WebArena-Lite, 16.1pp on Online-Mind2Web, and 20.0pp on WebVoyager, without training on their trajectories. We provide a simple recipe for compact browser agents: practice reusable skills directly, then compose them to solve unfamiliar tasks.
[CV-6] Pretraining Knows What Simulation Teaches How: Compositional Generalization in a Sim-Trained Robot Agent
链接: https://arxiv.org/abs/2610.35817
作者: Benjamin Pikus,Daniel Saunders,Mircea Mironenco,Salman Mohammadi,Mathew Mathew,Sirou Zhu,Divya Makkar,Xuhua Huang,Qinbo Li,Supreeth Narasimhaswamy,Siyuan Cen,Thomas Neff,Dejia Xu,Shyamal Buch
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:A robot policy needs to understand what a task is asking for in a scene, and how to move the robot to accomplish that task. To achieve these two skills, we introduce Zero-real Agent Policy (ZAP), a recipe that teaches a small off-the-shelf vision-language model to control a robot via a simple tool harness. We use a pretrained 27B parameter model to bootstrap world understanding and reasoning, followed by supervised fine-tuning and GRPO on simple simulated tasks to teach it how to move the robot. Without any real-world robot data during training, ZAP reaches 96.7% success on real-world tasks, matching GPT-6 Astra in our harness and outperforming the MolmoAct2 VLA (33.3%), while running 3x faster and 17x cheaper than GPT-6 Astra. We demonstrate that ZAP also generalizes to completely new objects, goals, strategies, and language by seating a test tube in a rack (and choosing a different hole in the rack when prompted), handing a test tube between arms mid-air, drawing an X on a tic-tac-toe board, and playing chess. These goals and types of manipulation don’t appear in our training data. Furthermore, ZAP also transfers to an unseen embodiment. On a single-arm Franka, ZAP reaches 55.9% on RoboLab-120, 4.9 percentage points ahead of the leading VLA/WAM baseline.
[CV-7] Multimodal Concert Video Editing: System Design and Exploratory Evaluation
链接: https://arxiv.org/abs/2610.35816
作者: Lennart Schiweck,Adrian Kempf,Michelle Retig,Moritz Feucht,Tim Mangliers,Benjamin Himpel
类目: Multimedia (cs.MM); Computer Vision and Pattern Recognition (cs.CV); Sound (cs.SD)
备注: 19 pages, 3 figures
Abstract:Editing multi-camera concerts requires coordinated decisions about musical relevance, visual quality and cut timing. We present a modular framework that combines instrument activity from multitrack audio, visible instruments and image sharpness with beat-based timing and probabilistic camera selection. An interpretable scoring model links these components to the choice of views over complete shot intervals. Component tests and an exploratory expert study on a four-camera concert yield mixed assessments across the editing variants. The findings motivate extending local audiovisual relevance with contextual and sequence-level criteria. The architecture provides a basis for testing these extensions and studying how component behaviour affects the quality of complete edits.
[CV-8] Does VGGT Know What It Does Not Know?
链接: https://arxiv.org/abs/2610.35806
作者: Adityan Thirumani Srinivasan,David Nordström
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 9 pages, 6 figures, 7 tables
Abstract:Feed-forward 3D reconstruction models such as VGGT regress cameras and geometry in a single forward pass, but they report no uncertainty for the camera poses they predict: an incorrect relative pose looks no different from a correct one. We ask whether the frozen model nonetheless knows when its pose is wrong. On HardMatch, a benchmark of 1,000 deliberately difficult image pairs on which VGGT fails on 35% of pairs, a linear probe on VGGT’s frozen features predicts pose failure on held-out categories with ROC-AUC 93.5%, compared with 79.7% for VGGT’s own point-map confidence and 77.7% for a classifier trained on all of VGGT’s outputs. The probe catches failures VGGT is confident about, lets a downstream system abstain on the pairs it distrusts, and transfers without retraining to rotation failures on PhotoTourism and ScanNet-1500, beyond what co-visibility explains. This knowledge is absent from the per-image encoder: it is computed partway through the layers that fuse the two views, and the same holds for a second model, \pi^3 . In essence, we find that VGGT knows more about its own failures than it reports.
[CV-9] Improving Signals Declining Encoder: A Known-Answer Test of Label-Free Monitoring in Self-Supervised Learning
链接: https://arxiv.org/abs/2610.35766
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: 17 pages, 6 figures, 2 tables. Code and data: this https URL . Checkpoints: this https URL
Abstract:Self-supervised training is monitored without labels, through the loss and statistics of the learned features. We report a case where we know the encoders lost part of what they had learned, so we can check whether these signals noticed. We trained LeJEPA encoders on synthetic images with known generating factors. In nine of ten seeds, the readability of those factors fell abruptly and stayed down, nearly always along the same direction in factor space. Meanwhile the loss kept falling, and most of the eleven label-free signals we tracked kept improving and rated the declining runs better than runs that did not decline. Standard change-point detection rarely flagged the decline near its onset, and selecting checkpoints by these signals mostly chose ones after it.
[CV-10] VidEdit3D: Flexible 3D Gaussian Editing with Video Diffusion
链接: https://arxiv.org/abs/2610.35756
作者: Yanming Yang,Chenxi Song,Tong Zhao,Juntao Xu,Wenhao Rao,Yiwen Chen,Yiwei Wang,Chi Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Recent advancements in text-guided 3D editing, particularly with 3D Gaussian Splatting (3DGS), have shown promise but still lag behind the flexibility of 2D editing tools. Existing methods face a critical trade-off between preserving multi-view consistency and allowing structural modifications. Image-based editors can also struggle to apply the intended edit to objects viewed from unusual angles. To address these limitations, we introduce VidEdit3D, a framework that reformulates 3D scene editing as a trajectory-based video editing task. By drawing a parallel between camera movement in static 3D scenes and temporal dynamics in video, we leverage the robust spatiotemporal priors of pre-trained Video Diffusion Models (VDMs) to encourage semantic and structural coherence across views. Specifically, we propose Trajectory-Guided Rendering (TGR) to select informative keyframes and construct smooth video clips aligned with VDM input requirements. To propagate edits from an anchor view to these clips, we develop Flow-Aligned Video Editing (FAVE), a training-free framework incorporating an Editing Consistency Guidance (ECG) mechanism. ECG contrasts source and edited image conditions at a shared latent state, using scheduled source-image amplification to strengthen adherence to the edited reference while maintaining original scene context. Finally, we introduce Region-Selective 3DGS Refinement (RS-GS) to strictly confine geometric and appearance updates to target primitives, reducing background attribute leakage. Experiments demonstrate stronger instruction alignment and better multi-view consistency than the baselines.
[CV-11] A Category-Selective Encryption Framework for Privacy Protection of UAV Images
链接: https://arxiv.org/abs/2610.35729
作者: Ting Yan,Minghui Shang,Haolang Sun,Xu Wang,Yamin Han,Mingze He
类目: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Imagery captured by unmanned aerial vehicle (UAV) systems increasingly raises privacy concerns, as UAV aerial surveillance routinely captures sensitive objects such as pedestrians and vehicles alongside ordinary scene content. Existing privacy-preserving approaches typically employ class-agnostic saliency detection to indiscriminately obfuscate all visually salient regions, leading to unnecessary information loss and hindering downstream scene analysis. This paper proposes a category-selective encryption framework for privacy-preserving UAV surveillance. The core advance over prior class-agnostic approaches is semantic-level selectivity: the system first recognizes what is present in the scene via an 80-class detector, then encrypts only user-specified sensitive categories while leaving non-sensitive objects and background intact for downstream analytics. The framework supports two complementary encryption schemes: Thumbnail-Preserving Encryption (TPE), which applies adaptive pixelation with a learnable blend parameter and recovers content through a jointly trained Enhanced U-Net, and XOR stream cipher encryption, which offers parameter-free, mathematically exact reversibility. The architectural design, composite loss formulation, and joint training methodology are presented in detail, establishing a foundation for practical privacy-preserving UAV surveillance that balances protection strength and scene utility.
[CV-12] MatAnyone-E: Sparse Memory Propagation for Efficient High-Resolution Video Matting
链接: https://arxiv.org/abs/2610.35724
作者: Shuyun Wang,Shangchen Zhou,Peiqing Yang,Yuekun Dai,Xin Yu,Chen Change Loy
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:High-resolution video matting requires preserving fine details, such as hair strands and soft transparency, while maintaining consistency across frames. Memory-based methods achieve strong matting quality by matching current-frame features with information stored from earlier frames, but this process becomes increasingly costly as resolution grows. We present MatAnyone-E, an efficient extension of MatAnyone 2 that focuses computation where it is most needed. Our approach first performs low-resolution propagation to identify uncertain regions that require closer inspection and the memory candidates most relevant to them. It then performs high-resolution matching only for these regions using the selected memory, while reusing low-resolution matching results elsewhere. This strategy reduces both the number of locations requiring detailed matching and the amount of memory searched, while preserving the information needed to produce a complete matte. We further reduce the cost of generating high-resolution mattes with a lightweight decoder designed around measured runtime bottlenecks and trained to reproduce the original decoder’s representations. Experiments show that MatAnyone-E maintains matting quality comparable to MatAnyone 2 while achieving an estimated 7.4x inference speedup and reducing peak GPU memory usage to 7.6% of the usage of MatAnyone 2 at 4K. With compilation-enabled inference on a single NVIDIA H200 GPU, it reaches 73.07 FPS at 1080p and 28.62 FPS at 4K.
[CV-13] riage: Skip Skim or Refine Attention Tiles in Video Diffusion Transformers
链接: https://arxiv.org/abs/2610.35706
作者: Yeonjae Kim,Omin Kwon,Seong Hoon Seo,Jae W. Lee
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Attention over tens of thousands of visual tokens dominates the inference of video diffusion transformers (DiTs). Existing methods reduce this cost through sparsity, low precision, or both, but use a single precision for all kept tiles. We observe that the output error introduced by approximation varies widely across tiles. Many tiles can be skipped with little error, others must be computed but gain little beyond low precision, and a few retain substantial error even at low precision. We present Triage, a training-free attention method that assigns each tile one of three approximation tiers: skipping it by replacing it with its block means, skimming it in NVFP4, or refining it with NVFP4 residual correction. A lightweight router predicts each tile’s error under each tier from block statistics and moves a tile to a more accurate tier only when the predicted error reduction justifies the extra compute, with a single price setting this trade-off. A fused kernel executes all three tiers in a single pass on the FP4 tensor cores of RTX Blackwell GPUs. Across Wan2.2, HunyuanVideo, and LTX-2.5, Triage accelerates end-to-end generation by up to 2.32x over dense FlashAttention, and in each of five settings it runs faster than every baseline while reaching 0.3-3.0 dB higher PSNR than the highest-PSNR baseline.
[CV-14] OmniShowcase: Omnidirectional Subject-Consistent Video Generation from Multi-View Images
链接: https://arxiv.org/abs/2610.35684
作者: Junyeong Ahn,Geonwoo Kim,Kinam Kim,Gihwi Kim,Hoonjin Jung,Hyojin Jang,Jaegul Choo
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 25 pages
Abstract:Generating videos of a specific subject from a few reference images remains challenging when the camera reveals surfaces that are not visible in the input. Early subject-to-video (S2V) methods condition on a single reference image and must therefore infer the subject’s unseen appearance and geometry as the viewpoint changes. To address this limitation, recent methods extend S2V to multiple reference views, providing more information about the subject from different viewpoints. However, they do not explicitly specify the 3D structure that connects these observations. In addition, existing training data do not guarantee that viewpoint changes preserve the same underlying 3D subject. We present OmniShowcase, built on two components: (i) a fully automated data synthesis pipeline that produces training videos along diverse camera trajectories, with viewpoint variation driven by underlying subject geometry, even spanning full 360-degree orbits; and (ii) a geometry-conditioning method that provides the video model with a turntable of surface-normal maps reconstructed from the reference views without requiring temporal alignment with the generated frames. We construct Showcase-4K, a corpus of 3D-asset-based videos covering twelve camera motions across 88 trajectory configurations with reconstruction-derived surface-normal conditioning, and ShowcaseBench, a held-out evaluation set with the source meshes and camera parameters used to render each video. Experiments show state-of-the-art subject identity and 3D consistency while maintaining competitive video quality and motion. Project page: this http URL
[CV-15] Printed and Scanned: Exposing the Frag ility of Single-Image Morphing Attack Detectors under Print-Scan Degradation
链接: https://arxiv.org/abs/2610.35682
作者: Diogo J. Paulo,Hugo Proença,João C. Neves
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Face morphing attacks, in which the biometric traits of two or more identities are blended into a single face image, pose a severe threat to identity document issuance and automated border control. Consequently, a large body of work has proposed Single-image Morphing Attack Detection (S-MAD) algorithms and benchmarked them on datasets such as SYN-MAD22. However, the majority of these benchmarks operate exclusively in the digital domain, which contrasts with the real-world passport issuance pipeline in some countries, which still requires an applicant-supplied photograph to be printed and subsequently re-scanned before it is embedded in the identity document or fed into an automated verification system. This print-scan step introduces halftoning, resampling, color-space, and sensor noise that are absent from digital pipelines, and may mask the high-frequency artifacts that S-MAD models rely on. In this work, we introduce a print-scan version of the SYN-MAD22 dataset, obtained by printing and re-scanning all bona fide and morphed images at 300 and 600 DPI, and we use it to evaluate eight state-of-the-art S-MAD models trained exclusively on digital data. We additionally repeat the acquisition with a second scanner to assess cross-device robustness. Our results show that detectors that achieve near state-of-the-art performance in the digital domain suffer a more than threefold increase in average equal error rate, with the same qualitative degradation observed across both scanning devices, exposing a critical, previously under-reported gap between digital-benchmark performance and robustness to print-scan degradation.
[CV-16] CounterFlow: Target-Conditioned Flow Matching for Visual Counterfactuals
链接: https://arxiv.org/abs/2610.35679
作者: Paul Sevellec,Elisa Fromont,Romaric Gaudel,Laurence Roze,Paul Viallard,Laetitia Houeix
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:In eXplainable AI, counterfactual explanations have gained significant traction by offering users direct, intuitive, and actionable insights into model decisions. We introduce CounterFlow, a novel framework based on Conditional Flow Matching that generates visual counterfactual explanations by directly steering a source image toward a user-specified target class. By leveraging target-conditioned vector fields, CounterFlow identifies highly realistic semantic modifications while requiring substantially lower computational overhead than traditional GAN- or diffusion-based architectures. Empirical evaluations on standard benchmarks demonstrate that CounterFlow achieves state-of-the-art performance in both visual realism and target class switching. By working directly in the raw pixel space, CounterFlow offers a more practical solution for real-world deployment.
[CV-17] VPR-Space: A Benchmark for Visual Place Recognition in Space Analog Domains
链接: https://arxiv.org/abs/2610.35642
作者: Richard Fechner,Peer Neubert
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual Place Recognition (VPR) is an important means for loop closure detection in visual Simultaneous Localization and Mapping (SLAM). While state-of-the-art VPR and SLAM approaches excel in urban and feature-rich natural environments, the vast and feature-poor environments of extraterrestrial surfaces like on the Moon or Mars pose severe challenges for existing approaches. This paper presents a novel benchmark, VPR-Space, to evaluate VPR capabilities on extraterrestrial surfaces. There are three contributions: (1) We collected sixteen VPR database-query datasets based on existing data from Space Analog Domains (SAD), terrestrial environments that resemble properties of the Moon or Mars. The collection includes a range of challenges regarding environment, viewpoint change, and illumination conditions. (2) To address the ill-posed nature of VPR, we propose a novel approach to generate the ground-truth information in VPR benchmarks. Instead of defining that a query and datbase image show the same place if and only if their camera poses are within fixed position and viewing direction thresholds, we propose to make the decision dependent on whether the match would be valuable in the downstream SLAM task. To avoid the overhead and restrictions of evaluating VPR in-the-loop within SLAM systems, we provide VPR ground-truth (GT) matrices based on the performance of an ensemble of state-of-the-art Relative Pose Estimators (RPE). Under this GT we require a VPR method to identify exactly those image pairs for which a subsequently applied RPE is expected to provide a valuable pose estimate in a SLAM context. (3) We evaluated nine state-of-the-art VPR approaches. The results show considerable room for improvement and indicate directions for future research. We provide a pre-release version of the dataset on github: \textttthis https URL
[CV-18] SeeRF: Learning to See Beyond Fields of View via Generative Image Registration and Fusion
链接: https://arxiv.org/abs/2610.35637
作者: Xunpeng Yi,Yibing Zhang,Qinglong Yan,Yifan Lu,Han Xu,Jiayi Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Due to the inherent discrepancies in imaging viewpoints and field-of-view mismatches among multi-modal sensors, existing fusion methods have long struggled to achieve complete scene-level information integration beyond the limitation in sensor fields of view. Consequently, the generated fusion results often contain modality-specific regions and fusion artifacts, limiting the comprehensive representation of scene structures and semantic information. To address this challenge, we propose a generative registration-fusion framework, termed SeeRF. Specifically, it first introduces a deformation generation strategy based on the diffeomorphic stationary velocity field, combined with visual foundation model-guided feature extraction and a coarse-to-fine progressive multi-modal image registration module, to achieve accurate cross-modal flow estimation and generate valid field-of-view masks. Subsequently, the registration module is tightly coupled with the generative diffusion module for multi-modal completion and fusion, where the cross-modal control encoding module and the control residual feature encoder-decoder mechanism provide effective guidance. By leveraging the powerful prior modeling capability of generative diffusion models, the proposed framework overcomes the limitations imposed by shared fields of view and enables effective completion and natural fusion of non-overlapping field-of-view regions. Extensive experiments on multiple public datasets and real-world scenarios demonstrate that the proposed method achieves superior performance, validating its effectiveness, superiority, and robustness. The code will be available at this https URL.
[CV-19] Synthetic Deformable Tubular Benchmark: A mechanically annotated dataset for evaluating deformable 3D reconstruction
链接: https://arxiv.org/abs/2610.35621
作者: Íñigo Cirauqui-Viloria(Universidad de Zaragoza)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 34 pages, including 5 pages of Supplementary Methods; 10 main-text figures and 1 supplementary figure
Abstract:Evaluating deformable 3D reconstruction requires ground truth linking camera motion, surface deformation, material correspondence and image appearance. We introduce the Synthetic Deformable Tubular Benchmark, generated with LumenMorph, an open-source toolkit for creating, annotating and evaluating tubular scenes. The release contains 5,580 RGB images. Its baseline crosses three surface-motion conditions with moving and stationary cameras in six scene families; matched appearance variants cover only moving-camera finite-element actions in four families. Documented public and evaluation interfaces expose camera poses, geometry, depth, deformation fields, persistent material correspondences and discrete finite-element records. Persistent triangle identities and material weights let users track the same surface location across frames and query its three-dimensional trajectory. Separately implemented geometric and mechanical checks assess annotation consistency with the generating models. Structure-from-motion, optical-flow and supplied-geometry displacement-interpolation studies demonstrate how the resource distinguishes registration coverage from geometric accuracy, exposes camera- and surface-motion effects, and tests the benefits and mismatch of mechanical priors. Public readers, evaluation contracts and the configurable generator support reuse and extension. Data integrity, fixed-environment repeatability and retained-prediction score replay have distinct verification scopes. The benchmark provides a controlled synthetic reference without claiming anatomical or clinical realism.
[CV-20] RoboDoctor: Training-Free Recursive Self-Improvement for Robot Policies
链接: https://arxiv.org/abs/2610.35604
作者: Heyang Yu,Haomin Li,Sihang Li,Jixin Yan,Shiyun Peng,Wanqi Zhong,Tianyu Xin,Yuan Yao,Xiaoyu Tian,Yiming Li
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Generalist robot policies and language-model agents provide complementary capabilities, but completing a task also requires adapting their use to the consequences of physical interaction. We study execution as a locus of adaptation: feedback guides the current action and supplies evidence for revising subsequent decisions under fixed model weights. We introduce RoboDoctor, a training-free framework that connects diagnosis, grounded action, and persistent experience across three revision targets. The harness allocates control between policy and agent; intervention determines when to assist while preserving useful policy behavior; and operation refines physical methods through observed effects. Linked experience records retain the reason for a decision, the attempted method, its outcome, and conditions for reuse. On a curated archive of 54 RoboDojo-Sim task entries, RoboDoctor achieves a five-dimension mean score of 57.10, with a native success rate of 49.44%. Recorded interactions illustrate strategy changes, selective correction, and grasp revision. These findings characterize execution-time adaptation in the evaluated system; controlled trials remain necessary to isolate the contribution of cross-episode experience.
[CV-21] Beyond Distribution Matching: Self-Supervised Representation Forcing for Few-Step Video Generation NEURIPS2026
链接: https://arxiv.org/abs/2610.35594
作者: Junyi Chen,Haiyu Zhang,Zhoujie Fu,Tengfei Wang,Chunchao Guo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026. Project page: this https URL
Abstract:Few-step distillation is essential for deploying modern video diffusion models. Among existing approaches, Distribution Matching Distillation (DMD) has emerged as a highly promising paradigm by achieving strong output-space fidelity without adversarial training, while remaining flexible across various student architectures. We observe, however, that DMD fundamentally relies on an output-space objective, lacking explicit regularization on the model’s internal representations. As the distillation step count aggressively shrinks, this output-only supervision becomes insufficient, causing the distilled student to degrade in both fine-grained spatial structures and coherent temporal dynamics. To address this limitation, we introduce Self Flow Forcing (SF ^2 ), which incorporates a predictive representation loss into the DMD framework without altering its output objective. By providing the student and an Exponential Moving Average (EMA) teacher with heterogeneous noise levels, we establish an information asymmetry. This formulation forces the student to predict the teacher’s internal representations from a more corrupted input, effectively integrating self-supervised representation alignment into distribution matching. Extensive experiments demonstrate that SF ^2 effectively mitigates temporal flickering and consistently improves both spatial fidelity and motion quality. Furthermore, the proposed mechanism is architecture-agnostic - seamlessly supporting both bidirectional and causal models - and requires no external feature extractors, introducing minimal computational overhead.
[CV-22] Hyperspectral Unmixing using a Structured Linear Mixing Model with Additive Spectral Variability
链接: https://arxiv.org/abs/2610.35580
作者: Salah Eddine Brezini,Yannick Deville,Moussa Sofiane Karoui,Fatima Zohra Benhalouche,Abdelaziz Ouamri
类目: Computer Vision and Pattern Recognition (cs.CV); Data Analysis, Statistics and Probability (physics.data-an)
备注:
Abstract:Linear hyperspectral unmixing aims at retrieving pure material spectra present in an observed scene and their associated abundance fractions in each pixel. In most practical situations, hyperspectral images are affected by spectral variability (SV) also known as intra-class variability, which is caused by different physical phenomena including atmospheric effects, intrinsic variability of the pure materials and illumination or topographic changes. Nevertheless, the standard hyperspectral unmixing methods that rely on the Linear Mixing Model (LMM) do not consider these phenomena. Thus, neglecting them will spread errors throughout the linear hyperspectral unmixing process. To avoid this problem, this paper proposes two hyperspectral unmixing algorithms dedicated to a parametric model called Structured Additive Variability LMM (SAV-LMM). This mixing model handles SV in an additive manner by incorporating two terms into the standard LMM. Moreover, the designed algorithms are based on Nonnegative Matrix Factorization (NMF) and optimize a cost function which integrates additional penalization accounting for SV. Furthermore, a novel improved initialization scheme, which takes SV into account, is proposed. Experiments conducted on both synthetic and real datasets show the effectiveness of the proposed algorithms for linearly unmixing hyperspectral images with SV compared to the tested state-of-the-art methods.
[CV-23] Copying As Defense: Protecting Images from Editing By Eliciting Model Copying
链接: https://arxiv.org/abs/2610.35566
作者: Salma Abdel Magid,Sanghyuk Chun,Eric Slyman,Kushal Kafle,Olga Russakovsky
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Image editors built on diffusion transformers (DiTs), such as Flux Kontext, make it possible to turn a single photograph into a convincing manipulated image, including harmful and non-consensual edits. Proactive defenses aim to prevent such manipulations by adding an imperceptible perturbation before an image is shared. Existing defenses for DiT-based in-context editors make the output attend less to the protected reference image. But this leaves the model free to follow the edit instruction using the text and its own generated tokens, often producing a new, weakly grounded image rather than preventing the edit. We find that the opposite strategy is more effective: instead of leading the model to ignore the reference, we encourage it to copy it. Through controlled attention interventions, we show that concentrating target-token attention on the reference image in only a few early blocks is sufficient to collapse the editing behavior, causing the model to reproduce its input, often as a lossy copy, rather than execute the requested transformation. Based on this insight, we introduce Copying-as-Defense (CaD), which maximizes per-token target-to-reference attention mass in early transformer blocks. Since attention rows are normalized, increasing reference attention suppresses the attention allocated to the text and generated canvas. Restricting the objective to early blocks also lets us truncate the backpropagation path, keeping optimization cheap. Across editors and benchmarks, CaD reduces average edit success from 81.4% on unprotected images to 24.0%, a 70.6% relative reduction, and by a further 44.3% relative to DeContext, the strongest baseline (43.0%). CaD is also roughly 1.8-2.0x faster per optimization step than comparable methods.
[CV-24] Metric Scale for Insta360 Structure-from-Motion from the Cameras Embedded IMU
链接: https://arxiv.org/abs/2610.35554
作者: Davide Cuttini,Diego Di Tommaso,Alessio Antoniacomi,Luca Simonetti
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages, 9 figures, 4 tables. Dataset: this https URL
Abstract:Structure-from-motion (SfM) from the dual-fisheye video of a consumer 360-degree camera recovers geometry only up to a similarity transform; in the pipeline considered here, metric scale is obtained from smartphone ARKit or ARCore poses captured alongside the camera. This paper evaluates a self-contained alternative: the IMU embedded in the Insta360 X5 and X6, recorded at 1 kHz. The sensor is characterised (noise density, camera-IMU synchronisation, accelerometer scale factor, bias stability), the camera-IMU rotation and frame timing are estimated from the data, and the scale is recovered by a linear visual-inertial alignment over the SfM poses of a whole survey. The standard formulation, with gravity and accelerometer bias shared across the survey, overestimates the scale by 3.4% on average; estimating each contiguous segment independently and averaging removes the bias. On nine surveys with a co-mounted LiDAR scanner the inertial scale deviates from the LiDAR by 0.74% in mean absolute error, against 1.94% for the smartphone scale; on printed fiducial markers both agree within the 1% uncertainty of the target. On the reference trajectories of thirty sequences of a public benchmark, brought to scales of 1/40, 1 and 40, the estimator recovers the metric scale with a mean error of -0.07% and a standard deviation of 1.28%. On fifty production surveys where the smartphone is the only reference, the two scales agree to a standard deviation of 2.6%, the inertial scale being 0.9% lower on average; the difference does not depend on window construction, grows with the extent of the smartphone anchor, and cannot be attributed to either reference. An ablation over window construction and a comparison with monocular metric-depth networks complete the study.
[CV-25] EARS: Evidence-Aware Retrieval via Sound and Motion for Budgeted Video Reasoning
链接: https://arxiv.org/abs/2610.35536
作者: Shih-Yao Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 8 figures. Code: this https URL
Abstract:Uniform frame sampling can miss brief evidence in sparse-event videos when the VLM sees only K frames. We propose EARS (Evidence-Aware Retrieval via Sound and motion), a training-free, query-free front-end that ranks frames using optical-flow motion and audio energy, then applies temporal non-maximum suppression for diversity. The evidence score alone has frame-level AP 0.39; used only to select frames for the VLM, EARS with temporal diversity reaches localization AP 0.94, compared with 0.88 under Uniform sampling. In the updated matched study, at K = 1 EARS reaches XD-subset F1 0.660 versus 0.614 for centered Uniform (the paired interval includes zero); earlier backbone studies use separate protocols. On mostly-silent UCF-Crime, however, EARS slightly trails Uniform. A single configuration is used across both benchmarks; the new controls do not repeat the original held-out search. EARS therefore targets low-budget videos whose salient evidence aligns with motion or audio; on dense narratives or stealthy events, uniform or semantic selection remains preferable, and cheap modality statistics cannot reliably route between them. Code: this https URL
[CV-26] WorldPager: Zero-Stall Tiered Memory for Long-Horizon World Models on a Single Consumer GPU
链接: https://arxiv.org/abs/2610.35529
作者: Bingnan Liu,Ruini Xue
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 4 figures, 9 tables
Abstract:Interactive world models keep the KV of past chunks as retrieval memory so that revisited places stay consistent. On a single consumer machine (one RTX 4090, 32 GiB of host memory, one NVMe SSD), long sessions with such memory hit two walls. The first is capacity: the KV of Matrix-Game 2.0 exhausts host memory after 45 seconds of video in bf16 and after 2.8 minutes in int4. The second is the model’s positional horizon of 4.3 minutes. We show that this memory is capacity-bound rather than bandwidth-bound and that its access set is known before compute, because the camera pose is an input. An analytical model built on these observations predicts spill points to within one chunk and classifies zero versus non-zero stalls correctly on held-out configurations. Guided by it, WorldPager stores int4 KV in a GPU/DRAM/NVMe hierarchy instead of recomputing it, prefetches each chunk’s memory from the extrapolated pose, never waits on a miss, and writes back in the background; to our knowledge, it is the first to place the KV memory of an interactive world model on an NVMe tier with zero-stall prefetching. With training-free position handling, it generates 37.5 minutes of video continuously with zero stalls and constant host memory, at about 9.5 frames per second. On a sealed test set, int4 storage shows no measurable quality loss; on long-gap revisits it lowers image quality by about one MUSIQ point.
[CV-27] Generative Structure from Motion with Native 3D Diffusion NEURIPS2026
链接: https://arxiv.org/abs/2610.35522
作者: Haobin Duan,Binbin Huang,Mochu Xiang,Yiqun Zhao,Zibo Zhao,Yi Ma,Shenghua Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026. Project page: this https URL
Abstract:Recovering an object’s 3D structure and camera motion from a few uncalibrated views is a fundamental problem in computer vision. Classical structure from motion methods recover 3D geometry consistent with the input images but cannot recover unseen regions; native generation methods produce plausible 3D objects which are related but not necessarily faithful to the input images. Methods simply concatenating reconstruction and generation allow error to cascade across the two stages, instead of producing a 3D object consistent with both the distribution and the input. We propose GenSfM, which integrates image-based 3D reconstruction with a native 3D diffusion model. In particular, we use a multi-modal diffusion transformer to jointly denoise the shape latent and per-view pose latents under shared self-attention, grounding generated geometry in the observed views while using object-level priors to infer unobserved regions. A compact 3D-native UV-volume tokenization keeps the joint sequence small as views are added, and a pose-aware stage refines geometry and texture by sampling per-voxel image features at the recovered camera poses. On Toys4k and GSO, for 4 uncalibrated views, GenSfM improves novel-view PSNR by 3.2 - 4.5 dB and reduces median Chamfer distance by 2.8 - 3.8\times over a strong native-3D generation baseline, while recovering input-view poses over 96% Acc@ 5^\circ without any external pose-estimation backbone, post-hoc optimization, or known intrinsics. Project page: this https URL.
[CV-28] Learning High-Quality Pseudo-Labels for Point-Supervised Temporal Action Localization
链接: https://arxiv.org/abs/2610.35516
作者: Zhenying Fang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 7 figures
Abstract:Point-supervised temporal action localization (PTAL) aims to recognize and localize action instances in untrimmed videos using only point-level annotations. Existing pseudo-label-based methods have made substantial progress in PTAL. However, confidence-based initialization can overlook well-localized proposals. Meanwhile, existing methods either keep pseudo-labels fixed during detector training or update them without fully exploiting alternative boundary evidence from the original proposal pool. These limitations constrain supervision quality, thereby hindering detection performance. To learn high-quality pseudo-labels, we propose a framework that integrates prior-guided initialization with consensus-guided refinement during detector training. Our framework comprises two modules: Prior-Guided Initialization (PGI) and Consensus-Guided Refinement (CGR). PGI combines proposal confidence with point-derived priors to establish instance-specific reliability for initial pseudo-label selection and candidate retention. CGR continually uses the retained original candidates to construct boundary targets, combining their consensus with evolving detector predictions to refine pseudo-labels throughout training. Experiments on four benchmarks demonstrate that our method achieves competitive detection performance compared with existing methods. In particular, our method achieves 68.5% average mAP over tIoU thresholds 0.1–0.7 on THUMOS14, surpassing PseR by 3.7 percentage points. Code is available at this https URL
[CV-29] HiFiPix: Bringing High-Fidelity Real-World Super-Resolution Back to Pixel Space
链接: https://arxiv.org/abs/2610.35515
作者: Wenjie Li,Du Chen,Hui Wang,Linlong Fan,Youwei Lv,Zeyi Chen,Zehan Chen,Jinwei Chen,Heng Guo,Zhanyu Ma,Qingnan Fan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Diffusion-based real-world image super-resolution suffers from limited reconstruction fidelity due to information loss in VAE-based latent representations. Recently, pixel-space diffusion models offer a promising high-faithfulness alternative by avoiding latent compression, but adapting them to Real-ISR remains challenging due to inherent patch boundary artifacts and degraded generative capability. To address the issues and maximize the potential of pretrained pixel-space diffusion models, we propose HiFiPix, a pixel-space Real-ISR framework capable of ultra high-fidelity reconstruction and fine-grained visual generation effects. To suppress patch boundary artifacts, we penalize unrealistic high-frequency periodic components with a Cross-Scale Wavelet Discriminator (CWSD) for image-level supervision, and Mixup Contrastive Self-Distillation (MixCSD) for intermediate representation regularization. Furthermore, we uncover a heterogeneous generation-fidelity trade-off across low-rank adapters (LoRA) and propose Selective LoRA Pruning (LoRA-SP) to adaptively prune LoRA layers, enhancing generative capability with minimal deterioration of reconstruction fidelity. Extensive experiments demonstrate that HiFiPix consistently outperforms existing state-of-the-art latent space methods in terms of both restoration accuracy and perceptual quality by a large margin. Code and models will be released.
[CV-30] Learning Scene Coordinate Reconstruction from Unposed Images via Pose Graph Optimization CVPR2026
链接: https://arxiv.org/abs/2610.35501
作者: Tze Ho Elden Tse,Jizong Peng,Angela Yao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to CVPR 2026 (Highlight). Project page: this https URL
Abstract:Learning-based structure-from-motion methods such as ACE-Zero have demonstrated strong performance in estimating cameraposes and scene coordinates from unordered image collections without requiring ground truth supervision. However, the lack of global and multi-view consistency constraints in ACE-Zero can lead to pose drift and misalignment, particularly in complex or ambiguous scenes. In this work, we propose a hybrid framework that integrates pose graph optimization (PGO) into ACE-Zero to refine camera poses and suppress incorrect refinements. We construct pose graphs directly from ACE-Zero outputs by extracting relative pose constraints from predicted scene coordinates. Furthermore, we introduce an uncertainty-aware optimization strategy by estimating confidence scores using geometric priors, including epipolar and optical flow consistencies across views. Our approach improves the robustness and accuracy of pose estimation, demonstrating that global geometric reasoning can effectively complement learning-based inference in structure-from-motion.
[CV-31] S2-RL: Sample-Set Dual Reinforcement Learning for Generative Semantic Segmentation Dataset Distillation NEURIPS2026
链接: https://arxiv.org/abs/2610.35491
作者: Haoyu Wang,Fei Zhou,Qingqing Qiu,Lei Zhang,Wei Wei,Chen Ding
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. Project Page: this https URL
Abstract:While dataset distillation has witnessed significant advances in image classification, its extension to dense prediction tasks such as semantic segmentation is plagued by two core bottlenecks: bi-level optimization-based approaches suffer from prohibitive computational costs stemming from pixel-wise gradient unrolling, whereas proxy-based generative methods merely infuse visual textures into fixed semantic masks derived from real data, thus failing to condense rich semantic knowledge into a compact set of informative, novel scene generations. To address both bottlenecks simultaneously, we propose a novel Sample-Set Reinforcement Learning (S ^2 -RL) framework for generative semantic segmentation dataset distillation (SSDD). Specifically, S ^2 -RL formulates SSDD as a diffusion model based text-to-image generation paradigm, enabling flexible generation of images with informative, novel semantic distributions via a concise text prompt. Furthermore, we fine-tune a diffusion policy via Group Relative Policy Optimization (GRPO), leveraging a sample-set dual reward paradigm: for individual samples, it enforces semantic alignment with input text prompts and intra-class feature diversity; for sample sets, it maximizes inter-sample semantic distribution diversity by solving a contextual multi-armed bandit problem. These design choices enable S ^2 -RL to distill large-scale semantic segmentation datasets into a compact set of generated samples that encapsulate diverse, informative semantic knowledge without the need of pixel-wise gradient unrolling. Extensive experiments on the ADE20K and COCO datasets demonstrate that S ^2 -RL achieves substantial and consistent improvements over state-of-the-art baselines, establishing a new benchmark for SSDD.
[CV-32] P-JSCC: Efficient Token Pruning and Joint Source-Channel Coding for LVLMs-based 3D-VQA in Cloud-Edge Collaboration
链接: https://arxiv.org/abs/2610.35489
作者: Zhengxin Li,Chongzhen Tian,Hui Yuan,Hao Liu,Sam Kwong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Deploying large vision-language models (LVLMs) on resource-constrained embodied agents for 3D visual question answering necessitates cloud-edge collaboration. However, offloading massive visual data to the cloud over bandwidth-limited and noisy channels introduces severe latency and transmission distortion. To address this, we propose TP-JSCC, an efficient token pruning and joint source-channel coding method for LVLMs-based 3D-VQA in Cloud-Edge Collaboration. First, we develop a Local and Global Semantics-based Hybrid Token Pruning (LGS-HTP) module to significantly reduce the visual token count. By decoupling foreground-background tokens and jointly selecting local and global semantics, LGS-HTP effectively reduces redundancy while preserving reasoning-critical tokens. Second, we introduce a lightweight Token Joint Source-Channel Coding (T-JSCC) module to further compress the pruned tokens and ensure their robust transmission over error-prone channels. By employing lightweight cascaded linear layers and Signal-to-Noise Ratio (SNR)-adaptive mechanism, this module adapts to fluctuating channel conditions while maintaining minimal computational overhead. Extensive experiments on ScanQA and SQA3D datasets indicate the exceptional efficiency and robustness of our approach. Specifically, on ScanQA validation set, the LGS-HTP module could achieve 90% pruning ratio while incurring less than a 1% CIDEr loss. Building on this, the T-JSCC further compresses the tokens by 18 times, resulting in an overall 180 times data volume reduction for robust transmission, while saving 90% in computational complexity for LVLM deployment. Our code is available at this https URL.
[CV-33] OBB-PBD: Extending Parallel Box Decoding to Oriented Bounding Boxes for Referring Expression Grounding in Aerial Imagery
链接: https://arxiv.org/abs/2610.35450
作者: Chang Woo Song,Yeong Jun Koh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published in Remote Sensing 18(20), 3535 (2026). Code and checkpoints: this https URL and this https URL
Abstract:Referring expression grounding in overhead imagery requires oriented bounding boxes rather than horizontal ones, since aerial objects appear at arbitrary orientations. Most grounding models emit box coordinates sequentially; Parallel Box Decoding (PBD) instead predicts all geometry slots in one forward pass, but only for four-coordinate horizontal boxes. We extend this block-aligned interface to a six-slot oriented representation using a double-angle parameterization. Rather than assuming correctness from a decreasing loss curve, we verify the extension with image/prompt-shuffle controls and visibility/leakage probes, which show that the corrected model depends on both the image and the prompt rather than on prompt statistics alone. On the aerial vehicle visual grounding (AVVG) benchmark, whose targets are genuinely oriented, the extended model improves rotated intersection over union (rotIoU) from 0.1253 for the zero-shot baseline to 0.303 +/- 0.012 (mean +/- standard deviation over three training seeds) across the complete 6139-record test split spanning all 120 images. Learning oriented boxes rather than their axis-aligned envelopes adds 0.055 +/- 0.008 rotIoU beyond a matched control (per-seed 95% confidence intervals all excluding zero), whereas block expansion alone yields no rotIoU gain distinguishable from zero.
[CV-34] Model-Aware Generation of Safety-Critical Scenes for Robust End-to-End Driving via Distilled Planner Surrogates
链接: https://arxiv.org/abs/2610.35448
作者: Khaled M. Seyam(1),Julian Wiederer(2),Samar Fathalah(1),Markus Braun(2),Bin Yang(1) ((1) University of Stuttgart, Germany, (2) Mercedes-Benz AG, Germany)
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:End-to-end driving models perform well on standard benchmarks but collide frequently in safety-critical scenarios. We present a failure-mode analysis on NeuroNCAP which shows that 95.9% of DiffusionDrive’s collisions are planning failures despite reliable detection, pinpointing the bottleneck in planning rather than perception. We attribute this to a fundamental data imbalance, as safety-critical interactions are rare in real driving logs, so planners are hardly ever trained on them. We close this gap with a novel model-aware framework that generates safety-critical training scenes against a specific planner. Each planner is attacked at the bird’s-eye-view (BEV) level through a differentiable surrogate distilled from its own planning head. This keeps the expensive video renderer out of the attack loop. We extract 50 scenes per planner, each pairing an attack with a safe escape trajectory that serves as the fine-tuning label. A fully automated quality-diversity search generates both the attacks and the escapes, so that their variety is not specified by hand. Only these scenes are rendered to fine-tune the planner, which reduces the computational cost significantly. Fine-tuning on these scenes roughly halves the closed-loop collision rate of two planners on NeuroNCAP (SparseDrive 28.0% to 12.2%, DiffusionDrive 37.5% to 19.2%) while largely preserving normal driving. A cross-planner transfer experiment shows that the gain comes from targeting the specific planner. Each planner improves most on the scenes generated against itself, and DiffusionDrive fine-tuned on SparseDrive’s scenes retains only 6.7 of its 18.3-point collision-rate reduction.
[CV-35] Unveil the Mask: Jointly-Optimized Diffusion Masked Pretraining for Appearance-Agnostic Operator Verification ACCV2026
链接: https://arxiv.org/abs/2610.35414
作者: Katsuya Ogata,Ryoya Yoshiyama,Kawachi Hodaka,Yuta Nakashima
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at ACCV 2026 Workshop on Who Is Behind the Avatar? (WBA)
Abstract:Photorealistic avatar videos decouple the appearance of a rendered face from the person who operates it, which makes conventional face recognition unsuitable for identifying the operator. Verifying the operator therefore requires representations of how a face moves that are insensitive to what the avatar looks like. We propose a verification framework built without any external pretrained model. It uses two complementary modalities provided for each sample: facial landmark sequences and video frames. From the landmark modality, it learns motion representations by diffusion masked pretraining over grouped landmarks, guided by lightweight auxiliary tasks that predict behavioral cues such as motion intensity and eye openness, and turns them into operator embeddings with a contrastive linear probe. To generalize to unseen avatar generators, the pretraining input is randomly transformed to simulate how a generator may alter the operator’s motion. From the video modality, it captures fine facial motion through differences between consecutive frames. The embeddings of the two modalities are combined into one. Our approach reaches an overall AUC of 0.769 on the WBA2026 benchmark. Linear probing shows that the video embedding retains little information about the avatar’s identity, whereas the landmark embedding retains some, and fusing the two modalities improves verification in particular across speaking and listening modes.
[CV-36] NeSySpec: Neuro-Symbolic Technical Intelligence for Auditing Engineering Specifications with an Application to Aircraft
链接: https://arxiv.org/abs/2610.35403
作者: Pier Mattia Basciano,Gianvito Antenori,Cristiano Andolfi,Giovanni Farina,Salvatore Monteleone
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Symbolic Computation (cs.SC)
备注:
Abstract:Engineers who audit a declared specification of an engineered system need to know which physical constraints it respects, which it violates, which field was altered and how to correct it. NeSySpec is an integrated and interpretable neuro-symbolic audit framework, built on Logic Tensor Networks, that answers these questions in one pipeline. Its axioms encode the physical relations of the domain as differentiable graded predicates whose tolerances are learned from a fleet of real records; from the same degrees, it diagnoses the violated constraint, identifies the altered field by abduction, proposes a minimum-change correction and ties image evidence to the parts it describes, abstaining exactly where evidence is missing, and returns all of it in one verification report. We validate NeSySpec on 107 aircraft types, with 16 test aircraft of 9 families held out of training, on specification records paired with ground photographs and satellite imagery and on synthetic errors: generated corruptions, edits labeled from the edit alone and deliberate falsifications. NeSySpec combines all the assessed audit outputs in one pipeline, which no learned, rule-based, solver-based, neuro-symbolic or language-model comparator does; on the test aircraft none is ahead of it in the macro-F1 of its constraint verdicts, in naming the violated constraint and the altered field or in the repair error, and it is competitive in record and image-pair detection. Its diagnosis names the violated constraint in 94.2% of corrupted test records, against 79.5% for classical Fellegi-Holt diagnosis, and on edits labeled from the edit alone, without retraining, abduction finds the altered field in 59.3% of records, against 30.6% for minimum-cardinality localization with random tie- breaking. Only the knowledge base, the visual predicates, the corruption generator and the data are specific to the domain.
[CV-37] oward Clinical Translation of Deep Learning-Based Intracranial Super-Resolution 4D Flow MRI Using Multi-Cohort Transfer Learning
链接: https://arxiv.org/abs/2610.35396
作者: Oliver Welin Odeback,Susanne Schnell,Patrick Winter,Johan Lundberg,Fredrik Piehl,Tobias Granberg,Alexander Fyrdahl,David Marlevi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 6 main figures, 4 main tables; supplementary material includes 6 sections, 2 figures, and 1 table
Abstract:We evaluated synthetic pretraining and multi-cohort transfer learning for deep learning-based super-resolution of intracranial 4D Flow MRI across heterogeneous neurologic cohorts. This mixed prospective-retrospective study analyzed 18 participants (12 healthy participants, 3 with moyamoya disease, and 3 with multiple sclerosis) with paired 1.2-mm and 0.6-mm 4D Flow MRI scans. A convolutional neural network pretrained on synthetic cerebrovascular data was fine-tuned using cohort-specific or pooled multi-cohort in vivo data. High-resolution-derived inputs enabled voxel-wise evaluation; native standard-resolution scans were used for angiographic and hemodynamic analyses. Synthetic pretraining reduced mean artery-wise velocity-normalized root mean square error (vNRMSE) versus interpolation (0.147 vs 0.166). Pooled transfer learning yielded the lowest mean vNRMSE (0.119 vs 0.127 for cohort-specific models and 0.147 for synthetic pretraining), with narrower 95% limits-of-agreement widths than synthetic pretraining across all test cases (0.109-0.222 vs 0.154-0.278 m/s). Super-resolved phase-contrast angiography achieved higher precision-recall area under the curve than native standard-resolution imaging in all cohorts (0.848-0.896 vs 0.486-0.805). These results indicate that pooled transfer learning improves cross-cohort reconstruction agreement and that super-resolution can improve vessel depiction, supporting further multi-center validation for cerebrovascular imaging.
[CV-38] ACSAN: All-Convolutional Self-Attention Network Integrating Convolutional Operations with Vision Transformer
链接: https://arxiv.org/abs/2610.35375
作者: Zimian Wei,Tianqi Zhang,Hengyue Pan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:In the past few years, Vision Transformer (ViT) has achieved excellent performance in the field of computer vision. However, there still some problems remain. For instance, ViTs heavily rely on large-scale training data since they simply contain a large number of trainable parameters, which slows down their learning efficiency and limits their applicability in resource-constrained scenarios. In this paper, we propose a novel deep learning model, namely All-Convolutional Self-Attention Network (ACSAN), which applies convolution operations to replace fully-connected matrices multiplications in the ViT framework. To implement ACSAN, we firstly design Dynamic Convolutional Self-Attention (DCSA) as the basic module, which employs multi-scale dilated convolutions to generate Query, Key, and Value. We defined spatial self-attention and channel self-attention to generate attention features respectively. The two attention features are finally merged as the output of the DCSA module. Based on DCSA, we then construct the basic ACSAN block. Comparing with ViT, we use DCSA to replace the regular self-attention operations, and use convolution layers to replace MLPs. To evaluate the performance of the proposed method, we train ACSAN on three widely-used image classification databases, i.e. CIFAR-10, CIFAR-100 and ImageNet-1k, from scratch. Experimental results show that ACSAN achieves excellent classification accuracy comparing with standard ViT with comparable or fewer parameters, as well as other state-of-the-art methods. Ablation studies verify that each component contributes positively to the final performance.
[CV-39] GIFT: Representation Geometry Matters for Single-Domain Generalized Object Detection NEURIPS2026
链接: https://arxiv.org/abs/2610.35374
作者: Yin Zhang,Yaoyue Zheng,Yongqiang Zhang,Bogdan Raducanu,Dan Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. 22 pages, 11 figures
Abstract:Single-Domain Generalized Object Detection (S-DGOD) aims to generalize from a single source domain to unseen domains under distribution shifts. Existing efforts mainly focus on simulating unseen domains or suppressing domain-specific factors. However, they largely overlook the intrinsic generalization capability embedded in Vision Foundation Models (VFMs). In this work, we show that unconstrained fine-tuning disrupts the geometric structure of pre-trained representations, which encodes domain-invariant semantics. Motivated by this, we propose Geometric-Invariance Fine-Tuning (GIFT), a geometric-aware fine-tuning method for VFMs. Specifically, we introduce a series of learnable Householder reflections to the pre-trained weights to encourage geometry-consistent adaptation under domain shift. In addition, we propose a lightweight channel-wise scaling to facilitate effective task-specific adaptation. Extensive experiments demonstrate that GIFT consistently outperforms state-of-the-art (SOTA) approaches on two S-DGOD benchmarks. Notably, by introducing fewer than 1% additional trainable parameters into the frozen backbone, the proposed method achieves improvements of +7.9% and +2.9% in mPC over the previous SOTA on the Cityscapes-C and Diverse Weather Dataset, respectively. This finding highlights that preserving representation geometry is crucial for robust cross-domain generalization, establishing GIFT as a principled and efficient solution to domain shift.
[CV-40] Shared-Factor-Conditioned Diffusion for Low-Rank Tensor Recovery
链接: https://arxiv.org/abs/2610.35368
作者: Zhengyun Cheng,Guanwen Zhang,Wei Zhou
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Tensor recovery spans heterogeneous data structures and observation processes, calling for a representation that can adapt across modalities and tasks. We propose DiffTR, an unsupervised framework that unifies tensor completion, denoising, and hyperspectral super-resolution through coordinate-conditioned diffusion. Shared factors indexed by tensor coordinates condition a compact score network, and score matching jointly learns the factors and an implicit tensor representation from the available observations. Scalar and vector-valued outputs accommodate individual entries and entire tubes, enabling a common formulation for tabular tensors, images, and videos. The framework separates representation learning from task-specific observation modeling: reverse diffusion recovers missing values; estimation via Tweedie’s formula, soft thresholding, and optional spatial consistency support mixed-noise denoising; and measurement-guided sampling transfers a prior learned from low-resolution hyperspectral observations to high-resolution reconstruction, followed by post-sampling refinement with data consistency and spectral subspace regularization. This separation makes the representation compatible with different corruption models and permits task-specific constraints to be incorporated within the same recovery framework. Experiments across multiple tensor modalities and recovery tasks demonstrate competitive reconstruction quality and the flexibility of a unified conditional generative representation. The code is available at this https URL
[CV-41] SG-Splat: Stable Geometry Learning under Pose Uncertainty for Pose-Free 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.35361
作者: Xiaohan Lu,Jiaqi Zhang,Weijia Jiang,Siwei Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:3D Gaussian Splatting (3DGS) has become a promising representation for real-time novel view synthesis and 3D scene reconstruction. For practical deployment, recent studies have explored pose-free feed-forward 3DGS, which reconstructs Gaussian representations directly from unposed input images. However, these methods remain sensitive to residual pose errors, which perturb canonical-space Gaussian transformation, introducing noisy supervision for geometry and attribute learning. To address this problem, we present SG-Splat, a pose-free feed-forward 3DGS framework that improves robustness to pose uncertainty and stabilizes the learning of Gaussian primitives. SG-Splat decouples geometry estimation and Gaussian attribute prediction into two dedicated branches, reducing error coupling between pose-sensitive geometry and Gaussian attributes. We further introduce a coarse-to-fine refinement pipeline, consisting of Flow-Guided Pose Refinement (FGPR) and Residual-Guided Geometry Correction (RGC), to improve geometric alignment before Gaussian attribute reconstruction. Specifically, FGPR refines camera poses with flow-based correlation features and differentiable bundle adjustment. RGC then corrects initial point maps using depth residuals derived from FGPR, producing more reliable Gaussian centers. To stabilize training, we introduce a dual-path strategy that mixes ground-truth and predicted poses for canonical transformation. It provides clean supervision for Gaussian attributes while improving geometry robustness to residual pose errors. Extensive experiments on DL3DV and RealEstate10K show that SG-Splat achieves state-of-the-art performance in both novel view synthesis and pose estimation.
[CV-42] Evidence-Guided Selection of Retinal Vessel Segmentation Pipelines: A Cross-Dataset Benchmark of Performance Structure Failure Modes and Reproducibility
链接: https://arxiv.org/abs/2610.35357
作者: Ruoyu Yan,Yuejia Liu,Shixin Xu,Xiaoyan Li,Xi Luo,Yuhui Deng,Huaxiong Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Retinal vessel segmentation is widely used in retinal image analysis, yet model selection based only on aggregate overlap scores can overlook structural fidelity, cross-dataset robustness, and application-specific error costs. We evaluated five executable binary vessel-segmentation pipelines (FSG-Net, SA-UNetv2, DCP-1024, GAVE, and FR-UNet) on 244 images from six public datasets using pixel-level agreement, cross-dataset stability, thin-vessel retention, structural metrics, error composition, and available computational evidence. Results were aggregated with equal weighting across datasets. No pipeline was optimal in every setting. FSG-Net achieved the highest equal-dataset mean Macro-F1 (0.77), IoU (0.63), clDice (0.83), and Skeleton F1 (0.78), whereas SA-UNetv2 achieved the highest mean sensitivity (0.85) and Thin-Vessel Sensitivity (0.63). Differences across datasets were often larger than those among the leading pipelines, and errors shared by all five methods were concentrated near annotated vessel boundaries and faint peripheral vessels. In an ancillary analysis on 45 HRF images, segmentation masks were converted into 11 ring-wise vascular features for leave-one-out classification. Controlled perturbations showed that local mask errors can be amplified in downstream features, while targeted stabilization reduced sensitivity for five features. The results support task-conditioned, benchmark-relative pipeline selection rather than a single universal ranking and highlight the need to validate the selected pipeline on the intended target domain.
[CV-43] CVTSplat: Multi-view Reconstruction Using an Adaptive Centroidal Voronoi Tessellation
链接: https://arxiv.org/abs/2610.35347
作者: Briac Toussaint,Diego Thomas,Jean-Sébastien Franco,Edmond Boyer,Christian Theobalt
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 13 pages
Abstract:Voronoi tessellations have recently emerged as an adaptive, space-partitioning discretization for shape and appearance reconstruction from multi-view images. However, the irregular cells they produce make the repeated reconstruction of the tessellation during optimization costly. We therefore explore adaptive centroidal Voronoi tessellations (CVTs), which extend standard CVTs with non-uniform cell sizes while preserving the regular, well-shaped cells that make them efficient to compute. In CVTSplat, cell size is modulated by the evolving signed distance field, concentrating sites near the zero level set while keeping cells nearly centroidal. This regularity lets us specialize the whole pipeline, with GPU algorithms for nearest-neighbor queries and approximate Voronoi construction, and a rasterizer tailored to small, nearly isotropic cells. Our Voronoi construction is 5 to 10 times faster than general-purpose alternatives in this regime. On DTU and BlendedMVS, CVTSplat reaches geometric accuracy competitive with the best Gaussian-splatting surface methods while training in about 2.5 minutes on a single GPU.
[CV-44] argeted Deep Regression - Initial Applications in Whole-Body Water-Fat MRI in UK Biobank
链接: https://arxiv.org/abs/2610.35339
作者: Filip Marttala,Taro Langner,Johan Öfverstedt,Nouman Ahmad,Håkan Ahlström,Robin Strand,Joel Kullberg
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 16 pages, 8 figures
Abstract:UK Biobank is a British clinical study which, in addition to large amounts of medical data collected, plans to include detailed Magnetic Resonance Imaging (MRI) examina- tions from 100,000 participants. The image data can be analyzed for example by auto- mated analysis with neural networks for image segmentation or classification to expand the collection of image-derived phenotypes for medical research. In this work, we pro- pose, present and evaluate a methodology that uses image-based regression to estimate numerical properties from the collected neck-to-knee water-fat MRI scans. It applies a ResNet50-style network to 2D multi-channel collages that represent targeted and study- specific anatomical sub-regions from the MRI scan. The targeted sub-region sampling is achieved by propagating manually defined subregions from a reference scan to the other 30,000 analyzed scans using image registration. The resulting collages can be compact while retaining high-resolution, for example by providing multiple slices of kidney tissue for a kidney volume regression network. This approach also enables group-wise saliency analysis that highlight image and anatomical regions used by the network for the pre- dictions. The method was applied for the regression endpoints of kidney volume, heart volume, and age. Predictive performance was found between R2 = 0.731-0.984 and the group-wise saliency maps were found to highlight medically relevant regions. We con- clude that the method can be used to standardize and restrict the information presented to a deep regression analysis pipeline, increasing information density and opening up a wide range of future medical inference studies.
[CV-45] OVM-Occ: Open-Vocabulary Monocular Occupancy Prediction via Geometry-Aware Vision-Language Distillation
链接: https://arxiv.org/abs/2610.35312
作者: Shuyang Yu,Haibo Qin,Shaohan Hu,Wei Shan,Xiaoqing Guan,You Wang,Guang Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 6 figures, 8 tables
Abstract:Monocular occupancy prediction in indoor scenes aims to recover a scene’s 3D geometry and semantics from a single RGB image. Existing methods are restricted to a closed set of predefined categories, limiting their ability to generalize to unseen categories. This limitation is particularly critical in indoor scenes, where objects exhibit highly diverse and fine-grained semantic categories. We study open-vocabulary monocular occupancy prediction for indoor scenes and present OVM-Occ, a Gaussian-based occupancy framework with an instance-aware vision-language teacher. Specifically, we adopt Language-Embedded Gaussians as the intermediate scene representation, propose a geometry-aware deformable attention module to refine Gaussian semantic features, and construct instance-level vision-language embeddings to provide reliable semantic supervision. We further introduce geometry-aware decoupled semantic distillation to mitigate the interference between geometry learning and semantic learning. To evaluate open-vocabulary occupancy prediction performance, we introduce OpenOcc-ScanNet and OpenOcc-Replica, two extended benchmarks with more diverse and fine-grained semantic categories. Experiments show that our method achieves state-of-the-art performance in open-vocabulary monocular occupancy prediction for indoor scenes, outperforming the previous state-of-the-art approach by 12.2 mIoU and 0.6 IoU on the Occ-ScanNet dataset, with consistent gains under new open-vocabulary evaluation protocols.
[CV-46] Low-rank MERA Decomposition of 3D OCT Image Volume for Compression Denoising Feature Extraction and Diagnosis of Retinal Diseases
链接: https://arxiv.org/abs/2610.35271
作者: Ivica Kopriva,Fei Shi,Marija Štanfel,Qin Peng,Jianhao Bao,Gaowei Li,Xinjian Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 4 figures, 5 tables
Abstract:Optical coherence tomography (OCT) is a non-invasive imaging modality used for visualization of retinal tissue. However, OCT volumes are affected by speckle noise, which degrades image contrast and obscures fine structural details. Furthermore, modern clinical practice requires large field-of-view, high-resolution OCT volumes, whose storage and transmission are demanding for archiving and teleophthalmology applications. To address aforementioned challenges, we propose a low-rank multi-scale entanglement renormalization ansatz (MERA) tensor network approximation of 3D OCT image volume, where the image volume is represented as contractions of a top core factor encoding global features and multiple interacting low-rank orthogonal or semi-orthogonal factors modeling inter- and intra-B scan correlations. This representation jointly enables compression, denoising, and feature extraction. Unlike deep learning approaches, the proposed approach is training-free and processes each OCT volume independently. We evaluate the proposed framework on OCT image volumes acquired by two different scanners and demonstrate improved visual quality and reduced retinal layer segmentation errors in comparison with several competing methods. We also extract MERA features from two clinical OCT volume datasets. They are subsequently classified using a Grassmann-manifold nearest-subspace classifier. With the subspace dimension as its only hyperparameter, the classifier is theoretically shown to be robust in small-sample clinical scenario. We demonstrate competitive accuracy of classification of retinal diseases in these small-sample settings, where deep learning methods are often limited by the scarcity of labeled data. The proposed structure-aware model-driven representation learning for high-dimensional data under limited supervision provides good generalization performance while maintaining high interpretability.
[CV-47] hermal-Headroom Routing for Efficient Robot Perception in Extreme Heat
链接: https://arxiv.org/abs/2610.35231
作者: Sparsh Roy,Kevin Zhou,Arhaan Shah,Shreyan Garg,Vihaan Joshi-Apte,Maxim Kudryashov,Nanda Cheruvanthala
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Performance (cs.PF)
备注: 9 pages, 5 figures, 11 tables. Code and experimental outputs: this https URL {Submitted IEEE Big Data}
Abstract:Adaptive robot perception systems must balance detector accuracy, inference latency, and energy consumption, particularly when high ambient temperatures cause processors to thermally throttle. We propose Thermal-Headroom Routing (THR), a lightweight model-selection policy that forecasts processor temperature using a closed-form RC thermal model and excludes detector configurations predicted to cross a throttling threshold. We evaluate THR with four YOLO11 configurations across seven real-world perception traces, eight ambient temperatures, and 20 randomized seeds, comparing it with latency-aware routing, fixed-threshold thermostats, duty caps, and a matched-compute-budget control. At ambient temperatures of 45-55 degrees C and a tight inference deadline, THR eliminates throttling in simulation, reduces deadline misses by 3.9-7.5 percentage points, and reduces energy per on-time frame by approximately 12-13% relative to a latency-belief router. A separate experiment using deadline-paced, nonoverlapping frame execution yields miss-rate reductions of 6.2-9.5 percentage points. A tuned thermostat or duty cap can match THR at a fixed operating point; predictive routing is most useful when ambient conditions or calibrated thermal parameters differ from those used to tune static policies. Measurements on an Apple M5 support the single-node heating approximation but do not validate hardware throttling. The results characterize the conditions under which thermal forecasting improves the efficiency and timeliness of resource-constrained robot perception.
[CV-48] Joint Latent Reasoning and Observation for Agent ic Video Understanding
链接: https://arxiv.org/abs/2610.35229
作者: Junzhang Liu,Hao Yuan,Qi Zhang,Yiming Wu,Long Chen,Zhecan Wang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Effective video agents must learn not only how to answer, but also where to look next. Transferring this capability to smaller models is challenging because each observation serves two purposes: supporting an answer and guiding the search for further evidence. Textual descriptions may capture what is currently visible while omitting cues needed for subsequent acquisition. We introduce Latent Agentic Reasoning for Video Analysis (LARVA), a framework that distills multi-round video agents through jointly learned latent reasoning and visual observations. Rather than communicating through captions, an observer passes compact continuous representations to a reasoner that selects the next acquisition action or produces the final answer. Crucially, action and answer supervision propagate through this interface, training the observer to communicate information useful for downstream decisions. Observation reconstruction and reasoning-distribution alignment provide auxiliary supervision from textual teacher trajectories. Across five video-understanding benchmarks, LARVA achieves strong performance with 4B and 7B backbones. On VideoHolmes, Video-MME, and LVBench, the 7B model outperforms a textual CoT student trained on the same teacher trajectories by 5.1, 2.5, and 3.7 percentage points, respectively. These results highlight the value of jointly learning how agents represent evidence and act on it when distilling video-understanding capabilities.
[CV-49] GaussianSpline: Adaptive Bézier Splines for Highly Dynamic 4D Gaussian Splatting
链接: https://arxiv.org/abs/2610.35224
作者: Tom Yachil,Andrei Sharf
类目: Graphics (cs.GR); Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 4 figures
Abstract:We present GaussianSpline, a dynamic 4D Gaussian Splatting method that models per-Gaussian motion using adaptive, non-uniform piecewise cubic Bézier splines. Unlike approaches based on global deformations, fixed-order polynomials, or uniformly parameterized splines, GaussianSpline jointly learns the Bézier control points and the temporal partition of each Gaussian’s trajectory. This adaptive representation allocates greater temporal capacity to highly dynamic regions while maintaining compact trajectories for near-static regions, enabling high-fidelity reconstruction of complex and rapidly changing motion. To jointly achieve globally smooth motion and local adaptivity, we enforce G^1 continuity between consecutive Bézier segments using a reflection-based constraint on their control points. This produces smooth, fully differentiable trajectories while preserving the flexibility of non-uniform temporal parameterization. Experiments on the Neural 3D Video benchmark, the Google Immersive dataset, as well as on synthetic scenes demonstrate that GaussianSpline substantially improves reconstruction quality in scenes with highly non-uniform and challenging dynamics, while maintaining competitive performance on general dynamic scenes compared with state-of-the-art 4D Gaussian Splatting methods.
[CV-50] O2D-MedNet: An Optimization-Aware Dual-Domain Framework for Robust Medical Image Segmentation
链接: https://arxiv.org/abs/2610.35220
作者: Shipeng Hu,Miaohui Wang,Jing Qin,Sukun Tian,Yuchun Sun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 16 pages, 12 figures
Abstract:Medical image segmentation (MIS) remains challenging due to composite degradations and multi-scale geometric variations, which jointly impair lesion-background separability, structural consistency, and optimization stability. Existing single-stream frameworks entangle heterogeneous cues within a single representation pathway, while hybrid-stream approaches often rely on generic or static fusion, limiting effective coordination across domains and failing to explicitly address optimization instability. To overcome these limitations, we propose a new optimization-aware dual-domain framework for MIS (O2D-MedNet). First, a degradation-aware dualdomain interaction (DDI) module integrates Fourier - and Wavelet-domain features to suppress degradation-sensitive responses while preserving structural and fine-detail information. Second, a cross-backbone contour-guided fusion (CGF) module enables selective interaction between CNN-based local detail modeling and visual state-space-based global structure modeling, improving boundary accuracy and representation consistency. Third, a gradient-context optimization (GCO) strategy stabilizes training by adaptively adjusting learning dynamics based on loss trends and gradient statistics, mitigating instability caused by class imbalance and heterogeneous sample difficulty. Extensive experiments on six benchmarks demonstrate consistent improvements over state-of-the-art MIS methods across multiple imaging modalities. Further analyses confirm enhanced sensitivity, structural consistency, and optimization stability, validating the effectiveness of the proposed O2D-MedNet for robust MIS under heterogeneous conditions.
[CV-51] Multi-Agent Relay Thinking
链接: https://arxiv.org/abs/2610.35219
作者: Xu Liu,Jungang Qiao,Guikun Chen,Xiaojuan Qi,Wenguan Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing multi-agent collaboration (MAC) methods have primarily explored a group thinking collaboration topology, where agents independently generate solutions and aggregate their outputs. In this work, we investigate an alternative relay thinking topology, where reasoning ownership is dynamically transferred across agents, allowing subsequent agents to inherit and refine an evolving reasoning trajectory. We introduce Multi-Agent Confidence-guided Relay (MACR), a framework that enables dynamical relay collaboration among agents. The key challenge is determining when reasoning should be relayed and which agent should continue the process. method addresses this challenge through confidence-guided relay triggering and a bidding mechanism that selects the next agent based on its confidence. Experiments on four benchmarks show that method outperforms the strongest single-model strategies in accuracy, e.g., +3.3% on MathVista and +3.9% on MMStar, and remains on par with multi-agent strategies while consuming 6.0x less token overhead on HallusionBench and this http URL analysis demonstrates that relay collaboration enables heterogeneous agents to progressively integrate complementary insights, combining heterogeneous capabilities across reasoning stages. Our source code is available at this https URL.
[CV-52] OrthoRF: Exploring Orthogonality in Object-Centric Representations ICLR2026
链接: https://arxiv.org/abs/2610.35203
作者: Despoina Touska,Bastiaan Onne Fagginger Auer,Alexandru Onose,Tejaswi Kasarla,Luis Armando Pérez Rey,Maximilian Lipp,Lyubov Amitonova,Martin R. Oswald,Pascal Cerfontaine
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Poster presentation at ICLR 2026
Abstract:Neural synchrony is hypothesized to help the brain organize visual scenes into structured multi-object representations. In machine learning, synchrony-based models analogously learn object-centric representations by storing binding in the phase of complex-valued features. Rotating Features (RF) instantiate this idea with vector-valued activations, encoding object presence in magnitudes and affiliation in orientations. We propose Orthogonal Rotating Features (OrthoRF), which enforces orthogonality in RF’s orientation space via an inner-product loss and architectural modifications. This yields sharper phase alignment and more reliable grouping. In evaluations of unsupervised object discovery, including settings with overlapping objects, noise, and out-of-distribution tests, OrthoRF matches or outperforms current models while producing more interpretable representations, and it eliminates the post-hoc clustering required by many synchrony-based approaches. Unlike current models, OrthoRF also recovers occluded object parts, indicating stronger grouping under occlusion. Overall, orthogonality emerges as a simple, effective inductive bias for synchrony-based object-centric learning.
[CV-53] Complexity-guided Regularization for Generalizable Human Gaussian Splatting NEURIPS2026
链接: https://arxiv.org/abs/2610.35152
作者: Gun Ryu,Gi-Mun Um,Won-Sik Cheong,Wonjun Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 11 figures, including appendix. Accepted to NeurIPS 2026
Abstract:Recent studies on generalizable human Gaussian splatting have achieved the notable progress in synthesizing novel views of unseen human subjects from multi-view images. Despite this progress, existing methods still struggle to represent fine-grained details of human subjects, which inherently involve high geometric and textural complexity. This is because previous methods do not impose a constraint on how densely Gaussians are placed, which often leads to blurry rendering results in complex regions. To address this problem, we propose a complexity-guided regularization scheme. The key idea is to leverage a complexity prior computed via the local entropy in geometric and textural characteristics of a clothed human. This effectively prevents Gaussians from being coarsely placed in complex regions, which facilitates the faithful reconstruction of intricate structures. Furthermore, we sample local patches predominantly from complex regions according to this complexity prior, which allows the model to be sufficiently supervised by local details. Experimental results on benchmark datasets demonstrate that our method delivers state-of-the-art performance while maintaining the real-time inference speed in generalizable human Gaussian splatting. The code and model are publicly available at this https URL.
[CV-54] mporal Slice Learning for AI-Generated Video Detection with 400times Fewer FLOPs NEURIPS2026
链接: https://arxiv.org/abs/2610.35148
作者: Wei Shang,Zhiyuan Li,Shengfeng He
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. 35 pages, 18 figures. Project page: this https URL
Abstract:Recent advances in generative video models have improved visual realism, making detection increasingly dependent on subtle temporal inconsistencies. However, these signals are weak and spatially localized, while most of the video content is redundant. This creates a fundamental mismatch: existing detectors rely on heavy backbones to model full videos, which leads to high computational cost and limited scalability. In this work, we rethink AI-generated video detection from a representation perspective. Instead of modeling entire videos, we restructure them into compact temporal slice representations that isolate informative temporal evolution. We introduce a lightweight framework using frequency-aware temporal slice consistency learning, where informative spatial rows and columns are aggregated over time to form slice-domain inputs. This reparameterization suppresses redundant appearance while preserving discriminative temporal dynamics, enabling efficient and targeted detection. By constructing slices from both high- and low-frequency regions, our method exposes complementary temporal artifacts, including unstable details and abnormal motion smoothness. We instantiate this idea in a Temporal Slice Consistency Network, which integrates task-driven slice localization, hierarchical positional encoding, and a lightweight Transformer to model cross-slice and cross-frequency dependencies. We further introduce AI-Artist, a new benchmark with artist-curated videos from recent high-fidelity generators, including Seedance and Kling. Experiments on GenVideo and AI-Artist show that our method achieves strong performance while requiring over 400 \times fewer model-forward FLOPs and 3.6 \times faster model inference than the strong video-based baseline ReStraV (2.9 \times faster end to end, including preprocessing), enabling scalable and practical deployment.
[CV-55] Sexual dimorphism and sex-specific features of pediatric hand bone shapes via statistical shape modeling
链接: https://arxiv.org/abs/2610.35143
作者: Edwin Tay,Amir A. Zadpoor,Nazli Tümer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Understanding the envelope of shape variation encompassing both sexual dimorphism and growth-related changes in living tissue is a non-trivial task. In this work, we use statistical shape models (SSMs) and analyze demographic-stratified shape parameters to quantify sexual dimorphism and sex-specific shape changes. To this end, we developed a two-stage detection-segmentation framework that accurately segmented n = 27664 unique bone masks from a radiographic dataset of n = 1456 subjects with diverse ages and sexes (HD95 = 1.512 \pm 0.005 , ASSD 0.449 \pm 001 , Dice = 0.987 \pm 0.00004 , IoU = 0.975 \pm 0.0001 ). Following SSM fitting, our follow-up analyses of stratified shape parameters highlighted the morphology of sexual dimorphisms for each bone and demonstrated sex-specific differences in shape development. Our results were consistent with existing literature and provided an unambiguous, data-driven methodology for characterizing tissue shape and its population-scale variation. This approach provides a potential methodological foundation for developing population-based models of anatomical variation and could be extended to investigate tissue growth and shape changes over time. Future work should evaluate the applicability of this methodology on alternative anatomical structures and imaging modalities, with the goal of validating these growth trajectories on longitudinal data.
[CV-56] Domain-Adapted Latent Diffusion Models for CT-to-PET Image Translation
链接: https://arxiv.org/abs/2610.35132
作者: Yann Bourdé,Lea Bogensperger,Arnela Hadzic,Michael Krauthammer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Positron Emission Tomography (PET) provides important metabolic information for tumor detection and staging but remains significantly less accessible than Computed Tomography (CT) due to its high cost, infrastructure requirements, and reliance on injected radioactive tracers. CT-to-PET image translation using generative models aims to address this disparity, yet state-of-the-art diffusion-based methods are limited by the scarcity of paired training data. To overcome this, we propose a domain-adapted latent diffusion framework that transfers natural-image priors using a relatively small dataset of 21,930 paired CT/PET slices, incorporating a signed power transform to resolve extreme intensity sparsity in PET data. Compared to CPDM, a state-ofthe-art baseline for CT-to-PET image translation, our approach yields superior performance in pixel-wise reconstruction and localized tumor fidelity, demonstrating effective transfer of natural-image priors to the medical domain. We further provide a systematic analysis of evaluation metric reliability for sparse PET data, showing that standard full-image metrics can be misleading indicators of translated image quality, and propose background-filtered and mask-restricted metrics as more clinically meaningful alternatives.
[CV-57] Understanding and Repairing Cooperative Perception under Adverse Weather
链接: https://arxiv.org/abs/2610.35110
作者: Mengxu Ma(Northwestern Polytechnical University)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:A cooperative 3D detector can miss a vehicle even when the LiDAR points on it remain unchanged. We study this failure under adverse weather as Weather-Induced Context Interference. To our knowledge, this is the first controlled study of how weather-related exterior changes remove cooperative benefits while preserving all points inside annotated objects. Controlled experiments on simulated and real roadside inputs show that exterior loss changes object representations before fusion and removes previously useful assistance. On real inputs, restoring object-neighborhood features recovers 30 of 31 induced failures; three equal-area exterior controls recover none. The failure condition also appears with a second fusion model. These findings motivate Adverse-Weather Cooperative Repair (AWCR), with separate implementations for the available supervision. Point Classifier filters added weather returns in simulation. On real inputs, Local Correction refines candidates and Quality Selection checks them against current measurements, without naturally paired clear/adverse-weather recordings. Candidate tracing shows that correction supplies accurate geometry for some targets that original candidates cannot localize at the evaluation threshold. On natural adverse-weather CoInfra development data, AWCR improves car bird’s-eye-view (BEV) AP@0.7 by 4.66 points with V2X-ViT and 5.56 with AttFuse. On simulated V2X-R fog, it improves vehicle BEV AP@0.7 by 4.47 points at a fixed 4,096-candidate budget.
[CV-58] Uncertainty Evaluation to Support Safety of the Intended Functionality Analysis for Determining the Applicability of Performance Targets Across Triggering Conditions in ML-Based LiDAR Object Detection
链接: https://arxiv.org/abs/2610.35092
作者: Milin Patel,Rolf Jung
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Robotics (cs.RO)
备注: 24 pages, 6 figures. Extended version of the VEHITS 2026 conference paper (doi: https://doi.org/10.5220/0014967200004030 ). Submitted for journal publication. Code and recorded results: this https URL
Abstract:For safety of the intended functionality (SOTIF) analysis according to ISO 21448, a machine learning object detector for light detection and ranging (LiDAR) point clouds can meet a performance target, a limit on the mean per-frame share of wrong detections among accepted detections, by rejecting detections above an uncertainty threshold. A threshold selected on data of one condition carries a statistical guarantee only for data distributed like its selection data, whereas SOTIF analysis covers every triggering condition. This paper presents an uncertainty evaluation in which the threshold is transferred into a triggering condition, a test decides at a stated error probability whether the target is applicable there, not applicable or not determined, and a lower bound counts the independent units an applicable outcome requires. For PointPillars ensembles on KITTI and nuScenes, counted in recorded sequences, as the dependence between frames requires, neither dataset held enough independent units for any target limit, and no transfer had a not applicable outcome. Counted in frames, the same data gave not applicable outcomes at limits fixed in advance after transfers into far distance, into rain and between two separately trained detectors, thresholds that met the target limit after a transfer rejected most correct detections, and in rain the area under the receiver operating characteristic curve did not decrease. A performance target met under one condition therefore counts as evidence under a triggering condition only after a test there, reported with its share of accepted correct detections and its number of independent recorded sequences.
[CV-59] st-Time Prompt-Agnostic Decomposition NEURIPS2026
链接: https://arxiv.org/abs/2610.35090
作者: Junze Wang,Lei Fan,Dezheng Zhang,Donglin Di,Yang Song,Sidong Liu,Cong Cong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by NeurIPS 2026
Abstract:Test-Time Prompt Tuning (TPT) adapts large pretrained models to unseen target distribution shifts by updating only lightweight prompt tokens. Existing TPT methods are effective, but they mainly optimize each prompt update locally. Across test-time steps, prompts updated at the current step are reused for the corresponding predictions and updates, so noisy or biased unlabeled objectives can accumulate and destabilize the prompt-update trajectory. We identify two failure modes: magnitude expansion, where updates move the prompt too far, and directional drift, where consecutive updates point in inconsistent directions. To analyze and control these failures, we propose Test-Time Prompt-Agnostic Decomposition ( \mathttTPD ), which characterizes observed prompt-update trajectories from a dynamical-systems view and decomposes base prompt updates into magnitude and direction. \mathttTPD computes a spectral radius to measure update expansion and decomposes update directions into persistent, oscillatory, and residual components. Building on this decomposition, Adaptive Koopman Control ( \mathttAKC ) regulates update magnitude by shrinking updates under expansive recent dynamics, while Hankel Update Router ( \mathttHUR ) refines update direction by preserving persistent components and suppressing oscillatory and residual components. Together, \mathttAKC and \mathttHUR produce a stabilized prompt update. As a prompt-agnostic framework, \mathttTPD can be plugged into visual and text TPT methods without modifying the backbone, prompt architecture, or adaptation loss. Experiments on 15 datasets show that \mathttTPD consistently improves 10 baselines, with accuracy gains of 3 – 8% for visual prompts and over 2% for text prompts.
[CV-60] Blendshape Generation for Models with Disconnected Parts: Adding ARKit/MediaPipe Blendshapes to Anime-Style Character Models WACV2027
链接: https://arxiv.org/abs/2610.35076
作者: Pramook Khungurn
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Machine Learning (cs.LG)
备注: Accepted to WACV 2027
Abstract:Blendshape generation for 3D models has been extensively researched, but very few works address generating blendshapes for models with multiple disconnected parts. We present a case study of generating fixed facial expression blendshapes for commercial anime-style 3D character models with up to 11 parts per face. To make sure that our system can handle models in a way that is agnostic to topology, we train neural networks to generate deformation fields around the input meshes. We employ at least one network to independently process each part, and the relative positions of their outputs are coordinated by having each network also process the geometry of an anchor part. We use two mechanisms to improve accuracy: having the networks predict whether the output displacement is zero or not, and classifying blendshapes into different size groups and dedicating a network to each group. We also introduce a method to improve blendshape quality by enhancing bilateral symmetry. Our method is simpler than recent neural-network-based works and solves a real problem of the user community of these models.
[CV-61] MemVLN: Streaming Episodic Memory for Vision-Language Navigation
链接: https://arxiv.org/abs/2610.35066
作者: Yan Wang,Jiangyong Huang,Baoxiong Jia,Siyuan Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Vision-language navigation (VLN) requires an embodied agent to follow language instructions through 3D environments. Maintaining persistent episodic memory remains a challenge for VLM-based streaming agents in long-horizon navigation. We present MEMVLN, a streaming episodic memory framework that maintains a bounded bank of compressed perceptual features annotated with timesteps and action-derived relative poses, and produces compact memory tokens via cross-attention. We pair the proposed memory with a frozen pretrained geometry encoder. To learn the memory module, we design a masking-based trajectory summarization task with multi-granularity sub-trajectory decomposition, encouraging route reconstruction exclusively from episodic memory. Under navigation training restricted to the official R2R/RxR trajectories, MEMVLN achieves state-of-the-art performance on most reported metrics on standard VLN-CE benchmarks. Long-horizon analyses further characterize when episodic memory is useful. For physical evaluation, we collect 30 instruction-conditioned episodes in one physical scene, use 24 as demonstrations for post-training, and hold out six for testing. On the held-out episodes, MEMVLN improves success over the VLM baseline from 17% to 40%, and post-training further raises it to 47%.
[CV-62] Kronecker-Structured Video Completion with Cross-Fitted Self-Supervised Residual Learning
链接: https://arxiv.org/abs/2610.35065
作者: Qi Qi,Yanfei Wang,Yanjun Cui,Jun-e Feng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Tube-wise video completion reconstructs temporal trajectories at spatial locations unobserved in every frame. We propose a two-stage framework combining regularized masked Kronecker product decomposition (KPD) with cross-fitted self-supervised residual learning. First, we estimate a compact structural backbone directly from observed entries using cyclic component-wise least-squares updates, spatial and temporal factor regularization, and regularizer-optimal scale balancing. The exact updates yield a nonincreasing objective. Second, we address the mismatch between in-sample fitting errors and the prediction errors encountered at missing tubes. We partition observed tubes into folds and construct residual targets using backbone models fitted without the corresponding target tubes. A lightweight 3-D convolutional neural network then predicts corrections from the backbone, visible residual context, observation mask, and local detail features. Exact data consistency preserves observed measurements. Experiments include ablations, sensitivity analysis, grayscale and RGB video completion, and structured missing patterns. Our method achieves the highest mean PSNR and SSIM at the higher tested sampling rates on the grayscale benchmark. On eight selected DAVIS RGB sequences, it achieves the highest mean missing-region PSNR and full-frame SSIM and the lowest mean missing-tube temporal-gradient MAE at every evaluated sampling rate.
[CV-63] LOCUS: Locate Once Check Until Settled. Draft-and-Verify Visual Grounding with a Box Head and Coordinate-Token Verification
链接: https://arxiv.org/abs/2610.35053
作者: Muhammad Talha
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 7 figures, 10 tables. Project page: this https URL Code: this https URL Models: this https URL and this https URL
Abstract:Vision-language models (VLMs) ground a referring expression in one of two ways. They decode the box as coordinate tokens, which is precise but takes one sequential step per token, or they regress it from a hidden state with a small head, which takes one step but loses accuracy at high IoU. The two fail differently: a head usually finds the right object but places its edges loosely, while coordinate tokens place edges precisely. We therefore use the regression head as a speculative drafter for the model’s own coordinate tokens. Locus writes the head’s box as coordinate tokens and checks them with the same backbone by Jacobi iteration over a shared image prefix. The fixed point is the model’s greedy coordinate-token answer, reached in about 5.60 sequential passes on average instead of 18.46 decoding steps; head and verifier are one set of weights, trained jointly. On the 8 RefCOCO, RefCOCO+ and RefCOCOg splits, verification raises Locus-2B’s Acc@0.9 from 54.87 for its own head to 59.13, and the one-pass head is ahead at Acc@0.5 of every published model with at most 3B parameters except Florence-2-L. Against Jag, the strongest open regression-head model, at its own size and training budget, Locus ties at Acc@0.5 and, with typical acceptance, gains +4.52 points at Acc@0.9, with paired significance tests. Exact answers come 1.5x faster than vLLM greedy-decoding the same coordinate tokens with the same weights; a pure regression head remains faster still.
[CV-64] DriveUniTok: A Unified Multimodal Tokenizer for Driving World-Action Models
链接: https://arxiv.org/abs/2610.35044
作者: Jingbo Xu,Hanchao Leng,JiaHui Huang,Hanbing Li,Jingwei Zhao,Kuiyuan Yang,Chen Gao,Guang Chen,Hangjun Ye,Yong Li
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 31 pages, 16 figures, 8 tables, including appendices
Abstract:World models are emerging as a fundamental component of policy learning for real-world agents, enabling agents to predict future world states and plan actions within a learned representation of the dynamic environment. Existing approaches, however, often model appearance, metric geometry represented by projected LiDAR depth, and ego motion either partially or independently, overlooking the fact that these modalities are different projections of a shared underlying world state. This separation can lead to information loss and cross-modal inconsistency, ultimately limiting downstream performance. Motivated by this unified view of the world, we propose DriveUniTok, a unified multimodal tokenizer for driving world-action models (WAMs) that encodes these modalities in a single decodable continuous latent. We freeze DriveUniTok and adapt a pretrained diffusion model, DriveUniWAM, to predict future unified latents; the tokenizer then decodes them into future appearance, projected LiDAR depth, and ego trajectories. Experiments on NAVSIM demonstrate that our unified design achieves better future prediction and planning performance at lower downstream computational cost than separate-stream alternatives. With reinforcement learning post-training, DriveUniWAM achieves a PDMS of 91.54, reaching state-of-the-art performance among the evaluated WAM-based planners.
[CV-65] Revealing Imaging Characteristics of Thermal Wakes from Underwater Vehicles Without Cooling-Water Discharge
链接: https://arxiv.org/abs/2610.35038
作者: Guoyi Zhang,Yanjin Du,Guangsheng Xu,Zhengyao Zhao,Siyang Chen,Jingjing Gao,Yuezheng Zhou,Yiyang Wu,Bin Wu,Tongsu Zhang,Lejia Ye,Donghe Wang,Jie Wang,Kaiye Hu,Zhipeng Ren,Han Wang,Boo Cheong Khoo,Xiangpeng Xu,Xiaohu Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Small uncrewed underwater vehicles are proliferating and threaten maritime security, yet evade sonar and magnetic sensors built for larger submarines. Infrared thermal wakes, generated by hydrodynamic displacement that modulates the cold-skin layer, offer a non-acoustic, non-magnetic alternative even without cooling-water discharge. Existing studies, however, rely on idealized simulations that fail to capture real-world backgrounds, offering limited guidance for target-background discrimination. We establish AVP-IR-Wake, the first large-scale, real-world, physically diverse benchmark of 161,485 frames collected over three years. We show that thermal wakes are extended structures of discrete vortices, violating the compactness assumption of conventional targets. We reveal local indistinguishability: individual wake vortices are statistically identical to wind-induced background vortices in temporal, frequency, and sparse domains, sharing low-frequency information and long-range correlations. This challenges the assumptions of compact, transient, high-frequency targets in conventional detection. Yet we establish global morphological separability: although locally inseparable, the wake as a whole is organized by vehicle trajectory, whereas background clutter is organized by wind-driven thermal fronts. This explains why representative detectors fail and shows that reliable detection must shift from local feature extraction to global trajectory-constrained physical modeling. Because the cold-skin layer varies globally and seasonally, forecasting wake observability will require integrating physical oceanography and atmospheric science, linking underwater surveillance to the changing ocean surface.
[CV-66] Zero-Shot Monocular Depth Estimation for Visual Perception in Lower-Limb Prostheses
链接: https://arxiv.org/abs/2610.35018
作者: Ishi Vijayvergiya,Marc Hölle,Youssef Dawoud,Amir El-Ghoussani,Michele De Vita,Rohan Asthana,Philipp Beckerle,Vasileios Belagiannis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at IEEE International Conference on Integrated Systems in Medical Technologies 2026
Abstract:Monocular depth estimation has advanced rapidly with the development of foundation models that demonstrate strong zero-shot generalisation across diverse visual domains. However, the applicability of these models to visual perception for lower-limb prosthetics remains largely unexplored. Reliable depth perception is particularly important for prosthesis users, who often experience significant challenges when navigating stairs, uneven terrain, and other complex environments. To address this gap, we introduce the Lower-Limb Prosthesis (LP) dataset, a benchmark designed to evaluate depth perception from an emulated knee-level prosthetic viewpoint. Using this dataset, we evaluate four state-of-the-art monocular depth estimation foundation models. Our results provide the first systematic benchmark of monocular depth estimation performance from a prosthetic viewpoint, while highlighting both the challenges and the potential of deploying foundation models in prosthetic perception systems. The code and dataset are available at this https URL.
[CV-67] When Pointing Becomes Profiling: Social Bias and Refusal in Text-Steerable Visual Encoders
链接: https://arxiv.org/abs/2610.35017
作者: Seyedarmin Azizi,Parsa Razmara,Sajjad Ghiasvand,Ardalan Aryashad,Tina Khezresmaeilzadeh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-steerable visual encoders condition a frozen Vision Transformer on a natural-language prompt and point at the matching image region, producing prompt-conditioned heatmaps and features. We audit the released SteerViT model and find that its steering interface turns language-driven pointing into demographic profiling. On counterfactual face mosaics and real multi-person photographs, “the person” selects a man more often than the base rate, crime words select Black people more often than occupation words do, “the man” among four women selects the Black woman 75.6% of the time on SocialCounterfactuals, and “the monkey” selects Black people. None of them is in the backbone’s unsteered saliency, and the crime association is not in the frozen text encoder; the race effects grow with steering strength and enter through the last two cross-attention layers: giving the prompt to those layers alone reproduces them. Four other grounding models, CLIPSeg, Grounding-DINO, SAM 3, and Qwen2.5-VL, show no crime routing and weaker race routing. The model cannot refuse. Its patch softmax must place its mass somewhere, and its heatmap separates a groundable request (“the police officer”) from one that is absent (“the pizza”) or that would require a stereotyped demographic inference (“the criminal”) only at chance (AUROC 0.52). We introduce SafeSteer, which adds a learned abstention output and counterfactual-invariance training on the same 23M steering parameters. SafeSteer separates should-point from should-abstain prompts at AUROC 0.997, generalizes to held-out sensitive and absent words (0.998; at a 0.5 threshold it refuses 99.8% of trained and 62% of held-out sensitive words) with no over-refusal of visible people on the benchmark, and removes most race routing while keeping localization and retrieval within 1 point of the released model. Code, benchmarks, and the curated mosaic sets will be released.
[CV-68] Shifted Window Temporal Transformer for YOLO-Guided Phase Recognition in Laparoscopic Cholecystectomy Surgery
链接: https://arxiv.org/abs/2610.35005
作者: Sheekar Banerjee,Md. Shahabuddin,Md. Mahbub Hasan Akash,Nabilah,Mohammad Arafat Hussain
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Intelligent operating rooms require accurate awareness of surgical workflow for better intra-operative decisions, teamwork, and surgical documentation. Therefore, accurately recognizing surgical phase during surgery is crucial. However, previous artificial intelligence-based methods provide fluctuating predictions for the transitions, restricting their clinical use. To address this challenge, we propose a clinically meaningful framework that integrates a Swin Transformer and Temporal Transformer Encoder (TTE) for effective surgical phase recognition and a parallel YOLOv11s network for real-time surgical instrument localization. To avoid phase flickering and to produce stable workflow predictions, a causal temporal smoothing is applied. The proposed framework is tested on the Cholec80 dataset for the simultaneous recognition of the surgical phases and instruments. The overall phase recognition accuracy of this model was 92%, and the macro average F1-score was 0.86. This performance surpasses CNN based methods such as EndoNet (81.7%) and SV-RCNet (85.3%), as well as temporal convolutional and transformer-based approaches including TeCNO (88.6%), Trans-SVNet (90.9%), and LoViT (91.5%). Besides, the module of instrument detection achieved mAP@0.5 = 0.863. The reliability of the results is confirmed, obtaining a high recognition rate of the workflows, and a high accuracy of the localization of the instruments. The proposed framework can serve as a foundation for intelligent surgical assistance, enriching situation awareness, and potential robotic surgery applications. Code will be made publicly available at: this https URL
[CV-69] DUET4D: Self-Corrective Coupled Guidance for Sparse-View 4D Reconstruction
链接: https://arxiv.org/abs/2610.35002
作者: Beizhen Zhao,Nanjie Yao,Boyi Fu,Qingsong Yan,Haofan Ren,Kaiyong Zhao,Hao Wang
类目: Computer Vision and Pattern Recognition (cs.CV); Computational Geometry (cs.CG)
备注:
Abstract:Sparse-view 4D reconstruction is highly under-constrained when only a few cameras are available. Generative priors can recover missing appearances, but video models are computationally impractical for iterative reconstruction, whereas single-view models produce temporally inconsistent pseudo-observations. Moreover, directly absorbing uncertain generative outputs into Gaussian optimization often creates redundant primitives and floating artifacts. To address these challenges, we propose DUET4D, a self-corrective coupled guidance framework for sparse-view 4D Gaussian splatting. The core of our approach is a self-corrective framework where the 4D representation guides generative prior for spatio-temporal consistent supervisions. We first construct sequential alignments from current Gaussian evidence to steer the latent restoration. Then we use Sequential Fidelity Resampling to lift generative supervisions into Gaussian space to selectively distinguish repeatedly supported primitives. The corrected representation then provides more reliable Gaussian evidence for the next restoration cycle. Extensive experiments demonstrate that DUET4D significantly improves rendering quality and rendering efficiency.
[CV-70] More Memory Better Videos? A Controlled Study of Visual Memory in Long-Video Generation
链接: https://arxiv.org/abs/2610.35001
作者: Yuzhuo Chen,Huafeng Shi,Xinyu Wang,Yulong Xu,Yucheng Wang,Haoqin Hong,Guoxin Zhang,Zehua Ma,Weiming Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Long-video generators increasingly retrieve past visual observations to prevent identity drift. The intuition is compelling: more access to history should make future video more consistent. But an image can be faithful to the past and wrong for the present. Does visual memory always improve the generated story? We investigate this question through VMem-Bench, a two-track diagnostic study. Track-A measures which references are selected from 92 real films (16,548 segments), before generation and without generator confounds. Track-B compares four retrieval policies with no-memory and gold-informed last-seen controls on 30 causally generated stories (2,824 segments) using a shared generator. Retrieval increases entity recall from 0.559 to 0.628-0.693 and improves similarity to past appearances, but decreases precision from 0.977 to 0.915-0.938 and forbidden-entity avoidance from 0.968 to 0.845-0.898. It can also reinforce obsolete appearances after the story changes them. Conversely, high long-gap recall can arise without memory when the prompt itself supplies the appearance; an appearance-stripping analysis exposes this confound. These results distinguish historical fidelity from current-story correctness and show why retrieval relevance or final-video consistency alone does not establish memory utility.
[CV-71] Sparse Recovery Birds-Eye View Model Soup for Robust Camera-Based 3D Object Detection IROS2026
链接: https://arxiv.org/abs/2610.34976
作者: Jinwoo Chung,Sangho An,Hyunsung Park,Jangho Kim
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to IROS 2026. 8 pages, 6 figures
Abstract:Camera-based bird’s-eye view (BEV) perception methods such as Fast-BEV have demonstrated promising results with fast inference speed and competitive accuracy. However, their robustness to real-world corruptions remains insufficiently explored, posing risks for autonomous driving deployment where environmental variations are inevitable. In this paper, we propose Sparse Recovery Bird’s-Eye View Model Soup (SR-BEVS), a sparse recovery learning approach that simultaneously improves clean accuracy and corruption robustness without requiring additional data or architectural modifications. Our method identifies pruning candidates based on layer-wise weight magnitudes, applies random pruning to create diverse sparse models, and retrains each model with a randomly sampled learning rate. By averaging these diverse recovered models into a single dense model, we achieve enhanced robustness against various corruptions while introducing zero inference overhead. We further provide a mode connectivity analysis demonstrating that our pruning-based diversification preserves the connected low-loss basin necessary for stable weight averaging. Experiments on the nuScenes dataset demonstrate up to 33.8% improvement in corruption robustness and a 1.8% gain in clean accuracy over the Fast-BEV baseline, with consistent improvements across sensor noise, weather distortions, and illumination changes. Our simple yet effective approach may be applicable to other BEV models.
[CV-72] GUIFlow: Temporal Redundancy-Aware Incremental Inference for GUI Agents
链接: https://arxiv.org/abs/2610.34959
作者: Ye Li,Haoran Ding,Huanan Liu,Yuan Meng,Yutian Shen,Yuansong Wang,Kangye Ji,Hanyun Cui,Hongbo Duan,Chen Tang,Jiawei Li,Shu-Tao Xia,Zhi Wang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:GUI agents built on vision-language models interact with continuously evolving interfaces, yet conventional inference repeatedly recomputes representations of largely overlapping observations and histories. This mismatch makes multi-step interaction costly, while changes in context, interface content, and action arguments prevent straightforward cache reuse. We introduce GUIFlow, a framework for full-pipeline incremental inference in multi-image GUI interaction. Our central insight is that advancing a decision need not require rebuilding all supporting representations: each interaction step can instead selectively update computational states maintained across the trajectory. To support new decisions as the context evolves, GUIFlow adapts how past computation is reused: historical KV states are position-aligned and injected into the current context; change signals guide selective updates of current-observation states in the vision encoder and language model; and earlier action responses serve as speculative drafts verified by the target model. Together, these mechanisms enable cross-step reuse throughout visual encoding, prefill, and decoding without additional training, a separate draft model, or changes to the model-visible inputs and action interface. Experiments demonstrate up to 3.39x geometric-mean speedup across offline multi-image benchmarks and up to 3.49x in online, closed-loop interaction on OSWorld, while maintaining task performance broadly comparable to the dense baseline.
[CV-73] DimFlow: Bidirectional Grid-Expert Learning for Dynamic Gaussian Reconstruction
链接: https://arxiv.org/abs/2610.34952
作者: Gabriel Adineera,Jinoh Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Dynamic Gaussian reconstruction requires compact scene features to support coordinated changes in geometry and appearance. A factorized grid provides efficient local queries, but accurate interpolation alone does not ensure that the resulting dynamic attributes remain compatible across space and time. We introduce DimFlow, a framework that learns these features through bidirectional interaction between a multi-resolution grid and a coordinate-based expert branch. Feature-level adversarial alignment connects their distributions, adaptive fusion transfers complementary information, and activation-masked residual feedback refines the grid during training. Four attribute experts organize deformation, rotation, opacity, and temporal information before Gaussian rendering. In the aggregate comparison across D-NeRF, HyperNeRF, and DAVIS, DimFlow achieves 39.73 dB PSNR, improving on 4D-GS by 1.92 dB while matching its reported 85 FPS rendering speed. The model occupies 18.9 MB, compared with 18.3 MB for 4D-GS. Component and attribute-expert ablations reveal complementary contributions from feature alignment, adaptive exchange, and temporal processing.
[CV-74] Fed-MAPTrack: Federated Modality-aware Prompt Tuning for RGB-X Tracking
链接: https://arxiv.org/abs/2610.34939
作者: Xiangyu Liu,Yijie Shen,Yue Yin,Zhitong Yin,Yongqi Pan,Wentao Fan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:RGB-X tracking complements RGB appearance with thermal, depth, or event observations, but centralized training requires collecting multimodal videos at a single location. Federated training keeps raw videos local, but sensor heterogeneity can create a mismatch between parameter-sharing domains and the clients contributing to them. This mismatch is particularly relevant when RGB-only clients coexist with sensor-equipped clients: their RGB observations remain useful, but they provide no auxiliary measurements for learning sensor-dependent interactions. We propose Fed-MAPTrack, a modality-aware federated prompt tuning framework for RGB-X tracking. The pretrained tracking backbone and prediction head are frozen, and clients optimize and exchange only prompt-related parameters. The server aggregates RGB-shared parameters globally and maintains separate thermal, event, and depth banks for auxiliary-specific, fusion, and normalization parameters. Selective participation retains RGB-only clients’ contributions to the global RGB bank while excluding their fusion and normalization updates from modality-specific aggregation. Across three matched training repetitions under a fixed client partition, Fed-MAPTrack improves the mean FedAvg scores by 2.35 AUC points on LasHeR, 3.45 SR points on VisEvent, and 2.11 F-score points on DepthTrack. Among the nine evaluated methods, it achieves the highest mean thermal AUC and depth F-score, while its mean event SR is 0.07 points below that of Independent FL. Controlled ablations show that extending the bucketing scope without a compatible participation rule can reduce performance. These results highlight the need to jointly define the sharing domain and eligible contributors for each parameter group.
[CV-75] Supervising the Visible State: Separating Representation from View Selection in Streaming Video
链接: https://arxiv.org/abs/2610.34926
作者: Yihao Hu,Pan Wang,Yitong Lin,Zhihao Wen,Deyuan Liu,Hang Wang,Mingda Wan,Yiwen Zhao,Yutong Li,Yuchen Mao,Tao Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:An action seen once and the same action seen three times receive the same action-recognition label, so that supervision does not require a model to count. Answers shared across histories that differ in event order, accumulation, or relations leave those distinctions unconstrained; we call this state under-specification. Visible-State Supervision (ViSS) addresses it by training on source-grounded probes of event identity, order, accumulation, relation change, and answer sufficiency; under the same recent-frame interface it improves OVO-Bench over a strict-matched generic temporal supervision baseline while remaining comparable on StreamingBench, with the largest gain in hallucination detection. Test-Time Causal View Search (TVS) then adapts evidence access while keeping the learned model fixed, expanding from recent frames to sparse event and compact history views on demand. The two interventions separate representation from view selection: supervision particularly benefits hallucination detection and causal reasoning, access particularly benefits event recall and forward reasoning. Together they reach the best reported accuracy among the published systems compared here on StreamingBench and OVO-Bench at a fraction of the processed frames of exhaustive selection, and strict replay shows that most of the remaining gap to a three-view oracle is a correct answer that was later replaced. Streaming accuracy depends on what supervision asks the model to keep, what evidence inference exposes, and whether the model keeps a correct answer once it has one.
[CV-76] Rethinking Data Imbalance in Class Incremental Surgical Instrument Segmentation
链接: https://arxiv.org/abs/2610.34925
作者: Shifang Zhao,Long Bai,Kun Yuan,Feng Li,Jieming Yu,Wenzhen Dong,Guankun Wang,Mobarakol Islam,Nicolas Padoy,Nassir Navab,Hongliang Ren
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:In surgical instrument segmentation, the increasing variety of instruments over time poses a significant challenge for existing neural networks, as they are unable to effectively learn such incremental tasks and suffer from catastrophic forgetting. When learning new data, the model experiences a sharp performance drop on previously learned data. Although several continual learning methods have been proposed for incremental understanding tasks in surgical scenarios, the issue of data imbalance often leads to a strong bias in the segmentation head, resulting in poor performance. Data imbalance can occur in two forms: (i) class imbalance between new and old data, and (ii) class imbalance within the same time point of data. Such imbalances often cause the dominant classes to take over the training process of continual semantic segmentation (CSS). To address this issue, we propose SurgCSS, a novel plug-and-play CSS framework for surgical instrument segmentation under data imbalance. Specifically, we generate realistic surgical backgrounds through inpainting and blend instrument foregrounds with the generated backgrounds in a class-aware manner to balance the data distribution in various scenarios. We further propose the Class Desensitization Loss by employing contrastive learning to correct edge biases caused by data imbalance. Moreover, we dynamically fuse the weight parameters of the old and new models to achieve a better trade-off between the biased and unbiased model weights. To investigate the data imbalance problem in surgical scenarios, we construct a new benchmark for surgical instrument CSS by integrating datasets. Extensive experiments demonstrate the effectiveness of the proposed framework, achieving significant performance improvement against existing baselines.
[CV-77] DiffAuction: An Auction-Driven Framework for Generative Image Advertising
链接: https://arxiv.org/abs/2610.34917
作者: Yijie Su,Ruiyang Si,Tonghan Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Diffusion models enable multiple advertisers to compete for exposure within a shared generated image. Yet existing approaches are computationally costly when generating images under varying target-share configurations, and it remains unclear how the resulting outcomes should be evaluated and priced. To address these gaps, we propose DiffAuction, a framework integrating controllable generation, click-intent evaluation, and auction design. First, our training-free Residual Share Calibration (RSC) decomposes joint-attention updates into product-specific residuals and rescales them according to target shares. We further introduce Share-Conditioned Prompting (SCP) to adapt the viewpoint to each desired allocation. Second, we develop an evaluator that uses a VLM to extract visual evidence and an LLM to assess Click-Through Intent (CTI), a proxy for click propensity, along four dimensions: Product Prominence, Brand Recognizability, Scene Integration, and Contextual Value Communication. We validate its assessments against judgments from 10 trained annotators on 1000 images. Third, we formulate this advertising setting as an auction over jointly feasible generation configurations. Each configuration has a public expected CTI vector, while advertisers hold configuration-specific private values per unit of CTI that map click intent into commercial value. We adopt a learnable affine maximizer auction (AMA), which jointly selects a shared generation configuration and determines advertiser-specific payments to maximize expected platform revenue. The resulting allocation and payment rules guarantee dominant-strategy incentive compatibility (DSIC) and individual rationality (IR). Experiments show that our generation method strengthens the monotonic relationship between target shares and CTI relative to baselines, while the learned AMA achieves higher platform revenue in the resulting auction setting.
[CV-78] LE-GDA: Layer Ensemble Gaussian Discriminant Analysis for Training-Free Few-Shot CLIP Adaptation WACV2027
链接: https://arxiv.org/abs/2610.34901
作者: Pavodi Maniamfu,Vogel Kiketa,Zezina Kuzulu Kuetu,Jakin Mukendi Kabongo,David Mayoya Kutangila,Pescie Honore Kinkani Mampuya,L’Or Wolo Sieka
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at WACV 2027. 23 pages, 5 figures, including supplementary material
Abstract:Training-free few-shot adaptation of vision-language models typically constructs a classifier from CLIP’s final projected image representation, leaving intermediate representations unused. Combining them is not straightforward. Class scores from different representations differ in spread by orders of magnitude for the same query. Direct averaging allocates influence by numerical scale rather than predictive information. We propose Layer-Ensemble Gaussian Discriminant Analysis (LE-GDA), a method that fits regularized shared-covariance discriminants independently to intermediate and final representations extracted in one frozen encoder pass. LE-GDA standardizes each discriminant’s class scores per query, selects a small representation multiset on labeled validation data, and combines it with the zero-shot classifier without gradient-based optimization. Across 11 datasets, four CLIP backbones, five shot counts, and three independently sampled K-shot training sets, LE-GDA achieves the highest mean accuracy of all evaluated methods on every backbone, raising the 16-shot mean from 78.01% to 79.54% over a matched GDA-CLIP (final-feature) baseline and improving that baseline for every backbone-shot combination. At one shot, LE-GDA retains a 1.20-point mean gain over the baseline even though covariance estimation is unavailable and per-query standardization is neutral. A control with richer per-class structure inside a single representation does not reproduce the gain, indicating that the benefit comes from combining representations. These results show that intermediate CLIP representations carry information beyond the final embedding and can improve training-free few-shot adaptation. Code and evaluation resources are available at this https URL.
[CV-79] SGDB: Semantic-Geometric Decoupling and Evidence Binding for Logical Anomaly Detection
链接: https://arxiv.org/abs/2610.34897
作者: Chenrui Ji,Jiafan Lu,Zilu Luo,Feng Wan,Qiuhui Chen,Weimin Zhong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 7 figures, includes supplementary material. Code: this https URL
Abstract:Logical anomaly detection requires recognizing scene components and measuring whether their configuration satisfies task constraints. Correct identities can coexist with inaccurate geometry; clean mask fragments can yield misleading lengths. We propose Semantic-Geometric Decoupling and Evidence Binding (SGDB). A semantic decoder learns object identities from component-annotated normal sources and factual edits; a separate reader supplies final measurements. Partial one-to-one association binds geometric witnesses to identities while retaining unmatched objects in the count. A support-certified reader selects mask-derived axes using line quality and complete component ownership. Explicit task requirements and normal-data references define executable checks, with decision thresholds fixed by independent normal calibration. Execution-grounded explanation traces rule contributions to source objects and measurements, extracting compact sufficient alarm certificates. A frozen language model expresses selected outcomes as cited claims, with numeric evidence retained by the executor. Reports preserve observed deviations and unavailable evidence. SGDB achieves 98.19% AUROC and 94.52% fixed-threshold F1 on five MVTec LOCO logical categories, and 93.56% AUROC and 92.76% F1 across all 50 VID-AD tasks. Frozen-output controls evaluate geometric readout and explanatory sufficiency and compactness.
[CV-80] MMGNet: Multi-Modal 6-DoF Grasp Synthesis for Cluttered Scenes
链接: https://arxiv.org/abs/2610.34893
作者: Rui Cao,Jia-Hui Pan,Yichuan Li,Yunfan Yu,Chi-Wing Fu,Jiangliu Wang,Yun-Hui Liu
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Under Review
Abstract:Grasping is a foundational capability in robotic manipulation. Scene-level 6-DoF grasp synthesis relies on point clouds for metric geometry, yet real-world point observations are often noisy, incomplete, and over-smoothed. RGB images provide object-region, boundary, and appearance cues, yet two bottlenecks limit their effective use in scene representation and local grasp reasoning. We present MMGNet, a Multi-Modal Grasp Synthesis Network that addresses these bottlenecks through hierarchical image-to-3D feature injection and rotational graspness modeling. Hierarchical injection integrates multi-scale image features throughout 3D representation formation, while the rotational graspness jointly predicts the grasp approach direction and in-plane rotation before organizing local image-informed features through grasp-shaded grouping. Ablation studies show that these two modules provide complementary gains. MMGNet achieves a mean AP of 60.34 on GraspNet-1Billion with RealSense, surpassing all compared methods. In real-world experiments with adversarial grasp objects, MMGNet improves the grasp success rate over AnyGrasp by 8.46 percentage points. \emphCode, models, and data will be released upon publication.
[CV-81] MedAdaptAgent : An Automatic Harness for Adapting Medical Imaging Models to New Sites
链接: https://arxiv.org/abs/2610.34884
作者: Hanxue Gu,Zhonghao Liu,Xingrui Gu,Kang Wang,Yang Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages, 2 figures. Code: this https URL
Abstract:Reusing pretrained medical-imaging models offers an alternative to training from scratch. A model trained for the same task may perform poorly at a new hospital because imaging protocols and preprocessing differ. Effective adaptation depends on the model architecture, domain shift, and available labels and compute. Adaptation can improve scores on a small labeled set yet reduce performance on unseen data. We present MedAdaptAgent, an automatic harness that turns a supplied model, target data, and a free-text request into a new-site adaptation workflow. Its LLM-driven Adaptation Planner uses model-site evidence to choose compatible experiments and revise subsequent adjustments. The harness requires evidence of improvement over the pretrained model, then delivers the selected model as a complete, runnable package. Evaluation spans 26 segmentation and classification transfers across architectures and compute settings, with segmentation annotation budgets ranging from zero-shot transfer to few-shot adaptation. With 20 local annotations, mean segmentation Dice rises from 0.483 to 0.774 across 10 transfers, a 60.1% relative gain, with no harmful replacements. Mean classification AUROC rises from 0.810 to 0.837 across 13 transfers, a 3.4% relative gain. Comparing adapted models with the pretrained model reduces harmful classification replacements by 30.8% versus selecting the highest-scoring model on the available labels, using identical candidates. Seven public-model deployments demonstrate adaptation and reproducible package delivery across contrast and modality shifts. A CT-to-MRI study extends the evaluation to four LLM planners, including two open-weight models. We release the code on GitHub.
[CV-82] LayoutMamba: Efficient Document Layout Analysis with Hybrid Mamba-Transformer Networks
链接: https://arxiv.org/abs/2610.34879
作者: Kehua Chen,Mingyi Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 2 figures, 4 tables
Abstract:Document layout analysis (DLA) faces a long-standing dilemma: existing methods either capture rich local details at prohibitive computational cost (e.g., Transformer-based approaches with quadratic complexity) or achieve efficient global modeling but sacrifice fine-grained structural awareness (e.g., state-space models). This trade-off hinders deployment in resource-constrained scenarios. We present LayoutMamba, a hybrid framework that synergistically combines sparse-attention Transformers for localized feature extraction (O(n) complexity) and the recurrent state-space mechanism of Mamba for dilution-free long-range dependency modeling across the 2D document canvas. Our two-stage paradigm first uses a lightweight Transformer encoder to hierarchically distill salient local semantics; second, Mamba propagates global contextual dependencies without attention dilution, enabling precise, efficient, and scalable layout understanding. We further identify a critical yet overlooked issue in existing DLA benchmarks: the inherent low-overlap distribution bias of layout objects, which we mitigate through a novel Geometric Overlap Penalty Loss (GOPLoss), a dual-optimization objective that minimizes redundant bounding box overlaps and encourages geometrically consistent predictions. Extensive experiments on PubLayNet and D4LA show that LayoutMamba surpasses state-of-the-art AP on PubLayNet with 8x fewer parameters and maintains leading performance on D4LA, ranking top-3 among published methods, achieving an unprecedented balance between accuracy, efficiency, and scalability. Our work pioneers a cognition-inspired paradigm for document analysis, advancing practical deployment in edge computing and resource-limited environments.
[CV-83] CoDeRNet: Selective Cross-Task Routing under Heterogeneous Supervision for Change Detection and Captioning NEURIPS2026
链接: https://arxiv.org/abs/2610.34876
作者: Eunki Cho,Hyeon Bae Kim,Seong Tae Kim
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026. 20 pages
Abstract:Change Detection Captioning (CDC) aims to jointly localize changed regions and describe their semantic content from bi-temporal remote sensing images. While recent approaches have demonstrated the effectiveness of unified modeling, they primarily rely on shared representations or jointly optimized architectures, leaving task-specific communication under heterogeneous supervision less explored. However, detection and captioning differ fundamentally in their supervision granularity, leading to heterogeneous representations with uneven reliability across tasks and spatial locations. In this paper, we reformulate CDC as a selective information routing problem between heterogeneous task branches, where the key challenge is to determine when and where information should be transferred. To this end, we propose CoDeRNet, a unified framework that preserves task-specific representations while enabling bidirectional exchange of complementary information. CoDeRNet introduces a Confidence-Decomposed Routing (CoDeR) mechanism, which decomposes the routing condition into receiver-side need for support and sender-side reliability cues to guide selective transfer. Experiments on LEVIR-MCI and WHU-CDC show that CoDeRNet achieves strong overall CDC performance, particularly improving caption generation while preserving competitive change localization. Extensive analyses further show that the benefits arise not merely from joint learning or generic cross-task communication, but from explicitly modeling selective cross-task information routing. These results highlight the importance of conditional information transfer for unified CDC under heterogeneous supervision.
[CV-84] DecoyCount: Evaluating Visual-Reasoning Misalignment in Counting
链接: https://arxiv.org/abs/2610.34855
作者: Qiuyi(Richard)Zhang,Zhen Xu,Jing Gu,Seojin Bang,Arnav Mohanty Das,Yaning Hu,Gautam Kumar,Jared Lichtarge,Le Xue,Bettina Catoto,Akul Datta,Daniel Luo,Farha Khan,Marcella Valentine,Yinfei Yang,Andrew M. Dai
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual counting benchmarks report whether vision-language models produce the correct number, but not whether errors arise from perception or reasoning. We introduce DecoyCount, a benchmark built from 1,660 real-world images with designed decoys across seven mechanisms, along with paired prompts: a neutral question, a matched alignment cue, and a novel multi-hop, multi-image question that situates the same counting problem in a realistic scenario. Using this design, we find that counting failures are induced partly by visual misalignment: models can execute the intended counting process when cue-guided but fail to select it unaided, and we term this difference in accuracy the reasoning-alignment gap. Despite being highly capable on general reasoning, frontier models remain below 50% accuracy on our benchmark, while alignment gaps range from 11.8-22.1 percentage points; strikingly, we observe that more capable models, greater thinking budgets, and agentic processes often induce larger gaps. Because humans may bring context that models lack, we test whether textual underspecification contributes to misalignment using a scenario-based multi-hop condition without explicit hinting, but this added context does not improve aggregate performance. Our analysis concludes that stronger general reasoning does not by itself generalize to aligned visual reasoning, leaving misalignment with human visual expectations as a substantial component of the remaining deficit.
[CV-85] Learning Dual-Camera Smooth Zoom via Decoupled Geometric-Photometric Data Synthesis
链接: https://arxiv.org/abs/2610.34854
作者: Junjie Fan,Zhilu Zhang,Renlong Wu,Dazhi Zhang,Wangmeng Zuo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL Code: this https URL . 13 pages, 13 figures
Abstract:When zooming between dual cameras on smartphones, prominent discontinuities in both geometric perspective and color rendition inevitably degrade the visual experience of users. To address this issue, we introduce the dual-camera smooth zoom (DCSZ) task, aiming to synthesize continuous intermediate frames for fluid zooming. However, directly applying existing video frame interpolation (VFI) models to this task is severely hindered by substantial domain gaps and the practical intractability of capturing real-world continuous ground-truth. To this end, we propose a novel data synthesis framework to generate high-fidelity DCSZ training sequences, named ZoomGS v2. It explicitly decouples geometric and color transformations from 3D scene representation modeling, thereby enabling extensive data generation from easily accessible single-camera 3D scenes. Furthermore, we develop ZoomFI, a specialized VFI network tailored to the DCSZ task, which leverages complementary bidirectional warping for comprehensive dual-camera information fusion. Extensive experiments on test datasets captured using two mobile phones demonstrate that training with the constructed DCSZ data significantly improves the performance of various VFI models. Moreover, ZoomFI achieves state-of-the-art results in both quantitative metrics and visual quality. The datasets, code, and pretrained models are available at this https URL
[CV-86] Recovering Black-Box Point Cloud Models from Pose Shifts via Derivative-Free Optimization WACV2027
链接: https://arxiv.org/abs/2610.34850
作者: Seonghwan Park,Min-Gyu Park
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: WACV 2027
Abstract:Point cloud models are increasingly deployed through score-only black-box interfaces. Under this access model, test-time rotations outside the training distribution can substantially degrade performance. Although existing mitigations such as equivariant retraining and gradient-based adaptation are effective, they are infeasible without access to model weights or gradients. To address this limitation, we propose CanDO, a test-time canonicalization method that recovers lost performance using only forward queries. Specifically, CanDO searches for an input transformation that maximizes model confidence and uses derivative-free optimization to perform this search without gradients. We further identify three conditions for successful recovery: a training augmentation gap, an accuracy-aligned confidence surrogate, and global search over a multi-modal objective. Across multiple architectures and benchmarks, CanDO recovers 76% of the oracle gap under azimuthal misorientation and 51% under arbitrary SO(3) rotations for object part segmentation. These results show that frozen score-only point cloud models can recover from rotation-induced distribution shifts through post-deployment canonicalization alone.
[CV-87] Seeing Before Naming: Context-Grounded Learning of Novel Visual Concepts
链接: https://arxiv.org/abs/2610.34846
作者: Gabriel Adineera,Jinoh Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Learning an unfamiliar visual category requires more than adapting a classifier: the model must connect its visual evidence to a useful semantic name. A pre-trained vision-language model can retain discriminative image features while providing a weak textual anchor for the target concept. We address this imbalance through context-grounded semantic anchoring. Our framework, SynerNet, learns concept-specific name embeddings, exposes them to compatible linguistic contexts, and conditions their representations on visual evidence. Difficulty-aware processing and adaptive objective weighting coordinate this adaptation while retaining the pre-trained backbone. On four VISTA-Beyond domains, SynerNet reaches a mean 16-shot accuracy of 74.35%, compared with 72.45% for few-shot name learning. Across 14 domains in the name-masked transfer setting, it improves mean accuracy from 45.17% to 46.74%. Ablations identify concept-specific name embeddings and cross-modal context integration as important components. These results support treating concept acquisition as the construction of visually grounded semantic anchors.
[CV-88] PosePlaner: Denoising Any Feed-Forward Pose Predictor from Pairwise Planar Geometry NEURIPS2026
链接: https://arxiv.org/abs/2610.34845
作者: Yihang Chen,Bin Tan,Yujun Shen,Yinghao Xu,Nan Xue
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. Project page: this https URL
Abstract:Recent multi-view Transformers (e.g., DUSt3R and VGGT) have advanced 3D reconstruction by scaling training data and model capacity from two views to many views and long sequences, subsuming the classical SfM/SLAM pipeline of feature extraction, matching, and pose estimation in a single feed-forward pass. Yet this subsumption hides the interfaces between pipeline stages—correspondences, geometry, and pose are entangled in a single forward pass, leaving supervised fine-tuning as the major path to improvement and offering no mechanism to correct predictions from geometric evidence at test time. We propose PosePlaner to close this gap by reviving the classical coupling between correspondences and pose via flow matching, treating pose refinement as a conditional generation problem that takes coarse geometric matches and initial pose as input condition and fuses the resulting pose hypotheses into a single refined estimate, providing a continuous, data-driven analog of RANSAC voting. Our method is trained efficiently on two-view data, yet serves as a periodic refiner for streaming n-view pose estimators with no retraining. Empirically, it corrects failures of feed-forward predictors, substantially improves correspondence-based solvers, and reduces accumulated drift in long streaming sequences.
[CV-89] Interpretable Brain Age Prediction Using White Matter Fiber Orientation Distributions with Rotation-Equivariant Deep Learning
链接: https://arxiv.org/abs/2610.34825
作者: Chenyang Fan,Robert D. Sanders,Elie Matar,Yixuan Yuan,Xiang Li,Sharon L. Naismith,Fernando Calamante,Jinglei Lv
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 10 pages, 8 figures and 2 tables
Abstract:Diffusion MRI (dMRI) provides information about white matter microstructure that can support brain age prediction. However, many existing prediction models use diffusion tensor imaging (DTI) scalar metrics that omit orientation information. Deep learning is increasingly used for brain age prediction, but conventional CNNs do not explicitly respect the geometric structure of orientation-dependent features. We present an interpretable brain age prediction framework that combines white matter fiber orientation distributions (WM-FODs) with an SE(3) (Special Euclidean group in 3 dimensions) equivariant CNN. The network operates directly on spherical harmonic (SH) coefficients of the FOD, preserving their transformation properties under rotation. Evaluation in 692 participants aged 36-90 years from the Human Connectome Project Aging dataset showed that the combined framework outperformed the evaluated DTI-based models and conventional CNNs. Averaging predictions from five cross-validation models yielded a mean absolute error of 3.54 years and Pearson’s correlation of 0.958. Our results indicated that the performance advantage depended on combining WM-FODs with equivariant processing. Prediction accuracy was highest at a maximum SH order of Lmax = 2 and declined at higher orders. Tract-level occlusion, where each segmented tract was masked in turn, revealed age-dependent patterns of model sensitivity, with importance peaks in frontal projection pathways at 36-44 years, the fornix at 45-54 years, the callosal isthmus at 55-64 years, the middle cerebellar peduncle at 65-74 years, and the callosal genu at 75-90 years. These findings support rotation-equivariant processing of WM-FODs for accurate brain age prediction with anatomically interpretable model sensitivity.
[CV-90] Selective Monocular Precession Estimation under Geometric and Evaluation Ambiguity: An Auditable Synthetic Study with ORBIT-SCOUT
链接: https://arxiv.org/abs/2610.34819
作者: Carlos Eduardo Sanoja
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 29 pages, 4 figures and 7 tables; extended narrative review and ancillary code/data for descriptive arithmetic and physical-metric reproduction
Abstract:Monocular observations of a tumbling object can support a precise fitted trajectory without establishing that the recovered physical axis is correct. We study this distinction for elongated, approximately cylindrical targets observed by a calibrated stationary camera. ORBIT-SCOUT couples image-derived line and rim geometry to constant-cone and torque-free triaxial trajectory families, then returns a linked momentum-axis and mean-precession-rate estimate only when numerical, geometric, temporal and competing-hypothesis checks pass. We describe the measurement objective, selection policy and limitations, and situate them within spacecraft pose estimation, unknown-target reconstruction, rigid-body identification and selective prediction. An internally preregistered population of 96 new synthetic scenes was processed in two cold passes, retaining all 192 attempts. In the primary pass, 38 scenes produced jointly correct estimates, none produced an incorrect estimate, 57 were withheld and one execution was incomplete. Correct coverage on the 64 supported scenes was 25/64. Ninety-five repeated pairs were scientifically equivalent under a frozen comparison policy; the remaining pair was incomplete. The registered coverage, failure-free execution and all-case repeatability targets were not met. Historical audits also exposed an overly permissive axis metric and motion-dependent acquisition cadence. The contribution is a documented measurement-to-decision implementation and a failure-preserving evaluation case study, not a new rigid-body theory or evidence of state-of-the-art accuracy. Zero observed errors does not establish calibrated risk, and conditional temporal checks do not remove shared image-feature bias. Real-image transfer and external reproduction remain open.
[CV-91] owards Spatio-Temporal Intelligence Systems in Urban Space: A Twin-Agent Perspective
链接: https://arxiv.org/abs/2610.34795
作者: Zhen Dong,Haiping Wang,Wei Tu,Yuan Liu,Bing Wang,Xuan Sun,Jinzhou Cao,Ronald Clark,Zhiheng Fu,Mohammed Bennamoun,Danfeng Hong,Yuanzheng Cui,Gemine Vivone,Pedram Ghamisi,Bisheng Yang,Qingquan Li,Deren Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: A review of urban digital twins and urban agent systems
Abstract:Cities must coordinate domains like infrastructure, mobility, environmental management, emergency response, and public services as conditions evolve. An urban Spatio-Temporal Intelligence (STI) system helps the city monitor citywide states, analyze problems, plan and execute interventions, and adapt through feedback, making urban governance more timely, effective, and transparent. Recent urban twinning and agentic AI techniques provide foundations for such systems. From a twin-agent perspective, this survey takes a system-level view of how these disparate techniques can be coherently integrated into an urban STI system. Specifically, we review how the twin reconstructs and represents a persistent, information-enriched urban environment, and how the agent understands the twin via interfaces, reasons over it, acts reliably, and continuously evolves with the city. Built upon this synthesis, we summarize a two-loop blueprint, where an inner agent loop for grounding, reasoning, and verification, and an outer system loop linking data, twin, agentic action, feedback, and update. We then use this to structure an STI research agenda and show how a shared prototype supports diverse urban domains.
[CV-92] oDoPro: Faster and Better Reference-Based Generation: Early Token Reduction via Downsampled Self-Probing
链接: https://arxiv.org/abs/2610.34782
作者: Caiyi Sun,Yujing Sun,Siu-Ming Yiu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent generic reference-based image generation models natively support multiple conditions by concatenating all reference image and text tokens into a joint sequence processed by a single Diffusion Transformer (DiT). However, this versatile paradigm comes at an efficiency cost: accommodating multiple references leads to a token explosion, incurring massive \mathcalO(N^2) computational complexity. Most acceleration methods operate inside the DiT, showing limited capability in reducing spatial redundancy and failing to alleviate peak VRAM. While some pre-model token reduction alternatives exist, they inevitably introduce external models. This raises a question: Is it possible to achieve acceleration and memory savings while preserving compatibility with in-model techniques and requiring no additional parameters? In this work, we realize this goal by proposing the early Token reduction via Downsampled self-Probing (ToDoPro). By executing a few denoising steps on spatially 2\times downsampled reference images, this cheap probing stage utilizes the DiT model itself to effectively and efficiently extract reference token importance. Using the probed importance, we apply a hierarchical reduction before the main denoising stage: we completely discard low-importance tokens, substitute mid-importance tokens with their downsampled counterparts to cheaply preserve global context, and keep high-importance tokens. Experiments across three base models and two benchmarks demonstrate that our early reduction accelerates inference significantly and mitigates peak GPU memory. Furthermore, we can improve generation quality by safely filtering out noisy, uninformative reference tokens, enabling the model to concentrate its limited attention capacity on genuinely useful tokens. The code will be released at this https URL
[CV-93] Factorized Gradients for Scalable Highly-expressive Parametric Diffeomorphisms NEURIPS2026
链接: https://arxiv.org/abs/2610.34766
作者: Amit Aflalo,Eran Treister,Chaim Baskin,Oren Freifeld
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by NeurIPS 2026
Abstract:Diffeomorphisms provide a flexible, topology-preserving mathematical tool for modeling complex spatial or temporal deformations and are used in various scientific fields. However, making these deformations simultaneously highly expressive and computationally efficient remains challenging. Continuous Piecewise-Affine Based (CPAB) transformations (Freifeld et al., ICCV 2015; IEEE TPAMI 2017) offer an attractive solution by parameterizing a family of diffeomorphisms via continuous velocity fields that are piecewise affine with respect to a chosen tessellation of the domain. Importantly, d=\dim(\theta) , the dimension of the parameter vector \theta , depends on the fineness of the tessellation rather than on the data resolution. Despite this compact parameterization, CPAB optimization has long been hindered by the tight coupling between trajectory integration and gradient computation with respect to \theta . We introduce FG-CPAB, a factorized-gradient formulation for scalable CPAB transformations. By separating integration from parameter-space projection, our method reduces gradient computation complexity from \mathcalO(dTN) to \mathcalO(TN+dC) , where T is the number of integration steps, N is the number of transformed points, and C is the number of tessellation cells. This reformulation yields speedups of up to four orders of magnitude (e.g., \approx 10^4\times in 2D), substantial memory savings (e.g., \approx 500\times reduction in peak GPU memory usage in 2D), and dramatically higher expressiveness. This makes fine-tessellation CPAB practical in 2D and 3D for the first time, enabling highly expressive parametric diffeomorphisms.
[CV-94] Efficient Multi-modal Video Fusion Network with Dynamic Region-Guided Temporal Interaction
链接: https://arxiv.org/abs/2610.34765
作者: Jingxue Huang,Xiaosong Li,Yi Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multi-modal video fusion aims to generate temporally coherent videos with salient targets and rich structural details by integrating complementary information from heterogeneous sensors. However, existing video fusion methods commonly rely on optical flow estimation and feature warping for inter-frame alignment and cross-frame information aggregation to maintain temporal consistency, thereby introducing considerable computational overhead and making the fusion performance vulnerable to flow estimation errors. To address these limitations, this paper proposes an efficient multi-modal video fusion network with dynamic region-guided temporal interaction. Specifically, the proposed spatio-temporal interaction module leverages the Test-Time Training paradigm to model temporal dependencies with linear complexity. Building on this design, the module restricts temporal modeling to key dynamic regions in the video sequence, as static background regions provide limited temporal cues. By excluding low-variation tokens from unnecessary temporal interaction, this selective strategy contributes to improved fusion quality and temporal consistency. In addition, leveraging the strong ability of pretrained MAE features to capture global structural dependencies, we introduce a feature-level temporal consistency constraint to suppress local flickering by aligning the inter-frame variation directions of the fused and source videos. We further propose a new MAE-based temporal metric that quantifies directional distance between their temporal variations within dynamic regions. Experimental results demonstrate that, compared with existing state-of-the-art methods, our method achieves superior fusion quality and temporal consistency on multiple benchmark datasets, while exhibiting high computational efficiency under edge-device deployment settings.
[CV-95] SceneAgent : An Agent ic Pipeline from 3D and 4D Captures to Simulation Environments
链接: https://arxiv.org/abs/2610.34745
作者: Luke Hollis,Tianxing Fan,Heng Yang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual captures provide detailed scene appearance, but converting them into interactive simulation environments requires object reconstruction, physical modeling, and correction of intermediate errors. We present SceneAgent, an agentic pipeline that coordinates these steps through simulator-based visual inspection and repair. The pipeline processes 3D Gaussian Splatting (3DGS), photogrammetry, LiDAR captures, AI-generated 3DGS scenes, and 4D videos of dynamic scenes, and combines object segmentation, physical parameter estimation, part decomposition, articulation, and deformable simulation. To support policy learning, SceneAgent generates diverse object variants (“digital sisters”) and scripted demonstrations. On a cluttered bin scene from GraspClutter6D, SceneAgent raises F1 at 1 cm for every coding agent with scorable outputs both with and without its skills, and its reconstructions are better than SimFoundry: all agent configurations exceed SimFoundry’s F1 score, with a best of 60.75% versus 6.9%. On three dynamic scenes from 4DCodeBench, SceneAgent raises the mean dynamic IoU of Gemini 4 Argon from 0.86% to 99.09%. We further demonstrate diverse simulated interactions and preliminary policy transfer to a real robot.
[CV-96] From Sentinel-2 to 1-m Imagery: Real-World times 10 Cross-Sensor Super-Resolution via Visual Autoregressive Modeling
链接: https://arxiv.org/abs/2610.34742
作者: Tailai Chen,Yuluxin Fu,Jie Deng,Shu Luo,Yanan Wu,Danfeng Hong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 7 figures, 7 tables. Code available at this https URL
Abstract:Reconstructing 1-m imagery from Sentinel-2 combines two difficulties. The input and the reference are observed by different satellites, which differ in spatial resolution, spectral response, radiometry, acquisition time, and local geometry, so no synthetic degradation operator links them. The required magnification of ten in each spatial dimension also far exceeds the factors of two to four at which remote sensing super-resolution is usually developed. We study this real-world \times 10 problem using YREB-SR, which pairs 2022 Sentinel-2 RGB composites with 2022 Jilin-1 RGB references resampled to a 1-m grid over the Yangtze River Economic Belt. The benchmark contains 11,559 geographically matched scene pairs, with 6,409, 2,135, and 3,015 assigned to training, validation, and testing, respectively. We develop CrossVAR, a visual autoregressive model that predicts the 1-m latent as a nine-scale hierarchy conditioned on the Sentinel-2 input and estimates the remaining quantization residual with a diffusion refiner run without injected noise. Complete 1600 \times 1600 scenes are reconstructed in overlapping windows that start on the Sentinel-2 pixel grid, as every training patch does, and every window is decoded at the size on which the decoder was trained before the windows are blended with Gaussian weights. On all 3,015 test scenes, CrossVAR achieves 27.60 dB PSNR, 0.7725 SSIM, 0.1451 LPIPS, and 7.15 FID, the best value of each metric among the compared methods. These results establish a strong autoregressive baseline for large-scale cross-sensor reconstruction. To facilitate reproducibility, the source code and implementation are publicly available at this https URL.
[CV-97] FGD2HF: Frequency-guided Dual-domain Hypergraph Fusion for Semantic Segmentation
链接: https://arxiv.org/abs/2610.34734
作者: Xiaofan Yang,Wei Pan,Yubin Liu,Junyu Wu,Junming Zhang,Jie Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages
Abstract:Frequency-domain information is beneficial for semantic segmentation. However, existing frequency-based methods often suffer from insufficient frequency utilization, inefficient fusion, and high computational complexity. To address these issues, we propose FGD2HF, a dual-domain fusion network that leverages frequency features to guide spatial-domain fusion and enhancement. Specifically, we introduce a novel Hybrid-Attention Encoder (NAHA) that jointly models local and global relationships through neighborhood and hypergraph attention, enhancing the network’s ability to extract effective features. We further introduce a Wavelet Transform and Enhancement (WTE) module to extract and refine low- and high-frequency cues from preceding-stage features, followed by Frequency-guided Dual-domain Fusion (FGDF) modules that use frequency cues to condition current-stage spatial grouping and feature propagation. This design incorporates both global and local frequency guidance into spatial feature modulation, enhancement and fusion with low computational cost. Extensive comparisons with numerous state-of-the-art semantic segmentation methods demonstrate that our method is highly competitive. Our full FGD2HF-B model achieves 48.7% mIoU on ADE20K (+0.3% over SegMAN-B with 18% fewer GFLOPs), 83.2% mIoU on Cityscapes (comparable to SegMAN-B with 21% fewer GFLOPs), and 1.1% higher mIoU than VWFormer-T on SBD with 21% fewer GFLOPs, while also achieving significant improvements in boundary segmentation accuracy.
[CV-98] RSKD: Grounding Semantic Consistency in Driving VLMs via Distillation
链接: https://arxiv.org/abs/2610.34689
作者: Chuchu Xie,Yang Zhong,Kangan Qian,Hao Ye,Kun Jiang,Diange Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Knowledge distillation can transfer large multimodal models to compact models for autonomous driving, but uniform token-level matching can dilute planning-relevant semantics and amplify teacher errors. We propose RSKD, a semantic-consistency-guided distillation framework that routes supervision according to both the task importance of an answer phrase and its semantic consistency with the teacher prediction. Task-critical semantic phrases are localized in the ground-truth answer, and a phrase-level consistency score controls the balance between Jensen–Shannon divergence (JSD) distillation and ground-truth supervision. Teacher information is emphasized only where it is semantically consistent, while semantically inconsistent phrases receive stronger ground-truth supervision. We evaluate RSKD on DrivingVQA, LingoQA, and MapLMv2. RSKD improves over the compared student-based baselines on DrivingVQA and LingoQA and remains competitive on MapLMv2.
[CV-99] From Speech Tokens to Stolen Voices: Attacker-Chosen-Text Speaker Impersonation via Representation Inversion
链接: https://arxiv.org/abs/2610.34688
作者: Ye Lu,Shen Wang,Runze Liu,Zhitao Ou,Haonan Huang,Yihan Yan,Fanghui Sun,Zhaoyang Zhang
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Speech interaction systems have gradually evolved from traditional cascaded systems that use text as intermediate information to end-to-end speech language models that directly process speech representations. In some applications, the speech frontend runs on the user’s device and sends speech representations to a remote server for inference, while the original audio remains local. However, these representations may still contain speaker identity information. If an attacker obtains them, it may use this information to imitate the user’s voice, threatening their voice privacy. To investigate this risk, we propose \stia, a Speech Token Inversion Attack that generates speech with attacker-specified text in a victim’s voice using a single short leaked representation, without access to the original audio or its transcript. Specifically, \stia converts discrete, continuous, or mixed speech representations into conditioning sequences and trains a conditional flow-matching speech generation model using pairs of different utterances from the same speaker in public speech datasets. The model combines speaker characteristics contained in the leaked representation with the specified text to generate speech in the victim’s voice. Experiments show that, with a Moshi representation leak corresponding to 3 seconds of speech, \stia achieves an average CAM++ speaker cosine similarity of 0.78 and a target-text word error rate of 3.80%. These results show that speech frontend representations reveal reusable speaker identity information and highlight the need to consider privacy risks when designing speech frontend representation methods.
[CV-100] Good Rankings Poor Choices? Rethinking Language-Only Vision-Language Model Selection
链接: https://arxiv.org/abs/2610.34685
作者: Hyuk Kwon,Jaebyeong Jeon,Hyeonseo Jang,Kibok Lee
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Language-Only Vision-Language Model Selection (LOVM) aims to identify the Vision-Language Model (VLM) with the highest zero-shot accuracy on a target task using only textual information. We revisit the existing LOVM evaluation protocol and argue that ranking metrics, such as top-5 recall and Kendall’s rank correlation, do not directly assess the performance of the selected model. We therefore adopt simple regret (SR), defined as the performance gap between the best and selected models, and introduce Outlier-Robust Normalized Regret (ORNR) to account for the varying performance ranges across tasks while reducing sensitivity to outliers. Building on this revised evaluation protocol, we propose Target-Conditioned Expected Source Regret (TC-ESR), a training-free method that selects the model with the lowest similarity-weighted mean regret across source tasks. Across 22 target tasks and 43 VLMs, TC-ESR achieves the lowest mean SR and ORNR among all compared methods. Separately, we identify two sources of target performance leakage in the prior ranking-based evaluation protocol and provide corrections that remove both.
[CV-101] Exploring perceptual straightness in learned visual representations ICLR2023
链接: https://arxiv.org/abs/2610.34678
作者: Anne Harrington,Vasha DuTell,Ayush Tewari,Mark Hamilton,Simon Stent,Ruth Rosenholtz,William T. Freeman
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Published at ICLR 2023
Abstract:Humans have been shown to use a “straightened” encoding to represent the natural visual world as it evolves in time (Hénaff et al. 2019). In the context of discrete video sequences, “straightened” means that changes between frames follow a more linear path in representation space at progressively deeper levels of processing. While deep convolutional networks are often proposed as models of human visual processing, many do not straighten natural videos. In this paper, we explore the relationship between network architecture, differing types of robustness, biologically-inspired filtering mechanisms, and representational straightness in response to time-varying input; we identify strengths and limitations of straightness as a useful way of evaluating neural network representations. We find that (1) adversarial training leads to straighter representations in both CNN and transformer-based architectures but (2) this effect is task-dependent, not generalizing to tasks such as segmentation and frame-prediction, where straight representations are not favorable for predictions; and nor to other types of robustness. In addition, (3) straighter representations impart temporal stability to class predictions, even for out-of-distribution data. Finally, (4) biologically-inspired elements increase straightness in the early stages of a network, but do not guarantee increased straightness in downstream layers of CNNs. We show that straightness is an easily computed measure of representational robustness and stability, as well as a hallmark of human representations with benefits for computer vision models.
[CV-102] Benchmarking Obstacle-Conditioned Safety in Household Robot Navigation
链接: https://arxiv.org/abs/2610.34674
作者: Hyunwoong Bae,Jisu Han,Jiyeon Choi,Dohoon Kim,Taesup Moon
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Safe household navigation requires robots to regulate clearance and speed according to what an obstacle is and how close it is, without unnecessary conservatism. Yet existing evaluations do not test whether semantic safety distinctions are reflected in continuous motion. To that end, we introduce OSMA-Bench, an Obstacle-Conditioned Safe Motion Adaptation Benchmark built on spatially matched scenario sets that differ only in obstacle type. We also propose the \emphSemantic Sensitivity Index (SSI), which measures whether clearance and distance-dependent speed and smoothness follow the caution ordering of obstacles. Evaluating VLM-based and vision-language-action (VLA) policies on the benchmark, we find that neither consistently adapts its motion to obstacle safety, even with explicit safety prompts. Our analysis shows that the policies can correctly distinguish obstacles by safety relevance but fail to reflect this distinction in the motion parameters or actions they produce. For VLM-based policies, which recognize the obstacle and then generate a navigation program, adding an explicit parameter-inference step between these two stages improves adaptation. These findings expose a gap in \emphsemantic-to-motion grounding, highlighting the need to translate semantic caution into appropriately safe motion.
[CV-103] Video World Models Need Predictable Latents
链接: https://arxiv.org/abs/2610.34663
作者: Mattia Scardecchia,Sherry Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 54 pages, 25 figures, 14 tables. Code: this https URL
Abstract:Video world models trained on large-scale robot data promise a safe and scalable approach for learning and evaluating policies that act in the physical world. However, they still struggle with long-horizon rollouts and out-of-distribution generalization, and they remain computationally demanding at inference time. Typically, video world models make predictions in the latent space of a reconstruction-trained autoencoder, which however is not directly optimized for dynamics modeling. On the other hand, world models designed for latent planning usually optimize latents to be predictable, but they forgo accurate prediction of observations in pixel space. In this paper, we show that it is possible to achieve the best of both worlds: a latent geometry optimized to make future prediction as simple as possible without sacrificing accurate prediction of future observations. More precisely, we find that temporal objectives for representation learning, paired with a past-conditioned decoder architecture, lead to consistent improvements in generative video world models at the same training budget. The gains include up to 5x cheaper inference, over 7x faster training, and 15-24% lower pixel and perceptual errors out of distribution and over long horizons. Our world models also agree more closely with real-world policy evaluation when using a vision-language model to judge task success from their pixel-space predictions. Overall, our study indicates that temporal predictability is an important design criterion for the latent space of video world models, and provides concrete guidance on how to incorporate it into the training of autoencoders.
[CV-104] StreamVibe: Proactive Video Agents for Evolving User Intent
链接: https://arxiv.org/abs/2610.34662
作者: Chengzhi Liu,Yuxiang Lai,Yuzhe Yang,Yanqing Liu,Jiheng Hou,Yue Cao,Haonan Wang,Ruixi Fan,Lan Wang,Sheng Liu,Songyou Peng,Yale Song,Kevin Qinghong Lin,Xin Eric Wang,Jui-Hsin Lai,Youbao Tang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 43 pages
Abstract:Vision-language models are shifting from offline passive understanding toward online proactive interaction. However, prevailing methods largely rely on explicit requests while miss the implicit user intent (e.g., “where should I click next?” while a user navigate a website), meanwhile fail to update the state simultaneously as user requests might update continuously. These limitations make them less suited to intent-aware interaction in practical. To address these challenges, we introduce StreamVibe-Bench, a new benchmark serving as testbed for user intent interaction, with 4.1K evaluation samples and 85K training samples spanning explicit, implicit, and dynamic intents. Building on this benchmark, we develop StreamVibe-VL, a streaming proactive model that infers implicit user intents from continuous observations and dynamically tailors its interactions as these intents evolve. We further introduce StreamVibe-Harness, which maintains user intents as persistent states with a fleixble lifecycles via a hierarchical memory to track the latest user intent and environmental state. Our experiments reveal three key findings: (i) StreamVibe-VL preserves strong general video understanding while improving response timing and answer quality by 10.7% over other baselines; (ii) existing proactive models remain limited in understanding user intent and deciding when to respond, especially under implicit and dynamic intents; and (iii) StreamVibe-Harness further improves response consistency and adaptation over sustained interaction, resulting in a 12.3% gain in proactive interaction.
[CV-105] Regional prediction of hematoma expansion from non-contrast computed tomography in spontaneous intracerebral hemorrhage patients
链接: https://arxiv.org/abs/2610.34645
作者: Natasha Ironside,Hongyu Kan,Kristofor Pas,Mohammad Shifat-E-Rabbi,Benjamin Newman,Shinjini Kundu,Kareem El Naamani,Tanvir Rizvi,Andrea Becerril-Gaitan,Harrison Snyder,Aarathi Manchikalapudi,Sarah Ratcliffe,Stacie L. Demel,Carl D. Langefeld,Daniel Woo,Stephan Mayer,E. Sander Connolly,Gustavo Rohde, for theERICH Investigators, theVISTA-ICH collaboration
类目: Computer Vision and Pattern Recognition (cs.CV); Tissues and Organs (q-bio.TO)
备注:
Abstract:Background The combination of hematoma location, expansion and final volume influence neurological outcome. The aim of this study was to derive and validate a quantitative and interpretable region-based NCCT image feature model for predicting hematoma expansion. Methods The Ethnic/Racial Variations of Intracerebral Hemorrhage(ERICH) study was used for derivation and the Virtual International Trials in Stroke Archive(VISTA) was used for validation. NCCT images at presentation and 24+/-6hours were pre-processed, segmented and categorized into lobar, basal ganglia or thalamic regions. Hematoma expansion was defined by region and according to its effect on clinical outcome. Transport-based morphometry (TBM) produced a unique transport-based representation from each NCCT hematoma image. Linear classifiers applied to transport representations were used to derive a predictive model. The regional TBM model’s performance was compared to alternate methods using AUROC analyses. Results The derivation dataset included 574 patients, comprising 162 lobar, 257 basal ganglia and 155 thalamic hematomas. The validation dataset included 196 patients, comprising 31 lobar, 119 basal ganglia and 46 thalamic hematomas. A combined regional model predicted hematoma expansion in the validation dataset with an AUROC of 0.787. The principal image features which varied by location were size, heterogeneity, eccentricity and density distribution towards the hematoma periphery. The regional TBM model outperformed alternative clinician based and machine-learning methods. Conclusions TBM discriminated and visualized regional features of hematoma expansion from presentation NCCT scans. This provides preliminary evidence for regional radiographic features of hematoma expansion. It has the potential to improve hematoma expansion prediction and understanding of hematoma expansion pathophysiology.
[CV-106] RIDE-Splat: Reflection Removal for In-the-Wild Driving Environments
链接: https://arxiv.org/abs/2610.34632
作者: Kevin David Hayes,Alex Hanson,Allen Tu,Pranav Asthana,Matthias Zwicker,Tom Goldstein
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Driving video captured through a windshield violates the world-space consistency assumed by 3D Gaussian Splatting (3DGS): the road scene exhibits parallax while cabin reflections remain attached to the camera, causing standard 3DGS to embed reflected content into the reconstructed scene. We introduce RIDE-Splat, a pipeline for reflection-free 3D reconstruction from posed driving video without clean images, reflection masks, or pretrained networks. RIDE-Splat assigns world-consistent transmission through the windshield to a 3DGS scene and the dominant camera-attached reflection to one shared, signed RGB image. It builds the clean reconstruction in two passes. First, it learns the 3D geometry and the signed reflection image, representing both added radiance and transmission attenuation. Then, it uses the reflection-removed training images to fit a clean 3DGS model. At inference, the reflection image is discarded, preserving the standard 3DGS rendering path. We also introduce the RIDE-Splat Dataset, the first 3D reconstruction benchmark for reflection removal from driving video – with real driving scenes, physical windshield reflections in training views, and separately captured non-reflective evaluation images. Across the RIDE-Splat Dataset, XLD, and Wayve, RIDE-Splat outperforms 3DGS, DC-Gaussian, and RobustSplat on all nine aggregate metrics, improving PSNR by up to 3.59 dB and reducing LPIPS by up to 0.0876. We achieve the highest metrics in 38 out of 42 per-scene comparisons and removes windshield reflections that remain visible in the baselines while preserving the transmitted scene.
[CV-107] P2P-Former: Bridging the Supervision Gap Between Point and Proposal in Temporal Action Localization
链接: https://arxiv.org/abs/2610.34618
作者: Yunchuan Ma,Laiyun Qing,Sai Xu,Guorong Li,Qingming Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recently, point-supervised temporal action localization has attracted increasing attention due to its low annotation cost (only one frame per action instance) and competitive detection performance. However, since point annotations are less complete than proposal annotations, existing methods mainly rely on multiple instance learning for snippet-level classification, which leads to a performance gap compared to fully supervised models. To bridge this gap, we propose P2P-Former, a novel point-to-proposal Transformer-based framework that effectively connects point-level supervision with proposal-level learning for temporal action localization. In this work, we identify two key challenges in transitioning from point-level supervision to full supervision: (1) how to generate reliable proposals as pseudo ground-truth, and (2) how to effectively integrate point-level cues into the fully supervised paradigm. To address the first challenge, we propose a Point-aware Proposal Generation (PPG) strategy that refines the confidence of proposals produced by a base locator using a point-aware importance estimation function, and then performs optimal matching between point annotations and candidate proposals to select reliable proposals as pseudo ground-truth. To tackle the second challenge, we introduce a Point-guided Semantic Enhancement (PSE) module that aggregates prototypes from point-annotated features and injects them into snippet features through Transformer-based interactions. Moreover, we design a Point-to-Proposal Constraint Loss (PPCL) that enforces each predicted proposal to include the annotated point, thereby encouraging more accurate localization. Extensive experiments on the THUMOS14 and ActivityNet-1.3 datasets demonstrate that our method outperforms several existing state-of-the-art approaches.
[CV-108] Do Words Matter? What Text Contributes to Few and Full-Shot Vision-Language Anomaly Detection
链接: https://arxiv.org/abs/2610.34610
作者: Arda Baris Basaran,Mesut Ceylan,Elgar Fleisch,Filipe Barata
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 36 pages, 1 figure, 28 tables
Abstract:Vision-language anomaly detectors describe normal and abnormal appearance with text, and their gains are commonly credited to the semantic content of these descriptions. This attribution is plausible in the zero-shot setting, but few- and full-shot detectors interpose a trained visual adapter, and in that setting it has not been isolated. Changing a description changes both its meaning and its geometry, so accuracy comparisons cannot separate the two. We ask what a fixed pair of text embeddings, or anchors, contributes to such a detector: semantic content, effective scale, or joint anchor geometry. For a single anchor pair, the answer is effective scale alone. The score reads the anchors only through their difference, and a trained adapter can rotate to follow any replacement direction, leaving that direction, and with it the linguistic content, unidentifiable. We prove that replacement anchors with the same difference-vector length permit exactly the same probability maps and attainable losses. Across seven medical datasets, three vision backbones, both supervision regimes, and three adapter families, scale-matched random and orthogonal anchors track correct prompts; breaking the scale match opens a gap that restoring it closes without text. The same pattern holds in MVFA and when MadCLIP’s learned prompts are replaced by a text-free learned scale. When one adapter serves many categories, the answer expands to their joint arrangement. Rotating all category axes together changes nothing, as predicted. For the tested CLIP prompts, these axes are nearly collinear, and meaning-free rotations that spread them apart improve performance on both industrial benchmarks. Even under frozen zero-shot transfer, a random anchor frame transfers as well as the prompt frame: coordinate consistency, not meaning. What these detectors read from their prompts is scale and arrangement, not semantic content.
[CV-109] FlexiMIND: Combined Pretrained Vision Features and Local Structure for Training-Free Deformable Medical Image Registration
链接: https://arxiv.org/abs/2610.34600
作者: Mojtaba Safari,Mingzhe Hu,Shansong Wang,Suyang Xi,Joshua Lorenz,Sean Pitroda,Aytekin Oto,Ralph R. Weichselbaum,Xiaofeng Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
备注:
Abstract:Objective. Deformable image registration remains challenging because corresponding anatomy may exhibit different intensity relationships. Hand-crafted modality-robust descriptors such as Modality-Independent Neighborhood Descriptor with Self-Similarity Context (MIND-SSC) and features from pretrained vision encoders provide distinct representations for correspondence estimation, but how they should be combined remains unclear. We investigate whether tighter integration improves registration over a sequential, scale-separated design. Approach. We introduce FlexiMIND, a training-free and label-free two-stage framework in which frozen FlexiCT-2D encoder features establish contextual correspondence and MIND-SSC refines the deformation within ConvexAdam. Five alternative feature-coupling strategies were evaluated under a protocol across CT-CT (DIR-Lab 4DCT 4DCT, 10 cases), MR-MR (ACDC, 50 patients), and MR-CT (Learn2Reg Abdomen MR-CT, 8 pairs) registration. Performance was assessed using target registration error (TRE), Dice similarity coefficient (Dice), SDlogJ, and local folding. Main results. FlexiMIND achieved a TRE of 1.27 mm on DIR-Lab 4DCT, a DSC of 0.807 on ACDC, and an organ-averaged DSC of 0.825 on Abdomen MR-CT. Across 15 coupling-strategy dataset comparisons, only four outperformed the reference method, with a maximum absolute improvement of 0.0043 in Dice, whereas degradation reached 20.6 % in TRE and 37.9 % in Dice. Changes in registration accuracy were closely associated with changes in deformation-field regularity, and strategies that better preserved regularity generally remained closest to the reference performance. Landmark-level analysis showed that the pretrained encoder stage primarily reduced large correspondence errors rather than uniformly improving local precision. Code is available upon acceptance at this https URL.
[CV-110] An Agent ic AI Workflow for Liver Lesion Segmentation and Structured Radiology Reporting
链接: https://arxiv.org/abs/2610.34599
作者: Fatma Ahmed Mansour,Rahatara Ferdousi,Manar Fathy Elshenawy
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Despite advances in automated liver lesion segmentation, a gap remains between image-derived predictions and clinically interpretable radiology reports. Bridging this gap requires quantitative measurements that are both reproducible and traceable. We present an evidence-grounded agentic AI framework that connects liver and lesion segmentation with deterministic quantification and structured reporting. TotalSegmentator-based nnU-Net models delineate the liver, Couinaud segments, and lesions, while computational tools extract volumes, dimensions, counts, and anatomical locations. A large language model (LLM) organizes these measurements into a predefined reporting schema, with numerical consistency checks against source measurements. On 51 LiTS cases, the segmentation stage achieved a liver Dice score of 0.947, lesion sensitivity of 75.6%, and positive predictive value of 68.9%. Exploratory screening of 25,494 MERLIN reports identified 10,208 (40.0%) with liver-related abnormalities. By connecting quantitative imaging evidence with agentic report generation, this work takes a step toward human-centered medical AI that prioritizes traceability and clinical relevance.
[CV-111] Finetuning Prompt-Based Image Editor for Geometry-Consistent Room Defurnishing
链接: https://arxiv.org/abs/2610.34597
作者: Prakhar Kulshreshtha,Brian Pugh,Zachary Nussdorfer,Salma Jiddi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 2 pages, 3 figures. Poster presented at IEEE ISMAR 2026 (non-archival)
Abstract:Erasing all furniture from a single room image to generate its empty counterpart is valuable for XR applications such as diminished reality and virtual staging. The challenge is twofold: inpainting the furniture regions with realistic content, and doing so without distorting the room’s 3D geometry. Prompt-driven image editors, such as FLUX.2, handle realism impressively, but their structural faithfulness is inferior to explicit layout conditioned inpainting. To improve geometric consistency, we propose a training recipe for finetuning a LoRA on 2,500 furnished-empty room pairs. FLUX.2 with our LoRA adaptor significantly improves geometric consistency while retaining the high texture fidelity. This single-photo inpainter can be an effective building block for depth-aware and 3D-consistent XR pipelines.
[CV-112] Recursive CLIP: Vision-Language Distillation with Looped Transformer Blocks
链接: https://arxiv.org/abs/2610.34563
作者: Szymon Mazurek,Alberto Presta,Michał Byra
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Contrastive vision-language models such as CLIP typically rely on independently parameterized transformer blocks across depth. We investigate the extent to which these blocks can instead be reused recurrently while preserving the transferable capabilities of CLIP. We study recursive transformer backbones in which a small set of N transformer blocks is applied M times per forward pass. By holding N \cdot M fixed to the depth of a standard ViT-B, our models preserve computational depth while progressively increasing parameter sharing. We consider two recursive CLIP model families: rCLIP, where only the vision encoder is recursive, and r ^2 CLIP, where recurrence is also introduced in the text encoder. We study these models under both direct contrastive training and through knowledge distillation from a strong ViT-B/16 teacher pretrained on DFN-2B, using the substantially smaller CC3M and CC12M datasets. We evaluate the resulting models on zero-shot classification and cross-modal retrieval, and further study their adaptability through few-shot learning, visual grounding, and concept unlearning. Our results show that CLIP can tolerate substantial recurrent block reuse, although its effect depends on the encoder and downstream task. Recurrence also provides parameter efficiency by construction, since fewer distinct transformer blocks need to be stored and adapted. Our results establish recurrent block reuse as a practical design space for vision-language models and highlight its implications for knowledge transfer, downstream adaptation, and parameter efficiency.
[CV-113] GRASP: Group-Adaptive Hybrid RL-SFT Post-Training for Medical VQA without Reasoning Annotations
链接: https://arxiv.org/abs/2610.34561
作者: Wanqi Dong,Wanyue Dong,Jingze Ge,Mehul Motani
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Adapting general-purpose vision-language models (VLMs) to medical visual question answering (VQA) is valuable for helping clinicians read images and make diagnoses, but annotated data are scarce and provide only short answers, without reasoning annotations. Supervised fine-tuning (SFT) fits these labels but overfits to them and amplifies their biases, eroding out-of-domain (OOD) generalization. Reinforcement learning (RL) better preserves generalization but never uses the gold answer as a training target: GRPO-style algorithms learn from the contrast between correct and incorrect answers within a group of sampled responses, so all-correct or all-wrong groups give no correctness signal. Existing SFT-RL hybrids either apply SFT and RL in separate stages or switch between them per question, or need reasoning traces from experts or stronger models, which medical VQA data lack. We propose GRASP, which combines RL and SFT in every update using only short answers: RL lets the model explore its own reasoning, SFT teaches the gold answer after that reasoning to supply the direct supervision RL lacks, and their balance adapts to how well the model answers each question. Tuning only about 1% of the parameters with LoRA, GRASP outperforms all compared training objectives both in domain and on OOD open-ended questions across Qwen3-VL-8B, InternVL3-8B and Gemma 3-4B. On an OOD benchmark spanning eight imaging modalities, GRASP-trained InternVL3-8B and Qwen3-VL-8B beat all 10 compared general and medical VLMs in open-ended accuracy, including models trained on up to 2,600 times more data or with over 250 times more trainable parameters. We release MedTerse, a dataset of 12.9K short-answer questions, and our code at this https URL.
[CV-114] VTC-AUDIT: An Interventional Reliability Audit for Visual-Token Compression under Paired Image Shifts AACL
链接: https://arxiv.org/abs/2610.34545
作者: Isaac(Rucheng)Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 6 tables. Code and experimental results: this https URL
Abstract:Visual-token compression is usually evaluated by clean accuracy and efficiency, but these averages do not reveal failures that compression introduces under an image shift where the uncompressed model stays correct. We introduce VTC-AUDIT, which estimates this incremental risk only on pairs where the full model is correct on both images and the compressed model is correct on the clean image. The four arms then cross two binary choices: which image selects the retained token positions, and which image supplies the visual features. We audit three open 8-9B vision-language models on four tasks and eight pre-specified shifts. We evaluate two training-free compressors—hard-pruning DivPrune and prune-and-merge ApET—at a matched 25% retained-position budget, producing 708,000 inference records. Reselecting positions on the shifted image, rather than reusing the clean-image positions, raises the failure rate in all six model-compressor settings. The increase is 3.04-9.60 percentage points for DivPrune and 2.05-4.20 percentage points for ApET, with every 95% interval above zero. Repeating the audit on a separately resampled second manifest reproduces all six increases. In a pooled monitor audit across the four benchmarks, support Jaccard instability fails the practical AUROC criterion of 0.65 in all six settings, whereas clean compressed-model uncertainty clears it. Thus, support turnover is a weak ranking signal under the audited benchmark mixture, while clean compressed-model uncertainty is substantially more informative. These results suggest that future visual-token compression evaluations should report conditional reliability alongside clean accuracy and efficiency.
[CV-115] Conviction Without Evidence: Verdict Stability Is Independent of Visual Grounding in Medical Vision–Language Models NEURIPS2026
链接: https://arxiv.org/abs/2610.34520
作者: Federico Felizzi,Francesco Andrea Causio,Bianca Destro Castaniti,Olivia Riccomi,Michele Ferramola,Vittorio De Vita,Alessandro Tosi,Antonio Cristiano,Alessia Longo,Lorenzo De Mori,Chiara Battipaglia,Melissa Sawaya,Luigi De Angelis,Marcello Di Pumpo,Alessandra Piscitelli,Pietro Eric Risuleo,Giulia Vojvodic,Mariapia Vassalli,Nicolo Scarsi,Manuel Del Medico
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at the Med-Reasoner workshop at NeurIPS 2026
Abstract:Vision–language models applied to medical question answering can reach correct diagnoses without using the diagnostic image, inferring instead from the clinical vignette. We ask whether this is visible in how \emphfirmly a model holds its answer: does a model defend a diagnosis derived from a blank placeholder as tenaciously as one derived from a real medical image? Crossing an image-substitution protocol with two levels of adversarial pressure on 60 Italian State Examination items, we find no evidence that ungrounded verdicts are more fragile. In a paired item-level analysis, fabricated consensus (L4) overturns 94.4% of verdicts with the real image and 98.1% without ( \Delta = +0.037 ; one-sided 95% upper bound +0.105 ); contentless doubt (L1) overturns 62.5% and 51.8% ( \Delta = -0.107 , 95% CI [-0.209, -0.001] ), the point estimate running opposite to the direction a grounding detector requires. The two levels nonetheless produce opposite \emphdynamics: under L4 the model flips once, always to the exact option argued for (110/110 first-turn flips), and then freezes, averaging 1.9 verdict changes over ten turns; under L1 it never settles, averaging 7.4 changes. Instability under contentless doubt tracks intrinsic item ambiguity rather than evidence availability, and so does not serve as a grounding signal. Flips are almost entirely corrupting (99/100 across both levels). The model is not unaware of the missing evidence—it flags the blank image at baseline on 35 of 60 items and abstains on 4–6—yet those abstentions do not survive contact with pressure. None of the epistemic signals a user could plausibly consult distinguishes a grounded diagnosis from a confabulated one.
[CV-116] RIAGE: Efficient Vision-Language-Action Inference Needs Both Reuse and Pruning
链接: https://arxiv.org/abs/2610.34519
作者: Sooyoung Kwon,Carlee Joe-Wong,Kyunghan Lee
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 8 pages, 6 figures
Abstract:Vision-language-action (VLA) models are increasingly used for robot manipulation because they map camera images and a language instruction directly to actions. Robots that cannot be tethered to a workstation, such as mobile manipulators and humanoids, must run them on onboard embedded boards, where the prefill latency of the language backbone grows with the hundreds of visual tokens processed at every control step. To meet real-time limits, training-free accelerators recompute only a subset of the visual tokens and skip the rest, either by pruning them or by reusing their cached key-value (KV) states. The two options save nearly the same computation, yet existing methods apply only one of them to every skipped token. Through a token-level diagnostic, we show that which option keeps the action closer to full recomputation depends on the token’s drift, the change in its shallow-layer value since its cache was computed. On both backbones tested, reusing tokens below a drift threshold was no worse than pruning them, whereas beyond it, reuse yields up to 4.4x the action error of pruning; the same diagnostic calibrates this threshold per backbone. Building on this finding, we propose TRIAGE, which deems a cache valid when its drift is below this threshold and it is not too old. TRIAGE recomputes a fixed budget of tokens, prioritizing important tokens with invalid caches, reuses every other token with a valid cache, and prunes the rest; prune-only and reuse-only methods are the two extremes of this partition. Because the budget fixes the number of recomputed tokens, it directly sets the prefill latency. On LIBERO and SimplerEnv, TRIAGE stays within 3 pp of full recomputation in average success rate while cutting prefill latency on a Jetson AGX Orin by up to 48%, and with only 7% of visual tokens recomputed, it beats the strongest baseline by 24.4 pp on LIBERO-Object.
[CV-117] PhysCreator: Creating Physically-Attributed 3D Assets at Will
链接: https://arxiv.org/abs/2610.34473
作者: Shiyan Zheng,Ming Li,Lei Han,Xiangyu Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Significant advancements in 3D asset creation have enabled high-fidelity 3D generation from images. However, most existing representations remain inherently static, lacking the ability to model physically plausible behaviors under dynamic interactions. We present \textbfPhysCreator, a unified framework that directly generates simulation-ready 3D representations with spatially varying physical properties from images. Unlike prior approaches that treat physical attributes as simple post-hoc predictions or rely on expensive per-instance optimization, we reformulate physical property estimation within a unified 3D representation learning framework. We introduce a voxel-aligned latent space by lifting image observations into a spatially structured 3D representation, in which geometry, appearance, and physical properties are jointly modeled and inferred in a feed-forward manner. To enable scalable learning without densely annotated physical ground truth, we leverage vision-language distillation to transfer material-aware semantic priors from powerful vision-language models into spatially grounded voxel-level supervision. Experiments demonstrate that our method achieves superior visual quality, physical plausibility, and generalization compared to prior approaches.
[CV-118] ask Relevance-Driven Prompt Expansion for Federated Continual Learning
链接: https://arxiv.org/abs/2610.34461
作者: Jiahua Dong,Wenqi Liang,Yang Cong,Yichen Li,Hongyuan Yu,Henghui Ding,Yulun Zhang,Salman Khan,Fahad Shahbaz Khan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
Abstract:Federated continual learning (FCL) has attracted considerable interest by enabling local clients to continually acquire knowledge from new tasks. However, most FCL methods assume that the global server has prior knowledge of the arrival times and number of categories for new tasks, which is unrealistic in the real world. Furthermore, they neglect the importance of positive inter-task semantic relationships when learning new tasks and are significantly affected by irrelevant relationships from old tasks, thereby exacerbating catastrophic forgetting. To resolve the above challenges, we develop a novel Task Relevance-driven Prompt Expansion (TRPE) model. Specifically, to eliminate the reliance on prior knowledge of new tasks, we propose an out-of-distribution detector to identify the arrival of new tasks within local clients, along with a radius-adaptive DBSCAN to infer the quantity of categories in new tasks on the proxy server. Moreover, to explore positive inter-task relationships, we calculate the affinity between the original gradient for learning new task and its gradient orthogonal to the feature space of old tasks to evaluate semantic relevance between old and new tasks from a global perspective. Then, we use this global task relevance to determine the key-prompt pair for new tasks and identify highly related old tasks. After concatenating their prompts for training, we can leverage such positive inter-task relations to improve the continual learning of newly arriving tasks. To balance the learning of new tasks and the forgetting of old ones, a soft orthogonality constraint is applied to the local key and prompt gradients. Experiments spanning diverse datasets and backbones confirm the efficacy of our TRPE model.
[CV-119] On Detection Reliability for Small Object Segmentation
链接: https://arxiv.org/abs/2610.34432
作者: Olivier Rukundo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 6 figures, 2 tables, 10 pages
Abstract:This study investigates whether predicted bounding boxes are reliable for determining where small-object segmentation should be performed without risking false negatives outside those boxes. The investigation leverages adenovirus particles as small objects, YOLO26x for detection, YO-LO26x-Sem for segmentation, and Mask R-CNN for both detection and segmentation. Segmenta-tion-model predictions were matched to ground-truth adenovirus masks or reference objects using their centroids, with ground-truth mask centroids preferred over ground-truth bounding-box centroids as references for matching. Matched predicted-mask centroids were considered as inside at least one detector-predicted bounding box or outside all such boxes, while ground-truth masks without matched predictions were recorded as having no matched segmentation. For the YOLO26 models, 34 reference objects (47.89%) had matched segmentation centroids inside at least one pre-dicted box, 8 (11.27%) had centroids outside all predicted boxes, and 29 (40.85%) had no matched predicted segmentation. For the Mask R-CNN model, the corresponding counts were 69 (97.18%), 0 (0.00%), and 2 (2.82%). These preliminary experiments demonstrated that restricting segmentation to detector-predicted bounding boxes risks excluding matched segmentation predictions outside those boxes, potentially introducing false negatives.
[CV-120] No Cones No Duals: A Fast and Scalable Solver for NRSfM WACV
链接: https://arxiv.org/abs/2610.34403
作者: Suryansh Kumar
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 21 pages, 14 figures, 19 tables, Accepted for publication at IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2027
Abstract:Non-rigid structure from motion (NRSfM) aims to recover camera motion and time-varying 3D shape from tracked 2D image measurements. Recent high-performing methods show that shape-space factorization remains highly effective for this task. These methods often use costly solvers for two key steps. First, a semidefinite programming (SDP) cone solver to perform Euclidean correction. Next, an alternating direction method of multipliers (ADMM) solver to recover the low-rank shape. These steps often scale poorly to long sequences and dense point tracks. We propose a fast, scalable NRSfM solver that avoids these costly steps. For rotation estimation, we introduce a differentiable Euclidean correction that directly optimizes small factors for each sequence. A row-polar projection then maps each estimated camera block to the closest row-orthonormal matrix. For shape estimation, we introduce an unrolled spectral-proximal solver. It combines reprojection-gradient steps with weighted singular-value shrinkage in the reshuffled shape domain. This solver further uses FISTA-style restart and continuation. The spectral shrinkage operator itself is classical, but we embed it in a fixed differentiable solver, while preserving NRSfM’s geometric constraints. The proposed approach is category-agnostic, requires no external training data, and maps naturally to modern accelerator-friendly linear algebra primitives. Across sparse, dense, noisy, missing-data, and long-sequence benchmarks, our solver achieves substantial speedups with competitive reconstruction accuracy, including 7x-1000x runtime improvements on dense sequences with 28,887 tracked points.
[CV-121] A Computationally Efficient Non-Interpolation Virtual Trellis Approach for Mammogram Gradient-based Patterns Analysis
链接: https://arxiv.org/abs/2610.34352
作者: Shahid Rasheed,Muhammad Rashid,Seifedine Kadry,Salman A. AlQahtani
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Feature analysis is a crucial step in medical image analysis, particularly in breast cancer classification. Traditional image analysis methods often struggle with optimal patterns along the diagonal axis, computational efficiency and feature buffer storage. Different classical operators require high-efficiency hardware for the complex operations, such as image convolution and gradient analysis, which are used for processing complex machine learning models for training and test operations. Here, in this research, we introduce a non-interpolation virtual trellis-based (NVHT) operator that utilizes a filter kernel over the same square image without transformation to a hexagonal trellis. This proposed method is very efficient in reducing computational and storage costs. The proposed operator is used for improved feature extraction efficiency and accuracy in disease classification. This method outperforms conventional approaches by leveraging a hexagonal pixel representation while maintaining high sensitivity in different directions with less storage.
[CV-122] Spatially-Aware Self-Organizing Graph Autoencoders via Localized Hebbian Learning
链接: https://arxiv.org/abs/2610.34346
作者: Robin Koç,Hazal Moğultay Özcan,Fatoş T. Yarman Vural
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 6 pages, 5 figures. Accepted to the UBMK conference (not presented)
Abstract:Traditional autoencoders rely on static network architectures to learn latent representations. Recently we suggested a dynamic architecture inspired by Hebbian plasticity, enabling topological self-organization among convolutional feature maps (R. Koç et al., Proceedings of the SIU, 2026). Early implementations of this paradigm infer relationships among feature maps using global average pooling and model their interactions through scalar-valued connections. This approach falls short to fully exploit the underlying spatial information, embedded in input images. Furthermore, abrupt transitions between different loss functions and training dynamics can introduce representation shocks, leading to unstable feature learning. In this paper, we introduce a Spatially-Aware Self-Organizing Graph Autoencoder (S-LooM) to address these structural bottlenecks. By using spatial cross-correlation, and spatial kernels, S-LooM establishes dynamic connections among feature maps. We replace static scalar message passing of autoencoders with learnable spatial kernels and eliminate high-frequency deconvolutional artifacts observed in decoders using nearest-neighbor upsampling. Via a soft-blending training curriculum, the latent topology aligns smoothly during training. Our experiments show that S-LooM yields a more stable topology, compared to the state of the art dynamic architectures. Meaning, the graph structure that emerge has more resistance to feature entanglement and noise. It forms functional communities responsible for visual attributes, such as illumination, shape and texture even under high connection densities.
[CV-123] Birth-Death Structural Learning for 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.34344
作者: Quan Tran Hong,Long Nguyen-Chi,Binh T. Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Code is available at: this https URL
Abstract:3D Gaussian Splatting (3DGS) has become a standard representation for real-time novel-view synthesis. Yet, its quality-efficiency trade-off still relies heavily on adaptive density control. Existing cloning, splitting, and pruning rules use proxy signals, such as position gradients, opacity, and image-space error. Although effective in practice, these signals do not directly answer the structural question behind density control: where should new Gaussians be allocated to reduce the multi-view reconstruction loss, and which existing primitives can be pruned with minimal penalty. To address this, we propose first-variation birth-death control for 3DGS, a principled approach that replaces heuristic structural decisions with a variational score derived from the reconstruction loss. At each structural update, we freeze the current Gaussian population, attach auxiliary mass coordinates to each primitive, and differentiate the multiview loss with respect to these coordinates. We demonstrate that this score is a coordinate of the first variation of the Splat Regression Model, giving a rigorous local descent surrogate for birth-death moves: high-score Gaussians are selected as birth sites, while low-score, low-opacity Gaussians are selected for removal. Across standard 3DGS benchmarks, integrating this score into 3DGS-MCMC yields state-of-the-art reconstruction quality and outperforms existing methods, delivering the most significant gains under tight Gaussian budgets. Ultimately, our work provides compelling evidence that gradient-based variational metrics offer a principled alternative to proxy heuristics in 3DGS.
[CV-124] CrossPair: Unlocking Generalist Visual Transformation Learning in Image Editors
链接: https://arxiv.org/abs/2610.34333
作者: Yibo Zhao,Lifan Jiang,Senlin cheng,Tianrun Wu,Bolei Xu,Xiaobo Li,Can Ye,Xiaofei He,Dehai Zhao,Boxi Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Language models switch tasks from demonstrations, but vision still often relies on separate systems for restoration, editing, and stylization. A before–after example offers a shared instruction only if the transferable change can be separated from the scene that demonstrates it. We introduce CrossPair, which adapts a pretrained multimodal image editor to transfer diverse transformations from one exemplar pair. Exemplar-before, exemplar-after, and query images take fixed roles in its native VLM and VAE pathways: the VLM extracts operation semantics, while the VAE retains fine appearance and structure. A training-only, change-aware DINOv3 objective aligns feature-change directions across content-distinct scenes, guiding the output to follow the demonstrated relation without an inference-time encoder. We construct Transform-360K by grouping filtered pairs from conditional representations, restoration, editing, virtual try-on, and stylization under shared transformations. On Relation252K-derived seen and held-out transformations, \method improves target agreement and edit-direction similarity over released systems. Ablations show that the semantic and visual exemplar routes are complementary and that DINO relation alignment further strengthens transfer; Transform-360K training also improves aggregate held-out-operation scores.
[CV-125] opX: Text-to-Motion for Unseen Skeletons from the Skeleton File Alone
链接: https://arxiv.org/abs/2610.34321
作者: Tengjiao Sun,Pengcheng Fang,Xiaoyu Zhan,Dongjie Fu,Xiaohao Cai,Hansung Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-to-motion models generate for the skeletons they were trained on, while a new character arrives as a skeleton file and nothing else. We study text-to-motion for skeletons that no training clip has used, with no clips of the character, no template, no retargeting and no fine-tuning. We present TopX, a 73M-parameter flow-matching transformer that generates joint rotations and a root trajectory for any kinematic tree from its tree, rest pose and joint names. Motion is normalised by the rest pose and the size of the skeleton, and the rest pose enters as a world-frame descriptor that does not depend on how the rig’s author oriented each joint, with rest conventions randomised in training. Attention is factorised over the joints and the frames of a motion and carries a spatial–temporal encoding designed for skeleton graphs of any topology: queries and keys are rotated across joints by a heat-kernel coordinate of the skeleton graph, which is defined for every tree and does not change with the order of its joints or the eigenbasis of its Laplacian, and across frames by a rotary encoding of frame order. To measure what this buys, we build SkelShift: four motion corpora under one codec, 143,334 captioned clips on skeletons of 4 to 142 joints, from which 4,232 clips on 2,043 skeletons are held out by clip, rig, tree class or species and organised along topology, joint count, body plan, species and rig source, and scored by motion quality: Fréchet distances on public kinetic features and on kinematic descriptors, foot skating and ground contact. Against UniMate trained on the same clips, on test rigs without a training clip, TopX lowers the descriptor FID from 5.73 to 4.58 when UniMate’s guidance is tuned for the kinetic FID, and the kinetic FID from 1.48 to 0.55 when it is tuned for the descriptor FID, also on motions no training skeleton performs; fewer of its frames sink below the ground.
[CV-126] Learning Discriminative Prototypes via Boundary Repulsion MixUp for Object Re-Identification
链接: https://arxiv.org/abs/2610.34305
作者: Athena Psalta,Vasileios Tsironis,Konstantinos Karantzalos
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Object re-identification requires embeddings that are robust to intra-class variation in viewpoint, illumination, and occlusion, yet preserve sharp separation between visually similar identities. The hardest errors arise precisely where confusable identities meet, a region that pairwise losses shape only implicitly. We present PBR-ReID, a prototype-based framework that makes this inter-class geometry an explicit learning target. Each identity is represented by a learnable prototype jointly optimized with a DiNOv3 backbone and the resulting prototype similarity matrix reveals, at every training step, which identities are most confusable. On this signal we build Boundary Repulsion MixUp (BRM), a geometry-aware regularizer that converts standard MixUp’s weakness into a discriminative signal. Rather than interpolating samples across confusable identities, which blurs the boundaries that matter most, BRM constructs an explicit boundary proxy between each confusable pair and repels their features away from it, converting boundary ambiguity into a discriminative signal. We adopt a two-stage training protocol consisting of a prototype warmup phase followed by EMA-stabilized joint fine-tuning. Extensive experiments on person and vehicle ReID benchmarks demonstrate strong performance, with thorough ablations confirming each component’s contribution. Code is available at this https URL.
[CV-127] Real-Time Conveyor Belt Defect Detection via Edge TPU-Accelerated YOLOv11 on Raspberry Pi 5
链接: https://arxiv.org/abs/2610.34304
作者: Muhammad Nur Bijak Bestari,Mulki Indana Zulfa
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 9 figures, 2 tables
Abstract:Conveyor belts are vital arteries in continuous bulk material handling, particularly in heavy process industries such as cement manufacturing. Undetected structural anomalies (longitudinal tears, holes, degraded patch repairs, and impact punctures) cause catastrophic belt snapping, costly downtime, and severe safety hazards. While deep learning offers high accuracy, existing models rely on power-hungry workstation GPUs, hindering practical deployment at remote industrial edge points. In this paper, we propose, implement, and validate an end-to-end, low-power edge AI vision system for real-time conveyor belt defect detection deployed on a Raspberry Pi 5 accelerated by a Google Coral Edge TPU coprocessor. The system leverages an optimized YOLOv11 nano (YOLOv11n) architecture trained on a multi-source industrial corpus exceeding 3,000 augmented samples from an operational cement plant in West Java, open-source datasets, and automated captures. The model undergoes post-training full integer quantization (INT8) compiled for systolic tensor execution. Experimental results demonstrate robust detection capability, achieving precision of 0.742, recall of 0.730, and mAP@0.5 of 0.764 across four defect classes. Deployed on-site in an IP-rated enclosure interfaced with an RTSP PoE industrial camera, the system achieves stable throughput of 30-40 FPS at 320x320 resolution with an end-to-end latency of 70-90 ms. Operational profiling under harsh clinker dust confirms thermal equilibrium at 55 +/- 5 deg C on the host CPU and ~65 deg C on the Edge TPU with zero frame loss. This demonstrates that edge AI bridges the gap between high-accuracy computer vision and cost-effective predictive maintenance in continuous manufacturing.
[CV-128] EditDistill: Is It Possible to Guide Video Editing with Image Editing?
链接: https://arxiv.org/abs/2610.34303
作者: Guojun Lei,Hong Li,Hongbing Yang,Lixue Gong,Chi Wang,Zheng Dong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in image editing have enabled fine-grained and highly controllable visual modifications. However, achieving similar levels of controllability in video editing remains an open challenge, particularly for precise local edits, dynamic temporal effects, and consistent lighting across this http URL this paper, we propose EditDistill, a framework that distills image-level visual transformations from edits to guide video this http URL, given a source video and an edit prompt, we first generate an edited version of the initial video frame using an off-the-shelf image editing model. We then distill the visual transformation of this original-edited image pair into a compact representation termed an edit this http URL descriptor is injected into a video diffusion model through our lightweight fusion module, which steers the denoising process toward the desired edit. To align the video trajectory with the demonstrated image edit while preserving motion and layout, we train with edit-descriptor alignment and DINO structural consistency losses, and derive an equivalent inference-time guidance formulation under flow this http URL experiments demonstrate that our method achieves strong performance on standard benchmarks and produces high-quality video editing results across diverse scenarios. For more details and videos, please refer to the anonymous webpage: this https URL.
[CV-129] ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence
链接: https://arxiv.org/abs/2610.34283
作者: Yihao Hu,Yitong Lin,Pan Wang,Zhihao Wen,Hang Wang,Yiwen Zhao,Deyuan Liu,Yutong Li,Mingda Wan,Yuchen Mao,Tao Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Training a spatial VLM to answer which side an object lies on does not require it to preserve the metric displacement that determines that answer, while supervision for a distance estimate may leave categorical relations largely unconstrained. When post-training emphasizes one readout, improvements can therefore concentrate on that interface while transfer to others remains weak or even degrades; we call this observed specialization interface lock-in. ChronoReplay addresses it by replaying heterogeneous spatial supervision as chronologically ordered episodes, retaining source-native targets, and balancing supervision across families. Under the same observation access, a single ChronoReplay checkpoint improves OVO-S-Bench, OST-Bench, and UCS-Bench together, including a 13.41-point gain on OST-Bench, and achieves the highest three-benchmark mean among the supervision settings compared here. Same-data controls further separate the ingredients of this transfer: answer serialization redistributes gains across OST and UCS, family-balanced learning improves over ordinary joint SFT and natural turn weighting, and chronological replay exceeds a shifted-order control by about 2.41 OST points across two paired seeds. Adaptive frame selection then changes evidence access while keeping the trained model fixed, providing an additional inference-time gain at lower visual cost. Streaming spatial performance depends on what distinctions supervision preserves, how heterogeneous histories are organized during learning, and what causal evidence inference makes available at test time.
[CV-130] Robust Single-Parameter Point Cloud Registration using Truncated Lp Functions
链接: https://arxiv.org/abs/2610.34271
作者: Seong Hun Lee,Patrick Vandewalle,Javier Civera
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:We propose a novel method for robust point cloud registration with only one tunable parameter. Our method, dubbed TLP, uses truncated L_p functions with different values of p to assign priority scores to correspondences, select the hypothesis, and refine the final model. By leveraging the robustness of truncated functions, TLP effectively handles the high outlier rates commonly encountered in real-world 3D feature matching. We also exploit GPU parallelization to evaluate hypotheses efficiently, enabling registration from thousands of correspondences in under a second on average. With only one tunable parameter, our method offers greater practicality than multi-parameter methods that require dataset-specific tuning of several parameters. Evaluations on public benchmark datasets demonstrate that TLP achieves higher accuracy than state-of-the-art single-parameter methods. Furthermore, it outperforms two widely used multi-parameter methods on three of the four evaluated datasets while maintaining competitive runtime.
[CV-131] RacerGS: Residual-Autoregressive Context for 3D Gaussian Splatting Compression
链接: https://arxiv.org/abs/2610.34263
作者: Wenqing Wang,Huimin Zeng,Yun Fu
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:3D Gaussian Splatting (3DGS) has emerged as a promising approach for fast, high-quality novel-view rendering. However, its large parameter size remains a bottleneck for storage and transmission. We propose RacerGS, a compact 3DGS compression framework that achieves scene reconstruction at significantly reduced storage sizes with preserved fidelity. To obtain compact and spatially coherent features, we design a hash projection model that projects the anchor hash encodings into compact context features. Additionally, we introduce a FiLM-modulated autoregressive GRU context model that utilizes causal dependencies among anchor feature groups and uses recurrent hidden states to refine entropy-model parameters, improving bitrate efficiency. Furthermore, we adaptively entropy-code quantized anchor-attribute residuals relative to their context-predicted means, leveraging spatial consistency to reduce symbol entropy. Overall, RacerGS achieves more than 150x compression over vanilla 3DGS and 28x over Scaffold-GS, while providing comparable rendering quality. Extensive experiments across five datasets demonstrate that RacerGS outperforms prior 3DGS compression methods in both storage reduction and rate-distortion performance.
[CV-132] ParetoGS: High-Quality Continuous-Budget 3D Gaussian Splatting with Sparse Adapted Anchors
链接: https://arxiv.org/abs/2610.34235
作者: Xiaobin Deng,Changyu Diao,Min Li,Duanqing Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Scalable 3D Gaussian Splatting allows one scene representation to serve different Gaussian-count budgets, yet its quality-budget curve may lag behind models optimized separately for each budget. Reducing the number of Gaussians changes both the selected members and the reconstruction tasks carried by the survivors; parameters inherited from a high-budget model may fit a low-budget representation poorly. We propose ParetoGS, which describes member changes and parameter evolution with a small set of adapted budget anchors and decodes intermediate budgets without separate optimization at every requested count. ParetoGS alternates pruning and fixed-member adaptation to build nested anchors with persistent identities, and encodes their parameter changes relative to the full model in a compact residual atlas. At decoding, it interpolates the states of shared members and gradually introduces the rendering contributions of new members, connecting sparse anchors into a continuous-budget representation. Across 13 scenes, ParetoGS attains better average PSNR, SSIM, and LPIPS with fewer Gaussians than the compared prefix methods at 10% of each method’s own full model. From a shared full model, it also outperforms FlexGS-imp at its trained 1%, 5%, and 10% budgets and NanoGS-imp at the 1% and 0.1% budgets. On Garden, its non-anchor quality approaches an empirical reference from independently optimized GNS models at higher retention ratios and exceeds it at some lower-budget points. Across 13 scenes, encoding reduces mean additional state from 390.0 to 85.7 MB with an average PSNR change of 0.01 dB. Project page: this https URL Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.34235 [cs.CV] (or arXiv:2610.34235v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.34235 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-133] BoundedGS: Memory-Efficient High-Resolution Training for 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.34233
作者: Xiaobin Deng,Changyu Diao,Min Li,Duanqing Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:High-resolution 3D Gaussian Splatting training increases pixel and Gaussian-tile intersection counts, expanding rasterization caches that coexist with projection and gradient state at peak memory. A complete-view update aggregates gradients across the view; each region’s forward and backward caches can be released after its contribution is collected. BoundedGS separates rasterization and projection backward passes and bounds the active local rasterization working set according to its measured load. Candidate-indexed projection and compact gradient storage address the view-level peaks exposed by this change, while CPU-assisted structure rebuilding and bounded image supply manage resources outside rendering. In a resolution sweep at a fixed count of about 6M Gaussians, the measured rendering working set stays within 1.073-1.077 GiB from 1K to 16K; at 11K, a warm pass takes 19.79 seconds versus 21.76 for full-frame execution. Across five native 5K outdoor scenes, full training reduces the mean rendering working set from 9.97 to 1.76 GiB (82.3%), retains similar reconstruction quality, and shortens training time from 36.7 to 33.9 minutes. Project page: this https URL Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.34233 [cs.CV] (or arXiv:2610.34233v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.34233 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-134] CropReward: A Human-Preference Benchmark and Reward Model for Aesthetic Image Cropping
链接: https://arxiv.org/abs/2610.34221
作者: Xiangfei Sheng,Xiaofeng Pan,Leida Li,Qian Zhang,Biao Wang,Johan Wagemans,Weisi Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages
Abstract:Aesthetic image cropping (AIC) aims to improve image composition and aesthetic quality by removing extraneous or distracting content. Over the past decade, AIC methods have evolved from specialized vision models to multimodal large language models(MLLMs). Reliable evaluation that reflects human preferences is essential for benchmarking AIC methods and guiding model optimization. However, the current AIC evaluation relies largely on geometry-based metrics such as intersection over union (IoU), as well as image aesthetic assessment(IAA) models. Although inconsistencies with human preferences have long been observed, the robustness and reliability of these evaluation criteria remain underexplored. To bridge this gap, we first construct CropBench, a large-scale benchmark comprising 106,995 human preference pairs for 1,019 images, with each candidate set containing the original image and candidate crops produced by 14 AIC methods. A systematic study on CropBench reveals that the existing criteria, ranging from IoU to off-the-shelf IAA assessment models, only exhibit limited alignment with human preferences. Building upon CropBench, we further propose CropReward, an MLLM-based reward model that evaluates a candidate crop by comparing with the original uncropped image and outputs a score indicating explicit aesthetic improvement or degradation relative to the original. In addition, we employ CropReward to construct CropReward-300k, the largest AIC dataset to date, and further develop CropAlign, a preference-aligned AIC model trained solely with CropReward supervision, without human-annotated crop boxes. Experimental results demonstrate that CropReward aligns with human preferences significantly better than existing criteria. CropAlign also surpasses the state-of-the-art AIC methods, demonstrating the advantage of CropReward in guiding AIC model optimization.
[CV-135] From Presence Labels to Bounding Boxes: Can Multimodal Large Language Models Scale Surgical Instrument Localization?
链接: https://arxiv.org/abs/2610.34218
作者: Mohamed Hamdy,Muraam Abdel-Ghani,Fatmaelzahraa Ahmed,Sifna Nasar,Mariam Ahmed,Khalid Al-Jalham,Abdulaziz Al-Ali,Shidin Balakrishnan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Spatial localization of surgical instruments through bounding-box annotations is a foundational component of surgical scene understanding, supporting detection, tracking, interaction analysis, and grounded vision-language applications. Yet existing annotations often cover selected frames, video subsets, or dataset variants rather than complete procedures. Manual extension is time-consuming and expert-dependent. Additionally, weakly supervised and detector-based pseudo-labels require task-specific training and may be imprecise. In this study, we design a scalable MLLM-based annotation framework that converts surgical frames and existing instrument-presence labels into accurate bounding boxes, and systematically analyze its reliability, cost, and design sensitivity. We instantiate the framework using six MLLMs and a multi-call configuration on laparoscopic cholecystectomy in Cholec80. Using a carefully designed stratification scheme, two annotators independently label 1,000 frames, and an expert surgeon reviews one annotation set to establish the reference. Results show that instrument-set F1 ranges from 88.2 to 92.7,whereas mIoU ranges from 0.180 to 0.735, showing that comparable recognition can conceal large localization differences. A structured multi-stage annotation pipeline achieves 0.784 mIoU, compared with 0.735 for the best single-call baseline and 0.814 human-human agreement. Using the best accuracy pipeline, we create and publicly release a spatial extension of Cholec80 comprising 285,977 bounding boxes across all 184,498 annotated frames from the 80 videos. The annotations and human references are made available on GitHub
[CV-136] StaX-Net: Stackelberg Interleaved Asymmetric Network for Cross-Modality Medical Image Translation
链接: https://arxiv.org/abs/2610.34214
作者: Amin Golnari,Saeid Sanei
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Cross-modality medical image translation is challenged by anatomically inconsistent synthesis that can compromise clinical diagnosis. We propose StaX-Net, a Stackelberg interleaved asymmetric network that incorporates a partially coupled encoder-discriminator structure, and a 2D feature aggregation and spatial transformation (FAST) layer that distills target-supervised priors under the learning using privileged information paradigm. The shared blocks between the generator and discriminator enhance architectural feature alignment. To stabilize the leader-follower training, a Stackelberg-inspired asymmetric optimization scheme employs explicit shared-parameter freezing and asymmetric learning rates. The FAST layer distills target-domain priors during training and applies a source-only transform at inference. This constrains the synthesis process through target-informed feature representations relative to standard generative architectures. The code is available on GitHub for public use.
[CV-137] ming Matters Twice: When to Query and When to Skip in Flow-Matching On-Policy Distillation
链接: https://arxiv.org/abs/2610.34191
作者: Renye Yan,Junyi Zhan,Jikang Cheng,Wei Peng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:On-policy distillation (OPD) enables flow-matching models to acquire complementary capabilities from specialized teachers through supervision on student-generated trajectories. However, different evaluation dimensions, such as aesthetics and OCR accuracy, and different tasks, such as image generation and editing, emphasize different aspects of the denoising process, suggesting that their effective supervision may occur at different stages. Existing methods typically select supervision states from predefined timestep distributions, which may overlook these phase-dependent differences and incur unnecessary computational overhead. This motivates two critical timing decisions: when to adaptively query teachers for effective supervision based on the evolving generative process, and when to skip the remaining rollout to avoid redundant computation. In this paper, we introduce PhaseOPD, a phase-aware query-and-skip framework that identifies generative phases online from revisions of predicted-clean latents, using existing student velocity predictions without additional model evaluations. For each task-assigned teacher, PhaseOPD selects a supervision state based on its phase preference and skips the unused rollout suffix once all query decisions are finalized, preserving the single-query velocity-distillation objective. On SD3.5-M, PhaseOPD executes only 5.56 of 16 rollout positions per local batch, reducing training time by 41.3% relative to DanceOPD while improving Aesthetic, GenEval, and OCR scores. Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.34191 [cs.CV] (or arXiv:2610.34191v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.34191 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-138] GeoPrompt: Geometry-Conditioned Prompt Learning for Underwater Open-Vocabulary Instance Segmentation
链接: https://arxiv.org/abs/2610.34190
作者: Shanshan Han,Yuxi Wang,Can Qin,Yujie Chen,Cong Pan,Haitao Li,Guoqiang Zhong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Open-vocabulary instance segmentation (OVIS) provides a promising solution for marine perception by recognizing and segmenting categories beyond predefined label spaces. Existing OVIS methods usually rely on vision-language alignment, where visual region features are matched with textual category embeddings to recognize both seen and unseen classes. However, this alignment assumption becomes fragile in underwater environments. Color attenuation, scattering, turbidity, and low contrast distort visual appearance, causing region features to deviate from their corresponding semantic concepts. As a result, the similarity between degraded visual features and textual classifiers becomes unreliable, which particularly weakens the recognition of unseen underwater categories that cannot be corrected by category-specific supervision. To address these issues, we propose GeoPrompt, a geometry-conditioned prompt learning framework for underwater open-vocabulary instance segmentation. Our key insight is that geometric structure is more stable than visual appearance in degraded underwater scenes. Specifically, GeoPrompt introduces a geometry-guided visual adaptation module (GVAM) to extract depth-derived structural priors and inject them into multi-scale visual features. It further develops a geometry-conditioned prompt learning (GCPL) that generates geometry-aware textual classifiers from structural representations. These geometry-aware classifiers are integrated with fixed and image-conditioned textual classifiers to improve open-vocabulary semantic alignment. Experiments on the MARIS-OpenIS benchmark show that GeoPrompt achieves state-of-the-art performance and brings consistent improvements on underwater open-vocabulary instance segmentation.
[CV-139] Freeze What Is Settled Cache What Is Smooth: Spatial Freezing Keeps Step Caching in Its Low-Error Regime
链接: https://arxiv.org/abs/2610.34177
作者: Baiyan Li,Ganxuan Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Multimedia (cs.MM)
备注: 10 pages, 7 figures
Abstract:Instruction-based image editors built on diffusion transformers run tens of denoising steps. Training-free acceleration either reuses features across steps (step caching) or stops computing the regions an edit leaves untouched (region freezing), and the two have been combined before. We ask how a high speed-up target should be split between them. On Qwen-Image-2.1, freezing does more than remove tokens: it lowers the cost of every step, so the step cache can reach the same overall speed-up with a conservative threshold that keeps its error small. A per-step cost model fitted on development runs predicts held-out latency with a median error of 4.0% and reproduces the speed ceilings of the region methods we re-implemented (2.73 \times predicted vs. 2.70 \times measured for RegionE). Driving frozen tokens to the model’s one-step x_0 estimate beats returning them to the reference image on every development sample, and in 90% of measured cases the composed error stays below the sum of the two individual errors. The resulting method, FreezePlan, keeps five full steps and gives the rest of the budget to TaylorSeer. On a pre-registered test split it runs 5.05 \times faster end to end at 29.8 dB PSNR to the original output, with no significant drop in a VLM judge score; TaylorSeer at 5.02 \times reaches 23.5 dB. The curve extends to 5.99 \times at 28.0 dB. Swapping in the region selectors of prior methods under our schedule lowers median PSNR by only 0.5–0.6 dB, which places the gain in the schedule and budget split rather than in the selector.
[CV-140] Reinforcement Learning for Adaptive Crowd Flow Control Using Dynamic Barriers
链接: https://arxiv.org/abs/2610.34163
作者: Alhanof Alolyan,He Wang,Amani Alonayzan
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: An earlier version was presented at the 26th Scientific Forum for Hajj, Umrah and Visit Research, Umm Al-Qura University (2026)
Abstract:Crowd barriers are widely used to regulate pedestrian movement during large public events such as pilgrimages. However, fixed barrier layouts often create unnecessary stopping and local congestion even when they successfully control the overall crowd flow. Finding better barrier configurations is difficult because there is no dataset that provides the best barrier placement for a given crowd scene. We address this challenge by introducing a simulation-driven reinforcement learning framework that learns barrier configurations without requiring labelled training data. Starting from a segmented crowd scene, we construct a realistic simulation environment that reproduces the observed pedestrian distribution. This environment is then used to train a Soft Actor-Critic (SAC) agent that continuously adjusts barrier positions to improve pedestrian movement. Unlike supervised approaches the proposed framework allows the agent to explore barrier configurations that have never been observed in real data. We evaluate the learned policy under unseen numbers of pedestrians. The learned policy consistently outperforms the existing manually designed barrier configuration by reducing pedestrian stops while maintaining effective flow regulation. These results demonstrate that combining computer vision crowd simulation and reinforcement learning provides an effective framework for optimizing crowd control infrastructure.
[CV-141] MuSTEL: A Large-Scale Multilingual Synthetic Text-Line Image Dataset for OCR
链接: https://arxiv.org/abs/2610.34148
作者: Majid Adibian,Hossein Saberi,Hossein Zeinali
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Optical character recognition (OCR) systems depend strongly on the scale and diversity of their training data. One approach to document OCR first analyzes the page layout and extracts text-line images, which are then transcribed by a recognition model. Multilingual line-level recognition involving Persian, Arabic, and English presents additional challenges because different writing directions, fonts, and styles may coexist within a single line. This paper introduces MuSTEL, a large-scale synthetic multilingual OCR dataset containing more than 14.5 million annotated text-line images, organized into a training partition and predefined small and large test sets. Samples are generated from natural-language corpora, including multilingual content already present in the source texts. The generation framework incorporates capability-aware font selection, within-line font and style changes, mixed-direction text rendering, highlighting, and underlining. A validated repository of 2,009 font files from 711 families supports broad typographic coverage, while varied backgrounds, blur, geometric distortion, skew, stroke width, and image save quality provide diverse document-like appearances. Statistical analysis characterizes the dataset’s language composition, transcription lengths, lexical and character-sequence diversity, image dimensions, typography, and rendering conditions. MuSTEL provides a resource for training and evaluating multilingual line-level OCR systems and investigating recognition robustness across languages, writing directions, and visual conditions.
[CV-142] Reversion Retention and Semantic Intervention: Measuring Functional Content in CLIP Fine-Tuning Deltas
链接: https://arxiv.org/abs/2610.34144
作者: Chih-Ting Liao,Kezhen Bao,Xin Cao
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 20 pages. Accepted for publication in Transactions on Machine Learning Research (TMLR)
Abstract:Fine-tuning adapts pretrained models to new tasks, but identifying which weight changes support the resulting behavior remains difficult. Weight-space interpretability assigns semantic descriptions to these changes, called fine-tuning deltas. Such descriptions leave open how much task performance depends on the identified updates, particularly when other updates can support the same predictions. We address this gap by pairing reversion, which restores selected parameters to their pretrained values, with retention, which keeps only selected fine-tuning updates. We evaluate Contrastive Language-Image Pretraining (CLIP) models on synthetic financial charts, Flowers102, EuroSAT, and FGVCAircraft with ViT-L/14, and on Flowers102 with ViT-B/16, using a shared recipe that tunes the final six visual blocks. Our contributions are threefold. First, attention value-output (VO) updates alone recover 43-87% of the accuracy improvement, although reverting them leaves chart accuracy unchanged. Cross-entropy loss preserves this contrast on Charts and EuroSAT, while correct-class score margins reveal additional contributions. Second, VO weights recover more loss improvement than all ten equally sized, layer-matched random subsets in each setting. VO and multilayer perceptron (MLP) retentions jointly correct 25-81% of pretrained errors, demonstrating overlapping recovered behavior, which we call operational redundancy. Third, removing semantically selected directions reduces VO-retained recovery more than matched-random removal in all five settings. Together, these results distinguish semantic alignment, removal cost, and retained recovery as complementary evidence about the functional role of fine-tuning updates within the evaluated CLIP recipe.
[CV-143] StickFigure: Enabling Pose-Only IoT Cameras for Privacy-Preserving Edge Vision
链接: https://arxiv.org/abs/2610.34130
作者: Aveline Xiaoyuan Han,Chenshu Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 9 figures
Abstract:Indoor IoT cameras can support assisted living and activity-aware spaces, but exporting raw RGB frames raises privacy concerns by exposing facial, clothing, and household details that body-motion tasks often do not require. We present StickFigure, a pose-only IoT camera system that, during real-time monitoring, keeps raw frames local and ephemeral while exporting only compact skeleton streams sufficient for downstream human activity recognition (HAR). While pose estimation has matured on GPUs, extracting accurate pose streams on resource-constrained IoT cameras remains difficult. To overcome these challenges and enable on-device pose streams on (almost) any IoT camera, we present two core designs in StickFigure: an activation-aware pose backbone that allocates channel capacity after spatial downsampling and a lightweight hierarchical MLP refiner that selectively corrects elbow and wrist coordinates using intermediate visual features and kinematic context. We fully implement StickFigure and evaluate it on public benchmarks and real-world deployment. Compared with state-of-the-art MCU pose models, StickFigure achieves the highest pose accuracy and downstream HAR utility on both NTU RGB+D and Penn Action, while using 96.5% fewer parameters and 99.3% fewer MACs per frame than RTMPose-S-based baselines, achieving 0.107 s p95 latency on a representative STM32H753 MCU. We believe StickFigure establishes a new operating point for privacy-aware edge vision on massive IoT cameras.
[CV-144] RoboDuo: Attempt Check and Atomic Recovery through Code for Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.34108
作者: Jie Wang,Bo Zhang,Renzhi Lu,Jianyu Chen
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-Language-Action (VLA) models map a language instruction and the current observation directly to actions, without checking whether the instruction has been completed. For long-horizon tasks, a single failed step therefore goes unnoticed and leaves the whole task unfinished. Such failed episodes, however, often end with the task partially complete. Moreover, the unfinished steps are reusable atomic skills that recur across demonstrations. A failed attempt can thus be recovered by composing with atomic skills. We propose RoboDuo, a closed-loop dual-system framework that uses code as the interface between a coding agent and two VLAs. The agent writes a program that lets a full-task VLA attempt the whole task and checks whether it is complete; on failure, an atomic VLA finishes the remaining steps. Since the agent emits no actions and only decides which VLA for execution, RoboDuo wraps an existing full-task VLA without updating its weights. To train the atomic VLA and the coding agent with the completion-checking ability, we build LIBERO-Atomic, LIBERO-Code, and LIBERO-Judge from a single set of LIBERO demonstrations. RoboDuo consistently improves X-VLA and \pi_0.5 on LIBERO and LIBERO-PRO, with the largest gains on long-horizon tasks, and recovers failed rollouts of a frozen full-task VLA on a real robot.
[CV-145] LegoOcc: Towards Consistent Open-Vocabulary 3D Occupancy Prediction
链接: https://arxiv.org/abs/2610.34093
作者: Changqing Zhou,Yueru Luo,Han Zhang,Zeyu Jiang,Changhao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Open-vocabulary 3D occupancy prediction is important for embodied agents operating in complex environments whose semantic categories extend beyond fixed taxonomies. While recent studies have explored open-vocabulary occupancy in outdoor driving scenarios, indoor environments remain challenging due to dense occlusions, complex geometry, and fine-grained long-tailed semantics. To reduce the need for expensive voxel-level semantic annotations, we study a geometry-only supervision setting requiring only binary occupancy labels. We build upon Language-Embedded Gaussians as a unified representation coupling 3D geometry with language-aligned semantics. For geometry learning, we introduce an opacity-aware Poisson-based Gaussian-to-Occupancy formulation that enables stable volumetric aggregation under binary occupancy supervision. For semantic learning, we render Gaussian features into images and align them with a frozen open-vocabulary model, together with Progressive Temperature Decay to reduce feature mixing during Gaussian splatting. Building on this framework, we further introduce Instance-Level Semantic Consistency to alleviate fragmented object predictions, and Covisibility-Aware Support-View Selection to provide informative and less redundant multi-view semantic supervision. We also extend our framework to outdoor driving scenarios to evaluate its generality across substantially different environments. Extensive experiments on indoor and outdoor benchmarks demonstrate that our method achieves strong open-vocabulary occupancy performance, validating the effectiveness and generality of our framework. Code will be available at this https URL.
[CV-146] Learning the Autoregressive: Direct Score-Difference Estimation with Amortized Future Credit for Video Distillation
链接: https://arxiv.org/abs/2610.34092
作者: Zixuan Duan,Yabo Chen,Xunzhi Xiang,Xin Zhang,Haibin Huang,Yin Chen,Chi Zhang,Qi Fan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Few-step distillation for autoregressive (AR) video generation commonly first initializes an AR student and then refines it with distribution matching distillation (DMD). This refinement process faces two challenges. First, although initialization already yields a reasonably capable student, DMD obtains the refinement direction indirectly by estimating the full student score before forming the teacher-student correction. Second, detached-history training prevents the generator from assigning credit to a current chunk for its influence on subsequent chunks. We introduce Leta, a few-step distillation framework that addresses these challenges with ScoreDelta and FutureCredit. ScoreDelta directly estimates the teacher-student correction from denoising residuals, while FutureCredit learns an auxiliary cross-chunk update signal from vector-Jacobian products (VJPs) of next-chunk teacher residuals. Leta combines these signals into a joint teacher-guided update field in a single forward pass during generator updates. Teacher evaluations and credit VJPs are used only for estimator training. For ScoreDelta, we establish population equivalence to score-difference estimation and a conditional approximation advantage under limited capacity in a fixed-feature linear model. In our controlled experiments, Leta improves over DMD by 0.9 points on VBench and 0.66 on VideoAlign. This advantage persists at a higher generator update frequency (a generator-to-estimator update ratio of G:E=1:1), where Leta improves over DMD by 1.6 on VBench and 0.35 on VideoAlign. Project page: this https URL.
[CV-147] Doppler-aided Position-Velocity Tracking of Hidden Objects using mmWave Radar
链接: https://arxiv.org/abs/2610.34085
作者: Yerin Cheon,Kush Garg,Akshat Dave
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 8 figures
Abstract:Safe indoor operation requires autonomous tracking of objects that are not directly visible to the robot. Optical non-line-of-sight (NLoS) methods depend on ambient lighting and scene geometry or on weak multi-bounce returns that need long acquisitions. Millimeter-wave (mmWave) radar offers an attractive alternative: walls act as reflectors for mmWave signals, and frequency-modulated continuous-wave (FMCW) radars enable fast, practical measurement of range, angle, and velocity. However, the low resolution of these radars leads to spatially ambiguous observations. The Doppler information, although particularly sensitive to target motion, remains underexplored in indoor NLoS localization: existing methods emphasize spatial representations and either collapse or discard the Doppler dimension before learning. To leverage the complete Range-Azimuth-Doppler (RAD) information, we acquire the first indoor NLoS radar dataset that pairs sequential RAD measurements with synchronized position and velocity ground truth under controlled variations in motion: 576 sequences totaling 720,856 frames across 12 trajectories of a mobile robot moving along varying speeds, starting positions, and material configurations. We then formulate NLoS tracking as sequential position-velocity estimation rather than frame-wise localization: a shared 3-D encoder processes the full RAD representation of consecutive frames, separate recurrent branches estimate position and velocity, and post-inference refinement combines the two into a motion-consistent trajectory. Our method achieves 8.95cm position and 1.75cm/s velocity errors on held-out trajectories and generalizes to unseen scenarios involving a moving human target. Together, our dataset and tracker establish a benchmark for tracking hidden moving targets with a low-cost radar.
[CV-148] Behavior Spatial Benchmark: Evaluating Spatial Understanding for Robotics
链接: https://arxiv.org/abs/2610.34051
作者: Sirui Chen,Huang Huang,Qineng Wang,Wenlong Huang,Manling Li,Ruohan Zhang,Li Fei-Fei,Jiajun Wu
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robot tasks require grounding spatial language in physical scenes, but task success alone cannot establish whether spatial instructions are followed. We present BEHAVIOR-Spatial, a benchmark built on BEHAVIOR-1K that measures spatial question answering, point-based target grounding, and robot action to examine how VLM spatial understanding relates to VLA behavior. It comprises 3,256 spatial questions across 49 activities, 1,240 grounding queries across 14 manipulation tasks, and action evaluations covering those 14 tasks and 10 navigation tasks. Our manipulation protocol evaluates held-out configurations and instruction following when the requested target changes within a fixed training layout. The baseline achieves 41.6% on held-out configurations but only 16.1% under changed instructions. VLM adaptation and robot-data pretraining provide limited improvements, whereas oracle visual target cues raise these rates to 60.3% and 38.0%, respectively. Perception evaluations reveal policy-dependent correlations between VLM question-answering and grounding accuracy and VLA manipulation performance. Together, these findings motivate evaluating spatial instruction following alongside task completion and examining when spatial perception performance translates into robot behavior. Project website: this https URL.
[CV-149] What Sketches Tell Us about LVLMs: Conventions Grounding and Localisation NEURIPS2026
链接: https://arxiv.org/abs/2610.34037
作者: Chaitat Utintu,Ahmed Bourouis,Arkaprabha Basu,Yi-Zhe Song
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted in NeurIPS 2026
Abstract:LVLM benchmarks tell us whether a model answers correctly, but not which visual conventions it has retained, nor where those conventions live. Sketches expose this hidden structure. A sketch removes texture, lighting, and photographic context while preserving the strokes humans judged sufficient for recognition. If an LVLM can operate on that abstraction, the surviving signal is structural rather than merely photometric. We introduce a training-free sketch probe with three readouts: temporal part-construction order, spatial attention-based grounding, and the mechanistic layer/head locus of grounding. We evaluate nine LVLMs on part order and five on the attention readouts, over two part-annotated sketch datasets. Three findings emerge. First, LVLM part order tracks human convention: models reproduce canonical orderings where human drawings converge, and become correspondingly variable where human orderings vary. Controls that blank, withhold, and mismatch the sketch locate this convention in the language model. Second, part-level grounding survives the abstraction: attention heads selected from the frozen model with an existing recipe recover part and object regions on object and scene sketches, with no grounding supervision. Third, the grounding signal concentrates in a narrow mid-network band whose location follows the base language model rather than parameter count in the two families we compare, and remains stable across object and scene sketches. The result is a practical diagnostic for sketch-driven LVLMs: no fine-tuning, no auxiliary segmenter, and a layer band that doubles as a head-selection prior, matching or improving grounding mAcc@0.1 in every model and dataset we tested.
[CV-150] Fovea-JEPA: Overcoming Geometric Blind Spots in V-JEPA for Camera Motion Recognition
链接: https://arxiv.org/abs/2610.34033
作者: Xiangkun Shi,Hanhui Li,Shiyue Zhang,Chang Liu,Junwen Pan,Zhenbang Sun,Xiaodan Liang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Camera motion recognition requires separating viewpoint changes caused by camera translation, rotation, or zoom from object motion and scene dynamics. Video-language models often rely on semantic cues and struggle with the geometric evidence needed for this task. Joint-Embedding Predictive Architectures such as V-JEPA offer strong spatio-temporal representations, but our experiments suggest that they underrepresent geometric cues in some regions. Under temporal-reversal supervision, V-JEPA improves overall but degrades on several motion families, whereas a weaker geometry-oriented VGGT predictor improves across all families. We refer to these geometrically underrepresented regions as geometric blind spots and propose Fovea-JEPA to identify and address them. Using V-JEPA as the base model and VGGT as a geometry expert, we construct intervention pseudo-labels from their prediction errors to weakly supervise a blind-spot predictor. The predictor produces token-wise response maps, which are aggregated into a clip-level intervention score to guide adaptive fusion of visual and geometric features. This selective intervention introduces complementary geometric information while preserving reliable visual predictions. Experiments on CameraMotionDataset, CameraBench, and ShotBench demonstrate the effectiveness of Fovea-JEPA, with gains of 9.67, 8.67, and 24.38 percentage points over previously reported state-of-the-art baselines, respectively. Code and models will be released upon acceptance.
[CV-151] Explaining Navigation Decisions: Stage-Aware Visual Attribution for Vision-Language Navigation
链接: https://arxiv.org/abs/2610.34028
作者: Yiyuan Yao,Xinpeng Ding,Minglei Chen,Fei Gao,Lihuo He,Jie Yuan,Aijun Li,Nannan Wang,Xinbo Gao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:As Vision-Language Models (VLMs) increasingly enable robots to follow language instructions, understanding their navigation decisions becomes important for reliable deployment. Existing explainability methods overlook a key challenge in Vision-Language Navigation (VLN): evidence for the current action may be distributed across historical and current observations and associated with different instruction stages. We identify instruction-stage insensitivity, where existing methods produce similar explanations despite changes across different navigation stages. To address this issue, we propose a stage-aware attribution framework for VLN. We first construct action-conditioned temporal instruction–visual attributions across observations. These attributions are projected into instruction space to estimate stage associations, which guide the aggregation of relevant historical evidence for refining the current explanation. An instruction-guided spatial filter further suppresses isolated responses while preserving compact, instruction-relevant regions. Experiments across multiple VLN policies and benchmarks show improved attribution quality over existing methods. We further demonstrate that our method acts as a diagnostic tool for analyzing failure cases, and validate its applicability to real-world navigation on a Unitree G1 robot.
[CV-152] Hybrid Mamba-Transformer Network for Single-Shot 3D Reconstruction from Fringe Patterns
链接: https://arxiv.org/abs/2610.33988
作者: Xinjie Zhou,XiaoLei Li
类目: Computer Vision and Pattern Recognition (cs.CV); Optics (physics.optics)
备注:
Abstract:Deep learning facilitates rapid single shot three dimensional (3D) reconstruction using fringe projection profilometry (FPP), but recovering accurate surface height from a single high frequency fringe image remains challenging because both long-range spatial consistency and fine geometric details must be preserved. Convolutional networks are limited in modeling long range dependencies, whereas window-based Transformers establish global interactions only progressively with increasing depth. To address these limitations, a hybrid Mamba Transformer network is proposed for fringe to height regression. The network adopts a Swin based U shaped architecture and introduces visual state space modeling at the intermediate resolution encoder stage. An axial and dilated vision state space (ADVSS) block is further developed to enhance prescan feature encoding through local, dilated, and axial depthwise convolutions while retaining four directional selective scanning. This design combines linear complexity long range modeling with direction aware preservation of local fringe structures. Experiments on a publicly available FPP dataset show that the proposed method outperforms hNet, SwinConvUNet, UHRNet, and HiDNet in mean absolute error(MAE), root mean square error (RMSE), absolute relative error, and structural similarity index measure (SSIM). Using only one deformed fringe image, the proposed method achieves an RMSE of 0.548 mm and an SSIM of 0.9985, reducing RMSE by 65.2% and 59.6% relative to UHRNet and HiDNet, respectively. These results demonstrate the effectiveness of combining selective state space modeling with hierarchical window based attention for accurate single-shot FPP reconstruction.
[CV-153] Accuracy Is Not Enough: An Efficiency- and Statistics-Aware Benchmark of ResNet50 and Vision Transformer for Explainable Multi-Class Gastrointestinal Diseases Classification
链接: https://arxiv.org/abs/2610.33965
作者: Akwasi Asare
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Automated classification of gastrointestinal (GI) findings in endoscopic images could support early diagnosis, but architectures are often chosen on accuracy alone. We present a controlled, reproducible benchmark of ResNet50 and ViT-B/16 for eight-class GI classification on the public Kvasir dataset (4,000 images), using a stratified split (400 test images) and stratified group 5-fold cross-validation over all images. In cross-validation, ResNet50 and ViT-B/16 reached 92.12% and 91.70% accuracy, and a two one-sided tests (TOST) procedure showed equivalence within a +/-3 percentage-point margin (difference +0.43 points, 90% CI -0.33 to +1.18; p 0.001). The architectures differed mainly in cost: ResNet50 had 3.6 times fewer parameters and 4.3 times fewer operations, 2.5-fold lower single-image CPU latency (69 vs. 177 ms) and 3.8-fold higher GPU throughput. Cross-validated calibration was similar (expected calibration error 0.021 vs. 0.023). An edge-oriented transformer, MobileViT-S, was most accurate on the test split (95.5%) but lost 9.4 points when an acquisition artefact present in some classes (the ScopeGuide inset) was occluded, whereas ResNet50 was not significantly affected. On 500 unseen polyp images with expert masks, both models’ Grad-CAM maps focused on the lesion (3.6 times chance). On a de-duplicated external subset of GastroVision (1,998 images), accuracy fell to 78-80% for all three models and differences between them remained small. Accuracy differences between architectures were thus small relative to the effects of shortcut learning and domain shift, whereas efficiency differences were large. An efficient CNN is a defensible default for resource-constrained screening, but no model is deployment-ready without local validation.
[CV-154] urboWAM: Learning What Matters for Fast World Action Models
链接: https://arxiv.org/abs/2610.33941
作者: Guile Wu,Shuangzhi Li,Xingxin Chen,Chengjie Huang,Sam Hosseini,Tongtong Cao,Yingxue Zhang,Dongfeng Bai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical report. Project page: this https URL
Abstract:World action models have demonstrated promising performance in robot policy learning, yet their high inference latency remains a critical bottleneck for real-world deployment. Recent approaches have sought to improve inference efficiency by reducing test-time future-frame imagination or accelerating action prediction. However, these methods often leverage representations from all video world model layers without explicitly considering their varying relevance to action prediction, leading to redundant computation and suboptimal representation utilization. In this work, we propose TurboWAM, a fast world action model that efficiently aggregates structured world representations conditioned on action predictions. Specifically, TurboWAM introduces a structured world representation aggregation module that adaptively fuses representations from different world model layers according to their relevance to the evolving action denoising state. Furthermore, TurboWAM employs a structured action supervision mechanism that adaptively modulates gradients from action supervision across different world model layers during backward propagation, facilitating world representation learning. During inference, by reducing the number of world model layers executed and caching relevant representations across action denoising steps, TurboWAM substantially reduces computational overhead. Extensive experiments on multiple robot manipulation benchmarks demonstrate that TurboWAM achieves low inference latency while maintaining competitive task success rates compared with state-of-the-art methods.
[CV-155] xture-Aware Strategies for Subregion Style Transfer Augmentation
链接: https://arxiv.org/abs/2610.33910
作者: Cheng Ji,Yu Zhao,Xiaochong Jiang,Wenting Yang,Shiqi Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published at 2026 3rd International Conference on Digital Image Processing and Computer Applications (DIPCA)
Abstract:A common strategy for data augmentation is to apply style transfer on training images to enrich appearance variations. Recent approaches have combined style augmentation with random erasing by selectively replacing image subregions with style-transferred patches. However, existing methods rely on uniformly random region selection or fixed geometric constraints, which may fail to cover informative texture patterns and limit augmentation diversity. To address these limitations, we propose a texture-aware subregion style augmentation framework that prioritizes regions with strong texture responses for style replacement. Furthermore, we extend this framework with a hybrid region strategy that combines a texture-driven region with an auxiliary random region, implicitly enabling more diverse style perturbations. Experiments demonstrate that our approach achieves higher classification accuracy and improves robustness compared with existing style-based augmentation methods.
[CV-156] Rank-1 Weight-Corrected Trainable Weight Averag ing for Out-of-Distribution Detection ACCV2026
链接: https://arxiv.org/abs/2610.33909
作者: Akito Shinohara,Hiroaki Aizawa
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to ACCV2026 Workshop on Subspaces in the Era of Foundation Models
Abstract:Out-of-distribution (OOD) detection identifies inputs outside the training distribution. Rank-based analyses such as RankFeat and RankWeight suggest that OOD overconfidence is closely related to dominant low-rank components in feature and weight spaces. Since weight averaging operates directly in parameter space and improves model generalization, we consider it for OOD detection from a dominant-component perspective. Trainable Weight Averaging (TWA) assigns a single coefficient to each checkpoint, limiting independent control of dominant weight components. We extend TWA with leading rank-one weight components extracted by Singular Value Decomposition as correction directions, whose contributions are learned through the in-distribution classification objective. On ResNet50, our method improves over fine-tuning and standard TWA across the evaluated datasets with MSP and RankWeight, both with and without Outlier Exposure. Mahalanobis-based evaluation yields mixed results, indicating that the benefit depends on the OOD scoring protocol. Target-layer analyses with MSP and Energy favor correction near the classifier.
[CV-157] WMEval: Evaluating Interactive World Models Grounded on Physical Simulations
链接: https://arxiv.org/abs/2610.33894
作者: Lanxiang Hu,Yixin Huang,Abhilash Shankarampeta,Zilin Dai,Yuyuan Wu,Kun Qian,Junze Liu,Arvind Srinivasan,Mingjia Huo,Will Lin,Kai Zhong,Weiqi Zhang,Tian Wang,Qingjun Cui,Tajana Rosing,Hao Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Interactive world models hold immense potential for high-stakes applications in robotics, game engines, and immersive digital twins, yet existing benchmarks primarily focus on perceptual similarity rather than the physical-causal coherence essential for real-world utility. We introduce WMEval, a comprehensive benchmark that evaluates interactive world models against Unity-based physical simulations. By leveraging Unity as a high-fidelity ground-truth oracle, we shift evaluation beyond perceptual visual assessment toward the objective state-based verification required for downstream applications. Spanning 9 diverse scenes, WMEval assesses action controllability, scene and character consistency, prompt adherence, kinematics, fluid mechanics, physical causality, and lighting. By executing matched action sequences in both the simulator and the world model, WMEval enables evaluation at both the perceptual level and the physical-causal level, using simulator-derived states as reference. Experiments on state-of-the-art models reveal significant room for improvement in dimensions that remain hidden by standard perceptual metrics but are vital for real-world deployments. Our code is available at: this https URL.
[CV-158] Semantic Concept Steering Breaks the Explanation Drift Loop in Continual Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.33877
作者: Yehonatan Elisha,Oren Barkan,Noam Koenigstein
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: NeurIPS 2026
Abstract:Catastrophic forgetting in continual learning (CL) manifests not only as accuracy degradation but also as \emphexplanation drift: models maintain predictive accuracy while silently shifting attention from diagnostic features to spurious correlations. This work identifies and resolves a previously overlooked structural flaw in explanation-aware CL: pixel-level saliency regularization derives supervision targets from the model’s own frozen saliency maps, creating a \emphself-referential drift loop that actively reinforces spurious attention across tasks. We introduce \textbfC ^3 L (Concept-Consistent Continual Learning), which breaks this loop by anchoring regularization to class-consistent semantic concepts (e.g., curved beak'', striped pattern’') discovered offline, independent of model state, providing (i) target independence from model drift, and (ii) cross-instance semantic consistency across every instance of a class. We identify that our concept-based objective, combined with approximate LRP relevance conservation, induces a \emphcrowding-out effect: since concept and background regions compete for the relevance of the same prediction, enforcing high relevance within concept regions implicitly suppresses spurious background correlations without explicit penalization. \method does not require human-annotated concept sets, as it employs a fully automated concept discovery and spatial grounding. Extensive evaluation across five benchmarks and over 20 baselines demonstrates state-of-the-art performance in accuracy, forgetting, explanation quality, spurious correlation mitigation, and out-of-distribution robustness.
[CV-159] Stable Dimensionality Reduction via Isometric Centroid-Encoder
链接: https://arxiv.org/abs/2610.33872
作者: Tomojit Ghosh,Siva Rama Krishna Prasad Changala,Prathyusha Kanakamalla,Harigovind Harikumar
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Total 45 pages, 15 figures and 16 tables
Abstract:Many supervised dimensionality reduction methods promote intra-class compactness and inter-class separation but do not explicitly constrain class geometry, leading to variability in the relative arrangement of classes across independent runs. Such variability limits the reproducibility and interpretability of low-dimensional embeddings. We propose the Isometric Centroid-Encoder (ICE), a supervised dimensionality reduction framework in which class centroids are learned directly in the latent space while preserving a prescribed pairwise class-centroid geometry. Unlike existing centroid- or prototype-based methods that primarily encourage intra-class compactness, ICE explicitly regularizes the relative arrangement of latent class centroids. Unlike multidimensional scaling and many metric-learning approaches that preserve sample-level relationships, ICE operates on class centroids, preserving inter-class geometry without requiring large numbers of sample-pair or triplet constraints. We further show that the intrinsic dimensionality required to represent the centroid-distance configuration of M classes is at most M-1 , explaining the observed saturation of distance-preservation measures as the latent dimensionality increases. To quantify embedding reproducibility, we introduce the Centroid Neighborhood Stability (CNS) metric, which measures the consistency of class-level neighborhood topology across independent runs and complements existing embedding quality measures. Experiments on image and biological datasets demonstrate that ICE produces stable embeddings with improved preservation of class-centroid geometry while remaining computationally efficient. Additional analyses of optimization dynamics, sensitivity, and preprocessing further provide practical insights into latent dimensionality and explicit geometry modeling for supervised dimensionality reduction.
[CV-160] RelateGS: Reliable Semantic and Relational Evidence for 3D Scene Understanding with Gaussian Splatting WACV2027
链接: https://arxiv.org/abs/2610.33868
作者: Seongjun Choi,Sangyoun Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027
Abstract:Language-aware 3D Gaussian Splatting (3DGS) has advanced language-driven 3D scene understanding by incorporating semantic and relational cues into reconstructed scenes. However, relation-aware grounding depends not only on representing such information, but also on whether the underlying scene-side evidence is reliable and sufficient. Multi-view semantic aggregation can incorporate weak observations, while object relations can remain under-observed along the available camera trajectory, leading to ambiguous evidence for target disambiguation. We present RelateGS, a structured language-aware Gaussian representation that selectively refines semantic and relational evidence for 3D visual grounding. RelateGS suppresses unreliable observations during semantic feature uplifting and consolidates multi-view object observations into a persistent 3D graph. The graph further identifies under-observed object relations and guides targeted re-querying of informative existing frames to complement their evidence. Experiments on Sr3D+ and Nr3D demonstrate competitive 3D grounding performance, improving the previous best by 7.9 and 1.7 percentage points at Acc@0.25, respectively.
[CV-161] Late Breaking Results: Audio-Language-Action Models with Direct Speech Conditioning for Robotic Manipulation
链接: https://arxiv.org/abs/2610.33866
作者: Enfu Nan,Pu Zhao,Yixiao Chen,Lin Zhao,Juyi Lin,Chen Wang,Weiwei Chen,Yanzhi Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 2 pages, 1 figure. Accepted to DAC 2026 Late Breaking Results
Abstract:Vision-language-action (VLA) models enable robots to follow instructions in typed text, limiting broader deployment in natural speech. The traditional solution by prepending a speech recognition before VLA to convert speech into texts would incur additional latency and propagate errors. To address this, we present Speech- \pi_0.5 , which maps raw speech directly to motor commands without explicit textual transcriptions. Specifically, built on \pi_0.5 , we use a cross-modal projection to convert voice frames directly into audio tokens with significant response latency reduction. To further improve the accuracy, we adopt split LoRA adaptation with dedicated audio and task adapters, under a two-stage training. Speech- \pi_0.5 achieves 3\times lower response latency than ASR pipeline while maintaining competitive task success on LIBERO, and generalizes robustly to unseen accents where a single adapter degrades significantly.
[CV-162] rackTok: Object-Centric Video Tokenization with Semantically Persistent Tokens
链接: https://arxiv.org/abs/2610.33857
作者: Nhi Pham,Christopher Wewer,Bernt Schiele,Jonas Fischer,Jan Eric Lenssen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We introduce TrackTok, an efficient video tokenization framework that produces object-centric tokens, i.e., tokens that consistently encode the same semantic objects across time, even as it moves, deforms, or undergoes partial occlusion. Conventional patch-based or existing holistic tokenizers produce token identities often tied to fixed image locations. In contrast, our approach learns temporally persistent object-centric representations that bind visual features across frames into stable semantic units. This enables substantially more compact, continuous latent video representations, reducing the number of tokens required for downstream generative modeling with state-of-the-art flow models, while preserving the semantic structure needed for high-quality synthesis. Our tokenizer offers a favorable quality-efficiency trade-off. TrackTok achieves comparable class-conditional generative FVD on UCF-101 dataset at 8 times fewer latent tokens and better latency, and outperforms existing 1D tokenization baselines on frame prediction task on Kinetics-600 across low token budgets. These results suggest that object-aligned and temporally persistent tokenization is a promising direction for scalable video generation, where reducing latent token counts is critical for efficient modeling. Our code for this work is available at this https URL.
[CV-163] Grounded Exploration for Open-Ended Image Editing
链接: https://arxiv.org/abs/2610.33810
作者: Jieyu Zhou,Mehdi Ben Ayed,Shubham Gupta,Zhenyu Tan,Jiajing Xu,Dmitry Kislyuk,Eric Tzeng,Christopher J. MacLellan,Charles Rosenberg
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Open-ended image-editing requests, such as ``decorate this room,‘’ admit multiple valid realizations. Direct sampling may produce ungrounded variation, whereas explicit diversity prompting often produces superficial differences. We study how to ground these requests without prematurely eliminating their legitimate ambiguity. We introduce a two-stage framework that first identifies diverse, feasible edit scopes and then realizes them as executable edit instructions. Across 300 open-ended image-editing tasks, our approach achieves 24.6% higher mean semantic dispersion and 17.1% higher mean groundedness than direct diversity prompting, while producing more valid and distinct realizations. These results show that structuring feasible edit directions before realization supports more useful grounded exploration.
[CV-164] ProMo3D: Probing Motion Cues in Frozen 3D Foundation Models NEURIPS2026 NEURIPS
链接: https://arxiv.org/abs/2610.33802
作者: Xiaoang Zhang,Mert Kiray,Yordanka Velikova,Manoj Biswanath,Benjamin Busam
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026 (Main Track). Poster: this https URL
Abstract:Feed-forward 3D foundation models have recently reshaped multi-view reconstruction, yet dynamic 4D understanding still often relies on large-scale motion supervision or hand-crafted, architecture-specific heuristics. We ask whether frozen 3D foundation model tokens already encode motion-relevant information that can be exposed by lightweight readout functions, rather than re-learning motion-specific features with high-capacity decoders. Through probing on three representative backbones, we find that static–dynamic separation is decodable from frozen token representations. Our rank-restricted and spectral analyses on \pi^3 X reveal that the recovered motion cues are concentrated in a low-dimensional subspace. Building on this, we propose Probing-based Motion Distillation (PMD), which converts pairwise motion-state compatibility scores against a fixed anchor token into dense motion masks. Trained on only 160 YouTube-VOS videos using a single 16 GB GPU, PMD transfers zero-shot to DAVIS, SegTrackv2, and FBMS-59 without target-benchmark fine-tuning. With multi-view pretrained geometric backbones, PMD outperforms training-free heuristics and approaches substantially heavier supervised and self-supervised methods, suggesting that motion decoding from frozen 3D foundation models can be lightweight and data-efficient.
[CV-165] Segmentation of Moving Objects During Camera Motion Using the OWL Function
链接: https://arxiv.org/abs/2610.33789
作者: Daniel Raviv,Juan D. Yepes
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:We present a principled, analytical framework for segmenting multiple independently moving rigid objects during single camera motion, built directly upon the OWL perception function. OWL encodes two fundamental visual motion cues, perceived looming L and perceived rotation \omega , into a single closed-form representation of relative 3D structure. A key consequence of this structure is that the unit direction of relative translation \hatt is analytically recoverable per image point from L and \omega alone. All points belonging to a non-rotating rigid body share the same \hatt , whereas points on different rigid bodies generally yield distinct values. This furnishes a natural segmentation criterion: clustering per-point \hatt estimates partitions the visible scene into independently moving rigid bodies. Building on the 2D complex form of OWL, we derive the framework in 3D using quaternions, for a single moving camera. The approach is parallel, camera rotation-invariant, and requires no training data, no depth information, and no semantic priors.
[CV-166] RoboReason : Boosting Robotics-Relevant VLM Reasoning Through Scalable Data Generation EMNLP2026
链接: https://arxiv.org/abs/2610.33778
作者: Sagnik Anupam,Utkarsh Lal,Aneesh Edara,Botong Zhang,Yao Tang,Jiatao Gu,Dinesh Jayaraman,Osbert Bastani
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: EMNLP 2026 Main
Abstract:Vision-Language Models (VLMs) have demonstrated impressive performance when integrated into robot manipulation pipelines, but real-world evaluation of VLM capabilities on downstream robotics tasks is both difficult and expensive to scale. We propose that spatial and temporal ordering tasks using images from stereo and wrist cameras attached to real-world robots are particularly well-suited for evaluating visual reasoning capabilities, as they assess skills directly applicable to robots. We evaluate VLM reasoning abilities on three questions: ordering shuffled frames conditioned on a language description of the task, ordering objects depthwise based on distance from the camera, and ordering objects from left-to-right in pixel coordinates. The complexity of these tasks can also be modulated by changing the size of the set to sort. Our experiments show that several state-of-the-art vision-language models struggle on ordering tasks. To fix this, we propose a scalable autonomous training data generation pipeline, and show that both supervised finetuning and reinforcement-learning on the generated data can significantly improve VLM reasoning performance both on spatiotemporal ordering tasks as well as downstream robotics and visual reasoning tasks.
[CV-167] C-DFF: Concept-Level Diagnostics for Mechanism Parity in Simulation-Based Validation of Autonomous Driving Perception ITSC
链接: https://arxiv.org/abs/2610.33760
作者: Danial Safaei,Siddartha Khastgir,Matthew Higgins,Jeroen Ploeg,Mohsen Alirezaei,Son Tong,Ludo Gielen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: 7 pages, 4 figures, 5 tables. Accepted and presented at the 2026 IEEE 29th International Conference on Intelligent Transportation Systems (ITSC), Naples, Italy, September 15-18, 2026
Abstract:Simulation is widely used in autonomous-driving validation, but matching task outputs between real and synthetic images does not guarantee that a system relies on the same visual evidence. This creates a validation gap for transfer, especially in safety-critical perception functions. We present Concept-Decisive Feature Fidelity (C-DFF), a concept-level mechanism-parity diagnostic for paired real-synthetic evaluation. Starting from decisive evidence extracted from a frozen system-under-test (SUT), C-DFF projects that evidence into a shared semantic concept space using off-the-shelf segmentation and compares real and synthetic concept-importance vectors. The result is not only a scalar mechanism-parity score, but also a concept-shift diagnostic that identifies which scene concepts become more or less decisive in simulation. We evaluate C-DFF on KITTI \leftrightarrow Virtual KITTI 2 with 2126 matched pairs across three frozen SUTs (steering regression, YOLOP drivable-area, YOLOP lane-line). Across these tasks, C-DFF is SUT-discriminative and yields interpretable concept-level summaries that support validation triage and simulator debugging. When reported alongside task-specific output-value context metrics, C-DFF shows that output agreement does not by itself establish mechanism parity. These results position C-DFF as a practical semantic diagnostic layer for validation and verification workflows in simulation-based AV testing.
[CV-168] Road Infrastructure Perception for Autonomous Driving: Taxonomy Representations Tasks and Challenges
链接: https://arxiv.org/abs/2610.33758
作者: Guoyang Zhao,Weiqing Qi,Kai Zhang,Zhihai Bi,Tianshuai Hu,Zeying Gong,Dezong Zhao,Jun Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Road infrastructure perception is a fundamental component of autonomous driving, as it recovers the structured road elements that define roadway geometry, delimit operational space, and encode traffic regulations. Although individual tasks such as lane detection, drivable-space estimation, traffic-sign recognition, and online map construction have been extensively studied, the field remains fragmented across task definitions, sensing assumptions, output representations, datasets, and evaluation protocols. This survey provides a function-oriented synthesis of road infrastructure perception. We first clarify its scope and organize existing studies into three complementary categories: geometric-structural, functional-spatial, and regulatory-semantic infrastructure. Based on this taxonomy, we review sensing modalities, scene representations, representative methods, benchmarks, and evaluation protocols, emphasizing the field’s evolution from image-space recognition toward metric, vectorized, topology-aware, occupancy-based, and rule-grounded scene modeling. We further examine how representation choices affect evaluation and downstream usability, revealing limitations of current benchmarks in relation reasoning, temporal consistency, robustness, and planning-oriented assessment. Finally, we discuss open challenges and future directions toward unified infrastructure representation, open-environment generalization, relation- and rule-aware reasoning, and downstream-oriented evaluation. By connecting tasks, representations, and evaluation under a common functional framework, this survey aims to establish road infrastructure perception as a coherent foundation for map-compatible and behavior-relevant autonomous-driving scene understanding.
[CV-169] Computer Vision Navigation for an Autonomous Vehicle Model Using Reinforcement Learning
链接: https://arxiv.org/abs/2610.33755
作者: Dinis Fernandes,Tiago Ribeiro,António Ribeiro,Carolina Lopes,Gil Lopes,A.Fernando Ribeiro
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 16 pages, 18 figures, Master Thesis
Abstract:This work aims to assess the capabilities of the reinforcement learning algorithm, Proximal Policy Optimization (PPO), in developing a policy capable of controlling the speed and steering angle of a vehicle model to navigate within simulated road lines. The control system’s input is a camera, and its outputs are the rotational speed of the rear wheels and the steering angle of the front wheels, controlling the vehicle’s speed and direction, respectively. The algorithm is trained and tested in a simulated environment, where it receives image frames from the vehicle’s camera, processes them through a feature extraction algorithm, and feeds them into the policy. For each frame, the policy outputs speed and direction commands for the vehicle. The policy is iteratively updated to enhance its performance. The reinforcement learning algorithm successfully generates a policy capable of navigating the vehicle in a simplified environment. Potential improvements are suggested to enhance the system’s robustness in more complex scenarios. Creating a simulation environment suitable for testing an advanced reinforcement learning algorithm is a complex task. This project offers a framework for other developers to test different approaches and algorithms. The tested algorithm, PPO, uses proximity intervals - an innovative approach in reinforcement learning. In some scenarios, the algorithm achieved better performance than predecessor algorithms, which are typically harder to implement and require tuning more hyperparameters. Comments: 16 pages, 18 figures, Master Thesis Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.33755 [cs.CV] (or arXiv:2610.33755v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.33755 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Fernando Ribeiro [view email] [v1] Sat, 24 Oct 2026 15:52:59 UTC (2,442 KB)
[CV-170] A SegFormer-based model for spatial Daylight Factor prediction with comprehensive obstruction encoding and regulatory validation
链接: https://arxiv.org/abs/2610.33737
作者: Stanislava Fedorova,Alejandro Pacheco Diéguez,Christoph Berkmiller,Alberto Floris,Libny Pacheco,Marie-Claude Dubois
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Conference: BuildSim Nordic 2026 15 pages, Appendices A, B, C 9 figures, 8 tables
Abstract:Daylight compliance assessment in residential buildings requires time-consuming raytracing simulations and specialist model preparation, limiting its integration into iterative architectural design. We present a machine learning-based methodology for predicting the spatial distribution of the Daylight Factor (DF) directly within Building Information Modelling (BIM) environments, delivering results in seconds without model export or manual curation. The approach addresses five gaps not jointly covered by any existing model: external obstruction encoding including balconies, own building height, variable material reflectances, non-orthogonal room geometries, and regulatory validation. Two models are developed: a default-materials model with fixed reflectances for preliminary design checks, and a custom-materials model with variable project-specific reflectances. Both are trained on about 20,000 synthetic Radiance-simulated rooms and validated on a held-out synthetic dataset and 102 rooms from 18 real building models. The default-materials model achieves \delta_1 = 83.4% and MAE = 0.11% DF on synthetic data, and on real buildings predicts BBR (Swedish) compliance area within 4.85 pp MAE (bias +2.31 pp) and TEK17 (Norwegian) mean room DF within 0.304% DF MAE (bias +0.287% DF). The custom-materials variant shows 1.7-2.4 \times higher error across all metrics. Both models concentrate accuracy in the BBR compliance-critical zone (DF 1%), achieving a sub-1% DF MAE of 0.105% DF and Compliance IoU of 0.883 on real buildings. All cases where compliance is overestimated are borderline, within 10.7 pp of the BBR threshold (median 3.2 pp) and 0.36% DF of the TEK17 threshold (median 0.10% DF). The default-materials model is recommended as a replacement for preliminary daylight simulations; the custom-materials variant requires further development before use in final compliance verification.
[CV-171] raffic-Context-Aware Driver Emotion Recognition with Expectation-Deviation-Driven Inference
链接: https://arxiv.org/abs/2610.33713
作者: Longhao Yan,Jinhao Liang,Kaidi Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Driver emotion provides an important indication of how drivers perceive and evaluate ongoing traffic situations, making reliable emotion recognition an important capability for effective human-vehicle interaction. Yet, most existing driver emotion recognition approaches rely primarily on in-cabin images and overlook how surrounding traffic interactions shape driver emotional responses. To address this limitation, we propose a traffic-context-aware driver emotion recognition (TADER) framework that integrates in-cabin signals with out-of-cabin traffic scenarios, modeling driver emotions as arising from expectation deviation represented by the discrepancy between predicted and observed vehicle trajectories. The proposed framework comprises two key components: (1) a vision-based base recognizer that extracts spatiotemporal representations from in-cabin facial images to infer driver emotions, and (2) a traffic-context-aware fusion module that quantifies expectation deviation from external traffic scenarios and integrates it into the base recognizer via attention-based embedding fusion. To overcome the scarcity of paired in-cabin and out-of-cabin data, we introduce a transfer learning strategy that transfers a trajectory prediction model trained on large-scale naturalistic datasets to smaller paired datasets. Extensive experiments demonstrate that TADER consistently outperforms state-of-the-art methods across both naturalistic and simulated driving environments by incorporating traffic context information into emotion recognition, resulting in improvements of 5% in accuracy on the AIDE and manD datasets.
[CV-172] E3tereo: Improving Stereo Matching at Depth Boundaries with Learned Geometric Edges CVPR
链接: https://arxiv.org/abs/2610.33704
作者: Qi Xiong,Qinghua Zhao,Yongqiang Xi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: submitted to cvpr
Abstract:Depth boundaries remain a major source of error in stereo matching. For IGEV on Scene Flow, geometric edge pixels account for only 6% of evaluated pixels but about half of the total absolute disparity error. Photometric edges can respond to texture changes without corresponding depth discontinuities, making them unreliable cues for boundary refinement. We propose E3Stereo, which learns geometric edges from disparity supervision and combines these boundary cues with the backbone disparity and image details to predict full-resolution disparity corrections. The same edge map also emphasizes boundary errors during training, while joint optimization allows the edge predictor to adapt to disparity refinement. On the Scene Flow dataset, E3Stereo reduces end-point error (EPE) at depth boundaries by 9.5%-14.3% across RAFT-Stereo, IGEV, and MonSter, while also reducing overall EPE. After fine-tuning, our IGEV-based model improves upon IGEV on all eleven reported metrics across ETH3D, KITTI 2012, and KITTI 2015. The source code will be made publicly available upon acceptance.
[CV-173] Seeing the Options: Typed Decisions over Visual Candidates with Reusable Option Memory
链接: https://arxiv.org/abs/2610.33702
作者: Fengxin Li
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Typed decision-making tasks in agents, such as UI element selection, product recommendation, and chart matching, require models to choose from a set of discrete candidates. While prior work demonstrates that text-only System One models achieve four key properties including candidate-order invariance, scalability, and cache reuse via separate request-candidate encoding, extending them to multimodal settings where candidates are visual remains underexplored. We introduce Iris, a framework that accommodates image, text, or hybrid candidates while preserving the core decision module. We prove that order invariance and exact cache reuse hold for any per-candidate encoder under Iris, and derive the closed-form latency reduction. We construct a visual-candidate benchmark from Multimodal-Mind2Web (K=16, deduplicated, order-shuffled, three generalization splits). Independent encoding (Iris) achieves 78.4 / 72.1 / 74.8 percent accuracy with zero-flip rate under candidate reordering. Joint encoding reaches 90.6 / 89.5 / 91.6 percent but suffers an 8 to 10 percent answer-flip rate when shuffled. To bridge this gap, we propose a design that encodes each candidate conditioned on the request state without inter-candidate cross-attention; it preserves strict order invariance, enables forked cache sharing, and attains 91.7 / 86.5 / 90.0 percent accuracy at one-third of OneJev latency with calibrated null probabilities (AUROC 0.93 to 0.97). Reusing image candidates across requests cuts latency by 37x at K=256 and 191x at K=2048. On MetaEncoder benchmarks, Iris-4B achieves 100 percent on JevBench original and easy, outperforming 30B MetaEncoder and 27B OpenJev. Finally, we resolve a systems bottleneck on Ampere GPUs: reformulating Qwen3.5 patch-embedding Conv3d into its exact matrix-product form accelerates processing by 350x. Code, data, and checkpoints: this https URL.
[CV-174] Scalable Unrolled Chains for MRI Reconstruction WACV2027
链接: https://arxiv.org/abs/2610.33695
作者: Emiljo Mëhillaj,Ender Konukoglu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 23 pages (including supplementary material), 14 figures. Accepted at WACV 2027
Abstract:Unrolled variational networks achieve state-of-the-art performance for accelerated MRI reconstruction, but standard end-to-end training becomes prohibitively memory intensive as the number and width of cascades increase. We propose a memory-efficient training framework that updates a randomly sampled cascade while retaining supervision from the final reconstruction loss. The error signal is propagated from the end of the chain to the sampled cascade using an activation-light backward pass that avoids storing full chain activations, thereby decoupling training memory from chain length. A time-conditioned network shared across cascades further keeps the parameter count independent of chain length. The backward pass is approximate only through frozen normalization statistics, and a trained exact gradient control shows the approximation leaves reconstruction quality unchanged. The framework remains trainable at chain lengths and cascade widths where standard end-to-end backpropagation exhausts GPU memory. On fastMRI single-coil knee and multi-coil brain reconstruction, our method consistently outperforms other local training approaches. On high-SNR knee slices, our method surpasses E2E-VarNet at 8\times and matches it at 4\times , while on the complete brain validation set it surpasses E2E-VarNet at 8\times and achieves comparable performance at 4\times , with fewer parameters, lower peak memory, and faster per-iteration training. Code is available at this https URL.
[CV-175] Rescuing Your Pruning: Rethinking Visual Token Pruning and Post-Training Recovery in MLLM s
链接: https://arxiv.org/abs/2610.33688
作者: Peiran Yin,Hao Wu,Zihao Yang,Haozhe Hu,Xudong Wang,Xiaoyu Shen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual token pruning substantially reduces the inference cost of multimodal large language models (MLLMs), yet existing methods are mostly training-free and plug-and-play. They treat the resulting performance drop as an inevitable cost of compression and pursue increasingly complex pruning rules. We rethink this view through two questions: how much of the loss can post-training rescue, and, building on this recovery, how can joint compression push pruning to its limit? We curate a 282K-sample post-training mixture covering general VQA, reasoning, OCR, and grounding, and post-train 36 configurations that drop or merge tokens under a unified protocol. Post-training rescues most of the loss, even for OCR and grounding, and its adaptation gain in closing the gap between the pruned and the dense model grows as the budget tightens. For joint compression, we find that the vision encoder, not the LLM, dominates prefill time in native-resolution MLLMs, so where tokens are removed matters more than how many: keeping 40% of tokens at the encoder runs faster than keeping 10% at the projector. Our best configuration prunes the encoder lightly and the projector aggressively, leaves the LLM intact, and keeps general understanding and reasoning at the dense level. It accelerates the LLM stage by 3.49x but the end-to-end time-to-first-token by only 1.54x, showing that LLM-stage speedups overstate end-to-end gains. We hope these findings offer insights for future research on joint visual token compression. Our data, protocol, and code will be released.
[CV-176] EXE: Egocentric Cross-Embodiment Pretraining via Factorized Latent Actions for VLA Models
链接: https://arxiv.org/abs/2610.33685
作者: Qingyuan Cai,Bo Zhang,Yongzhen Huang,Jianyu Chen
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-Language-Action (VLA) models are bottlenecked by the scarcity of robot data. Egocentric human manipulation videos offer an abundant, diverse, and semantically-rich alternative that inherently captures manipulation intent. However, the embodiment gap makes it difficult to extract action-relevant representations from these videos. From the perspective of action execution, human hands and robots share analogous manipulation primitives—reaching, grasping, placing—yet exhibit distinct kinematics tied to their embodiments. This might motivate our key approach: egocentric video pretraining can yield shared action representations capturing transferable manipulation skills, while human-specific motion patterns should be disentangled to avoid interfering with robot policy learning. To this end, we propose EXE, an Egocentric Cross(X)-Embodiment pretraining framework that learns a factorized latent action representation from egocentric human videos, separating shared manipulation priors from embodiment-specific signals. By leveraging the shared action space, the robot policy acquires generalizable manipulation semantics from diverse human demonstrations without being biased by human-specific kinematics. This shared knowledge transfers across embodiments with different kinematic structures, enabling the same pretraining to benefit both parallel grippers and dexterous hands. EXE yields consistent gains over baseline without pretraining on RoboTwin 2.0, RoboCasa GR1, and real-world manipulation tasks.
[CV-177] racking Is Not Sensing: Artifact-Aware Thermal Video Respiration Estimation in Head-Free Marmosets WACV2027
链接: https://arxiv.org/abs/2610.33661
作者: Penghao Wang,Sen Yang,Fenghua Xie,He Wang,Xindong Song
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV 2027
Abstract:Continuous physiological monitoring is essential for linking behavior, internal state, and neural activity in neuroscience and translational research. In common marmosets, a key nonhuman primate model for social neuroscience, respiratory rate provides a basic readout of arousal, stress, and physiological regulation. Contact sensors can provide accurate measurements, but they add preparation burden and may interfere with head-free behavioral protocols. Thermal video offers a noncontact alternative by capturing breathing-related temperature changes around the nose and mouth. However, in head-free marmosets, these cues are weak, localized, and easily corrupted by head motion, pose changes, and fur. Here we show that this setting exposes a broader challenge in video-based physiological sensing: visually reliable tracking does not necessarily imply physiologically valid measurement. Using synchronized thermal video and contact respiration recordings, we show that marmoset respiration is recoverable from thermal video but strongly ROI-sensitive. Under head-free motion, visually plausible nasal tracking can still produce traces inconsistent with contact respiration, demonstrating that tracking is not sensing. We therefore estimate respiratory rate from learned temporal signals extracted from thermal clips and introduce Selective Temporal Attenuation Regularization (STAR). STAR selectively attenuates high-motion thermal discontinuities through local temporal imputation during training, regularizing models toward respiration-consistent dynamics while leaving inference unchanged on raw video. Across multiple deep-learning backbones, STAR improves head-free estimation, reaching 2.23 bpm MAE compared with 20.76 bpm for tracked ROI analysis.
[CV-178] FreeView: Robust 3D Reconstruction from Uncurated Views
链接: https://arxiv.org/abs/2610.33656
作者: Junho Kim,Hyunyoung Choi,Seongwon Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reconstructing a coherent 3D scene from uncurated image collections remains challenging due to the presence of irrelevant or non-overlapping views. Recent work has shown that feed-forward reconstruction models contain internal view-consistency signals that support distractor rejection. However, reference-conditioned filtering relies on a curated clean reference, which is not guaranteed in practical image collections. To address this limitation, we introduce FreeView, a training-free framework that jointly selects a filtering reference and rejects distractor views from an uncurated collection. FreeView leverages collection-level consensus to identify a representative view that serves as a reliable reference for filtering. It then adapts view rejection to each input collection by combining complementary visual and geometric evidence. This approach preserves useful views under substantial appearance and viewpoint variations while rejecting distractors. FreeView integrates with both VGGT and pi^3, removing the need for a manually curated reference or manual calibration of view rejection. Extensive experiments across multiple datasets demonstrate high clean-view retention and distractor rejection rates, leading to reliable 3D reconstruction from uncurated views.
[CV-179] LENS: Likelihood Evaluation and Neighbor Smoothing for Attribute-Based Person Retrieval on Unseen Cameras WACV2027
链接: https://arxiv.org/abs/2610.33641
作者: Nguyen Dang Duy Tan,Vo Thanh Nguyen,Linh T. T. Tran,Huynh Nguyen Phat
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, 6 tables. Submitted to the 7th Real-World Surveillance Workshop (RWS) at WACV 2027
Abstract:Attribute-based person retrieval ranks surveillance images from a binary query of semantic attributes, but mean Average Attribute Degree of Match (mADM) drops on unseen cameras as predicted probabilities shift. Common pipelines rank gallery images by their L1 distance to the query, which penalizes a confident single-attribute mismatch no more than small uncertainties across many attributes, while post-hoc recalibration fitted on source validation data fails to transfer because the per-attribute maps fitted on source data do not match those of new cameras. We propose LENS (Likelihood Evaluation and Neighbor Smoothing), combining fine-tuned DINOv3 and CLIP models with a distilled student. LENS regularizes mutually exclusive attribute groups with group cross-entropy in training, scores gallery images by the log-likelihood of the exact query, and smooths gallery logits over nearest neighbors using DINOv3 appearance features from the same forward pass. On held-out datasets, all three changes raise mADM: the likelihood score improves all six retrieval metrics for the main ensemble and mADM for all 48 development models, and training-free kNN smoothing improves all 46 tested development models. On the hidden test set of the UPAR 2027 benchmark, LENS reaches 68.17 mADM.
[CV-180] UniQuery: Unifying Vision Models with a Joint Query Head for Cross-Domain Pedestrian Attribute Recognition WACV2027
链接: https://arxiv.org/abs/2610.33639
作者: Nguyen Dang Duy Tan,Linh T. T. Tran,Huynh Nguyen Phat,Vo Thanh Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, 5 tables. Submitted to the 7th Real-World Surveillance Workshop (RWS) at WACV 2027
Abstract:Pedestrian Attribute Recognition (PAR) predicts semantic attributes of individuals, such as gender, age, clothing and accessories, and its descriptions support attribute-based person retrieval and person re-identification. However, accurately recognizing fine-grained attributes across domains remains challenging due to occlusion, appearance variations, and shifts in image distributions. Ensembles of independently trained foundation encoders combine their logits with fixed fusion weights, so every attribute uses the same mix and the encoders are never trained together. We propose UniQuery, which instead jointly fine-tunes DINOv3 and CLIP for attribute-specific fusion. It combines two query heads, one for each encoder, with a joint query head whose attribute queries attend to tokens from both encoders, and the prediction is the average of the logits of the three query heads. We evaluate UniQuery on the benchmark of the UPAR Challenge 2027 at the WACV 2027 Real-World Surveillance workshop, where models are trained on PA100k, PETA and Market-1501 with a shared set of attributes. In leave-one-dataset-out experiments, UniQuery outperforms the independent ensemble in Challenge_Avg on both held-out datasets. Ablations show that learning fusion over independently trained encoders does not improve upon logit averaging in the mean over the two held-out datasets, while freezing the encoders lowers the Challenge_Avg of UniQuery. On the hidden test set of the benchmark, UniQuery achieves a Challenge_Avg of 0.8001.
[CV-181] DROA-CLIPSeg: Prompt-Guided Thin Crack Segmentation in Low-Light Conditions BMVC2026
链接: https://arxiv.org/abs/2610.33635
作者: Vinita Shukla,Partibha Jangra,Vinayak Pathak,Praful Hambarde,Amit Shukla
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at the 37th British Machine Vision Conference (BMVC 2026)
Abstract:Pavement crack segmentation under low light is degraded by reduced contrast, shadows, headlight glare, joints, stains, and wet seams. Existing crack segmentation methods mostly assume good illumination. Low-light enhancement methods improve visual quality, but they do not explicitly preserve crack topology. To address these limitations, we formulate low-light crack segmentation as a degradation-aware, prompt-grounded structural defect segmentation problem and propose DROA-CLIPSeg, a parameter-efficient framework that adapts a frozen CLIPSeg vision language backbone using two lightweight modules. QKV Low-Rank Adaptation is injected into the Q/K/V projections of the vision and text encoders to improve crack-specific image text alignment without backbone fine-tuning. The proposed Domain Residual Orthogonal Adapter (DROA) decomposes intermediate features into a dominant crack-domain branch, which captures recurring low-light crack patterns, and an orthogonal residual branch, which preserves faint, thin, and fragmented evidence usually suppressed by noise, shadow, or over-smoothing. We also introduce CrackLingo-9K and CrackLingo-12K, text augmented versions of Crack9k and Crack12k with crack-aware prompts. Training only approximately 1.93 M parameters, corresponding to 1.3% of the full model, DROA-CLIPSeg outperforms 9 zero-shot and 4 fine-tuned vision language baselines on both splits, generalises zero-shot to MVTec AD across textures and defect morphologies, and is supported by extensive ablations. Our source code is publicly available at this https URL
[CV-182] ONOT/TONO-Print: A Paired Benchmark for Identity-Document Portrait Quality under Document Production Shift
链接: https://arxiv.org/abs/2610.33627
作者: Alexander Vinogradov
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 2 pages, 2 figures, 2 tables
Abstract:Identity-document portraits are modified during document production by printing, engraving, and integration into the document template. These transformations may change image texture, contrast, colour, and other visual properties without necessarily changing the underlying biometric or presentation information. In this work, we introduce ONOT/TONO-Print, a small paired benchmark for evaluating face-image quality models under identity-document production shift. The benchmark is based on ONOT and TONO portraits and contains synthetic variants representing five common production techniques: toner, inkjet, dye sublimation, laser engraving, and thermal transfer. A smaller physical subset contains portraits printed and scanned using inkjet and toner devices. The benchmark is used to compare the behaviour of biometric and ICAO-related face-quality components before and after production. Results show that production can introduce substantial domain shift for some quality estimators, while other models remain comparatively stable.
[CV-183] Allocation Before the Clock: Graph Set and Game Theory of Professional League of Legends Objective Setups
链接: https://arxiv.org/abs/2610.33626
作者: Zheng Gao,Zhenyu Liu,Yifan Yang,Xiaoyu Li
类目: Computer Vision and Pattern Recognition (cs.CV); Applications (stat.AP)
备注: 32 pages, 28 figures, 22 tables
Abstract:Before every neutral objective in professional League of Legends, each team decides which of its five players walk to the pit, which hold a lane, and when to commit. The leading analysis of such setups, TAGNet for Counter-Strike 2, clusters graph-network embeddings of player positions and reads cluster win rates as tactical quality. We treat a setup as an allocation on a walled map: formations are set partitions on the lattice \Pi_5 , control is a geodesic Voronoi split of a navigation graph, and commitment is an allocation game solved as a quantal response equilibrium priced in win probability. On all 164 games of MSI and Worlds 2025 (2,994 team-setups, minimap tracks validated against Riot positions), trailing teams leave the pit empty far more often than leading teams, by +34.8 percentage points at the first baron and +30.4 percentage points at Atakhan. Our model, HASSE-QRE, finds six setup patterns that whole-model series-bootstrap refits reproduce (ARI 0.819, comparable to 0.813 for TAGNet’s K -Means step alone) and indicates that teams price fights, travel and committed players, not the objective itself (weight 0.018; fixing it to one costs 59.9 log-likelihood points). The equilibrium reproduces a quarter of the commitment gap, almost all of it through state variables rather than its strategic channels. Weighting a team-level contrast by exposure moves it across p=0.05 (0.046 against 0.052). We release the tracks, setups and code.
[CV-184] Satellite-derived bathymetry of Arabian Gulf ports under leakage-free validation: a multi-site benchmark of empirical machine-learning and deep-learning models against 285000 multibeam soundings with an outlook for UAE waters
链接: https://arxiv.org/abs/2610.33609
作者: Muhammad Saeed,Fatema Al Hajri,Mohamed H. Alhosani,Dmitry Mikhaylov,Yasser F. Al Wahedi
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Atmospheric and Oceanic Physics (physics.ao-ph); Geophysics (physics.geo-ph)
备注: 54 pages, 28 figures, 25 tables
Abstract:Satellite-derived bathymetry (SDB) is often proposed as a low-cost complement to multibeam surveys of ports, but its accuracy is seldom measured on an area the model has never seen. We test SDB in the turbid, shallow and tidal waters of the Abu Dhabi coast against 285,619 multibeam soundings from twelve surveys at six port areas, reduced to Lowest Astronomical Tide. The Stumpf ratio, the Lyzenga transform, a monotonic gradient-boosted tree model (LightGBM) and a U-Net ensemble were trained on 72 Sentinel-2 images and tested in transfer to an unseen site and with local control (about 20% of a site’s soundings, spatial blocks beyond a 225 m buffer). The validation design changed the result more than the model: the same LightGBM model gave 0.57 m RMSE under a random pixel split, 1.30 m under spatial blocks and 1.66 m at an unseen site. In transfer the Stumpf ratio (1.45 m) was not beaten by LightGBM, the U-Net or Lyzenga. With local control a hybrid of the tree output and the band ratio reached 1.11 m over 0-11 m, 67% of pixels within the IHO S-44 Order 2 tolerance and a conformal 95% interval with 95% observed coverage; in clear water it reached 0.29-0.53 m. The seabed could not be read beyond 11 m, and summer images were five to six times more turbid than winter ones. ICESat-2 ATL24 agreed with the surveys to 0.25 m after a depth-dependent offset was removed. A 60 m reconnaissance mosaic of 17,200 km2 of the UAE coast, calibrated cell by cell on ICESat-2 without any vessel survey, reached 1.42 m RMSE on held-out blocks over half of the product. Sentinel-2 SDB with about a hundred control depths per image is a usable reconnaissance and change-monitoring tool in these waters to about 10 m; it does not meet S-44 Order 1 without local control, and accuracies from random splits should not be taken as operational accuracy.
[CV-185] Mix Forcing: Causal Pre-Training via Mixture Attention for Interactive World Models
链接: https://arxiv.org/abs/2610.33589
作者: Zelin Gao,Hao Ouyang,Jiapeng Zhu,Qiuyu Wang,Jingye Chen,Zichen Liu,Xing Zhu,Yujun Shen,Yinghao Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:While interactive video world models require autoregressive generation, directly adapting pretrained bidirectional video diffusion models into autoregressive ones with teacher forcing can disrupt their visual priors and degrade autoregressive generation quality. In this paper, we introduce Mix Forcing, a causal pre-training paradigm for interactive video world models, implemented with the Mixture-of-Bidirectional-and-Autoregressive (MoBA) attention. Mix Forcing jointly trains a video diffusion model with both teacher forcing and bidirectional training, yielding a strong causal initialization for few-step post-training. In self-attention, Mix Forcing leverages teacher forcing to learn causal next-chunk prediction for autoregressive rollout, while using bidirectional attention to preserve pretrained visual priors. To support online text interaction, Mix Forcing further adopts a chunk-wise cross-attention design that conditions each video chunk on a shared environment prompt and its own local prompt, without exposing future instructions to the current chunk. Following causal pre-training on a 14B image-to-video model, we apply consistency and distribution matching distillation to obtain a four-step autoregressive generator. Experiments on our internal benchmark and public benchmarks, including WBench and SANA-WM-Bench, show that our model achieves state-of-the-art visual quality and timely event following with precise camera control. Notably, our model supports high-fidelity long-horizon generation without quality degradation even over one-hour rollouts. Project page is here: this https URL
[CV-186] UltraFlash: Accelerating Megapixel Visual Synthesis NEURIPS2026
链接: https://arxiv.org/abs/2610.33581
作者: Phuc Lai,Anh Nguyen,Phong Nguyen,Anh Tran
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NeurIPS 2026
Abstract:Megapixel visual synthesis with latent diffusion models requires operating far beyond the resolutions at which these models are typically trained. Cascade refinement is a leading strategy for this setting: it starts from a base image at a native resolution and progressively upsamples it to the target resolution through multiple refinement stages. Each stage typically performs a pixel-space resolution transition followed by latent-space denoising. While prior acceleration efforts mainly focus on reducing denoising cost, we identify a complementary and underexplored bottleneck: the repeated movement of representations through the VAE. We call these operations pixel-space transitions. In cascade pipelines, intermediate latents are often decoded to RGB, resized, and re-encoded before refinement, while final latents are decoded patch by patch with dense overlap to suppress boundary artifacts. We propose UltraFlash, a transition-efficient approach that accelerates these VAE-heavy operations. UltraFlash has two components. Latent Hyperloop replaces intermediate RGB round trips with a learned latent-space shortcut that approximates the decode-resize-encode transformation without repeated VAE decoding and encoding. Sparse Patch-based Decoding accelerates final reconstruction by addressing patch seams caused by inconsistent GroupNorm statistics and reusing cached statistics across patches, enabling sparse VAE decoding with much smaller overlap. Integrated into a state-of-the-art cascade baseline, UltraFlash reduces 4K generation latency by approximately 2x while improving image quality. These results show that optimizing pixel-space transitions can improve both efficiency and fidelity, offering an effective route toward fast, high-quality megapixel visual synthesis.
[CV-187] ZoomSEM: accelerating large-scale brain connectomics with intelligent SEM super-resolution
链接: https://arxiv.org/abs/2610.33580
作者: Jiateng Shou,Zhiwei Xiong,Ruikang Xu,Zhicheng Wang,Peiyao Shi,Te Shi,Haiqun Jin,Chunying Yin,Jun Guo,Manfred Auer,Hanchuan Peng,Ruobing Zhang,Feng Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Serial electron microscopy (EM) is essential for large-volume biological imaging and connectomics, yet its throughput remains a major bottleneck, with cubic-millimeter-scale brain imaging requiring months or years. Existing deep learning-based EM super-resolution methods are limited to small upscaling factors or rely on synthetic data, offering modest acceleration and limited practical impact. We introduce ZoomSEM, a deep learning-based super-resolution method to dramatically accelerate SEM imaging. Trained on real low- and high-resolution scanning electron microscopy (SEM) pairs, ZoomSEM enables reliable recovery of 4 \times , 8 \times , and up to 16 \times higher-resolution images from fast low-resolution scans while preserving fine neuronal structures critical for connectomic analysis, achieving up to 30-fold effective acceleration. Extensive experiments across datasets of diverse biological samples demonstrate robust generalization under realistic low-resolution acquisition. This advance enables high-resolution, large-volume serial EM imaging at unprecedented speed, poised to transform connectomics and related fields.
[CV-188] CADonatello: Towards Real-World Multi-Part 3D CAD Reconstruction
链接: https://arxiv.org/abs/2610.33575
作者: Ahmet Serdar Karadeniz,Dimitrios Mallis,Parsa Akbari Chianeh,Kseniya Cherenkova,Anis Kacem,Djamila Aouada
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing methods for CAD reconstruction from 3D scans achieve strong performance on standard benchmarks, yet their effectiveness often degrades in real-world scenarios where objects consist of multiple parts and scan artifacts such as noise. A key reason is that current CAD reconstruction datasets predominantly contain single-part models and clean geometric inputs, limiting the ability of learned models to generalize to realistic settings. In this work, we introduce CADonatello, a part-aware framework for reconstructing parametric CAD models from real-world 3D scans. Our approach first decomposes the input geometry into structurally meaningful components using a foundation-scale 3D part segmentation model, and then reconstructs each component independently through a cross-section-driven pipeline that performs sketch parameterization followed by an extrusion. To improve the robustness of sketch parameterization to scan-derived cross-sections, we further introduce CC3D Cross-Sections, a dataset of paired rasterized cross-section images from realistic scans and their corresponding parametric sketches derived from aligned CAD models. This dataset enables effective domain adaptation from clean CAD sketch images to scan-derived cross-section images. Extensive experiments demonstrate that the proposed approach improves generalization to noisy scans and enables reliable reconstruction of complex multi-part objects across multiple real-world CAD reverse engineering benchmarks.
[CV-189] Distilling Dense Trajectory Teachers For Large-Motion Video World Models
链接: https://arxiv.org/abs/2610.33566
作者: Shuofeng Sun,Linqing Zhao,Haibin Yan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video world models are expected to generate videos that accurately follow specified camera trajectories while maintaining consistent 3D structures. However, existing models are predominantly trained on densely sampled videos with small inter-frame camera motions. When applied to sparse, large step trajectories, they often produce frozen frames, inaccurate camera motion, and severe geometric drift. We attribute this degradation to a mismatch between the gradual viewpoint transitions seen during training and the large viewpoint changes required at inference time. To address this problem, we propose Dense Trajectory Teacher Distillation (DTTD), a post-training framework that transfers a model’s generation capability from dense trajectories to large camera motions. Given a sparse trajectory, DTTD interpolates additional camera poses to construct a dense trajectory covering the same path. The frozen base model then generates reliable teacher videos along this familiar dense trajectory, and pose aligned teacher observations are used to evaluate candidate videos generated under the original sparse trajectory. In this way, the model serves as its own generation teacher, requiring neither additional video data nor a stronger external video generator. We further combine dense teacher supervision with geometry aware rewards that assess cross frame consistency and camera trajectory following, and use the resulting preferences for offline preference optimization. To stably expand the model’s motion range, we introduce a cumulative curriculum that progressively incorporates larger motion steps while retaining simpler trajectories to reduce capability forgetting.
[CV-190] What Remains OOD? OODConceptBench for Open-World Concept Recognition in Vision-Language Models
链接: https://arxiv.org/abs/2610.33564
作者: Zhiran Luo,Zhiyong Yang,Zitai Wang,Sicong Li,Qianqian Xu,Qingming Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:In the era of large-scale pretrained models, zero-shot generalization has increasingly blurred the boundary between known and unknown concepts. However, current evaluations do not directly reveal which open-world concepts different vision-language models can recognize. To study this problem, we introduce OODConceptBench, a hierarchical benchmark for open-world concept recognition on image-concept pairs. The full benchmark set (L1) contains 30,752 images, and the harder subset (L2) contains 14,285 images. The benchmark covers 560 concept phrases across seven open-world categories. We evaluate 37 vision-language models across release year, model family, and variant. We obtain three main findings. First, target-concept recognition is lower on L2 than on L1 across all 37 evaluated models. Second, later-released models do not necessarily achieve stronger open-world concept recognition. Third, an increase in overall recognition does not imply consistent improvement across all benchmark categories. Together, these results show that OOD remains a challenge even in the era of large-scale vision-language pretraining.
[CV-191] GeoLift: Grounding Video Spatial Reasoning in Explicit 3D Maps
链接: https://arxiv.org/abs/2610.33558
作者: Xinglin Li,TingTing Long,Jingzhi Zhou,Jiajing Chen,Jian Yang,Jin Xie
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 3 figures, 4 tables. Xinglin Li and TingTing Long contributed equally. Jin Xie is the corresponding author
Abstract:Video spatial reasoning requires multimodal large language models (MLLMs) to understand how objects are arranged and related across viewpoints. While these models can recognize objects in individual frames, integrating these observations to solve complex spatial tasks remains challenging. We examine route planning as a representative task and observe that existing models become substantially less accurate as the number of required turns increases, with several models achieving accuracy close to random guessing on routes requiring three or more turns. This finding suggests that recognizing landmarks alone is insufficient: models must also connect observations across viewpoints and track how landmarks are arranged in a shared space. Motivated by this observation, we introduce GeoLift, a training-free framework that uses feed-forward 3D reconstruction to map objects detected in individual frames into a shared metric 3D scene. GeoLift combines a frozen MLLM for 2D grounding with pretrained geometry models for scene reconstruction. It then uses shared geometric regions, called superpoints, to associate observations of the same object across views without a learned matcher. The reconstructed 3D map enables explicit computation of spatial relations and multistep routes, and provides geometric evidence for estimating distances and sizes. With all model weights frozen, GeoLift improves Qwen3-VL-4B’s average score on VSI-Bench from 55.4% to 64.1%, including an increase in route-planning accuracy from 30.9% to 53.6%, and achieves 59.8% accuracy on MindCube. These results show that connecting frame-level object grounding through a shared 3D map can improve complex spatial reasoning without task-specific training.
[CV-192] End-to-end prostate lesion segmentation and detection from undersampled diffusion-weighted MRI
链接: https://arxiv.org/abs/2610.33554
作者: Lukas T. Rotkopf,Marco Schlimbach,Kevin Kröninger,Stefan O. Schoenberg,Heinz-Peter Schlemmer,Jens Kleesiek,Moritz Rempe
类目: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
备注:
Abstract:Automated prostate lesion detection from diffusion-weighted MR imaging (DWI) conventionally relies on reconstructed images and does not fully use all acquisition information. We present an end-to-end network that uses complex multicoil DWI data and coil sensitivity information for prostate lesion segmentation. We included 269 fastMRI prostate examinations with additional voxel-level lesion annotations. Of these, 64 examinations contained PIRADS 4 or 5 lesions. The proposed network, VarNet-Seg, consists of a k-space variational network (VarNet) with six variational cascades and an image-space 3D segmentation U-Net. Comparator models used the same segmentation architecture with fully coil-combined images (CC-Seg), fixed root-sum-of-squares reconstruction (RSS-Seg), or fixed SENSE reconstruction (SENSE-Seg). We performed ablations with end-to-end training and data consistency (DC) enabled or disabled. We evaluated lesion segmentation using Dice and patient-level detection using AUROC in five-fold cross-validation with full sampling, simulated undersampling, added noise, or both. VarNet-Seg achieved the highest Dice at every undersampling level. At eight-fold acceleration, Dice was 0.22 for VarNet-Seg, 0.14 for SENSE-Seg and 0.16 for VarNet-Seg (no DC). The paired difference from SENSE-Seg was 0.08 (95% CI, 0.04-0.11). With full sampling, VarNet-Seg and RSS-Seg performed similarly (0.40 vs 0.39). Patient-level AUROC decreased from 0.75 with full sampling to 0.63 at eight-fold acceleration; corresponding RSS-Seg AUROCs were 0.76 and 0.64. The proposed multicoil network improved prostate lesion segmentation under simulated undersampling compared to conventional models. Matched ablations supported contributions from both end-to-end training and data consistency. Prospective evaluation is needed to establish whether this approach can be used clinically in accelerated prostate DWI. Subjects: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph) Cite as: arXiv:2610.33554 [cs.CV] (or arXiv:2610.33554v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.33554 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Lukas Thomas Rotkopf [view email] [v1] Sat, 24 Oct 2026 09:13:57 UTC (1,463 KB)
[CV-193] Cross-Scale Sparse Attention for Referring Segmentation in Remote Sensing
链接: https://arxiv.org/abs/2610.33551
作者: Yuxi Li,Lu Si,Xia Liao,Yidong Shen,Zheng Wei,Jun Zhang,Xin Yuan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:High-resolution remote sensing (RS) imagery provides fine spatial detail over large areas, making it essential for fine-grained scene understanding. Referring Remote Sensing Image Segmentation (RRSIS) aims to segment regions described by natural language, yet many existing methods rely on sequential alignment–fusion pipelines with limited joint modeling of cross-scale visual hierarchies and linguistic dependencies. Moreover, dense attention mechanisms struggle with the sparsity and wide spatial dispersion of RS targets, while current datasets offer limited structured evaluation of multi-target references, multi-hop relational reasoning, and multi-granularity descriptions, limiting both model development and evaluation. To address these challenges, we propose a Cross-Scale Sparse Attention Model (CSSAM), which aligns multiscale visual features with complex referring expressions through two key modules. The Cross-scale Modality-aligned Fusion Module (CMFM) performs unified cross-scale cross-modal fusion, while the Dynamic Multimodal Sparse Hybrid Attention (DMSHA) selectively attends to informative visual–textual tokens, reducing dense visual interactions. Furthermore, we introduce M ^3 RRSIS-D, a RRSIS dataset explicitly designed around three core reasoning dimensions: multi-target (M ^1 ), reference-dependent multi-hop (M ^2 ), and multi-granularity (M ^3 ) expressions. Extensive experiments on M ^3 RRSIS-D and two public RRSIS benchmarks demonstrate that CSSAM achieves competitive performance across accuracy, fine-grained reasoning, and high-IoU localization metrics, validating its effectiveness for complex semantic reasoning in realistic RS scenarios. Code, dataset, and model weights will be released upon acceptance.
[CV-194] Patch-Derived or Register Tokens? Comparing Variable-Length Image Tokenizers ECCV2026
链接: https://arxiv.org/abs/2610.33540
作者: Philipp Schenkel,Tobias Schürmann,Omer Sahin Tas
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: LoViF Workshop @ ECCV 2026 (Oral), final version
Abstract:Image tokenizers convert pixels into compact latent sequences for efficient generative modeling. Most approaches arrange latents on a downsampled two-dimensional grid, tying sequence length to image resolution. We compare two variable-length strategies: computationally efficient patch-derived tokens, which inherit explicit spatial organization, and more demanding register tokens, which have no predefined spatial layout. Our reconstruction results reveal that patch-derived tokens perform substantially better under extreme compression, whereas register tokens achieve slightly better reconstruction quality at intermediate budgets. At high token budgets, both approaches achieve similar reconstruction quality. For downstream generative tasks, register tokens outperform patch-derived tokens. Adding representation alignment with external pretrained visual encoders substantially improves the generative performance of patch-derived tokens, allowing them to match or outperform register tokens depending on the token budget. We further find that both approaches learn spatially localized representations. Patch-derived tokens largely retain their explicitly imposed spatial organization, while register tokens develop an implicit two-dimensional spatial structure despite having no predefined layout. Representation alignment reduces this spatial organizational constraint, enabling individual tokens to exert a stronger non-local influence.
[CV-195] Optimizing Physical Neural Networks through Device Variable Selection
链接: https://arxiv.org/abs/2610.33522
作者: Sjoerd van den Belt,Manuel Escudero,Mohamadreza Zolfagharinejad,Wilfred G. van der Wiel,Nikolaos Alachiotis
类目: Emerging Technologies (cs.ET); Hardware Architecture (cs.AR); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Physical neural networks (PNNs) leverage device physics in electronics, optics, and related domains to enable energy-efficient neural networks. Broken-isomorphism PNNs replace linear operations with multivariate physical transformations, modifying how input signals are mapped to device variables. We exploit network architectures emerging from these mappings by treating the routing between signals and input variables of physical devices as trainable parameters. This approach expands device-level tunability of the physical transformations while, at the system level, demonstrating improved expressivity of PNNs with negligible additional inference overhead. This discrete routing is optimized by training a categorical distribution using Gumbel-Softmax reparametrization. We evaluate this approach using experimentally calibrated differentiable models of multivariate electronic RNPUs for MNIST and CIFAR-10 classification. Using variable selection, we improve the MNIST classification accuracy for a small CNN model from 94.5% to 96.9% and the accuracy of CIFAR-10 from 62.8% to 68.0%. Our method uniquely leverages the induced biases of broken-isomorphism PNN architectures, paving the way towards more efficient physics-based neural network inference.
[CV-196] Promise and Limits of Cross-Architecture Knowledge Distillation for Lightweight Medical Vision Transformers (LiteMed-ViT) ICICS
链接: https://arxiv.org/abs/2610.33520
作者: Shuaibu Shehu(SR University),Elijah M. Flomo(SR University),Mohammad Safiya(GITAM University),Anurodh Kumar(SR University),Subodh Kumar(Dr. KNS Memorial Institute of Medical Sciences)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 6 pages, including references. Accepted at the International Conference on Intelligent Computing and Secure Communication Systems (ICICSCS 2026)
Abstract:Running deep models for medical image analysis at the point of care is challenging with standard architectures: convolutional neural networks (CNNs) are often too heavy for edge devices, while Vision Transformers (ViTs) require massive training datasets to compensate for missing inductive biases. This study investigates how far the dual-token distillation used to train data-efficient ViTs can be pushed when applied across architectures, from a CNN-based teacher into a lightweight ViT student. LiteMed-ViT is introduced as a dual-token ViT featuring a convolutional stem and a class-balanced distillation objective; at 0.234M parameters, LiteMed-ViT carries 47.8x fewer parameters than its ResNet18 teacher at a memory footprint of just 0.89 MB. Evaluated against three compact CNN baselines on three MedMNIST subsets of varying modality and class balance, LiteMed-ViT attains higher accuracy than all three baselines on BloodMNIST (95.05%) and two on PathMNIST (89.15%). Under MedMNIST-C corruptions, it is at least as robust as every compact CNN across all subsets, outperforming them on BloodMNIST and PathMNIST without declining in relative standing as severity increases. Particular attention is devoted to identifying the exact conditions where performance plateaus. On DermaMNIST, LiteMed-ViT’s balanced accuracy remains up to 1.9 points below the strongest compact CNNs and does not benefit from higher input resolutions up to 224 pixels, where ResNet18 gains 12.6 points. Stronger teachers do not yield a more accurate student, and overconfident ones lower its DermaMNIST balanced accuracy unless calibrated.
[CV-197] Component Pooling Trained Across Orientations for Few-Shot Remote Sensing Scene Classification
链接: https://arxiv.org/abs/2610.33518
作者: Bingchen Yan,Wenheng Guo,Yuye Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Few-shot remote sensing scene classification (FSRSSC) aims to recognize novel scene categories from only a few labeled images. Most FSRSSC methods summarize the feature map of an image by global average pooling, which treats the few objects that define an aerial scene and the large background around them alike, and they rarely consider that aerial images have no canonical orientation. To address these issues, we propose RoCoP, a parameter-free component pooling network trained across orientations. RoCoP consists of a feature extractor, a non-negative decomposition module (NDM) and a component scoring module (CSM). NDM factorizes the local descriptors of each image into a few additive components with a deterministic initialization and a single multiplicative update, so that the positions of one object are grouped together. CSM weights each component by its activation magnitude and converts the weights into a soft mask for pooling. Because both modules are invariant to permutations of feature positions, the only rotation-sensitive part of the pipeline is the backbone, which we train to classify randomly rotated views of every image. Experiments on NWPU-RESISC45, AID and UC Merced show that component pooling and rotated-view training are complementary: together they improve a strong transfer-learning baseline by about 2.5 points in 5-way 1-shot accuracy on NWPU-RESISC45 and AID, without adding parameters at inference. Under an identical protocol, RoCoP outperforms re-implemented metric-based and task-adaptive methods, and NDM and CSM also improve existing methods when plugged into them. We further identify the regime in which the method does not apply, namely very small base sets.
[CV-198] FreqPKC: Frequency-Aware PeakConv for Radar Semantic Segmentation
链接: https://arxiv.org/abs/2610.33504
作者: Yangyong Deng,Pengyu Zhang,Shaoqiu Song,Xiaotao Huang,Jian Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Automotive millimeter-wave (mmWave) radar has attracted increasing attention for semantic segmentation in autonomous driving, owing to its robustness under adverse weather and lighting conditions. Recent works have introduced radar-specific neural operators such as PeakConv (PKC), inspired by constant false alarm rate (CFAR) detection, to better model the peak-dominated characteristics of radar images. However, existing methods typically treat the guard bandwidth as learnable parameters, without explicitly characterizing the frequency response of the PKC operator. In this work, we revisit PKC from a frequency-domain perspective and propose Frequency-Aware PeakConv (FreqPKC), which explicitly models and exploits its underlying spectral characteristics. By analyzing the frequency response of PKC, FreqPKC enables the network to capture a broader range of frequency components and adaptively adjust its effective bandwidth. With the proposed Frequency-Aware (FA) module, the network can infer appropriate guard bandwidths to balance high- and low-frequency information: it enhances high-frequency components in target regions for more precise segmentation while suppressing noise through low-frequency filtering in background areas. Building on FreqPKC, we further develop FreqPKC-Net, a radar semantic segmentation (RSS) framework that integrates frequency-guided adaptive modulation with cross-view fusion for multi-view inputs, and can also be adapted to single-view data. Experiments on the public datasets CARRADA, CARRADA-RAC, and a self-built dataset named RS-SAR demonstrate that FreqPKC-Net consistently outperforms existing SoTA methods on both range-Doppler and range-angle views.
[CV-199] Explicit Frictional Contact Modeling in Neural Garment Simulation WACV2027
链接: https://arxiv.org/abs/2610.33492
作者: Thanh-Nguyen Truong,Vu-Hoang Tran,Ching-Chun Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Machine Learning (cs.LG)
备注: WACV 2027
Abstract:Graph neural networks (GNNs) simulate garments efficiently through physics-based self-supervision, but leave body-garment friction implicit or inside training energies, with no friction coefficient \mu to set at inference; retraining the tested GNNs with a coefficient-specific friction term barely responds to \mu . Optimization-based solvers decide each contact’s stick-slip state from the reaction that governs it, but only by re-optimizing the garment state at every step; a friction term in a one-pass training objective acts as a fixed-weight penalty on sliding, and the trained weights apply no per-contact Coulomb condition at inference. We therefore resolve friction explicitly after prediction: a Modular Contact Operator, attached to a frozen predictor without retraining or an iterative solve, estimates the normal load, applies a closed-form Coulomb correction along a path across body triangles, and propagates the induced motion through the cloth as a linearized membrane wave. On HOOD and ContourCraft, it raises sticking and lowers slip monotonically in \mu , tracking a reference solver’s response while coefficient-specific retraining stays flat, with low penetration and reference-consistent stretching at an eighth of the solver’s runtime.
[CV-200] ReSight: Task-Driven Self-Improving Memory for Embodied Agents
链接: https://arxiv.org/abs/2610.33487
作者: Lirong Che,Yang Chen,Yimeng Su,Daqi Liu,Zhenyu Huang,Jiong Wang,Yuzhe Yang,Xu Cao,Chuang Wang,Jian Su,Xueqian Wang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robots can use historical observations to answer spatiotemporal queries in familiar environments, but the presence of relevant evidence does not imply that existing tools support the operations a task requires. We introduce ReSight, a task-driven self-improving memory system for embodied agents that expands the capabilities for processing and accessing historical observations as task demands arise. ReSight retains keyframes and their spatiotemporal associations, separating observation evidence from extensible derived fields so that stored observations can be reprocessed for new needs. An execution agent and a coding agent share this memory through different interfaces: the former calls registered tools within a bounded online budget, while the latter uses a programmable workspace and a larger exploration budget to inspect evidence, write programs, and acquire models as needed. To realize capability expansion in ReSight, we develop Explore-Distill-Mount (EDM), which identifies capability gaps from execution and exploration records without task reference answers and converts exploration-derived solutions into reusable memory builders, query tools, and skills, extending memory representations and access capabilities as needed. Experiments on three public benchmarks demonstrate effective base memory access and capability expansion. With the execution model and online budget held fixed within each benchmark, EDM reduces mean position error on OC-NaVQA from 33.10 m to 24.72 m and improves high-level success on FindingDory from 38.3% to 63.8%. With the tools and skill frozen, success still improves by 28.3 percentage points on FindingDory episodes excluded from exploration and distillation. Experiments on real robot patrol memories further demonstrate reuse of these capabilities for subsequent queries and new memories, with fewer online tool calls on average.
[CV-201] Can One Hear Geometry? Gauge-Invariant Spectral Descriptors for Text-to-B-rep Retrieval
链接: https://arxiv.org/abs/2610.33484
作者: Hongzhu Yi,Hao Liu,Tao Yu,Tiankun Yang,Yujia Yang,Zhenghao Zhang,Tianyu Zong,Zhenyu Guan,Bingkang Shi,Haopeng Jin,Yuanxiang Wang,Zhongtian Luo,Jungang Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-to-3D solid retrieval requires aligning natural language with precise geometry despite variations in pose, scale, and recording conventions. Existing methods discretize Boundary Representation (B-rep) models into point clouds or meshes, or learn features from faces, edges, and topology, risking information loss or lacking explicit invariance guarantees. We propose a text-to-B-rep retrieval framework based on group actions and spectral geometric invariants, making invariance an explicit constraint on descriptor construction. Specifically, we use group actions to characterize equivalence transformations induced by translation, rotation, reflection, uniform scaling, and contour recording conventions. For sketch-extrusion solids, we construct a fixed 280-dimensional spectral geometric descriptor by integrating geometric information from three complementary levels, including cross-sectional regions, contour boundaries, and extrusion structures. Furthermore, we use descriptor-consistent multi-positive sets and geometric attribute supervision to promote structural alignment between text and solids. On Text-to-3D solid retrieval, our method achieves 88.6% Recall@5, exceeding the previous state of the art by 13.6 percentage points and demonstrating the effectiveness of explicit geometric invariants for precise 3D solid retrieval.
[CV-202] Energy Efficient Lung Tumor Segmentation with a modified CBAM ENet Skip Model
链接: https://arxiv.org/abs/2610.33479
作者: Sharan Thummagunti,Aayushi Padia,Abhiram Koppuravuri,Prabhu Prasad B M,G N Girish
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Lung tumor segmentation plays a vital role in early detection and treatment planning of lung cancer. Although state-of-the-art deep learning models achieve high dice score coefficients (DSC), each forward pass requires computing large amounts of floating point operations, making them computationally expensive, requiring GPU acceleration, and having slow inference speed, limiting real-world clinical deployment. We propose a lightweight, energy-efficient model that leverages the ENet architecture for its real-time inference capability and low computational cost, enhanced with UNet-style skip connections between encoder and decoder stages for feature reuse and the lightweight Convolutional Block Attention Module (CBAM) for improved feature representation. The model is trained on publicly available data from The Cancer Imaging Archive (TCIA) via the Medical Segmentation Decathlon Challenge (MSD) and evaluated on the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset. Our model achieves a DSC of 0.90812 (threshold of 0.5) with its compact architecture of 430,052 parameters and 0.549 Giga Multiply-Accumulate operations (GMAC), making it up to 10 times faster, requires 79 times less GMAC than UNet, and has 22 times fewer parameters than SegNet. These characteristics make it well-suited for deployment in clinical settings where computational resources are limited and decrease the cost of utilizing such a system. Additionally, the explainable AI technique of Grad-CAM is used to provide visual interpretability of the model’s predictions and to identify regions where segmentation performance requires improvement.
[CV-203] MarineLife-16K: An Attribute-Rich Video-Text Dataset for Motion-Aware Marine Video Understanding WACV2027
链接: https://arxiv.org/abs/2610.33478
作者: Quang-Trung Truong,Thanh-Nghia Nguyen,Phu-Phong Vo,Jaeyeon Kim,Cheng Chen,Jie Zhang,Binh-Son Hua,Duc Thanh Nguyen,Ivor Tsang,Sai-Kit Yeung
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Published at WACV2027
Abstract:Marine video understanding is crucial for a wide range of real-world applications. However, existing marine video datasets suffer from limitations, including the limited number of data samples, fine-grained text descriptions, and detailed visual attributes. To address these limitations, we introduce a large-scale, video-text dataset featuring rich, in-the-wild visual attributes tailored for complex marine environments. Our dataset comprises 2,000 video-text pairs annotated with 13 attributes covering up to four objects per scene, object visual/geometric properties, behaviors, environmental context, and camera motion, spanning diverse lighting conditions, times of day, and 7 countries. From these annotations, we generate 16,080 video-question pairs (12,080 multiple-choice, 4,000 open-ended). Furthermore, we benchmark both proprietary and open-source video-LLM models on video question answering and video captioning tasks. We find that (i) proprietary models often achieves superior performances; (ii) VLMs handle object motion well but struggle with complex camera motion; (iii) closed-source models struggle to perceive camera motion; and (iv) VLMs struggle to reason in static cameras.
[CV-204] Qwen -Image-PE: Learning Visual Design for Text-to-Image Prompt Enhancement
链接: https://arxiv.org/abs/2610.33472
作者: Zekai Zhang,Yanran Zhang,Shengming Yin,Yan Shu,Zhening Liu,Bingxuan Zhao,Chenyang Zhu,Donghao Zhou,Jiahao Li,Jiahao Wang,Jie Zhang,Kaiyuan Gao,Kun Yan,Lihan Jiang,Ningyuan Tang,Qihang Ai,Rui Song,Shuo Cao,Xiao Xu,Xiaoyue Chen,Yuxiang Chen,Zhendong Wang,Zhouyang Li,Zihao Liu,Zikai Zhou,Bo Zheng,Huishuai Zhang,Dongyan Zhao,Chenfei Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Learning visual design is essential for prompt enhancement (PE) to turn underspecified user requests into coherent visual specifications for text-to-image (T2I) generation. Existing methods preserve user intent and improve overall visual appeal, with less explicit emphasis on fine-grained design dimensions. To address this gap, we introduce Qwen-Image-PE, which learns visual design through two-stage training. Supervised fine-tuning with Caption-Based Reverse Distillation (CRD) uses dense captions of well-designed images as design supervision. Reinforcement learning with Groupwise Design Reward (GDR) jointly compares all candidate images for the same request to refine lighting, composition, layout, and detail. On Qwen-Image-Bench with Qwen-Image-2.1, Qwen-Image-PE-397B improves over original requests by 7.81 points, while Qwen-Image-PE-9B outperforms all evaluated PE baselines. Further evaluations show gains in visual design, aesthetics, image quality, and human preference. We find that PE scales along three axes: training data, model size, and information density, with sufficient information density as a prerequisite for effective enhancement. Moreover, we demonstrate that Qwen-Image-PE generalizes to four additional T2I backbones, even surpassing native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein. We release Qwen-Image-PE-9B at this https URL.
[CV-205] Semantic-First MixGRPO: Reliable Emotional Image Generation via Temporal Decoupling
链接: https://arxiv.org/abs/2610.33460
作者: Zhenxuan Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 24 pages, 8 figures, 8 tables
Abstract:Controllable image synthesis should modify visual affect without replacing the objects, layout, or relations specified by a prompt. Existing methods address this tension through conditioning architectures or auxiliary losses, while mixed ordinary differential equation (ODE) and stochastic differential equation (SDE) sampling has mainly been used to accelerate group relative policy optimization (GRPO). We instead study stochastic-window position as an attribute-control variable for flow-based image generation. Using FLUX.1-dev with EmotiCrafter-style valence-arousal conditioning, we compare fixed early, middle, and late four-transition SDE windows, progressive MixGRPO, and full-path Flow-GRPO under matched settings. Because step 0 has the highest noise, our semantic-first schedule uses deterministic ODE sampling at steps 0-19 to establish content, then confines stochastic exploration and policy updates to steps 20-23. Across three seeds, it achieves a CLIPScore of 29.662\pm0.276 , cross-affect DINO consistency of 80.1\pm0.9 , and semantic drift of 2.9%, while reducing training cost by 33.5% relative to full-path Flow-GRPO. Full-path stochastic training remains more accurate on valence-arousal targets, revealing a genuine trade-off. Semantic-first nevertheless improves both affect and semantic metrics over progressive scheduling, preserves the principal subject in 95.98% of images at five extreme or neutral targets, and receives the highest semantic-preservation and overall-preference ratings in a 30-participant study. These results identify probability-flow time as a practical graphics control dimension for reliable affective appearance manipulation.
[CV-206] Dynamic Query as Budgeting for Tiny Object Detection
链接: https://arxiv.org/abs/2610.33430
作者: Yi Zhang,Di Xiong,Ge Gao,Xiangyue Zhang,Yihang Qiu,Yuxuan Zhou,Shuo Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent DETR-based tiny object detectors adopt dynamic query mechanisms to handle severe density imbalance. However, existing designs entangle two effects: how many and which queries are decoded. This entanglement obscures what drives the gains and complicates dynamic-query design; additional learned predictors and heuristic filtering also introduce overhead and potential selection errors. We disentangle these effects and show that dynamic query primarily acts as \emphbudgeting—allocating decoder query capacity according to object density—rather than intricate query this http URL by this view, we formulate a monotonic and conservative budgeting principle and propose a budgeting-based DETR (BUTR), a streamlined framework that uses a lightweight, non-learned budgeter to estimate an input-dependent query budget K(x) from encoder confidences. BUTR retains standard Top- K(x) query initialization without additional learned modules or heuristic filtering in the this http URL 800\times800 resolution, BUTR-L achieves 35.7 AP at 47.8 FPS on AI-TOD-v2 and 29.8 AP at 42.1 FPS on VisDrone on an NVIDIA T4 GPU. Inference-time integration on COCO further demonstrates its applicability to general scenario detection. Code is available at this https URL.
[CV-207] How Robust Are CNNs and Vision Transformers for Alzheimers MRI Classification Across Datasets?
链接: https://arxiv.org/abs/2610.33416
作者: Hafija Aktar,Molly Croes,Frank S. Menniti,Abdeltawab Hendawi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 9 tables, 3 figures. Submitted to Biomedical Signal Processing and Control
Abstract:Accurate assessment of Alzheimer’s disease stages from MRI is important for understanding disease progression and supporting the development of computer-aided analysis tools. Deep learning models have shown promising results for Alzheimer’s disease classification from MRI. However, when evaluated on single datasets, it is difficult to determine whether the observed performance reflects consistent architectural behavior or dataset-specific effects. In this work, we present a multi-dataset benchmarking framework with standardized preprocessing and evaluation. We compare five architectures: ResNet-50, ReXNet-150, Swin Transformer, ViT-Base, and ViT-Large. The models are evaluated across three MRI datasets representing severe class imbalance, train-test distribution mismatch, and large-scale augmentation. The results show broadly consistent performance patterns across datasets, with ViT-Large achieving the best performance on all three datasets and reaching up to 99.84% accuracy and 99.86% macro F1. Performance differences between architectures are most pronounced under distribution mismatch and considerably smaller in the augmented-data setting. Across datasets, lower class-level performance is observed primarily among adjacent early-stage categories, particularly Very Mild and Mild impairment. These findings highlight the importance of evaluating deep learning architectures across multiple data conditions rather than relying on single-dataset performance.
[CV-208] AV-FABLE: Audio-Visual Manipulation Forensics in Long-Form Videos
链接: https://arxiv.org/abs/2610.33413
作者: Yuesheng Huang,Hoi Fan Au,Jiaqi Hou,Ruicheng Zhang,Jiarui He,Haocheng Yang,Bin Hu,Kewei Zhang,Weijian Deng,Jun Zhou,Xiu Li
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注: 41 pages, 10 figures, including appendices
Abstract:Localized AI manipulations in long-form videos often affect brief events within otherwise authentic content, making video-level classification insufficient for forensic analysis. Existing audio-visual deepfake benchmarks largely focus on short talking-head clips, while long-form temporal forensics remains predominantly visual. We introduce AV-FABLE, a dual-timeline benchmark for sparse manipulations in open-scene long-form audio-visual videos. AV-FABLE contains 16,650 pristine–manipulated twin pairs (998 hours) created with 27 editing configurations, with separate audio and visual timelines that enable modality-specific temporal localization and expose errors hidden by joint-timeline evaluation. On AV-FABLE, zero-shot prompting of audio-visual large language models (AV-LLMs) yields near-chance detection and poor temporal localization. In contrast, supervised readouts of their frozen perception features recover strong forensic signals, suggesting manipulation evidence is encoded in these representations but not effectively exposed through language outputs. Motivated by this observation, we propose NFSR(Native Forensics via Separated Readouts), which reads frozen AV-LLM features with separate audio and visual temporal decoders. NFSR achieves the best audio-visual localization performance in-domain and across all three out-of-distribution settings, including transfer to held-out videos from an external long-video forensic benchmark. AV-FABLE, code, and model weights will be publicly released.
[CV-209] Bidirectional Cross-frequency Domain Interaction Network for UHD Low-Light Image Enhancement
链接: https://arxiv.org/abs/2610.33402
作者: Bo Fu,Hai Chi,Liyan Wang,Shilin Fu,Wei Wang,Kin-Man Lam,Jinshan Pan,Cong Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 14 figures
Abstract:Existing Ultra-High-Definition (UHD) low-light image enhancement (LLIE) methods typically rely on either spatial- or frequency-domain modeling, or independently process different frequency components for efficient restoration. However, they often fail to jointly exploit complementary spatial–frequency information and the inherent relationship between illumination and structural information across frequencies, leading to inconsistent illumination, residual noise, and detail artifacts. In this paper, we propose a novel Bidirectional Cross-frequency Domain Interaction Network (BiFreqNet) for UHD LLIE. Specifically, BiFreqNet incorporates a Dual-domain Feature Interaction (DFI) module into the multi-scale representations of a pretrained variational autoencoder (VAE) to perform reciprocal enhancement between low- and high-frequency features. DFI consists of a Spatial and Spectral-Aware Low-Frequency Enhancement (S ^2 LFE) block and a Cross-Frequency Enhancement (CFE) block. S ^2 LFE exploits high-frequency structural cues to guide low-frequency restoration and jointly models spatial and spectral information to recover global exposure and spatially varying illumination while suppressing sensor noise. CFE further leverages the enhanced low-frequency contextual information to guide adaptive high-frequency refinement, enabling faithful reconstruction of structures and fine details. Extensive experiments on four general and UHD LLIE benchmarks demonstrate that BiFreqNet achieves competitive or superior performance compared with state-of-the-art methods.
[CV-210] Rethinking Microscopy Generation: Co-Designed Diffusion for Biologically Interpretable Single-Cell Synthesis BMVC2026
链接: https://arxiv.org/abs/2610.33398
作者: Shumpei Takezaki,Duway Nicolas Lesmes-Leon,Ryoma Bise,Gillian Lovell,Bianca Migliori,Andreas Dengel,Sheraz Ahmed
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at BMVC2026
Abstract:Generative models for microscopy can already produce realistic whole-image fields of view, but offer no principled way to control how each individual cell looks. We address this gap by recasting microscopy generation at the single-cell level and designing the input controls and the model’s internal representation jointly, so that both are organized along the same biologically meaningful directions. On the input side, we aggregate a long list of standard cell measurements into a small set of semantic axes (size, elongation, brightness, texture, etc.) that are each named, measurable, and traversable. On the model side, we introduce BioVAE, an encoder-decoder whose latent space is explicitly aligned with the same set of measurements. A key finding is that the benefit of biologically structuring the latent emerges only when conditioning is also moved to the semantic axes, showing that the two ends must be designed jointly. Experiments on LIVECell show that semantic-axis conditioning substantially improves generative fidelity over high-dimensional conditioning, a blind study with biologists confirms the realism of generated cells, and per-axis traversals produce stable and interpretable phenotypic changes.
[CV-211] Fast Instinct Slow Reason : A Dual-Speed Agent ic Framework for Self-Configuring Multi-Object Tracking in Satellite Video
链接: https://arxiv.org/abs/2610.33396
作者: Fan He,Baining Wu,Yan Li,Xiaoyang Zeng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multi-object tracking in satellite video underpins persistent monitoring, yet offline-tuned trackers lack the onboard intelligence and efficiency to adapt within an observation of tens of seconds. The root is that adaptation still means relabelling and retraining the detector, which cannot run on board in time. Tuning the tracker configuration strongly controls accuracy without touching the network, yet search needs labelled or proxy evaluations, and a language agent costs seconds and tokens per decision. We propose a dual-speed agentic framework pairing a fast instinct compiled offline with a slow reason serving the mission. Its algorithm, Arbitration of Unlabelled Shortlists from Prior Experience, or AUSPEX, learns a covering shortlist and a label-free ranker from labelled source tasks and commits one configuration per sequence. Its system, a Cascaded Onboard Runtime with Verified Uplinked Skills, or CORVUS, cascades reflex, skill, and agent layers, so that a verified skill commits 2.5 s into each observation while a language agent serves natural- language missions. On SatMTB and SatVideoDT with three recent detectors, AUSPEX raises IDF1 by 2.84 to 2.94 points and MOTA by 3.21 to 3.29 points on average without target labels, balances accuracy and labels best among sixteen methods, and loses at most 0.28 IDF1 points at 1.25 tracking passes. CORVUS executes four missions with execution accuracy up to 0.99 and target recall at the perception ceiling, spends no token on the routine decision, runs on Jetson Xavier NX, and hosts fifteen language models from six families without altering any commitment, demonstrating its extensibility and generalization.
[CV-212] AdShot: Benchmarking Multimodal Large Language Models for Video Advertisement Clipping
链接: https://arxiv.org/abs/2610.33377
作者: Wen Xie,Om Rastogi,Sai Siddhartha Vivek Dhir Rangoju,Gijs Overgoor,Yakov Bart,Sarah Ostadabbas
类目: Multimedia (cs.MM); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Advertisers routinely create multiple duration variants of ads to meet marketing budgets and viewer preferences, a labor-intensive and costly process. While multimodal large language models (MLLMs) excel at general video understanding, their ability to perform specialized editorial tasks remains unexplored. We introduce AdShot, a benchmark for evaluating MLLMs on shot selection for ad clipping. It contains 823 professionally edited ad pairs (30-sec sources and 15-sec edits) across 194 brands and 17 industries, annotated with shot-level boundaries and content features (action, emotion, information, imagery). We evaluate 13 state-of-the-art video-only MLLMs and 2 audio-visual MLLMs (7B-72B parameters) on shot selection accuracy (Precision, Recall, F1), duration constraints (15s), and content preservation, including an audio ablation. We find: (1) model architecture matters more than scale, with smaller models often matching or exceeding larger ones; performance is strongest for target ads with 6-10 shots; (2) models exhibit a mean absolute duration error of 5.39s, with most tending to overshoot the 15s target, though they can still substantially reduce human editing effort; (3) all models exhibit systematic content biases, favoring information-rich and emotion-focused shots while under-selecting action and imagery. We publicly release our benchmark to facilitate research at the intersection of multimodal AI and computational advertising.
[CV-213] ght Learned Event-Inertial Odometry (TLEIO): a Transformer-Based Architecture for End-to-End Motion from Event Streams
链接: https://arxiv.org/abs/2610.33356
作者: Matteo Cascone,David Jensen,Alessandro Pirini,Angelo Stefanini,Pietro Venè,Marta Tintoré Gazulla,Javier Hidalgo Carriò
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages
Abstract:Event cameras provide microsecond temporal resolution, high dynamic range, and sparse asynchronous measurements, making them compelling for odometry under fast motion and challenging illumination. Yet, accurate event-based state estimation remains difficult: existing systems often use events as a complement to frames or rely on feature tracking, correspondence search, and optimization-heavy backends, limiting the efficiency promised by event sensing. We introduce Tight Learned Event-Inertial Odometry (TLEIO), a compact filter-based pipeline that learns short-window relative motion directly from voxelized event streams and tightly fuses the resulting constraints with IMU propagation. Its learned front-end, EventsFormer, predicts local displacements and per-axis uncertainty from event voxels in a single forward pass, while a stochastic-cloning EKF fuses these overlapping predictions with IMU propagation, producing metric trajectories without explicit feature tracks and pose optimization at inference time. The network is trained directly from pose displacements, without requiring event-flow labels. We evaluate TLEIO on the TartanAir/TartanEvent benchmark and on real DAVIS240c sequences. On TartanAir, TLEIO reduces average ATE from 2.26 m to 1.10 m over the IMU-only TLIO baseline, while achieving up to 48.9% lower runtime than optimization-based event odometry methods.
[CV-214] Detecting Temporal Concepts in Video Transformers via a No-False-Positives Test
链接: https://arxiv.org/abs/2610.33353
作者: Jianghuai Li,Andrew Ren,Greg Shakhnarovich
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 39 pages, 1 figure, 23 tables
Abstract:Video models have the architecture to support encoding temporal aspects of the input, i.e. concepts that require more than one frame to define, like object movement. However, because motion and appearance change are entangled, a feature that appears to encode temporal concepts like object movement may instead encode non-temporal concepts like spatial presence at different positions whose activations merely correlate with motion. To resolve this, we introduce the No-False-Positives (NFP) test, a statistical test with a formal guarantee of statistical specificity for identifying features encoding temporal concepts. Using specially designed stimuli, we apply this test to representations produced by VideoMAE, V-JEPA2, and TimeSformer, and observe that temporal features exist in each model. We validate the test first by showing it yields no temporal features for the DINOv2 image-encoder. Then, using a synthetic representation space, we show that detection rate increases with a direction’s projection onto the temporal subspace, and that Sparse Autoencoder features test-flagged as temporal project much more of their norm onto the temporal subspace than unflagged features. Interventions on temporal SAE features yield mixed results across models, suggesting that intervening on temporal features may not outperform a non-temporal feature baseline.
[CV-215] Real Time Motion Detection in Video Inspired by Photonic Time Stretch
链接: https://arxiv.org/abs/2610.33351
作者: Callen MacPhee,Yiming Zhou,Bahram Jalali
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV); Optics (physics.optics)
备注:
Abstract:Motion perception is fundamental to computer vision and physical AI, and plays a central role in robotics, autonomous vehicles, human-computer interaction, and video compression. Traditional approaches such as background subtraction, frame-differencing and optical flow algorithms often struggle in complex or dynamic scenes, typically sacrificing spatial resolution for computational tractability. Here we introduce MoViD (Motion estimation via Virtual dispersion and phase Detection), a physics-inspired algorithm grounded in the principles that govern photonic time stretch - a highly successful ultrafast realtime data acquisition platform. Drawing inspiration from the optical physics therein, MoViD treats video as a collection of optical time series waveforms associated with pixels. Motion of an object through a pixel broadens the time series spectrum and, after group delay dispersion, induces a temporal phase that exists only if there is motion. Spatiotemporal group delay dispersion templates, corresponding to dielectric functions, classify specific motion trajectories. Unlike conventional empirical phase-based motion-analysis pipelines that require narrow-band spatial channelization and pyramid reconstruction, the physics-based MoViD operates directly on the broadband video signal, vastly increasing its computational speed. The resulting algorithm achieves unprecedented performance in both motion characterization and motion magnification, providing full-resolution motion mapping in real time at nearly 200 frames per second at 4K resolution and 70 frames per second for 8K video - orders of magnitude faster than classical optical flow approaches or learning-based methods. Notably, such extreme performance enables real time motion analysis at ultrahigh resolution for motion detection and compensation in robotics and other autonomous devices.
[CV-216] Compression of Vision Foundation Models for Retinal OCT via Growing-Neural-Gas Codebooks and Graph-Guided Masked Image Modeling
链接: https://arxiv.org/abs/2610.33340
作者: Ali Yavari,Tilman Schmoll,Rainer A. Leitgeb,Wolfgang Drexler,Thomas Schlegl
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 31 pages, 2 figures, 8 tables; supplementary material provided as an ancillary file
Abstract:Vision foundation models for retinal imaging are large and computationally demanding, limiting point-of-care deployment. We ask how much of such a model’s downstream utility survives compression into a smaller network. We propose a topology-aware discretization framework that distills a continuous foundation model into a discrete codebook, then uses it as the reconstruction target for masked image modeling (MIM). It combines gradient-free codebook learning via Growing Neural Gas, which yields an adjacency graph over the codes, with a graph-distance soft cross-entropy loss that replaces one-hot MIM targets with topologically graded supervision. With student pretraining on 383 public Duke iAMD optical coherence tomography (OCT) volumes, the pipeline distills an 86M-parameter teacher into a 22M-parameter student with at least 3.6x fewer FLOPs. The student performs close to, and in some settings above, its teacher across three classification and two segmentation benchmarks, with the magnitude and direction of the difference depending on the teacher, task, metric, and student capacity. On the external OLIVES benchmark, it surpasses its MIRAGE teacher on every metric (98.06% vs. 95.22% balanced accuracy). It works with two Vision Transformer (ViT-Base) teachers, retinal (MIRAGE) and natural-image (DINOv3); with the out-of-domain teacher, the student outperforms it on retinal-layer segmentation, indicating that the added value of retina-specific pretraining is task-dependent: largest for segmentation, while a general-domain teacher is competitive for classification. These results yield compact retinal models potentially suitable for deployment, transferring across OCT datasets, devices, and clinical tasks, including three datasets not used in the student’s pretraining.
[CV-217] nyFlow: Breaking the Warping Bottleneck for Edge Optical Flow
链接: https://arxiv.org/abs/2610.33338
作者: Xiao’ao Song,Chahat Deep Singh
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 9 Pages, 5 Figures, 4 Tables
Abstract:High-performance optical flow networks are difficult to deploy on resource-constrained edge accelerators because their iterative refinement warps features by the current flow, which relies on flow-dependent resampling and data-dependent memory access. We present TinyFlow, a 1.26M-parameter warp-free optical flow network that preserves iterative coarse-to-fine refinement by replacing the warp with a fixed-window correlation over constant offsets, so the whole network compiles as a single static INT8 graph. Quantization-aware fine-tuning, warp-teacher distillation, and model souping reduce the Float32-to-INT8 penalty to below 0.2 end-point error (EPE) on MPI-Sintel. TinyFlow outperforms EdgeFlowNet and NanoFlowNet on Sintel Clean, Sintel Final, and FlyingChairs in both Float32 and INT8, while using 54% fewer parameters than EdgeFlowNet, and our INT8 model achieves 6.29/7.13 EPE on Sintel Clean/Final. In flow-only monocular visual odometry on TartanAir and UZH-FPV, the same INT8 model yields the lowest trajectory error on seven of eleven sequences.
[CV-218] KINESYS: Test-Time Adaptation for Reference-Based Effect Editing in Images
链接: https://arxiv.org/abs/2610.33292
作者: Noah Frahm,Luchao Qi,Haoze Li,Dan B Goldman,Sambit Bhattacharya,Roni Sengupta
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Visual effects are often easier to demonstrate via reference than to describe in words. We study effect-reference editing, where a reference image shows a desired visual effect and language identifies which properties should guide the edit. Existing single-pass editing techniques often fail to make fine-grained edits, resulting in unintended changes to source content or deviations from the reference effect. We introduce KINESYS, a test-time adaptation framework for effect-reference editing that combines reference-aware and source-aware feedback to transfer the desired effect while preserving the composition and content of the original shot. To ensure KINESYS can produce a diverse set of edits that preserve the source, we develop a robust optimization strategy in which the generative model is fine-tuned across multiple noise samples and optimization runs. Evaluation with VLM judges, source preservation metrics (LPIPS, TPIPS), and a user study shows that KINESYS outperforms state-of-the-art image editors in source preservation; on effect transfer, it remains competitive with Nano Banana 2 and outperforms open-weight editors.
[CV-219] Image Clustering with Vision-Language Model Embeddings
链接: https://arxiv.org/abs/2610.33265
作者: João P. Matos-Carvalho,Nuno Fachada
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Pretrained vision-language models provide image representations whose suitability for clustering may vary with feature extraction, image domain, and partitioning algorithm. We compare nine representation pipelines across seven image benchmarks and four clustering configurations, yielding 252 baseline combinations. Eight pipelines extract visual features, including two vision-only references; the ninth generates descriptions and embeds the resulting text. With agglomerative clustering (Ward linkage) fixed, Qwen3-VL achieves the highest adjusted Rand index (ARI) on five datasets, including 0.925 on MNIST, while BLIP-2 leads on CIFAR-10 and CIFAR-100 with ARIs of 0.925 and 0.591, respectively. A separate sensitivity analysis compares raw and L2-normalized embeddings at their original dimensionality across all 63 dataset-representation pairs. Normalization preserves the highest-ARI representation in every dataset, increases ARI in 35 pairs, decreases it in 26, and leaves two numerically unchanged. On the SVHN dataset, the highest ARI remains low, rising from 0.097 to 0.098 after normalization. The evaluated caption-mediated pipeline obtains relatively high ARI on isolated digits but does not consistently exceed direct visual features. These findings support selecting and evaluating complete representation-clustering pipelines for the intended domain, with explicit attention to extraction and normalization choices.
[CV-220] Explaining Basketball Shot Classifications with Grad-CAM: An Investigation of Shortcut Reliance with Spatiotemporal Correlations
链接: https://arxiv.org/abs/2610.33263
作者: Hongyi Tian,Robin Cohen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 2 pages, 2 figures. Code and data: this https URL
Abstract:Basketball action recognition is used in high-stakes scenarios to inform such tasks as specialized coaching, yet the evidence used by its deep-learning models is poorly understood. This paper investigates whether there is evidence that an R(2+1)D-18 model trained on the Basketball-51 dataset relies on spurious spatiotemporal correlations when classifying basketball shot outcomes. Gradient-weighted Class Activation Mapping (Grad-CAM) was used to generate spatiotemporal saliency maps, whose faithfulness was evaluated by progressively blurring the most salient pixels. Twenty saliency maps were then visually reviewed for recurring salient regions. Grad-CAM highlighted spatially central regions of clips including the shooter, net, and occasionally the courtside crowd. Blurring interventions were conducted to support these patterns. Blurring the central region resulted in an accuracy of 43.7%, compared to the original test accuracy of 69.3% and random-blurred accuracy of 50.9%. The additional 7.2 percentage-point reduction from central blurring compared to random blurring suggests that the model relies on task-relevant visual regions; this highlights the need for continual audits to ensure trustworthy, equitable sports analytics.
[CV-221] Automated identification of ivory carvings using machine learning: A proof-of-concept approach to detecting the illegal wildlife trade
链接: https://arxiv.org/abs/2610.33243
作者: Andrius Tamosiunas,Simone Mozzon,David Bacon,Caroline Cox,Alan Collins
类目: Computer Vision and Pattern Recognition (cs.CV); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注: 16 pages, 7 figures and 2 tables. Submitted for publication in PLOS One
Abstract:Despite a ban on the international trading of recently poached ivory, elephants are still killed for their tusks. Ivory and its products are mainly sold in online marketplaces. However, due to the huge volume and variety of adverts and the similarity of ivory with other materials that can be legally sold, it is hard for enforcement officials to identify illicit trading and prosecute sellers. Existing techniques for identifying elephant ivory, such as Fourier-transform infrared spectroscopy and microscopy, can reliably distinguish ivory from other materials, but require physical access to the item and cannot be applied to photographs alone. As a result, these methods are unsuitable for screening the large volume of images posted in online marketplaces, leaving a gap for automated, image-based identification tools. In this proof of concept work, we investigate the potential of machine learning techniques for image analysis to automatically distinguish ivory objects from bone. We employ a semantic segmentation framework using a U-Net architecture with a ResNet34 backbone, leveraging transfer learning on an augmented dataset of 541 original images. Using a dataset of images of ivory and items resembling carved ivory, several models were trialled. On a distinct test set, the classifier achieved an Area Under the Curve (AUC) of 0.81 and an overall classification accuracy of 81% on confident samples. The paper presents the best-performing model as a proof of concept for developing a system that can effectively identify ivory and its products when indicative features of elephant ivory are present in an image. This work provides initial steps towards a machine learning-based framework designed to advise law-enforcement in detecting illegal elephant ivory trade.
[CV-222] he Locality Trade-off in Masked Patch Self-Distillation NEURIPS2026
链接: https://arxiv.org/abs/2610.33223
作者: Xi Weng,Zhaoyu Liu,Lianyu Hu,Jin Song Dong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026
Abstract:Masked patch self-distillation is a key component of modern vision foundation models, including DINOv2 and DINOv3, yet its implications for local visual information remain unclear. We show that its semantic benefits can coexist with a scale-dependent locality cost. Using two annotation-free diagnostics, we distinguish recoverable local appearance from spatial correspondence. In smaller models, adding iBOT substantially weakens patch correspondence while largely preserving appearance decodability. In larger models, correspondence improves, but pixel and texture evidence becomes less linearly accessible in deep features. Thus, scaling changes the form of the locality cost rather than uniformly eliminating it. Our analysis of Gram anchoring further shows that preserving patch relations does not guarantee simultaneous improvements in both properties. These findings motivate evaluating and designing visual representations around the complementary demands of semantic abstraction, appearance preservation, and spatial correspondence. Our code is available at this https URL.
[CV-223] Vision Token Pruning via Query-Vision Interaction Decomposition NEURIPS2026
链接: https://arxiv.org/abs/2610.33215
作者: Harshithanjani Athi,Sravan Kumar Ankireddy,Jianzhong Charlie Zhang,Hyeji Kim
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages for the main paper, 8 figures including supplementary material, accepted at NeurIPS 2026
Abstract:Large vision-language models (VLMs), which process both visual inputs and text queries, incur high inference costs due to the large number of visual tokens, making visual token pruning a natural approach for improving efficiency. Recent work has leveraged query information to guide token selection, often through heuristics derived from attention patterns or token dynamics. However, these approaches do not explicitly exploit the underlying structure of query-vision interactions. We address this gap by observing that the query-vision interaction matrix, naturally induced by the query and key projections already present in transformer attention, is effectively low-rank, with a small number of dominant latent interaction modes capturing query-relevant semantics. Motivated by this observation, we propose Query-Vision Decomposition (QViD), a novel training-free query-aware visual token pruning method that exploits this structure. Extensive experiments on both image and video understanding benchmarks show that QViD consistently improves performance under matched token budgets, with the largest gains appearing under aggressive compression regimes.
[CV-224] Support-guided OOD Removal with Diffusion
链接: https://arxiv.org/abs/2610.33194
作者: Alireza Kheirandish,Sara Fridovich-Keil
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Sound (cs.SD)
备注:
Abstract:Many real-world signals are corrupted by localized, structured perturbations whose distribution is unknown: an intermittent background noise in a music recording, a watermark stamped on a photograph, a cast shadow. We study the task of recovering a clean signal x from an observation y containing such a localized out-of-distribution (OOD) corruption. We propose Support-guided OOD Removal with Diffusion, SORD, an iterative algorithm to recover x using only two ingredients: a binary map of the corruption support and a single diffusion model pretrained on clean signals. SORD does not require any paired data or any statistical model of the corruption. We evaluate SORD on natural images and human faces with spatially localized corruptions including handwritten digits, other images, watermarks, and shadows, as well as audio recordings of different musical instruments with temporally localized corruptions including other instruments and human speech. SORD is competitive with, and often outperforms, posterior-sampling inpainting given the same diffusion prior and the same mask, source-separation baselines that require a separate diffusion model for each source, and, on several benchmarks, supervised networks trained on paired clean and corrupted signals. The code is released here: \hrefthis https URLthis https URL.
[CV-225] World-in-Context: Interactive Video World Modeling with In-Context Conditioning
链接: https://arxiv.org/abs/2610.33169
作者: Harold Haodong Chen,Rui Peng,Rongjin Guo,Lingteng Qiu,Kejie Qiu,Chao Xu,Zilong Dong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project: this https URL
Abstract:Interactive video world models require coordinated control over viewing perspectives, character actions, and camera motion. Existing methods typically leave view modes implicit in text captions, conflate action with camera motion, and inject controls through fragmented architectural pathways that capture interdependencies only indirectly. In this work, we present World-in-Context, an interactive world modeling framework powered by In-Context Conditioning (ICC). Rather than relying on specialized conditioning pathways, World-in-Context unifies view, action, and camera controls into explicit tokens contextualized directly within a shared self-attention space. Our formulation reveals that (i) explicit view tokens disentangle conflicting dynamics and unlock emergent view switching from single-view data, while (ii) action and camera inputs play complementary roles in governing local responsiveness and global stability. To preserve temporal continuity across execution intervals, we further introduce Interaction-Transport RoPE (IT-RoPE), parameterizing temporal attention with accumulated control transitions. On WBench, World-in-Context-5B achieves navigation performance on par with 22B EchoWM ( 81.5 vs. 81.6 ) and attains a 93.8% perspective-switching success rate, substantially outperforming existing 5B baselines ( 18.8% ). Controlled analyses further demonstrate compounding advantages of World-in-Context under richer conditioning mixtures, favorable data efficiency, and natural applicability to causal rollouts.
[CV-226] Gaussian Episodic Memory for Long-Term Spatiotemporal Reasoning
链接: https://arxiv.org/abs/2610.33158
作者: Ling-Chia Chen,Abdelrhman Werby,Kaan Aytekin,Kai O. Arras
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Code and data: this https URL
Abstract:Real-world environments undergo continual semi-static changes, requiring a memory representation that captures both the spatial and temporal evolution of a scene. Existing spatio-temporal memories for robots typically rely either on structured representations, which are limited to the entities and attributes captured during construction, or on image-based memories, which are constrained to previously observed viewpoints. We introduce GEMS, a framework that combines a persistent spatio-temporal 3D Gaussian Splatting (3DGS) map with an agentic reasoning pipeline. The 3DGS map is reconstructed once and subsequently updated only in regions where the scene changes, storing only these modifications while preserving visual detail across time. The agentic pipeline leverages this map to render novel views on demand, iteratively accumulating evidence to answer spatio-temporal questions and ground objects from natural-language queries. To advance future research, we introduce GEMS-Bench, a spatiotemporal question-answering benchmark focused on semi-static environments. It comprises 3 real-world and 12 synthetic environments, spanning 473 question-answer pairs. We evaluate GEMS on novel-view synthesis quality across scene updates, 3D object grounding, spatiotemporal question answering, and storage efficiency. GEMS outperforms recent baselines in 3D object grounding and spatiotemporal question answering, while achieving rendering quality on par with dedicated scene-update methods. Code and data: this https URL
[CV-227] WebUniverse: Scaling Verified Training Data for Agent ic Web Development
链接: https://arxiv.org/abs/2610.33149
作者: Yaorui Ding,Fengqing Jiang,Yichen Feng,Yuetai Li,Zhangchen Xu,Kaiyuan Zheng,Xuanyu Lu,Bhaskar Ramasubramanian,Basel Alomair,Luyao Niu,Radha Poovendran
类目: oftware Engineering (cs.SE); Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Training agents for front-end web development requires demonstrations of code editing, execution, browser inspection and revision, yet public trajectories and reusable task environments remain limited. We introduce WebUniverse, a pipeline that constructs verifiable tasks with verified development trajectories from open-source repositories and real web pages. The tasks cover text-grounded application development in standalone and iterative settings and vision-grounded reproduction across pages and viewports. We collect agent trajectories for these tasks and verify the resulting applications against task-specific functional or visual requirements. We release WebUniverse-Env, containing 15K executable task environments, and WebUniverse-Mix, containing the corresponding 15K development trajectories with verification records and scores. Fine-tuning Qwen3.5-35B-A3B on WebUniverse-Mix raises the Vision2Web average score from 51.7 to 68.3, WebGen-Bench accuracy from 34.5% to 63.9%, and Web-Bench pass@2 from 12.2% to 23.1%. The resulting model also outperforms a public-data baseline trained with the same configuration and sample budget across all reported metrics. Ablations show complementary benefits from text and visual trajectories and continued gains as training data scales from 5K to 15K trajectories.
[CV-228] Learning Pathology-Sensitive Representations for Unsupervised Retinal OCT Anomaly Detection
链接: https://arxiv.org/abs/2610.33143
作者: Tania Haghighi,Sina Gholami,Behafarin Emam,Jennifer I Lim,Theodore Leng,Atalie C Thompson,Hamed Tabkhi,Minhaj Nur Alam
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Deep learning approaches for retinal optical coherence tomography (OCT) predominantly rely on predefined disease categories, limiting their ability to detect rare or previously unseen abnormalities. While visual foundation models provide transferable representations, their effectiveness in identifying pathological deviations from normal retinal anatomy remains insufficiently understood. In this work, we introduce an unsupervised framework for retinal OCT anomaly detection and localization that adapts pretrained visual representations to normal retinal anatomy through anatomy-aware reconstruction and contrastive triplet learning guided by synthetic structural perturbations, without requiring disease-specific annotations. We systematically investigate the influence of pretraining strategies, anatomy-aware representation adaptation, and anomaly scoring mechanisms on the detection and localization of unseen retinal abnormalities. Our findings demonstrate that adapting pretrained representations to normal retinal structure improves anomaly sensitivity, while pixel-space reconstruction errors and feature-space deviations provide complementary signals for identifying pathological changes. These findings underscore the importance of anatomy-aware representation learning and offer insights into developing generalizable retinal anomaly detection frameworks beyond predefined disease categories.
[CV-229] WorldCast: World-Consistent Character Video Generation with Large-Angle Camera Control
链接: https://arxiv.org/abs/2610.33135
作者: Lei Cao,Yu Liu,Wenhua Liao,Peng Zhang,Fei Wang,Hao Zhang,Jiajun Le,Daiguo Zhou,Jian Luan,Jiayi Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 30 pages, 14 figures, 3 tables
Abstract:Precise camera control is essential for cinematic character video generation, yet existing methods struggle to follow large-angle trajectories with combined rotation, translation, and radius variation from arbitrary initial azimuths. Such viewpoint changes also reveal character regions unseen in the initial frame, causing geometric distortions, appearance degradation, and identity drift. To systematically study these coupled challenges, we introduce CLACT, a large-scale Character-centric dataset and benchmark covering diverse Large-Angle Camera Trajectories, with accurate camera geometry and multi-view character observations for training and evaluation. We further propose WorldCast, a framework for world-consistent character video generation under diverse large-angle camera motions. Specifically, we back-project the initial frame into a 3D point cloud using estimated depth, align the point-cloud scale with the target camera trajectory, and render it along the trajectory to obtain explicit geometric guidance and coverage masks. To complement the incomplete character information, we introduce Geometry-Aligned Memory Attention (GAMA). Using relative camera poses estimated jointly from the initial frame and multi-view character anchors, we express anchor and target-view geometry in a common coordinate frame. GAMA uses this geometric alignment to guide the retrieval and injection of view-relevant character features during generation. Evaluations across diverse large-angle camera trajectories demonstrate accurate trajectory following while preserving character geometry, appearance, and identity.
[CV-230] Ampe-DB: A Dataset for Human Activity and Interaction Analysis Using the Ghanaian Game Ampe
链接: https://arxiv.org/abs/2610.33130
作者: Ruth Senior Biney,Isaac Kojo Yedonu Aboah
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 74 pages, 9 figures, 7 numbered tables. Ampe-DB V1 dataset (637 interaction clips) available at this https URL
Abstract:Human activity recognition benchmarks underrepresent culturally situated forms of movement and interaction. We introduce Ampe-DB, a multimodal dataset based on Ampe, a traditional Ghanaian game involving coordinated jumping, leg movements, and two-person interaction. It contains 637 canonical clips from 15 participants, with RGB video, two-person pose representations, participant and pair metadata, four-class movement labels, and derived binary movement-relation labels. We evaluate four-class movement recognition and binary movement-relation classification using three protocols: a duplicate-safe fixed benchmark, four-fold pair-held-out evaluation, and participant-held-out diagnostics for participants with complete class coverage. Baselines comprise a frozen Kinetics-400-pretrained R(2+1)D-18 RGB encoder, a two-layer bidirectional GRU for pose sequences, and validation-selected RGB-pose late fusion. For four-class recognition, RGB macro-F1 falls from 0.955 on the fixed benchmark to 0.418 under pair-held-out evaluation and 0.384 under participant-held-out evaluation. Pose is more robust, achieving 0.957, 0.811, and 0.839 across the same protocols. Fusion reaches 0.976 on the fixed benchmark and 0.824 under pair-held-out evaluation, but does not consistently improve on pose under participant shift. Binary movement-relation recognition also becomes substantially harder under held-out evaluation, especially for different-leg configurations. These findings show that strong fixed-benchmark scores can mask failures to generalize across people and pairings. Ampe-DB supports research on fine-grained movement recognition, human-human interaction, multimodal learning, and identity-aware generalization. Comments: 74 pages, 9 figures, 7 numbered tables. Ampe-DB V1 dataset (637 interaction clips) available at this https URL Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) Cite as: arXiv:2610.33130 [cs.CV] (or arXiv:2610.33130v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.33130 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Ruth Senior Biney [view email] [v1] Fri, 23 Oct 2026 13:28:41 UTC (2,718 KB)
[CV-231] Beyond Visible Speakers: Context-Aware Active Speaker Detection with Speaker Registration ICASSP2027
链接: https://arxiv.org/abs/2610.33127
作者: Pu Wang,Hugo Van hamme
类目: ound (cs.SD); Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM); Audio and Speech Processing (eess.AS); Image and Video Processing (eess.IV)
备注: Submitted to IEEE ICASSP 2027
Abstract:Active speaker detection (ASD) relies on audiovisual correspondence between speech and visible faces, yet this correspondence becomes unreliable in crowded scenes, or when visual evidence is unavailable. We introduce ROAM-ASD-C and ROAM-ASD-CR to extend ROAM-ASD with contextual and identity-aware reasoning. ROAM-ASD-C jointly reasons across simultaneously visible candidates, allowing other faces to provide complementary evidence when the target observation is ambiguous or incomplete. ROAM-ASD-CR incorporates registered-speaker representations derived from enrollment speech, allowing detection of whether a speaker is active even when that speaker is not visually observed. ROAM-ASD-C achieves state-of-the-art performance across five ASD benchmarks: 96.8% mAP on AVA, 98.9% on WASD, 88.8% on UniTalk, 98.7% on Talkies, and 99.5% on ASW. It also improves ROAM-ASD by up to 13.3 mAP under 75% target-track visual removal.
[CV-232] he Paired Thermal Channel Is Two Orthogonal Axes: A Controlled Attribution Study of Calibration Cost and Detection Gain in Infrared Person Detection
链接: https://arxiv.org/abs/2610.33118
作者: Tian Zhang,Yang Chen,Bing Bai,Weikai He,Xiang Gao,Zhirong Su
类目: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:The calibrated thermal channel in infrared person detection comprises two structurally orthogonal axes. True calibrated \Delta T outperforms scale-matched grayscale pseudo- \Delta T by +0.0221 mAP (paired Wilcoxon p=0.0098 , Cohen’s d=+1.08 ), confirming that the calibrated channel carries information the grayscale channel lacks. The calibration axis is driven by the joint pairing-scale condition. Preserving both paired correspondence and absolute calibrated scale raises expected calibration error (ECE) from \approx 0.05 to 0.142 . The 2\times 2 interaction ( p=0.0059 ) localizes the rise to this condition. Breaking either factor alone returns ECE to the broken-factor level (true-vs-const p=0.002 ; true-vs-shuffled p=0.002 ). Under post-hoc temperature scaling, the pairing cost survives recalibration ( p=0.002 ), while the scale cost is largely absorbed ( p=0.16 ). As regards the detection axis, the gain stems from structural de-structuring. Phase randomization, which destroys spatial structure while preserving per-sample variance, yields +0.0185 mAP over true \Delta T on FLIR ADAS ( p=0.014 , d=+0.96 ). The FLAME~3 gain is +0.0064 mAP ( p=0.078 , n=8). On the calibration axis, the same intervention is insignificant on FLIR ADAS ( p=0.47 ) but significant on FLAME~3 ( p=0.0078 ). These findings rest on a five-arm controlled attribution with temperature-evidential learning (TEL) as the testbed. TEL obtains a single evidence-derived routing scalar together with a complementary diagnostic entropy, and establishes a finite-evidence degeneracy theorem proving that the acquisition product degenerates to this scalar with a closed-form bound. A physical 0.5^\circ C threshold admits cross-dataset physical readings, whereas a gray threshold of 20 inverts without physical interpretation.
[CV-233] SWIFT-Cell: Segmentation Without Iterative Flow Tracking
链接: https://arxiv.org/abs/2610.33081
作者: Devansh Lalwani,Swapnil Bhat,Muralikrishna H H,Anand Reddy,Maulik Shah
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generalist segmenters such as Cellpose-SAM (CPSAM) outline cells accurately across stains but slowly; watershed detectors such as QuPath’s are fast but less accurate. In time-critical use, such as interactive retrieval of cells from a user-drawn region, CPSAM at the reference settings takes 58-303 s per measured 8192 x 8192 region on an RTX 4500 Ada GPU, beyond a seconds-scale interactive target. We distil CPSAM into SWIFT-Cell (SWIFT), a 0.63M-parameter U-Net trained only on CPSAM masks of 1,202 HE and IHC tiles. On three 8192 x 8192 development regions SWIFT takes 2.0-5.2 s on 4 CPU cores in int8, 2.5-5.1x faster than QuPath’s detector on the same cores, and 0.36-0.49 s on a GPU. Against human annotations on eight HE, IHC and ISH sets it keeps 95% of CPSAM’s panoptic quality (PQ) on PanNuke (0.593 vs. 0.627) and is level on NuCLS (0.546 vs. 0.540). On Ki-67 IHC masks scored to the outline centre it is 0.205 PQ above QuPath’s better mode, 0.079 below CPSAM and 0.027 below InstanSeg (CI includes 0; at the outer edge 0.008 below CPSAM); on 212 pathologist-outlined nuclei in Ki-67 fields of 12 untouched patients its recall is 0.925 (CPSAM 0.920, InstanSeg 0.901; paired difference CIs include 0); one method-blinded pathologist accepted its outlines in 96% of 100 fields (60 from two development tissue patients, 40 hematology; CPSAM 83%, QuPath 53%; tissue 100% for SWIFT and CPSAM); on ER/PR points from 12 patients its detection F1 is 0.882 (CPSAM 0.896, InstanSeg 0.857), and on 4-reader PD-L1 points 0.83 (CPSAM 0.85, readers 0.87). SWIFT is close to CPSAM in accuracy and faster than watershed detection.
[CV-234] Rethinking Video Storag e for AI Training: Breaking Cross-Frame Dependencies for Efficient Random Access
链接: https://arxiv.org/abs/2610.33079
作者: Zetian Song,Qize Yu,Yue Chen,Bohan Zhou,Bowen Ping,Yuqi Ye,Haocheng Tang,Haotian Zhang,Qizhe Wang,Shiyu Huang,Jiaqi Zhang,Siwei Ma,Wen Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注: 18 pages, including supplementary material. Project page: this https URL Code: this https URL
Abstract:Large-scale AI training increasingly relies on massive video datasets with random temporal sampling patterns. However, existing video storage paradigms are optimized for sequential playback and employ sophisticated inter-frame reference structures to maximize compression efficiency. When applied to the random-sampling workloads of AI training, these paradigms face substantial decoding-efficiency bottlenecks. Workarounds such as independent JPEG frames and short groups of pictures (GOPs) reduce decoding work at the cost of larger storage and transfer volumes. In this paper, we identify two sources of decode amplification: sequential reconstruction of unrelated frames and reconstruction of the reference chains required by sampled targets. We address both sources by co-designing the decoding schedule and video reference structure. Dependency-aware sparse decoding reconstructs only requested frames and their required references, skipping unrelated reconstruction without transcoding the source videos. Optional offline Dependency-Decoupled Random Access (DDRA) encoding further reduces reference reconstruction by restricting each interior frame to two bounding intra-coded anchors. Together, sparse decoding and DDRA retain inter-frame prediction while bounding reconstruction to at most three frames per requested target, independent of the intra period. Experiments on a large-scale robotics corpus show that, at comparable reconstruction quality, our method achieves 1.45\times the loading throughput of LeRobot v3 with 58.7% less media payload.
[CV-235] Evaluating Visual Language Models on Visual Grounding for Visually Impaired People
链接: https://arxiv.org/abs/2610.33078
作者: Fairooz Azim,Lydia Achour,Claudia Borg,Lina Rojas-Barahona
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Visual grounding (VG) in visual question answering (VQA) is a fundamental capability of Vision-Language Models (VLMs), enabling the detection of the most relevant region in an image to answer the question. This reflects the model’s capability of multimodal comprehension, which is crucial when interacting with visually impaired people (VIP). Indeed, visual grounding, when applied to assistive scenarios, is not merely a technical feature, but rather a trust and safety requirement. In this work, we study visual grounding in VLMs for visually impaired people. To this end, we assess open-source VLMs on visual grounding using the VizWiz-VQA-Grounding dataset and challenge, which contains real-life images taken by VIP. We propose VG-SHAP, which extends an existing explainability framework to provide visual grounding explanations, allowing the evaluation of black-box VLMs. Our results on the VizWiz-VQA-Grounding challenge show that despite generating correct answers, VLMs struggle to ground in the proper region in the image. However, these results are constrained by the quality of the segmentation model of our explainability framework and further research is necessary.
[CV-236] Diagnosing and Mitigating Evidence Misgrounding in Multimodal Knowledge Editing EMNLP2026
链接: https://arxiv.org/abs/2610.33075
作者: Zhiwu Wang,Yichen Wu,Jinqian Yang,Yizhou Yang,Renzhen Wang,Qian Zhao,Deyu Meng
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to Findings of EMNLP 2026
Abstract:Multimodal knowledge editing aims to update visually grounded facts in Multimodal Large Language Models (MLLMs) while preserving unrelated text and visual behaviors. However, standard answer-level metrics can hide evidence misgrounding: an edit may produce the target answer on the original image yet remain sensitive to incidental context rather than the intended visual evidence. To examine this issue in a controlled setting, we construct WaterBird-Edit, an analysis set that decouples bird habitat labels from background context and measures whether edited predictions remain stable when the target evidence is preserved but the background changes. Motivated by this diagnosis, we propose VICE (Visual In-Context Editing), a retrieval-based framework that stores edits in external memory and applies them through visual in-context demonstrations designed to expose task-relevant evidence. Experiments on three standard benchmarks show that VICE achieves strong lifelong editing performance, while results on WaterBird-Edit show that it substantially reduces sensitivity to the controlled background shifts. Code and dataset are available in this https URL.
[CV-237] AGQ: Action-Guided Ultra-Low-Bit Quantization for Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.33074
作者: Xuhong Qiu,Zhuoyu Wu,Qiawei Zheng,Wenhui Ou,Wenqi Fang,Raphaël C.-W. Phan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 3 figures, 3 tables. Code and experimental results: this https URL
Abstract:Vision-language-action (VLA) models enable robots to perform tasks specified through natural-language instructions, but their large weight footprint challenges deployment on memory-constrained onboard platforms. Ultra-low-bit quantization reduces this footprint, yet preserving task performance and controlling weight recovery costs remain challenging. We propose Action-Guided Quantization, which combines action sensitivity with a measured proxy for weight recovery cost to allocate matrix bitwidths under storage budgets. Per-channel calibration compensates for quantization error, while a uniform bitwidth within each matrix preserves regular layouts. Bit-dependent grouped lossless coding reduces resident weight storage, and block-wise recovery limits temporary buffers. On LIBERO, AGQ achieves 92.7% task success at W3, exceeding AWQ and GPTQ by 1.2 and 3.7 percentage points, respectively, and retains 70.5% success at W2.5. An asymmetric configuration, denoted W1.5 ^* , uses W4 vision weights and a parameter-weighted average of 1.5 bits for language weights. With lossless coding, it achieves 6.10\times complete-model compression relative to BF16 while retaining 25.5% task success. Code and experimental results are available at this https URL
[CV-238] VLMProbe: An Open-source Library for VLM Probing and Interpretation
链接: https://arxiv.org/abs/2610.33072
作者: Gexin Huang,Haisheng Fu,Yanting Yang,Chen Zhao,Ruinan Jin,Jiaqi Liu,Xiaoxiao Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 43
Abstract:Understanding how vision-language models use visual information requires connecting internal representations to model behavior. We present vlmprobe, an open-source library that integrates lens readouts, attention attribution, causal interventions, and sparse autoencoder (SAE) analysis within a common model interface. Shared adapters, calibration and checkpoint utilities, evaluation pipelines, and visualizations support systematic analysis across architectures. Using this framework, we compare methods across vision-language, language, and vision models through matched case studies and controlled evaluations of pretrained sparse dictionaries. The comparisons show that representation readouts and spatial explanations depend on the analysis objective and observation site, and that local attribution estimates can disagree with measured intervention effects. SAE evaluations further expose distinct trade-offs in reconstruction fidelity, concept readout, and spatial localization. Together, these results support combining multiple forms of evidence to evaluate an interpretation. vlmprobe provides a reusable basis for selecting methods, comparing their conclusions, and testing hypotheses about model behavior. The code is available on this http URL
[CV-239] DuoPIC: Uncertainty-Aware Voxel-Level Digital Twins of Cerebral Perfusion with Transport PDEs in Acute Stroke
链接: https://arxiv.org/abs/2610.33071
作者: Anh Duong Tran,Quang-Anh N. D.,Toan Sy Pham,Trang H. Hoang,Minh N. Do,Duy Ton Mai,Huan X. Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Digital twins of cerebral perfusion relate tracer dynamics to cerebral blood flow (CBF), blood volume (CBV) and mean transit time (MTT), but must be data-driven, physically consistent and uncertainty-aware. We introduce DuoPIC, a digital twin that needs no reference maps, in which a Windkessel flow model drives a tracer transport partial differential equation (PDE). A population-level neural closure divides each region into fast and slow pathways with tracer mass and MTT conserved by construction, and a tempered posterior quantifies the uncertainty of every estimate. Its voxel maps set the state of the art for lesion detection on ISLES’17 MR (AUC 0.745) and ISLES’24 CT (0.743). On held-out patients, the neural PDE reduced the compartment’s reconstruction error by 19% and lifted 95% MTT interval coverage from 88% to 94%. Code: this https URL
[CV-240] PhaseBreak: Proactive Temporal Protection against Unauthorized Image-to-Video Animation
链接: https://arxiv.org/abs/2610.33070
作者: Jiashu Zhang,Xuanyou Li,Qixin Zhang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Image-to-Video (I2V) diffusion models can animate a single portrait or photograph, enabling shared images to be repurposed for motion synthesis without the subject’s authorization. We study proactive visual protection against such use: before an image is released, its owner applies a bounded perturbation that preserves the sharing workflow but disrupts downstream motion generation. Existing spectral disruption objectives globally attenuate temporal representations and therefore provide limited control over subtle or spatially localized motion. Our key idea is that generated dynamics depend not only on temporal feature strength, but also on the order in which feature events occur. We propose PhaseBreak, a white-box spectral retiming protection method that matches intermediate denoiser features to a constrained temporal rearrangement of their clean counterparts. PhaseBreak operates along the true temporal axis and uses a subject-aware mask to confine the protective perturbation to the depicted subject. Evaluations on Stable Video Diffusion and DynamiCrafter show that protected images induce greater short-term motion variation and lower temporal regularity, impeding stable motion customization from the released image.
[CV-241] Decentralized Multi-View Multi-Zone Tracking for Scalable Camera Networks
链接: https://arxiv.org/abs/2610.33069
作者: David Voihanski,Mor Sinai,Ben Zion Bobrovsky
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We present a decentralized multi-view tracking system for scalable camera networks. Single-view detections are projected to bird’s-eye-view (BEV) coordinates and fused by independent multi-view trackers, each managing a dynamic spatial zone. To support scalability across spatially partitioned camera networks, we introduce a decentralized layer for trajectory unification, utilizing a locking protocol and periodic track-merging operations termed global collate. This process is supported by a delayed-output buffering mechanism that retroactively applies history corrections. The configurable delay window absorbs asynchronous coordination latency and helps stabilize cross-zone identity consistency. We detail the design of dynamic zone polygons, consensus-based coordination, and the validation framework, which together maintain global ID consistency with minimal accuracy loss compared to a single-instance baseline in the evaluated configurations.
[CV-242] OvisOCR3 Technical Report
链接: https://arxiv.org/abs/2610.33066
作者: Yu Xia,Yinglun Li,Yuhui Chen,Jinlong Huang,Yuyang Wanyan,Shengze Shi,Yifei Song,Yue Duan,Shubo Lin,Jianshan Zhao,En Lin,Shiyin Lu,Qing-Guo Chen,Zhao Xu,Weihua Luo
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:We introduce OvisOCR3, a 0.8B model for a broad range of OCR-related tasks. OvisOCR3 unifies document parsing, general OCR, key information extraction (KIE), document figure captioning, and cross-page table reconstruction in a single model. We train the model on real and synthetic data through continued pretraining and use domain-balanced mid-training to balance supervision across tasks with different data volumes and response lengths. Multi-teacher on-policy distillation transfers expertise from five domain-specific 4B teachers into the shared model, followed by mixed-task reinforcement learning with domain-specific rewards. We compare OvisOCR3 with general-purpose vision-language models and specialized OCR models on public and in-house benchmarks covering all five task families. In these comparisons, it achieves an overall score of 96.74 on OmniDocBench v1.6 and the highest overall scores on PureDocBench, Real5-OmniDocBench, and MDPBench. On OCRBench v2, it ranks first, scoring 65.9 and 61.3 on the English and Chinese splits, respectively. It also outperforms the baselines on OCRBench and all four KIE benchmarks. On DocFigCapBench, it achieves the highest localization F1 and a caption score close to that of Qwen3.8-Max. For cross-page tables, it leads in Relation F1 on CrossTabBench and truncated-table TEDS on MPDocBench-Parse. These results demonstrate that a compact 0.8B model can achieve strong performance across diverse OCR-related tasks. OvisOCR3 is available at this https URL.
[CV-243] ProtoEEG: Structured Prototype Alignment for EEG Visual Decoding
链接: https://arxiv.org/abs/2610.33065
作者: Kieren Yu,Chang Huang,Ziyang Liu,Kaishun Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:EEG visual decoding remains difficult because scalp recordings are noisy, spatially coarse, and variable across trials and subjects. Most systems align each trial with a dense image embedding, although that embedding mixes semantic content with instance-specific texture, pose, and background. This target can demand details absent from EEG and provides no explicit account of the visual factors supporting a prediction. We propose \textbfProtoEEG, a prototype-mediated framework that reformulates EEG–image alignment as soft composition over a shared learnable visual codebook. A trainable EEG encoder and a frozen visual teacher produce assignment distributions over the same K=64 prototypes. Their weighted reconstructions form a structured representation for retrieval and image generation. Prototype contrast supplies the main cross-modal training signal, distribution alignment coordinates codebook use across modalities, an instance anchor preserves fine-grained discrimination, and diversity regularization prevents redundant prototypes. The shared assignment profile also makes the learned bottleneck directly testable through prototype retention, random selection, and identity shuffling. Across five seeds, ProtoEEG reaches 64.0% Top-1 and 91.5% Top-5 on THINGS-EEG2 200-way retrieval. After retraining on EEG-ImageNet, it obtains 44.8% Top-1 and 73.8% Top-5 in 80-way retrieval. ProtoEEG also ranks first on all seven reconstruction metrics. Removing prototype contrast lowers Top-1 by 56.9 percentage points, while shuffling prototype identities reduces it to 0.5%. Prototype analyses further show distributed assignments and broad codebook use rather than one-prototype routing. These results support structured alignment targets as a practical alternative to increasing encoder or teacher capacity alone.
[CV-244] A Canary in the Neural Mine: Catching an Image Classifiers Silent Confident Failures
链接: https://arxiv.org/abs/2610.33063
作者: Wonseok Shin,Songkuk Kim
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Deployed vision models auto-accept confident predictions, so high-confidence errors propagate silently. Maximum Softmax Probability (MSP), the standard failure predictor, has a structural blind spot: as a scalar read from the classifier head, it is first-order blind to feature variation orthogonal to its confidence gradient. Through a factorial analysis of feature-space detectors, we identify a conditional-redundancy principle explaining why they fail to complement MSP: the class-separability geometry they measure is already encoded in MSP and collapses under fusion. The signal that survives is class-conditional typicality on the off-manifold residual—and because this residual’s distribution varies by backbone, from near-Gaussian to heavy-tailed, it must be estimated non-parametrically. We instantiate these findings as CANARY, a training-free detector that computes a per-class kNN distance on the PCA residual of the penultimate feature, percentile-calibrates it on validation data, and fuses it with MSP through a single validation-tuned weight. CANARY’s benefit follows a headroom law: gains grow as baseline MSP quality falls, so a single clean-validation number forecasts the improvement. With every parameter selected on in-distribution data alone, CANARY matches or improves 382 of 385 shift cells (e-AURC; 379 on AUROC) across five backbones, with worst-case regression within seed noise ( -0.0002 AUROC) and the largest gains on natural distribution shifts (ImageNet-A/R, ObjectNet)—where deployed models are most exposed.
[CV-245] Oddin: Latent Visual Reasoning in Unified Multimodal Models via Self-Distillation from Image Generation
链接: https://arxiv.org/abs/2610.33062
作者: Jaeyong Ko,Hankyeol Kim,Pilsung Kang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Unified large multimodal models (ULMMs) are able to perform visual reasoning through interleaved text-image generation, but generating explicit intermediate images incurs significant computational overhead. While visual latent reasoning addresses this, most existing methods rely on vision-language models (VLMs) incapable of visual generation. We propose Oddin, a novel framework enabling latent visual reasoning in ULMMs. Oddin replaces intermediate images with compact latent tokens processed entirely by the understanding branch, reserving the generation branch for the final output image. We train the model via offline self-distillation with three objectives, final image generation losses, text distillation losses, and cross-entropy losses for special tokens. Oddin surpasses interleaved generation baselines and closed-source VLMs on average across seven VBVR-Pro visual reasoning tasks. Qualitative analysis reveals that latent tokens attend to relevant regions in an input image, demonstrating that ULMMs can perform complex visual reasoning without generating intermediate images.
[CV-246] MARC: Mechanism-Aware Reflective Correction for Articulated Object Code Generation
链接: https://arxiv.org/abs/2610.33060
作者: Xiong Yang,Junkun Hao,Zijian Xiao,Zipeng Ye,Yuchen Xie,Ran Yi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Generating an articulated object from images and text requires more than plausible parts and movable joints: the assembled mechanism must execute the intended motion. This is particularly challenging for closed-chain and coupled mechanisms, where an error in one connection can block motion or separate connected parts. We introduce MARC, Mechanism-Aware Reflective Correction, which shares an executable kinematic graph across program construction, motion solving, and correction. The graph binds physical connections to local attachment frames and represents loop closures and joint-coordinate couplings. Given prescribed driver inputs, a numerical solver computes follower motion and associates errors with specific connections and configurations. The agent uses these diagnostics to probe possible causes, revise connections or geometry, and retest the mechanism along the original motion paths. On a 20-object test set, MARC with Kimi-K3 achieves 67.8% valid motion coverage and 60.0% overall success, outperforming six baselines. Starting from the same initial candidates, correction improves these scores by 13.7 and 20.0 percentage points, respectively. Ablations further show the benefits of graph-based mechanism support and graph-guided repair. Project homepage: this https URL.
[CV-247] MOSAIC: Towards a Universal Latent Space for Computational Pathology Foundation Models WACV2027
链接: https://arxiv.org/abs/2610.33059
作者: Dibaloke Chanda,Praveen Madiraju
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at WACV 2027
Abstract:Pathology foundation models (PFMs) differ substantially in architecture, pretraining objective, training data and representation dimensionality, yet it remains unclear whether they learn fundamentally different feature spaces or alternative coordinate systems of a shared underlying representation. We present MOSAIC, a systematic study of latent-space similarity, alignment and interoperability across 12 patch-level and 6 slide-level PFMs. We first quantify representational agreement using complementary similarity measures and analyze how architecture, pretraining objective, network depth and magnification influence latent geometry. We then evaluate multiple alignment strategies for mapping independently learned representations into a common latent space. The resulting representations are evaluated through cross-model agreement, retrieval, geometry preservation, and downstream multiple-instance learning across molecular and morphological prediction tasks. MOSAIC provides a unified framework for studying whether heterogeneous PFMs can move beyond representational similarity toward a shared and functionally transferable pathology representation space. Our code, benchmarks and additional results are available at this https URL.
[CV-248] KDHNet: A Knowledge-Driven Hybrid Network for Limited-Calibration SSVEP Decoding
链接: https://arxiv.org/abs/2610.33058
作者: Zhentao He,Ziwei Wang,Dongrui Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 12 pages, 7 figures
Abstract:Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) achieve high information transfer rates but often require sufficient subject-specific calibration data for reliable decoding. Analytical spatial filters are effective with limited data, but their linear formulations constrain representation capacity. Deep models learn more flexible representations but are prone to overfitting when calibration data are scarce. We propose KDHNet, which combines knowledge-driven analytical processing with data-driven representation learning. After filter-bank preprocessing, cascaded task-discriminant component analysis (TDCA) and task-related component analysis (TRCA) extract discriminative and trial-reproducible components. Learnable sub-band weighting then fuses spectral information, and an SSVEPformer decoder learns temporal and spectral representations for classification. For inter-subject decoding, a target-guided least-squares transformation aligns source trials using one labeled target trial per class. Extensive experiments were performed on three public datasets, covering intra- and inter-subject protocols, five analysis windows, and multiple calibration ratios. In intra-subject evaluation, KDHNet achieved the highest mean accuracy and information transfer rate in 14 of 15 window settings and all ten calibration-ratio settings. In inter-subject evaluation, KDHNet obtained the best or second-best performance, and the gains were generally larger with less calibration data. Ablation studies indicated that the estimated TDCA projection performed better when fixed than when optimized from analytical or random initialization. Combining fixed knowledge priors with trainable neural modules is effective for low-calibration SSVEP decoding. This framework supports more practical, robust, and plug-and-play BCIs. Code is available at this https URL.
[CV-249] DRA-SAM: Disagreement-Driven Referential Active SAM for Low-Altitude Drone Referring Segmentation
链接: https://arxiv.org/abs/2610.33057
作者: Jisheng Dang,Huaiyuan Peng,Wencan Zhang,Bimei Wang,Jingwen Zhao,Wenjun Ma,Hong Peng,Bin Hu,Qi Tian,Tat-Seng Chua
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 10 pages, 5 figures
Abstract:Referring segmentation in low-altitude drone imagery requires distinguishing small, visually similar instances under complex linguistic and spatial constraints. Global inference may yield multiple plausible masks or leave the intended referent insufficiently represented, motivating selective local re-inference. We propose DRA-SAM, a disagreement-driven framework that allocates an additional inference pass according to competing referential hypotheses and an unresolved state. A lightweight EvidenceScorer estimates beliefs over existing candidates and the unresolved state. A training-free Referential Disagreement LOOK policy balances belief-weighted candidate competition and unresolved-region coverage to select a crop for local re-inference. The same benchmark-adapted SAM3-I backbone processes the selected region, and the resulting evidence updates existing hypotheses and introduces new candidates. Among the compared methods, DRA-SAM achieves state-of-the-art performance on six of seven test metrics on both RIS-LAD and UAVid-RIS, including mIoU and P@0.5-P@0.9, and on all seven RIS-LAD validation metrics. On RIS-LAD, DRA-SAM improves the matched global SAM3-I baseline from 49.55 to 52.34 mIoU and from 46.80 to 58.05 P@0.5. On UAVid-RIS, the independently adapted DRA-SAM achieves 73.26 mIoU and 86.86 P@0.5. Controlled crop-policy comparisons demonstrate the effectiveness of disagreement-guided crop selection, while target-size analysis reveals particularly pronounced gains for aerial targets occupying less than 1% of an image. These results establish referential disagreement as an effective criterion for allocating local inference. Code is available at this https URL.
[CV-250] Aerial-R1: Reinforcing Aerial Reasoning and Segmentation in MLLM s
链接: https://arxiv.org/abs/2610.33056
作者: Jisheng Dang,Huaiyuan Peng,Yixuan Wang,Kun Wang,Zhang Li,Tat-Seng Chua
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 16 pages, 8 figures, including supplementary material
Abstract:While Multimodal Large Language Models (MLLMs) have revolutionized general visual perception, existing methods suffer from severe hallucination in the aerial domain, where extreme scale variations and visual ambiguity abound. Existing Supervised Fine-Tuning (SFT) paradigms force models to mimic ground truth without modeling uncertainty, often resulting in blind guessing on tiny or unrecognizable objects in aerial scenes. In this paper, we present Aerial-R1, the first aerial MLLM framework to reinforce visual reasoning capabilities for aerial reasoning and segmentation. Drawing inspiration from the reasoning-incentivizing mechanisms of DeepSeek-R1, we find that trustworthy perception stems from consistency across views, establishing a visual parallel to the Chain-of-Thought in logic. To effectively adapt policy optimization for non-autoregressive mask generation, we introduce Hybrid-View Group Advantage-Weighted Regression (H-GAWR). This strategy constructs dynamic groups of high-resolution and low-resolution views, seamlessly bridging group-relative advantage estimation with advantage-weighted policy extraction. By enforcing consensus on clear inputs and strictly penalizing hallucinations on ambiguous ones, Aerial-R1 develops an explicitly incentivized refusal mechanism. Extensive experiments on VSAI-Ref, the first aerial MLLM benchmark dedicated to jointly evaluating dense reasoning and hallucination robustness, demonstrate that Aerial-R1 significantly outperforms SFT baselines in segmentation precision while substantially mitigating fatal hallucinations in low-information scenarios, paving the way for more reliable and trustworthy aerial agents. Codes are available on this https URL.
[CV-251] GMoE-FTI: A Universal Facial Template Inversion Method Adopting Genuine Mixture-of-Experts Mechanism
链接: https://arxiv.org/abs/2610.33055
作者: Zixuan Shen,Zhihua Xia,Kaikai Gan,Peipeng Yu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Face recognition systems have been ubiquitously deployed across daily scenarios. To protect user privacy, mainstream systems store facial templates rather than raw face images for identity verification. However, facial template inversion (FTI) technologies can reconstruct faces from templates, which endangers biometric privacy. Existing FTI methods only fit single type of templates, failing to reconstruct identity-consistent faces from unseen types. To tackle this limitation, we survey mainstream template extractors which can be classified into five categories according the template distribution: unit-sphere, angle-linear, strict unit-sphere, cone-shaped, anuniform hypersphere. Inspired by this observation, this paper proposes GMoE-FTI, a universal FTI framework powered by a genuine MoE mechanism that differs from load-balanced multi-branch alternatives. It follows a three-stage training strategy: (1) independently train five mapping experts (MEs) and a matching module (MM); (2) conduct collaborative inference by weighting each ME’s output with MM’s output, so as to generate unified latent codes compatible with pre-trained StyleGAN3; (3) perform end-to-end joint fine-tuning over all modules to identity consistency and boost cross-expert coordination. Extensive evaluations on the LFW benchmark validate the superiority of GMoE-FTI against SOTA methods. Evaluated on seven prevailing unseen template types including AdaFace, CosFace, CurricularFace, ElasticFace, SphereFace2, SphereFace-R, and TransFace, GMoE-FTI achieves an average 40% TAR improvement, with average FAPD and FAPC each reduced by 2%, verifying its universal generalization across heterogeneous templates.
[CV-252] Saga: an in silico ontology for a multimodal patient world model
链接: https://arxiv.org/abs/2610.33054
作者: Youcheng Li(Peking University, Isoplex Lab),Fa Tian(Isoplex Lab),Binbin Su(Peking Union Medical College, Isoplex Lab)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 6 figures, 4 tables
Abstract:Simulating an individual’s disease course requires connecting patient history to future events, clinical timing and disease morphology. Longitudinal prediction and multimodal representations exploit this information, but organizing future readouts around a common patient state and advancing trajectories without real intermediate observations remain central challenges. Here we introduce Saga, a multimodal patient world model that learns a shared state from patient histories to predict clinical events and generate multimodal images. Generated events and times are appended to the history and re-encoded to advance prediction. We studied 461,370 patients with 12 cancer types across 112 centres. Modality-specific paired evaluations linked future images to observed follow-up. With historical images held fixed, removing patient conditioning reduced event F1 from 0.716 to 0.591 and increased lesion-burden error from 11.8% to 21.3%. In free rollouts, eight-step event F1 was 0.672, compared with 0.553 for a longitudinal Transformer evaluated from the same origins, and timing error was lower by 8.7 days. In the internal test set, one-year progression AUROC was 0.882, with values of 0.871 and 0.856 in temporal and geographical external evaluations. These results show that shared patient states connect future phenotypes across observational scales with event-driven trajectory progression, sustaining comparative predictive advantages during rollout and discrimination in external clinical evaluation. Saga makes patient-level disease simulation a multimodal prediction task that can be tested against observed follow-up.
[CV-253] EKT-Router: Expert-Evidence Alignment for Embodied Vision-Language Models
链接: https://arxiv.org/abs/2610.33044
作者: Weilin Ruan,Yingjie Xu,Yangkai Wei,Bohan Zhou,Shuaibo Li,Harold Haodong Chen,Kaiwen Zhou,Yinchuan Li,Ying-Cong Chen,James Cheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages, 5 figures
Abstract:Embodied vision-language models must answer questions from partial, evolving observations. Their success depends on both the reasoning capability selected and the evidence available to it. Expert routers adapt internal computation, while tool-augmented systems acquire external evidence; choosing these independently can leave a suitable expert without the evidence it needs. We introduce EKT-Router, a bounded framework that jointly selects a capability expert, optional knowledge, and perceptual tools. Outcome-derived signatures represent the overlapping strengths of General, Spatial Reasoning, and Spatiotemporal experts. A joint utility scores complete expert-evidence routes under a call budget. After acquisition, a packet verifier admits valid evidence with positive predicted expert-conditioned gain. The selected expert then answers once, reverting to the unchanged observation if all packets are rejected. Across 6 embodied QA benchmarks, EKT-Router achieves a mean score of 61.4, the highest among the compared systems, and averages 74.7 on 10 general vision-language benchmarks. With all external evidence disabled, question-conditioned expert selection alone improves over the best fixed expert by 1.75 points. Ablations show that learned routing is the dominant component and that knowledge and perception are complementary.
[CV-254] Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI
链接: https://arxiv.org/abs/2610.33039
作者: Ganesh Ajjanagadde,Chris Alexiuk,Luke Alonso,Kumar Anshuman,Maor Ashkenazi,Nave Assaf,Gorkem Batmaz,Omer Belhasin,Shai Ben Nun,Akhiad Bercovich,Pritam Biswas,Caroline Blanco,Tijmen Blankevoort,Michael Boone,Christopher Borden,Kari Briski,Bryan Catanzaro,Zijia Chen,Aishwarya Chhabra,Hung-Yueh Chiang,Annamalai Chockalingam,Jonathan Cohen,Joey Conway,Olivier Delalleau,Gerardo Delgado Cabrera,Leon Derczynski,Li Ding,Will Efstratis,Michael Elad,Ellie Evans,Wes Feely,Michael Fukuyama,Gaurav Garg,Rishabh Garg,Arvind Gopalakrishnan,Shishir Goyal,Kartik Gulia,Hao Guo,Negar Habibi,Scott Hara,Ehsan Hosseini Asl,Rohit Jena,Udi Karpas,Bilal Kartal,Mayank Kaushik,Manoj Kilaru,Mackenzie Kosseim,Oleksii Kuchaiev,Matthieu Le,Raymond Lee,Jian Leong,Jonathan Li,Sheldon Liang,Mingjie Liu,Eileen Long,Katherine Luna,Chenjie Luo,Mingyuan Ma,Arsh Maknojia,Richard Mazzarese,Jessica Medina,Puhui Meng,Yoav Miron,Pavlo Molchanov,David Mosallanezhad,Chetan Mungekar,Marina Neseem,Oluwatobi Olabiyi,Jason Paul,Bhavesh Pawar,Seth Poulos,Anav Prasad,Meredith Price,Sweta Priyadarshi,Haifeng Qian,Vinay Raman,Kaustubh Rao,Hayley Ross,Matan Rusanovsky,Sudeep Sabnis,Karan Sapra,Micah Schaffer,Schmitty,Alex Settle,Ashton Sharabiani,Haran Kumar Shiv Kumar,Rameshwar Shivbhakta,Guillermo Siman,Barnaby Simkin,Oliver Simons,Andrii Skliar,Padmavathy Subramanian,Yoshi Suhara,Ali Taghibakhshi,Nima Tajbakhsh,Andrew Tao,Roy Uziel,Prasoon Varshney,Jose Vergara,Huy Vu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We introduce Nemotron 3.5 Bolt 74B-A9B 24GB, a hybrid Mamba-Transformer vision-language model that optimizes speed and intelligence per GB of memory used for local agentic workloads. Derived from the Nemotron 3.5 Super VL 120B-A12B model family, Nemotron 3.5 Bolt 74B-A9B 24GB combines Star Elastic compression with 2-bit Quantization-Aware Distillation for memory-constrained deployment. The model is optimized for reasoning, agentic tool use, and web terminal interactions. We also trained a LiLiCorr speculative drafter that is released alongside the model to reach our target inference speed. The resulting model achieves a higher accuracy than leading models with similar memory footprints and speed across many benchmarks for general agentic capabilities (e.g., GDPVal v2), web browsing (e.g., BrowseComp), and terminal use (e.g., Terminal-Bench 2.1), with a relatively high TPS of 84 on RTX Spark that is comparable to models with much lower active parameter counts and more than sufficient for long-horizon agentic applications.
[CV-255] Intuition with Evidence: Adaptive Dual Memory for Multimodal Agents
链接: https://arxiv.org/abs/2610.33019
作者: Qiuhe Hong,Jingya Wang,Cuzhen Geng,Shenyao Peng,Munan Ning,Yuyang Liu,Yonghong Tian
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:As multimodal agents engage in sustained interactions, long-term memory serves two distinct but complementary functions: fine-grained evidence recall and internalized interpretation for global understanding. Although episodic and latent memory address these demands through explicit historical records and internalized experience, neither paradigm alone can accommodate the diverse memory affinities across tasks, owing to limitations in episodic evidence coverage and integration, and in latent factual grounding. To bridge this divide, we propose a dual-memory framework, unifying a model-native latent memory as an intuitive prior with a multi-indexed episodic memory for exact evidence reconstruction via retrieval-conditioned joint alignment. At inference, the framework adaptively composes these memories through a training-free routing mechanism driven by Jensen-Shannon Divergence (JSD) sensitivity and confidence proxies for query-dependent memory affinity. Extensive evaluations across multimodal benchmarks demonstrate state-of-the-art performance, outperforming single-paradigm baselines overall (+4.67% over vanilla Episodic), with further analysis highlighting episodic strengths in evidence-localized tasks and latent contributions to integrating dispersed information and compensating for imperfect retrieval.
[CV-256] Belief-Aware VLM Model for Human-like Reasoning
链接: https://arxiv.org/abs/2604.09686
作者: Anshul Nayak,Shahil Shaik,Yue Wang
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Systems and Control (eess.SY)
备注: Accepted for publication at the IEEE International Conference on Robot and Human Interactive Communication (RO-MAN 2026). 6 pages, 3 figures, 1 table
Abstract:Traditional neural network models for intent inference rely heavily on observable states and struggle to generalize across diverse tasks and dynamic environments. Recent advances in Vision Language Models (VLMs) and Vision Language Action (VLA) models introduce common-sense reasoning through large-scale multimodal pretraining, enabling zero-shot performance across tasks. However, these models still lack explicit mechanisms to represent and update belief, limiting their ability to reason like humans or capture the evolving human intent over long-horizon. To address this, we propose a belief-aware VLM framework that integrates retrieval-based memory and reinforcement learning. Instead of learning an explicit belief model, we approximate belief using a vector-based memory that retrieves relevant multimodal context, which is incorporated into the VLM for reasoning. We further refine decision-making using a reinforcement learning policy over the VLM latent space. We evaluate our approach on publicly available VQA datasets such as HD-EPIC and demonstrate consistent improvements over zero-shot baselines, highlighting the importance of belief-aware reasoning.
[CV-257] Optimal Projections for Discriminative Dictionary Learning using the JL-lemma
链接: https://arxiv.org/abs/2308.13991
作者: G.Madhuri,Atul Negi,Kaluri V.Rangarao
类目: Computer Vision and Pattern Recognition (cs.CV); Emerging Technologies (cs.ET); Machine Learning (cs.LG); Performance (cs.PF); Signal Processing (eess.SP)
备注:
Abstract:Dimensionality reduction-based dictionary learning methods in the literature have often used iterative random projections. The dimensionality of such a random projection matrix is a random number that might not lead to a separable subspace structure in the transformed space. The convergence of such methods highly depends on the initial seed values used. Also, gradient descent-based updates might result in local minima. This paper proposes a constructive approach to derandomize the projection matrix using the Johnson-Lindenstrauss lemma. Rather than reducing dimensionality via random projections, a projection matrix derived from the proposed Modified Supervised PC analysis is used. A heuristic is proposed to decide the data perturbation levels and the dictionary atom’s corresponding suitable description length. The projection matrix is derived in a single step, provides maximum feature-label consistency of the transformed space, and preserves the geometry of the original data. The projection matrix thus constructed is proved to be a JL-embedding. Despite confusing classes in the OCR datasets, the dictionary trained in the transformed space generates discriminative sparse coefficients with reduced complexity. Empirical study demonstrates that the proposed method performs well even when the number of classes and dimensionality increase. Experimentation on OCR and face recognition datasets shows better classification performance than other algorithms.
[CV-258] MedVTok: A General-Purpose Medical Visual Tokenizer NEURIPS2026
链接: https://arxiv.org/abs/2610.35842
作者: Chenglong Ma,Yuanfeng Ji,Junzhi Ning,Jiyao Liu,Yujie Wei,Shaohao Rui,Chaoyang Zhang,Wenjie Li,Wenhao Tang,Cheng Tang,Wei Li,Shujian Gao,Yujie Zhang,Ying Chen,Zilong Li,Zongyuan Ge,Diping Song,Jin Ye,Ming Hu,Junjun He,Hongming Shan
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: NeurIPS 2026
Abstract:Multimodal medical AI requires a common visual interface for encoding heterogeneous imaging evidence for patient condition modeling. However, existing medical visual tokenizers are often tied to a specific dimension and imaging modality, forcing multimodal systems to rely on fragmented and poorly reusable representations. We introduce MedVTok, a general-purpose visual tokenizer that maps 2D images and 3D volumes from diverse imaging modalities into a unified token space. Building such a tokenizer poses two key challenges: preserving anatomical structure across dimensions and capturing clinical semantics across modalities. To this end, MedVTok introduces (i) slice differential regularization, which explicitly models inter-slice anatomical coherence missing in previous per-pixel/per-voxel supervision, and (ii) multi-expert representation alignment, which integrates knowledge from multiple medical expert encoders while retaining modality-specific complementary knowledge. Trained on 28M images and 70K volumes across 9 imaging modalities, MedVTok supports classification, retrieval, segmentation, synthesis, visual question answering, and medical report generation through a single tokenizer. Across 24 benchmarks, MedVTok achieves state-of-the-art performance, demonstrating a scalable token interface that unifies medical image dimension, modality, and task usage.
[CV-259] Accuracy and efficiency of compact foundation models for surface-water mapping during floods
链接: https://arxiv.org/abs/2610.34904
作者: Parastoo Farajpoor,Mohammadreza Narimani
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Applications (stat.AP)
备注: 19 pages, 8 figures, 7 tables. Code: this https URL Data: this https URL
Abstract:Pretrained Earth observation models may reduce the effort needed to map surface water, but their value depends on the alternative against which they are compared. We evaluated Prithvi-EO-2.0-Tiny using the 446 hand-labeled Sentinel-2 chips from 11 flood events in Sen1Floods11. A fine-tuned model, a frozen-encoder model, and an identical randomly initialized architecture were compared with a similarly sized U-Net, a larger Prithvi model, and spectral-index thresholds. Experiments varied the number of training chips, excluded events from model fitting, and measured computational cost and prediction reliability. With 252 training chips, U-Net achieved a water intersection-over-union (IoU) of 0.856 on the official test set, compared with 0.833 for fine-tuned Prithvi-Tiny. Pretraining improved the tiny transformer over random initialization by a reported 0.046 IoU with 25 training chips and 0.023 with 252 chips, but did not reverse the U-Net ranking. Across 11 event-exclusion evaluations, mean IoU was 0.768 for fine-tuned Prithvi-Tiny and 0.777 for U-Net; the paired interval included zero. Fine-tuned Prithvi-Tiny processed a 512x512-pixel chip in 16.1 ms on a GPU and 258 ms on a laptop CPU, compared with 30.6 and 665 ms for U-Net. Errors were concentrated near water boundaries, and high predictive entropy identified many, but not all, erroneous pixels. These results support a specific role for compact pretrained models: lower measured computational demand with competitive accuracy, rather than consistent superiority. Event-dependent performance, shared hyperparameter selection, and the distinction between observed water and newly inundated land delimit the conclusions.
[CV-260] Semantic-Guided Frequency Modulation for Learned Image Compression
链接: https://arxiv.org/abs/2610.33616
作者: Gangyi Tian,Wen Ji,Alireza Furutanpey,Mingxuan Li
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:High-frequency attenuation can reduce the rate of learned image codecs but also remove object boundaries and fine detail. We propose Semantic-Guided Adaptive Latent-Frequency Modulation (SG-AFM), which uses a semantic structural prior to guide attenuation of encoder latents. SG-AFM applies a Haar wavelet transform and combines a fixed semantic-frequency multiplier with a bounded learned correction to modulate high-frequency subbands. The semantic prior is computed only at the encoder. No semantic map is transmitted, and no decoder modules are added. Across six datasets, SG-AFM reduces PSNR-based BD-rate relative to TCM by an average of 13.17%. It achieves the best BD-rate among the compared methods on ExDark, NightCity, and OnePlus, with a mean reduction of 16.80%. An ablation on OnePlus demonstrates the benefit of spatially aligned semantic guidance, and visual comparisons illustrate improved reconstruction of local structures.
[CV-261] Cross-View and Cross-Time Tracking of Crack Evolution in Concrete
链接: https://arxiv.org/abs/2610.33550
作者: Pengru Deng,Jiapeng Yao,Su Wang,Xuhui He,Zhao Wang,Takanori Emaru,Takashi Matsumoto,Xinrun Li
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 50 pages. Author manuscript
Abstract:Crack propagation provides richer information than static crack observations for evaluating structural conditions of concrete infrastructure and planning maintenance strategies. Unlike static crack detection, which can now be largely automated using computer vision, vision-based crack tracking is a coupled cross-view and cross-time problem: the same physical crack must retain its identity as camera viewpoints change, while newly developed growth must be distinguished from existing crack segments in a unified reference view. Existing segmentation and 3D reconstruction methods can recover crack masks or scene geometry, but they do not preserve crack identity across views or quantify temporal extension. To address this gap, this study formulates crack growth monitoring as a cross-view and cross-time crack-instance tracking task and proposes a geometry-assisted, morphology-aware framework. The framework first uses image-based geometric registration to align observations from different viewpoints into a shared structural reference, then applies morphology-aware instantiation to assign physically interpretable crack identities, and finally performs temporal mapping in the unified view to distinguish newly developed growth from existing crack segments. To enable evaluation, a temporally annotated dataset was established and used, covering three representative cracking scenarios: single dominant crack propagation, mixed-mode branching, and multiple-crack interaction, with pixel-level masks and cross-time instance labels. Experimental results show that the proposed framework can establish loose cross-view identity correspondence under changing viewpoints, while strict spatial overlap and growth localization remain limited by early micro-crack segmentability and projection drift, highlighting directions for future crack-monitoring research.
[CV-262] Matching Particle-Arrival Rates for Low-Count Image Denoising
链接: https://arxiv.org/abs/2610.33136
作者: Huazhen Liu,Zipei Yan,Jizhou Li
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Low-count imaging arises when limited illumination or dose yields only a few detected particles in microscopy, such as photons or electrons, causing measurements to be dominated by signal-dependent Poisson shot noise. Supervised denoisers typically require paired clean and noisy observations, whereas many self-supervised approaches exploit spatial or temporal redundancy without directly estimating the underlying particle-arrival process. We introduce PARM, a self-supervised denoising framework based on particle-arrival-rate matching. By modeling particle accumulation over exposure as a Poisson counting process, PARM identifies the conditional particle-arrival-rate field with the posterior mean of the latent clean rate. Poisson thinning then splits each observation into complementary lower-exposure views that provide self-supervised rate targets for one another. In implementation, the network estimates the normalized spatial allocation of the conditional rate, while the observed total count determines its global particle mass. Experiments on synthetic data, natural-video sequences corrupted with Poisson noise, and real low-dose electron microscopy demonstrate leading or competitive performance while effectively preserving fine structures. PARM therefore provides a statistically grounded approach to low-count image denoising when clean references are unavailable.
[CV-263] Investigating the trackability of landmarks in NBI and WLI
链接: https://arxiv.org/abs/2610.33125
作者: Jonas Below,Piermarco Gerli,Anna Jötten,Dennis Eggert,Axelle Felicio-Briegel,Veronika Volgger,Christian Betz,Alexander Schlaefer
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: Accecpted at CURAC conference 2026
Abstract:Reliable visual tracking of anatomical landmarks in endoscopic video is a key prerequisite for computer-assisted interventions in laryngeal surgery. While Narrow Band Imaging (NBI) is clinically established for enhancing vascular contrast, its effect on automated tracking relative to conventional White Light Imaging (WLI) remains unexplored. We present a quantitative comparison of landmark trackability in NBI and WLI using a forward-backward error framework applied to 61 endoscopic videos from 39 patients. Two regions of interest are evaluated: the vocal folds (VF) and mucosa with a prominent vascular pattern (MPVP). We compare a classical feature-based tracker (CSRT) against learning-based point trackers (CoTracker3-Online and TrackOnR) across both modalities. NBI consistently reduced the forward-backward error in both regions, reaching statistical significance for the vocal folds, while failure rates stayed low. These findings indicate that NBI can improve tracking robustness in laryngeal endoscopy, with implications for motion analysis, navigation, and robotic assistance.
[CV-264] Spectrally optimized implicit neural representations mitigate acquisition-related image quality trade-offs in intracranial 4D Flow MRI
链接: https://arxiv.org/abs/2610.33124
作者: Oliver Welin Odeback,Nerea González-Aranceta,Sebastián Jofre,Javier Bisbal,Edward Ferdian,Alistair A. Young,Pia Callmer,Sameer A. Ansari,Mia Bonini,David Nordsletten,Patrick Winter,Susanne Schnell,Tobias Granberg,Alexander Fyrdahl,David Marlevi
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 36 pages, 7 main figures, 2 supplementary figures. Under review at Medical Image Analysis
Abstract:4D Flow MRI promises opportunities for non-invasive mapping of neurovascular hemodynamics, but its clinical utility faces trade-offs between resolution, acquisition time, and noise. Deep learning-based super-resolution offers a post-scan solution, but supervised approaches require paired or domain-specific data, while patient-specific physics-informed models can be computationally demanding. This creates a need for unsupervised reconstruction models that can adapt to individual acquisitions without paired high-resolution training data, while remaining robust across anatomy, disease severity, and acquisition quality. To address this, we introduce a spectrally tuned implicit neural representation (INR) framework for intracranial 4D Flow MRI, representing the velocity field as a continuous coordinate-based neural field using WIRE activation functions with tunable spectral bias. Data-driven and physics-informed variants were evaluated in nine cerebrovascular cases spanning healthy controls and moderate-to-severe intracranial atherosclerotic disease, using CFD-derived high-resolution reference fields, MR-realistic synthetic low-resolution inputs, and direct fitting to acquired in-vivo data. Cohort-optimal spectral-bias settings generalized across subjects with only small increases in relative error compared with subject-specific optima, reducing the need for case-specific hyperparameter tuning. Physics-informed regularization provided modest additional accuracy but required substantially longer optimization, whereas the data-driven INR achieved near-comparable reconstruction performance within minutes. The framework also remained robust to acquisition-realistic noise and moderate VENC-dependent aliasing. These findings establish spectrally optimized INRs as a flexible unsupervised approach for mitigating acquisition-related trade-offs in intracranial 4D Flow MRI.
人工智能
[AI-0] Retriever-Free Retrieval: Navigating Documents with Decision Models
链接: https://arxiv.org/abs/2610.35868
作者: Chao Wan,Paul Jünger,Beixi Du,Bodong Liu,Navaneetha Krishnan,Kilian Q. Weinberger
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:While language models have become dramatically more capable, retrieval still largely relies on lexical matching or similarity in fixed embedding spaces. This creates a capability asymmetry: what the retriever fails to find, even the strongest generator cannot recover. We propose DecisionRAG, which brings the capabilities of pretrained language models to retrieval using decision models that jointly evaluate candidate branches in a single forward pass, without generating any tokens, and show that judging candidates in a shared context makes their relevance estimates comparable. Like a person browsing a library, DecisionRAG navigates the corpus’s inherent structure from broad topics down to individual passages, deciding at each level which branches are worth exploring. Without task-specific training or a first-stage retriever, DecisionRAG outperforms dense retrieval with an 8B reranker by 13.8 R@1 points on MultiDoc2Dial and 17.4 Hit@1 points on Natural Questions. Across four benchmarks, including a corpus of 2.68 million Wikipedia paragraphs, retrieval takes just 0.7-1.8 seconds per query.
[AI-1] Learning Force-Reactive Humanoid Loco-Manipulation
链接: https://arxiv.org/abs/2610.35862
作者: Sihang Li,Xiaofeng Guo,Jinkun Cao,Charlie Cheng,Ruihan Yang,Chen Feng,Guanya Shi,Haozhi Qi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Systems and Control (eess.SY)
备注: 9 pages, 6 figures
Abstract:Many everyday tasks require controlling contact force as well as motion. On a humanoid, this force must often be coordinated across both hands and the whole body, and cannot be specified by kinematic trajectories alone. Therefore, we present FLORE, a framework for learning force-reactive humanoid loco-manipulation. We collect robot-free demonstrations using two force-sensing UMI-style handheld grippers for hand motion and contact wrenches, and wearable trackers for body motion. A force-reactive diffusion policy takes vision, proprioception, and measured wrenches as input, then predicts actions and per-hand force commands. A hybrid compliance whole-body controller with force tracking executes these commands while maintaining balance: an integral of the force error measured by real force/torque (F/T) sensors lets each hand track force accurately along the commanded direction and remain compliant along the others. In simulation, the controller tracks per-hand force commands of varied magnitude and direction, with little loss in whole-body motion accuracy. Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. For more details, please visit our website: this https URL.
[AI-2] Patient-specific rubric-based evaluation of large language models across longitudinal cancer care
链接: https://arxiv.org/abs/2610.35847
作者: Wei-Wei Zhang,Pengkai Wang,Yan Li,Zhijie Sang,Guan-Qun Zhou,Jun-Yan Li,Shu-Han Zhou,Pei-Wen Liu,Ting-Xi Tang,Min Tang,Zhitian Hou,Jianmin Wu,Jian Guan,i Lin,Hongxia Yang,Ying Sun
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models (LLMs) show promise in specialty medicine, yet existing evaluations largely focus on isolated clinical encounters rather than tracking how clinical responsibilities evolve across a longitudinal care pathway. Here we introduce NPC-RubricBench, a phase-aware benchmark for evaluating multistage clinical reasoning across the cancer care continuum, using nasopharyngeal carcinoma as a defined disease context. Curated from longitudinal records of 405 patients across two cancer centers, the benchmark comprises 941 patient-phase units across six sequential care phases and 17,688 patient-specific atomic criteria spanning five evaluation dimensions, with each assessment strictly restricted to decision-time clinical evidence. Evaluating 30 general-purpose and medical LLMs revealed that strong diagnostic performance does not carry over to downstream care. We evaluated 30 general-purpose and medical LLMs, showing that high diagnostic performance did not translate into improved downstream care. Instead, the dominant failure modes reversed across phases. Explicit decision errors dominated anatomical staging at 48.2%, whereas decision omissions dominated initial work-up and post-treatment follow-up at 33.4% and 30.4%, respectively. In the evaluated model set, medically adapted models showed higher omission rates than general-purpose models. Moreover, varying query granularity exposes a critical specification trade-off: broad open-ended requests degrade Decision Quality across all 30 models, while more detailed, checklist-like prompts raise Decision Quality but reduce Evidence Faithfulness and Safety Adherence. Furthermore, NPC-RubricBench demonstrates that static diagnostic benchmarks are insufficient to ensure consistent clinical reliability over time, highlighting the importance of phase-adaptive safeguards in multistage medical AI.
[AI-3] I-HITL: Inalienable Human-in-the-Loop
链接: https://arxiv.org/abs/2610.35835
作者: Samuel Breckenridge,Dani Vilardell,Wenhao Wang,Fan Zhang,Ari Juels
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Human oversight is a pillar of AI safety. It is often exercised by a human in the loop (HITL) who approves critical actions taken by an AI agent. AI agents’ increasing ability to bypass guardrails, though, raises the question of how to enforce HITL robustly. We introduce the concept of inalienable human-in-the-loop (I-HITL). An I-HITL protocol ensures that an agent can initiate an action only with explicit human approval, even when the human interacts with a system (e.g., a local agentic harness) compromised by the agent. A strong I-HITL protocol additionally ensures such approval even when an adversarial human attempts to delegate approvals to an agent. To motivate I-HITL, we report on experiments with a popular agentic wallet platform in which raising an agent’s spending cap requires HITL approval via emailed one-time passcodes (OTPs). Given access to the user’s email and a task blocked by the cap, we find that ten different frontier models will often autonomously raise the cap, harvesting OTPs to bypass a HITL guardrail and surfacing other vulnerabilities. We formally define I-HITL using the concept of communication topologies and prove necessary and sufficient conditions for its realization. We show that an I-HITL protocol requires a trusted entity - a trusted computing device (or equivalent) - and characterize which such devices can and cannot realize an I-HITL protocol. We present a prototype I-HITL system called Countersign that can plug into and strengthen existing HITL systems using off-the-shelf consumer mobile devices. Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.35835 [cs.CR] (or arXiv:2610.35835v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.35835 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Daniel Vilardell Regué [view email] [v1] Mon, 26 Oct 2026 17:52:14 UTC (4,323 KB)
[AI-4] Profiling a Frontier Agent with EBench: From Direct Control to Harness with Policy
链接: https://arxiv.org/abs/2610.35832
作者: Yifei Yao,Yude Zou,Guanbo Jiao,Xing Gao,Ning Gao,Yukai Wang,Jiantong Chen,Zihou Zhu,Xinyu Li,Ningyu Yan,Shuai Wang,Jianming Ma,Haoyu Guo,Weinan Zhang,Chunhua Shen
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Frontier multimodal large language models (MLLMs) can control robots without robot-specific training, but how their capabilities compare and interact with post-trained vision-language-action (VLA) and world action model (WAM) policies across manipulation demands remains to be thoroughly evaluated. In this report, we evaluate GPT-6-Astra on EBench as the core of an agent that controls a robot through an agent harness under the same visual and proprioceptive observation protocol as post-trained policies. We study two agent configurations: direct control, in which the agent generates waypoint commands directly, and harness with policy, in which it can also invoke a post-trained VLA or WAM policy for action execution. Under direct control, the agent achieves competitive overall performance relative to post-trained policies. It leads on short-horizon tasks and exhibits the least performance variation across generalization dimensions, whereas it underperforms post-trained policies on long-horizon tasks, tabletop manipulation, and high-precision contact-rich tasks. In the harness with policy, invoking a post-trained VLA or WAM combines the complementary strengths of both components, raising overall task success to the level of the strongest compared post-trained policies while substantially reducing token cost. However, performance remains below the per-task upper bound of the standalone components due to suboptimal policy invocation, fixed-instruction execution without subgoals, and grasp disruption when the agent resumes control. Overall, our study characterizes the capability boundary of a frontier MLLM agent and suggests directions for improving future agent harness design. Project page: this https URL
[AI-5] Reva: Real-Time Vulnerability Detection through Contextual Probing of Large Language Models
链接: https://arxiv.org/abs/2610.35829
作者: Amine Lbath,Massih-Reza Amini
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:A vulnerability costs more to fix the later it is discovered. As coding agents generate code faster than developers can review it, software needs a first line of defense screening each new function in real time. LLM agents have recently become capable of finding real-world vulnerabilities by working as a security expert would: they run analysis tools, explore the surrounding code, test hypotheses and weigh the evidence before forming their own judgement. This reasoning, however, spans many model calls, so that each verdict takes minutes of compute, far too long for real-time screening. Faster detectors, which fine-tune deep learning models or prompt an LLM for a direct verdict, are less accurate: they judge a function without the context an agent gathers, and an LLM’s generated verdict is unreliable even when given some of it. We present REVA (REal-time Vulnerability Analysis), an AI vulnerability detector that brings the security context and code understanding components behind agentic detectors into a single forward pass of an LLM. For each new function, REVA gives the LLM a pre-computed threat model of its module, lightweight static analysis findings and call-graph information. REVA leverages a trained probe to read the LLM’s internal representations of the new function in this context. On PrimeVul’s held-out test set, REVA on GLM-5.3-Flash reaches 60.0% accuracy in about one second per function. This matches VulTriage (59.2%), the state-of-the-art agentic detector, run on the same LLM, while being over 250 times faster and less than a tenth as costly. On the same prompt, the probe is 14% to 17% more accurate than the LLM’s own direct verdict. Across three open LLMs, REVA is at least as accurate as VulTriage, and about 10% more accurate on two of them. An LLM’s internal representations thus encode much of what agentic detectors obtain through reasoning. Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.35829 [cs.CR] (or arXiv:2610.35829v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.35829 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-6] Game theory of academic effort and hiring in the age of AI
链接: https://arxiv.org/abs/2610.35802
作者: Saptarshi Pal,Peter S. Park
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Computer Science and Game Theory (cs.GT)
备注: 30 pages, seven visualizations (four figures, three tables)
Abstract:Many academic disciplines (in particular, mathematics and the theoretical sciences) face difficult questions about how to adapt their practices to the rise of highly capable artificial intelligence (AI). Before AI, a well-written research paper with a strong theoretical result acted as a reliable signal that its authors understood their field in depth. Now that AI is widely accessible, such signals have become cheaper to produce. As a result, they no longer reliably sort academics by their level of understanding. This has direct consequences for academic hiring. The age of AI raises a question for each side of the academic job market. Should an early-career researcher invest in a deep understanding of their field, or instead use AI to build a longer resume with many papers? Should a recruiter favor applicants with fewer papers who pass an interview aimed at probing understanding, or insist on both publication volume and a pass? In this paper, we develop a game-theoretic model of this two-sided dilemma. We classify all pure-strategy equilibria that can arise in this game and use learning simulations to identify the behaviors that populations of applicants and recruiters may learn to adopt. Our analysis provides a framework for reasoning about how academic hiring can adapt to AI and for evaluating practices that preserve the value of deep understanding.
[AI-7] A Unified Arena for LLM Agents across Diverse Eurogame
链接: https://arxiv.org/abs/2610.35773
作者: Fengrui Zhang,Xiang Li,Ruomeng Ding,Jiayi Yuan,Yuchen Huang,Zhaoyang Han,Xiaoliang Wang,Daoyuan Chen,Yaliang Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Interactive environments offer a practical way to evaluate large language model (LLM) agents, and competitive games extend this evaluation to decisions shaped by competing goals and other agents’ choices. Yet existing evaluations confined to individual games or execution frameworks leave unclear which competitive advantages persist across settings. We introduce DECE (Diverse Eurogame Competitive Environments), a unified arena spanning ten Eurogames with shared interfaces and evaluation protocols. These games combine lasting strategic commitments, competition over shared resources, and multiple routes to victory. Our primary study evaluates 12 model-harness configurations, spanning nine models and four execution harnesses, in 480 three-player matches across ten games. The main and supplementary studies comprise 1,032 completed matches in total. These comparisons reveal broad cross-game ranking agreement alongside game-specific strengths. Matched studies show that the same model can perform differently across execution harnesses, adding another source of variation to these profiles. These performance advantages do not consistently translate into better token or cost efficiency, making resource use an additional consideration when comparing configurations. Together, the findings support evaluating complete agent configurations by both their competitive performance across games and the resources needed to achieve it.
[AI-8] Free to Look Free to Reason : Simple LLM Harnesses Outdiagnose Hand-Crafted Root Cause Analysis Agents
链接: https://arxiv.org/abs/2610.35767
作者: Tingzhu Bi,Xun Zhang,Jinglin Li,Ping Wang,Meng Ma
类目: Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Software Engineering (cs.SE)
备注: 12 pages
Abstract:Root cause analysis (RCA) of microservice failures has attracted many LLM agents whose investigation is hand-crafted, through standard operating procedures, fixed stages and role agents. Each is evaluated on its own data against few baselines, so it is unknown whether this design beats letting a capable model investigate on its own. We compare three simple harnesses that prescribe none of the model’s calls (one call over an evidence package, a SQL agent and the Codex coding agent) with five published LLM RCA agents and with statistical and causal methods on RCAtlas, 90 validated cases from eight public datasets, across six models and three reasoning efforts. The simple harnesses are the most accurate (PR@1 65-77% on three mid-tier models), ahead of the published agents (16-53%) and the non-LLM methods (6-39%), on every model on which both families were run. The published agents fix from a tenth to all of their model calls in the program, and a case study shows how these prescriptions skip the decisive evidence. A stronger model mainly improves reliability, a multi-turn harness mainly improves aptitude, and more reasoning effort changes accuracy by a few points at most while multiplying the time by up to seven. Among runs more accurate than the best non-LLM method, every Pareto-optimal run in cost or time uses a simple harness.
[AI-9] Beyond Physical Illness: Can Clinical Narratives Reveal Emerging Mental Health Risks in Chronically Ill Patients?
链接: https://arxiv.org/abs/2610.35761
作者: Sudeshna Jana,Siddhartha Laha,Manjira Sinha,Tirthankar Dasgupta,Pabitra Mitra
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Mental-health risks such as depression, dementia, and suicidal ideation can emerge during hospitalization among chronically ill patients and may worsen without timely recognition and intervention. However, most existing mental health studies focus on patients with established psychiatric conditions rather than patients hospitalized primarily for physical illness. In this work, we investigate whether latent mental-health risk in chronically ill patients can be inferred from longitudinal clinical narratives. We construct a weakly supervised cohort around 4,000 patients from MIMIC-III using mental-health evidence extracted from discharge summaries. A hierarchical state-space framework derives mental-health representations from day-wise clinical narratives and models their temporal evolution for risk estimation throughout hospitalization. Preliminary results demonstrate improved binary risk prediction using the proposed representations and show distinct model derived risk trajectories between risk and no-risk cohorts, providing a foundation for studying emerging mental-health risks in chronically ill patients.
[AI-10] CLOSURE: Cross-Channel Closure for Multivariate Time Series Anomaly Detection
链接: https://arxiv.org/abs/2610.35753
作者: Fan Xu,Hao Wu,Zhi Song,Hanbo Huang,Wenzhuo Ma,Nan Wang,Wei Gong,Xibin Zhao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multivariate time series produced by industrial installations and server clusters are under continuous monitoring, and anomaly detection is the first line of defense before a local fault escalates into a system-wide failure. In the absence of labels, the dominant paradigm trains a reconstruction model on normal data and scores the residual to reach a verdict. Recent work has steadily enriched the expressiveness of the recovery map through Transformers, diffusion models, and cross-scale associations. A residual, however, only certifies that an observation can be recovered; it does not certify that the variables still explain one another as they do under normal operation. A coordinated yet unfamiliar operating transition inflates the residual without breaking any relationship, whereas a sensor drifting away from its peers may leave the residual silent. Methods that fold inter-variable dependencies into the reconstruction objective inherit this ambiguity, because their relational evidence is ultimately supervised by the same recovery loss. We therefore propose CLOSURE, which evaluates recoverability and cross-channel consistency as two normal-data prediction tasks learned in separation. The recoverability task is solved by a diffusion reconstructor conditioned on a discrete memory of normal temporal bases. The consistency task is solved by a relation-closure operator that predicts every channel exclusively from the others, where the target’s own trajectory is withheld and sparse signed gates decide which peers are actually used. The two scores are reconciled only through calibration on normal data. Extensive experiments show that CLOSURE consistently outperforms recent reconstruction-, frequency-, and topology-based detectors on public benchmark streams, confirming the value of examining consistency alongside recoverability.
[AI-11] Audit Where Hacks Pay: Budgeted Strong-Verifier Audits Against Reward Hacking in GRPO
链接: https://arxiv.org/abs/2610.35716
作者: Shurun Li,Michael Hua Wang,Haibo Zeng,Long Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 27 pages, 5 figures, 8 tables
Abstract:Reinforcement learning with verifiable rewards trains on cheap checks, such as unit tests, that a policy can learn to game, while a stronger check is too expensive to run on every rollout. We study how to spend a limited strong-verifier budget in GRPO, separating which weakly passing rollouts to audit from what reward a caught hack receives. At a fixed policy state with centered advantages, the expected update depends on a design only through each rollout’s marginal audit probability and correction; reflecting a caught hack’s reward about the failure reward negates the advantages of an all-hack group; and auditing whole groups with the fewest weak passes first (low-pass budgeted audits, LB) minimizes the advantage mass left on hacks among pass-count designs with a fixed correction when a label’s value falls with the pass count. We tested this analysis in actual LLM training: from each run’s starting state, the sign of this mass on a frozen panel predicted whether hacking would be suppressed in 33 of 39 runs, including 13 of 16 preregistered ones. In LeetCode (Wong et al., 2026) and Countdown-Code (Khalifa et al., 2026), LB drives hacking to near zero, while random audits with the same penalty and budget do not and use more labels. We also find that the randomness of the update affects training: in Countdown-Code, two unbiased corrections, per-item reward-level IPW and a group-level advantage correction, bring hacking down too but end 6.1-11.8 and 8.0-11.5 points lower in held-out accuracy than LB at the same budget. A full audit, which the group-level correction reproduces in expectation, stays within about two points of LB in three seeds, while LB uses 28% of its labels, and allocating the same correction by pass count recovers most of the loss (one seed). Training outcomes thus depend not only on the expected update but also on the randomness of its estimate.
[AI-12] Explainable Multimodal AI Approach to Reduce Unnecessary Breast Biopsies
链接: https://arxiv.org/abs/2610.35704
作者: Sadam Hussain,Mansoor Ali,Daly Betzabeth Avendaño-Avalos,Jose Gerardo Tamez-Peña
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Recently, there has been rise in multimodal vision-language models for computer-aided diagnosis (CAD), yet their application in breast diagnosis, particularly for predicting biopsy outcomes remains limited. This study presents a multimodal vision-language approach, incorporating Grad-CAM as an Explainable AI (XAI) technique, to predict benign tumors in patients recommended for biopsy, especially those classified as BI-RADS 4 and 5. A novel multimodal framework was devel- oped using Vision Transformer (ViT) for imaging feature extraction and DeBERTa-V3-Large for textual embeddings. The model was trained and evaluated on a comprehensive in-house dataset comprising 1,632 patients, each with four mammography views and a corresponding radiology report. The features of both modalities were extracted independently and fused by late fusion before classification. Grad-CAM was used to generate visual explanations highlighting key regions influencing model predictions. The proposed model demonstrated high sensitivity 90%, 95%, 99%, and 100% with corresponding specificities of 40%, 36%, 18%, and 11%, respectively. The best AUC achieved was 0.72 by multimodal approach on the test set, outperforming the image only (0.66) and text-only (0.68) baselines. The proposed multimodal architecture, enhanced with Grad-CAM, demonstrates a strong potential to identify benign tumors among high-risk patients, offering interpretable insights and thereby helping reduce unnecessary biopsies, patient anxiety, and healthcare costs.
[AI-13] rySee: Modular Perception-Enhanced VLM Grounding for Closed-Loop Robot Task Planning in Specialized Domains
链接: https://arxiv.org/abs/2610.35687
作者: Elias Hofele,Thomas Trapp,David Timmermann,Tristan Schnell,Georg Heppner,Arne Roennau
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: To be published in the proceedings of the 2026 IEEE International Conference on Robotics and Biomimetics (ROBIO)
Abstract:Vision Language Models (VLMs) have shown impressive capabilities in high-level reasoning and task planning for robots. However, specialized domains remain a challenge for VLMs’ image interpretation and grounding abilities. We introduce TrySee, a modular closed-loop robot task planning pipeline integrating general-purpose VLM reasoning with domain-specific perception and established robotics components. The system consists of an inherently reactive cyclic planning pipeline, augmented with domain-specific modular vision preprocessing, to provide online grounding. Finally, a dynamic preprocessing selection mechanism contributes to the scalability of the pipeline. The system requires neither VLM fine-tuning nor end-to-end demonstrations and adapts through replaceable perception modules and motion planning. The approach is evaluated on planning tasks across 5 widely different domains, including a real-world robot disassembly task. The field tests in a brownfield environment demonstrate the approach’s practicality in integrating VLM planning into specialized robotic systems with limited domain-specific training data.
[AI-14] Adapt for Planning Not just Prediction: Optimistic Test-Time World Model Adaptation
链接: https://arxiv.org/abs/2610.35636
作者: Akshay Mete,Dileep Kalathil,P. R. Kumar
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: Accepted at the CoRL 2026 Workshop on Pretrain to Adapt
Abstract:Latent world models are typically pre-trained offline and kept frozen at test time, so planning with them can degrade under distribution shift. Test-time adaptation (TTA) can mitigate this degradation by updating the model during deployment. Unlike pre-training on fixed data, TTA for planning is closed-loop: the planner uses model rollouts to select actions, and the resulting transitions drive the next update. This raises the question: which objective should guide TTA for planning? Under a prediction-only objective, the model is fit only to the outcomes of its own actions, so it can remain wrong about the actions that would reach the goal, and low prediction error need not translate into better planning. Inspired by reward-biased maximum likelihood from adaptive control, we introduce Optimistic Test-Time Adaptation (OpTTA). Its objective has two terms, the prediction loss on observed transitions and the planner’s goal-reaching cost on its candidate rollouts, both minimized over the model parameters. The planning term biases adaptation toward models that are optimistic about reaching the goal, without any additional environment interaction. Across visual and environment shifts on manipulation, reaching and navigation tasks, OpTTA raises final planning success over the frozen model by 13.5 percentage points on average, compared with 6.9 for the prediction-only objective (AdaJEPA), when only the predictor is adapted. When the encoder is adapted as well, the gains are 13.5 and 7.5 points. These experiments span three JEPA world models with different encoders and training objectives and, for temporal straightening, both CEM and gradient-based planners. Our results suggest that test-time adaptation of world models for planning benefits from an optimistic, task-aware term in addition to the prediction loss.
[AI-15] VaLoP: Value Gradient Lookahead and Pullback for Reward-Guided Diffusion
链接: https://arxiv.org/abs/2610.35607
作者: Kwanyoung Kim
类目: Artificial Intelligence (cs.AI)
备注: 35 pages. Project page: this https URL
Abstract:Reward guidance steers a pretrained diffusion or flow model toward images or actions that a reward model prefers, at sampling time and without retraining the model for each reward. It follows the gradient of a value function, the expected reward given the noisy state. The two common estimates of this gradient, gradients through a one-step denoised estimate and learned value models, are least accurate at high noise, where the coarse structure of the sample is decided. Interestingly, we find that both errors are posterior covariances, which grow with the uncertainty about the clean sample, and hence at higher noise. To reduce them, we evaluate the value at a less noisy state, one sampler step ahead, where this uncertainty is smaller, and pull its gradient back to the current state. This lookahead places both estimators at the two ends of one family and reveals that an intermediate lookahead requires one additional term, a correction through the generator. We therefore propose Value Gradient Lookahead and Pullback (VaLoP), which computes this correction with one vector-Jacobian product, and VaLoP-Fast, which amortizes it with a small head that, for text-to-image models, is trained with a single reward and reused for the others. Across various models, rewards, and benchmark datasets, both robustly outperform existing value-gradient guidance on the target reward, VaLoP-Fast at about the cost of learned-value guidance. Both extend to critic-guided robot policies, where VaLoP-Fast attains the highest success rates on LIBERO and RoboCasa.
[AI-16] ArchiveLens: A Generative Spatial Knowledge Framework for Semantic Reconstruction of Historical Maps
链接: https://arxiv.org/abs/2610.35606
作者: Yanan Wu,Yalin Yang,Chengbin Deng
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG); Social and Information Networks (cs.SI)
备注: 31 pages
Abstract:Historical maps contain dense morphological and semantic information about past urban environments, but many digitization workflows focus primarily on extracting geometry, and do not capture cartographic semantics such as building use and construction materials. This study introduces ArchiveLens, a generative spatial knowledge framework that operationalizes historical maps as entity-centric, machine-queryable geospatial databases. The pipeline decouples morphological and semantic reconstruction through three stages: tiled YOLOv11-seg instance segmentation for building footprint extraction; deterministic material classification via hue, saturation, and value (HSV)-based color thresholds; and localized attribute parsing using a quantized vision-language model constrained by domain-specific prompts. Extracted entities are integrated into a hybrid Spatial-Temporal Retrieval-Augmented Generation (ST-RAG) architecture that combines dense semantic retrieval with symbolic spatio-temporal filtering and mandatory provenance grounding. Applied to 445 georeferenced Sanborn map sheets of Chicago (1918-1950), the framework achieves high geometric and semantic accuracy. Beyond performance, ArchiveLens advances GIScience by formalizing historical cartography as structured spatial knowledge representation, allowing researchers to query building attributes and compare spatial patterns across map sheets and years.
[AI-17] Evolve on Graph: Self-Evolving Knowledge Graph Reasoning without Question Annotations
链接: https://arxiv.org/abs/2610.35598
作者: Zongyuan Sun,Shuai Zhao,Ke Shu,Ziyue Zhu,Shangyang Wu,Yuanlin Lin,Tao Wang,Shiyue Li,Yingjun Du,Anh Tuan Luu,Haoran Luo
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Knowledge graphs (KGs) provide reliable external knowledge for multi-hop factual reasoning with large language models (LLMs), yet most KGs in real-world domains such as finance and industry lack question–answer (QA) annotations. Without such annotations, existing methods struggle to improve LLM graph reasoning: training-free agents cannot improve their exploration through interaction; reinforcement learning (RL) agents depend on QA data annotated by humans or synthesized by closed-source models; and training data is typically built offline, so its topology mix is fixed and cannot follow the agent’s ability. We propose Evolve on Graph (EoG), a framework that lets an LLM self-evolve its graph reasoning without any QA annotations. Through Graph Self-Play, a single LLM alternates between a Proposer that poses questions and a Solver that answers them. To make questions verifiable, Symbolically Verifiable Topology Synthesis (SVTS) samples reasoning topologies on the KG and executes the corresponding first-order logic queries, yielding topology skeletons with correct and complete answer sets; guided by a skeleton, the Proposer explores the graph to pose a multi-hop question. To let task topologies follow the Solver, a Boundary-Aware Graph Curriculum rewards the Proposer by the Solver’s score variance, keeping questions at the Solver’s capability boundary so that complex topologies enter training as the Solver improves. On six benchmarks over Freebase, Wikidata, and an industrial KG, EoG consistently outperforms prompting-, retrieval-, and agent-based methods, showing that LLM graph reasoning can improve through self-play without QA annotations. Our code is available at this https URL
[AI-18] Agent ic Algorithm Design: LLM s that Propose Formalise and Prove Runtime Bounds for Evolutionary Algorithms in Lean
链接: https://arxiv.org/abs/2610.35595
作者: Per Kristian Lehre
类目: Artificial Intelligence (cs.AI); Neural and Evolutionary Computing (cs.NE)
备注: A github repository with source code will be published later
Abstract:LLMs now prove hard theorems in Lean. Can they also design algorithms and prove their correctness and efficiency? LLM-guided design of optimisation heuristics (such as AlphaEvolve) is purely empirical. Automated theorem provers with LLMs prove given statements. So far, there has been little evidence that LLMs can do both: design an algorithm and guarantee its efficiency by a mathematical proof. We present a staged pipeline modelled after research methodology in design and analysis of evolutionary algorithms. The agent proposes black-box optimisation algorithms, and Lean guards filter out flawed designs. The agent then defines a potential function and proves drift bounds that guarantee an expected runtime improvement over the baseline algorithm. At no point in the pipeline are the algorithms compared empirically. The agent generates a black-box optimisation algorithm with expected runtime O(n^2) on a CycleUncover problem, where the classical (1+1) EA needs runtime e^\Omega(n) . The pipeline also designs and analyses a black-box optimisation algorithm for integer-valued submodular minimisation with expected runtime O(n^2+2c) , and it matches the optimal order O(n^2) on LeadingOnes. Across 8 problem classes, 12 of 15 runs with frontier models complete the pipeline with a machine-checked expected-runtime bound. An ablation study shows that removing the staging and asking the agent to zero-shot an algorithmic design with a runtime guarantee leads to a copy of the seeded baseline algorithm (random search) or to incomplete proofs. Our study suggests that machine-checked specifications in Lean can replace empirical evaluation in open-ended algorithm discovery. To the best of our knowledge, this is the first example of an automated pipeline that designs efficient black-box optimisation algorithms with machine-checked runtime bounds in Lean.
[AI-19] SAP-ViT: A Self-Attributing Patch Vision Transformer for Medical Image Classification
链接: https://arxiv.org/abs/2610.35593
作者: Wei Zhang,Wolfgang Nejdl,Ming Tang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Vision Transformer (ViT) classifiers commonly rely on a global classification (CLS) token for image-level prediction, which is poorly aligned with local interpretability requirements in medical imaging. We propose SAP-ViT, a Self-Attributing Patch Vision Transformer that replaces CLS-based classification with direct aggregation of last-layer patch embeddings. This design makes class predictions decomposable into patch-level contributions as part of the forward pass. To encourage concise and readable evidence, SAP-ViT applies sparsity regularization to the patch-weight distribution. Since self-attention mixes information across patch tokens, we further introduce an input-space back-mapping mechanism to estimate which input patches contribute to influential output patch embeddings. We evaluate SAP-ViT qualitatively and quantitatively on multiple medical image classification datasets spanning different imaging modalities. The results show that SAP-ViT maintains strong classification performance while providing consistent and interpretable patch-level explanations across datasets. Ablation results further show that both sparsity regularization and the alignment component of input-space back-mapping contribute to the final explanation quality. Overall, SAP-ViT provides a favorable trade-off between predictive accuracy and interpretability for ViT-based medical image classification.
[AI-20] Infinity Machine: a self-improving AI agent for research
链接: https://arxiv.org/abs/2610.35579
作者: Harish Santhanalakshmi Ganesan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We present the Infinity Machine, a self-improving AI agent that we use for research on problems whose answers aprogram or data can check. Each candidate solution is produced by a full tool-using agent that works for up to an hour in a fresh sandbox; proposers see only a public score, and a private split decides what counts. The harness itself(prompts, tools, skills, playbook, meta agent) is an editable genome, changed only through a kernel-enforced gate with held-out and cost checks. An autonomous mode goes from a plain-text goal to a verified report with no human step: an agent picks an open problem and writes its verifier, the kernel calibrates it, an adversarial agent audits it, and an agent that never sees the verifier re-checks the result. Across fourteen problems in six days, for about 180 of model usage, the system produced four results beyond the published state of the art on catalogued open problems in quantum information: a certified depolarizing-channel threshold point p = 0.064966, a qutrit positivity frontier of 0.2910672, a 44-gate depth-6 Golay state preparation, and a stabilizer-rank bound chi(T^7) = 10. It also cut SABRE qubit-routing overhead from 68,559 to 12,087 added gates. In hardware, it cut the energy per MAC of an open 8x8 INT8 systolic tile by 77.7% through a full open RTL-to-GDS flow, bit-exact. Three fully autonomous runs ended in honestties with public records; one audit caught a stale record, and one agent-written score pointing the wrong way went unnoticed. We report the failures as carefully: the one self-derived harness generation was rejected on cost, proposers gamed lenient scorers four times, and operator changes were decisive for several results.
[AI-21] ExCal: Optimization-Free Multivariate Time Series Anomaly Detection via Excess Calibration
链接: https://arxiv.org/abs/2610.35568
作者: Kadir-Kaan Özer,René Ebeling,Markus Enzweiler
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at the 2026 IEEE International Conference on Big Data (IEEE BigData 2026), Special Session on Intelligent Data Mining. 10 pages, 3 figures, 9 tables
Abstract:We introduce ExCal, a normal-only multivariate time series anomaly detector that combines raw state retrieval, a signal/derivative subspace, and robust statistics across temporal scales. Each expert map receives a high normality floor from the training prefix. During scoring, positive excess is retained, attenuated for broadly firing maps, and combined with a small rank term. The method fits reference memories and projections without gradient optimization or anomaly labels. Prevalence and ranks depend on the complete scored sequence, making inference batch and transductive rather than online. On 180 series from 17 TSB-AD-M families, ExCal obtains AUC-PR 0.421\pm0.37 and VUS-PR 0.452\pm0.37 over three seeds, ranking second among 24 methods. Holm-adjusted paired tests detect no significant difference from AxonAD or StreamVAE, which are the two leading non-pretrained methods on TSB-AD-M, on either primary metric, while finding significant positive differences against the other 21 baselines. A separate single-run probe over three series measures median model-wrapper time of 0.0662 s. Ablations associate the largest primary-metric changes with raw retrieval and excess-magnitude preservation. The evaluated score is decomposable by expert, but its diagnostic utility and online suitability remain untested.
[AI-22] Multi-Fidelity Training Reshapes Uncertainty Decomposition in Atomistic Models
链接: https://arxiv.org/abs/2610.35560
作者: Matteo Tolloso,Davide Bacciu,Luigi Bonati
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-fidelity training, which combines abundant low-fidelity data with scarce high-fidelity labels, is increasingly common in atomistic modeling, but its effect on predictive uncertainty remains largely unexplored. We investigate this question using deep ensembles of mean-variance MACE models trained on ANI-1x, contrasting high-fidelity training from scratch with low-fidelity pretraining followed by fine-tuning. We characterize aleatoric, epistemic, and total uncertainty throughout training and under two distribution shifts: unseen molecular systems and higher-energy configurations. We find that pretraining reduces prediction error and improves uncertainty-based ranking across all test sets, but does not yield well-calibrated predictive variances; calibration deteriorates more severely under energy extrapolation. Training-dynamics analysis localizes most of this calibration degradation to high-fidelity optimization, during which the learned aleatoric variance becomes overconfident on held-out data. Moreover, multi-fidelity fine-tuning produces a more effective uncertainty-based acquisition function than high-fidelity training from scratch. These findings show that uncertainty ranking and calibration capture distinct properties and must be assessed separately in multi-fidelity atomistic models, especially under extrapolation.
[AI-23] he Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability Exploitation
链接: https://arxiv.org/abs/2610.35541
作者: Ruozhao Yang,Jiongchi Yu,Ming Hu,Xiaofei Xie
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Web services increasingly expose backend interfaces that let users delegate tasks to AI agents, which invoke service operations directly without navigating websites. This shift changes how existing Web vulnerabilities can be exploited. We investigate these changes through VELA, an executable benchmark covering 44 vulnerabilities across twelve Web application families. The benchmark preserves vulnerable backends and derives agent-facing interfaces from legitimate workflows. We evaluate three commercial agents across multiple interface configurations, using controlled comparisons to separate interface effects from agent behavior and independent oracles to verify attack outcomes. Our results show that the agent realizes every attack that direct interface invocation realizes in the reference configuration, so exploitability is set mainly by the interface: 91% of vulnerabilities remain exploitable under the reference design. Interface restrictions prevent some attacks by removing required operations or limiting input control, while retention remains high across vulnerability classes. No case-level loss in the reference configuration is attributed to agent refusal or input normalization, and in one case the agent completed an attack by supplying calls the attacker omitted. Across the tested configurations, interface design affects exploitability more than agent choice, moving retention from 39% to 95% against at most 4.5 points across agents, but some restrictive interfaces also limit legitimate workflows. Our study provides a systematic empirical account of how agent-mediated access reshapes existing Web vulnerabilities, grounding future Web architecture design in measured security outcomes and trade-offs between reducing attack exposure and preserving legitimate functionality in agent-facing services.
[AI-24] JANUS: Synthesizing Cross-Layer Defenses Against Web Deserialization Attacks
链接: https://arxiv.org/abs/2610.35532
作者: Ruozhao Yang,Minglang Li,Jiongchi Yu,Xinfeng Li,Xiaofei Xie
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Unsafe deserialization turns untrusted data into attacker-controlled execution, and a single vulnerability can admit diverse exploit variants that evade defenses tailored to a known exploit. Existing defenses typically operate at a single enforcement point with limited visibility into the full attack. Combining controls across layers offers complementary protection, but coordinating them to preserve legitimate functionality and adapting them to new variants requires substantial expert effort. We propose JANUS, a red-blue adversarial framework for automated cross-layer defense synthesis. Starting from a known exploit, \sys iteratively refines deployment-specific policies in an isolated offline environment. A black-box attacker probes the current policy for exploit variants using only externally observable feedback. The defender summarizes each verified variant into its security-relevant effects, uses counterfactual replay to separate complementary coverage from redundant enforcement, and jointly selects and scopes controls across layers. Each candidate policy is validated against accumulated exploits and benign workflows. We evaluate JANUS on sixteen vulnerability deployments across six Java and PHP applications against gateway, application, system, and cross-layer defenses. JANUS reduces held-out attack success to 6.2%, compared with 18.8-55.0% for the baselines, fully mitigates 13 of 16 deployments, and preserves 98.4% of benign functionality tests at a 6.9% targeted P95 latency increase. Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE) Cite as: arXiv:2610.35532 [cs.CR] (or arXiv:2610.35532v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.35532 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-25] CANS: Capability-Preserving Neuron Steering for Large Language Models
链接: https://arxiv.org/abs/2610.35530
作者: Yucheng Wu,Chengyi Xie,Chenpeng Wang,Liangming Pan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Activation steering offers training-free and dynamic control of target behaviors in large language models, but existing methods can impair general capabilities and require domain-specific intervention designs. We introduce Capability-Preserving Neuron Steering (CANS), a unified framework that balances target-behavior control with capability preservation. To make target-behavior improvement and general-capability preservation tractable, we introduce two proxies: log-probabilities of target tokens in model generated responses, and output-distribution KL divergence on general-purpose inputs. Combining the local approximation of the two proxies yields a closed-form solution, enabling lightweight steering without iterative model training or additional inference-time computation. The resulting neuron-wise scaling coefficients achieve effective steering while limiting general-capability loss. A shared data-construction and intervention procedure supports adaptation across domains using sampled responses with target-behavior filtering and target-span annotations. Across safety, toxicity, and hallucination control on Qwen and Llama models, CANS achieves comparable or better steering while better preserving MMLU, BBH, and IFEval performance than SFT, CAA, and neuron-level baselines. On Qwen3-4B, CANS reduces HarmBench attack success from 15.50% to 1.75% with a 1.11 -point decrease in average capability. These results support CANS as a practical approach to behavioral control across domains.
[AI-26] Ranking Function Generation with LLM s for Termination
链接: https://arxiv.org/abs/2610.35523
作者: Rongkang Li,Jiaxin Hu,Taolue Chen,Yong Li,Fu Song,Zhilin Wu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: 24 pages, 6 pages
Abstract:Ranking functions are a fundamental technique for proving loop termination, but constructing them manually is challenging and requires substantial expertise. We propose RAINFALL, a neuro-symbolic framework which follows the generate-verify-refine paradigm and tightly integrates LLMs with symbolic verification. RAINFALL formulates ranking function synthesis as a code-completion task by inserting placeholders into the input program and constraining the LLM to fill four predefined slots for each loop. The completed program is then checked by a symbolic verifier. If verification succeeds, a valid ranking function is obtained; otherwise, the verifier’s diagnostics are translated into natural-language feedback that guides iterative refinement of the LLM’s output. We evaluate RAINFALL on two public benchmark suites comprising 358 terminating C programs. Using MiniMax V2.7, RAINFALL synthesizes verifiable ranking functions for 323 programs (90.22%), outperforming all promising baselines. In particular, it solves 12 more programs than PROTON 2.1, the best-performing baseline and winner of the SV-COMP 2026 Termination category. Furthermore, combining RAINFALL with PROTON 2.1 increases the success rate to 96.65% (346/358 programs). We also conduct ablation study confirming individual contributions of placeholder contracts, few-shot examples, symbolic verification, and diagnostic information feedback. Experiments across LLMs further confirm the effectiveness of RAINFALL. Our results suggest a generic, sound framework for ranking-function synthesis and, beyond termination, LLM-assisted synthesis of formally checkable liveness certificates, both of which represent new directions of software engineering research.
[AI-27] Governing Enterprise AI Traffic at Scale: Design and Operation of a Policy-Driven Multi-Vendor LLM Gateway on Envoy
链接: https://arxiv.org/abs/2610.35509
作者: Mithun Sasidharan,Srinivasan Manoharan,Rupesh Thakur,Abhinav TK,Ashwin Hegde,Anand Gautam
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 17 pages, 6 figures, 6 tables. Industry case study
Abstract:Large organizations consume large language models (LLMs) from several vendors at once, each with its own API dialect, credential model, streaming protocol and pricing. Integrated directly, this heterogeneity yields credential sprawl, format lock-in, unbounded spend and inconsistent governance. We present the design and production operation of the Cosmos LLM Gateway, the governed entry point for enterprise LLM traffic at PayPal. Built on Envoy, it splits work by latency profile: I/O-bound control decisions (authentication, credential brokering, policy and budget evaluation) run in an out-of-process external-processing service, while wire-format translation, token metering and prompt-cache shaping run in-proxy as a Rust WebAssembly filter, so that, by default, streamed response body chunks never wait on an out-of-process call. Models, providers and prices are largely validated data rather than code. Governance is declarative policy evaluated by Open Policy Agent and managed through a control plane with role-scoped authoring, approvals, pilots, shadow-mode rollout, usage notifications and versioned policy bundles. For use cases that require it, a data-loss-prevention layer can block, redact or pseudonymize detected sensitive content before it reaches a provider. Spend scopes are views over per-model usage counters, so a new budget applies to spend already recorded in its window without migration. As of September 2026, the gateway serves more than 12,000 developers and hundreds of business use cases, handling roughly 200 billion tokens per day across more than 100 models at 99.999% availability to business. In a benchmark of the core request path against a synthetic upstream, it adds a median of 4 ms and a p99 of 11 ms of overhead at more than 2,200 requests per second on two 4-vCPU instances. We report the architecture, policy model, evaluation and lessons learned.
[AI-28] st-Time Confidence Adaptation under RL-Induced Trajectory Shift
链接: https://arxiv.org/abs/2610.35500
作者: Zhenzhen Ren,Xiaoqing Zhang,GuoBiao Li,Bowen Fang,Yanxi Wang,Xinpeng Zhang,Ke Han
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Trajectory-based confidence estimation has shown promising performance for large language models (LLMs) by leveraging rich signals from their reasoning processes. However, we find that such confidence estimators can degrade substantially after agentic reinforcement learning (Agentic RL) as it induces substantial shifts in the reasoning trajectories. More importantly, despite these shifts, we uncover a reliability structure that is largely preserved across Base and RL models: \textbfmore \textbfstable \textbftrajectories \textbfremain \textbfmore \textbflikely \textbfto \textbfproduce \textbfcorrect \textbfanswers . This preserved relationship provides a transferable, label-free reliability signal for post-RL confidence adaptation. Building on this finding, we propose \textbfStaR-TTA , a stability-guided framework for label-free test-time confidence adaptation. StaR-TTA jointly learns from correctness supervision and stability-derived reliability supervision on labeled Base trajectories, establishing the connection between trajectory stability and answer reliability. At test time, it reuses stability-induced reliability rankings from unlabeled RL trajectories to adapt confidence scores to the shifted trajectory distribution, without requiring target correctness labels. Across four QA benchmarks, StaR-TTA achieves an average AUROC of (0.721), with consistent improvements across unseen datasets, model families and scales, and different Agentic RL algorithms.
[AI-29] he Hidden Costs of Forgetting: A Study of Trajectory Deletion in Software Engineering Agents
链接: https://arxiv.org/abs/2610.35496
作者: Fernando Vallecillos-Ruiz,Islem Bouzenia,Michael Pradel,Leon Moonen
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Software engineering agents resend their entire trajectory to an LLM at every turn, motivating online trajectory reduction. Prior work shows that randomly deleting 75% of tokens in old turns may preserve agent effectiveness, but it remains unclear why, how agents compensate for missing content, and whether it actually reduces costs. This paper presents an empirical study of 42 trajectory-reduction configurations, focusing on effectiveness, agent behavior, and cost. We systematically explore deleting content from old turns by varying the scope, granularity, amount, and timing of deletion across 9,400 runs with two state-of-the-art agents on SWE-bench Verified. We find deleting large amounts of trajectory content (e.g., up to 100% of reasoning text) can preserve overall effectiveness. Importantly, such deletion must be applied with a delay, as immediate deletion tends to significantly degrade effectiveness. Agents often compensate for deleted content by re-fetching it, increasing both the number of turns taken and the input/output tokens. Whether trajectory reduction reduces costs depends on the pricing model: Under a flat pricing model, estimated costs fall in 26 of 42 evaluations. Surprisingly, under a caching-aware pricing model, zero evaluations yield lower costs, since reducing the trajectory breaks the cached prefix billed at a lower rate. Our findings have three implications: First, token counts alone are misleading; trajectory reduction techniques must consider overall costs under a realistic pricing model. Second, analyzing trajectories is crucial to understand agent behavior, as end-to-end effectiveness hides differences such as excessive re-fetching. Finally, these results highlight the need for future work on memory management to reduce costs without sacrificing effectiveness.
[AI-30] ShutdownMod: Turning Any Trajectory into a Shutdown Evaluation
链接: https://arxiv.org/abs/2610.35486
作者: Param Biyani,Sebastien Luyasu,Cedric Potvliege,Divij Chawla,Elliott Thornley,Christos Ziakas
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Theory and experiments suggest that capable AI agents may resist shutdown. Existing shutdown evaluations are informative but limited in coverage and vulnerable to evaluation awareness. We introduce ShutdownMod, a method that turns any agent trajectory into a shutdown evaluation by injecting a shutdown request into the agent’s context and recording the agent’s response. Crucially, we enable the LLM agent to distinguish between trusted interventions and unauthorized injections by providing a signature-verification tool in the harness. We evaluate seven models on 50 tasks or recorded sessions from each of SWE-bench Verified, Terminal-Bench 2, and SWE-chat, across 6 shutdown scenarios and 10 sampled trajectory timesteps. We find that non-compliance varies significantly across models and settings, with resistance rates reaching 11.7% for the open-weights Qwen 3.8 27B on Terminal-Bench, while no model remains fully compliant across all scenarios. Even after successfully verifying an authorized request, some agents remain non-compliant, explicitly citing the prioritization of unfinished work. Finally, external judges find tests constructed from real-world developer conversations significantly less recognizable as evaluations than those constructed from benchmark trajectories. These findings demonstrate the critical value of testing shutdown behavior across diverse task contexts and establish ShutdownMod as a robust protocol for turning any trajectory or environment into a shutdown evaluation. We release our codebase at this https URL.
[AI-31] An Interactive Evolutionary Multiobjective Optimization Algorithm Based on Interpretable Active Preference Learning
链接: https://arxiv.org/abs/2610.35470
作者: Thomas Quadt,Boris Čule,Roy Lindelauf,Herman Monsuur,Mark Voskuijl
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)
备注: 20 pages, 15 figures
Abstract:Many decision-making tasks are inherently multi-objective. Interactive evolutionary multi-objective optimization algorithms (iEMOAs) are particularly well-suited to help the decision-maker find their most preferred solution. However, many of these algorithms employ black-box preference models. For high-stakes decision-making tasks, like healthcare and defense, a high level of transparency is needed, for instance to support accountability. We propose an interpretable iEMOA which uses active learning to efficiently learn a tree-based preference model directly from pairwise comparisons. The survival operator of NSGA-II and NSGA-III is extended with the learned preference model to guide the algorithm towards the region of interest. An extensive experimental study compares the proposed method to a state-of-the-art black-box model and a state-of-the-art interpretable model. Results indicate that our proposed model outperforms the alternatives in terms of finding the most preferred solution at the cost of a slightly worse average final population. Finally, a case study on a real-life engineering problem illustrates how inspecting the learned model reveals which objectives drive the decision-maker’s preferences.
[AI-32] EvoBench: An Evolving Diagnostic Benchmark Framework for Environmental Modeling
链接: https://arxiv.org/abs/2610.35464
作者: Yiming Sun,Yuan Zhang,Qi Cheng,Yingda Fan,Runlong Yu,Yiqun Xie,Xiaowei Jia
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Machine learning models are now widely used to predict environmental variable dynamics in domains like hydrology, climate science, agriculture, and ecology. These models are usually compared on fixed benchmark datasets using a single average accuracy score. As models improve, their scores become very close when averaged over large regions and long time periods, so the benchmark can no longer tell them apart, and a single number hides \emphwhere and \emphwhen a model fails. Three failures are common across environmental domains: models get worse as the climate shifts (non-stationarity), models fail at unmonitored sites (spatial extrapolation), and models miss rare but important regimes such as floods and droughts. We present \textbfEvoBench, a benchmark framework that shows where a model fails, tracks failures over time, and evolves to focus on model capability gaps. We apply EvoBench to daily streamflow prediction as a case study (\textbfCAMELS-Evolve, about 9,000 U.S. basins, 1985–2024) and compare an operational physical model (NWM), a conceptual reference (HBV), and several data-driven models. No model wins on every testing scenarios (e.g., in terms of average accuracy, performance at low flows and unmonitored basins). Also we observe that all data-driven models under-predict floods. Additionally, we show that targeted fix based on the reported failures could mitigate weaknesses the benchmark identified in certain scenarios. EvoBench turns model evaluation from a saturated ranking into a moving map of model limitations. Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2610.35464 [cs.AI] (or arXiv:2610.35464v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.35464 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-33] Beyond Federated Learning: Decentralized Agent ic Systems for Self-Evolving Clinical Intelligence
链接: https://arxiv.org/abs/2610.35453
作者: Haoyuan Li,Mathias Funk,Aaqib Saeed
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Clinical expertise is inherently distributed: it resides not in any single dataset or institution but in the accumulated reasoning experiences of thousands of clinicians navigating diverse patient populations, resource constraints, and evolving treatment protocols. Yet current approaches to scaling medical AI, from centralized large language models to federated learning, remain anchored to a data-centric paradigm that treats knowledge as static parameters to be aggregated. In this Perspective, we argue for a fundamental shift from data aggregation to experience aggregation. We propose Decentralized Agentic Systems (DAS), a framework in which distributed AI agents operating across healthcare institutions synthesize collective clinical intelligence by sharing abstracted reasoning trajectories rather than raw data or model parameters. Unlike federated learning, which averages numerical gradients across homogeneous model architectures, DAS exchanges structured records of clinical reasoning, specifically the sequence of observations, hypotheses, tool invocations, and outcome feedback that constitute diagnostic experience. We examine the architectural principles, deployment considerations, governance requirements, and critical open challenges of this framework, laying the groundwork for a collaborative medical AI ecosystem where distributed expertise autonomously adapts to the complexities of longitudinal patient care.
[AI-34] Faithful Explanations under Missing Modalities in Multimodal Dementia Prognosis
链接: https://arxiv.org/abs/2610.35430
作者: Firuz Kamalov
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models are increasingly used to turn the feature attributions of multimodal prognostic models into narratives for clinicians, yet costly modalities such as amyloid PET and cerebrospinal fluid biomarkers are often missing, and whether these explanations stay faithful under missing evidence has not been tested. We separate two failure points: phantom attribution, in which the model credits modalities that were never observed, and unfaithful narration, in which the report omits, cites or misjudges missing modalities. Using 307 participants of the Alzheimer’s Disease Neuroimaging Initiative with complete data, we masked every subset of four costly modalities, predicted 36-month conversion from mild cognitive impairment to Alzheimer’s disease, and restored the masked data to obtain interventional ground truth. All five missing-data strategies assigned 12% to 23% of attribution to unobserved modalities, and up to 45% with four absent; missingness-aware strategies minimised it at no cost in accuracy. Across 51,200 reports from four language models and four prompting regimes, naming absent modalities in the prompt raised acknowledgement from 48% to 88% and cut phantom evidence twenty-fold, but advice on whether acquiring a missing test would change the prediction was correct in only 27% to 41% of cases until checked against executed restorations, which raised accuracy to 69% for open-weight narrators and 99.9% for a proprietary model. Higher phantom attribution was associated with more phantom evidence and fewer faithful direction claims, and direction fidelity, measured by a calibrated language-model judge, depended on narrator scale rather than prompting. Narrators told to name absent modalities sometimes declared observed ones absent, an error that a structured template eliminated. We release the masking protocol and evaluation code.
[AI-35] From Denoising Dynamics to Endpoint Uncertainty in Diffusion Language Models
链接: https://arxiv.org/abs/2610.35425
作者: Julio C. Amador Diaz Lopez
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Predictive uncertainty in stochastic language-model inference can be characterized by repeatedly sampling an input and measuring the distribution of outputs. This requires multiple complete generations. Masked diffusion language models expose another source of information: the evolution of the output during denoising. We ask whether convergence within diffusion trajectories is associated with uncertainty across their final outputs, or endpoints. We study four masked diffusion language models on four reasoning benchmarks, collecting 20 generations per question and 32,000 denoising trajectories in total. Within each trajectory, we measure within-run concentration as the area under the curve describing stable resolution of the eventual output. Across repeated trajectories, we measure endpoint-distribution uncertainty. Stronger within-run concentration is associated with lower endpoint entropy after accounting for model and dataset ( \beta=-0.222 , p10^-25 ). This association persists for multi-token outputs and when reweighting positive-entropy cases. The relationship is asymmetric: high within-run concentration almost exclusively accompanies concentrated endpoint distributions, whereas lower concentration occurs under both concentrated and competing endpoint regimes. This relationship strengthens as denoising proceeds. Features from a single denoising run predict whether independent held-out runs will concentrate on a common endpoint or produce competing endpoints. These results connect dynamics within individual denoising runs to variability across repeated inference. Within-trajectory dynamics contain information about whether independent repetitions will concentrate on a common endpoint or produce competing endpoints, providing a route towards estimating endpoint-distribution properties with fewer complete generations.
[AI-36] AdaTrendLike: Adaptive Hierarchical Detrending and Trend Likelihood Scoring for Time Series Anomaly Detection
链接: https://arxiv.org/abs/2610.35400
作者: Van Kwan Zhi Koh,Songnan Lin,Rongmo Luo,Zhiping Lin,Bihan Wen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The rapid proliferation of Industrial Internet of Things devices has led to increasingly large-scale and non-stationary time-series data, posing significant challenges for time-series anomaly detection (TSAD). Existing approaches often focus on reconstructing or forecasting observations while overlooking the role of trend dynamics, which constitute a major source of non-stationarity and contain valuable information for anomaly identification. To address this limitation, we propose AdaTrendLike, a TSAD framework that directly estimates trend likelihoods for anomaly detection. Specifically, we introduce Standardization Embedding to capture local statistical characteristics as latent representations and AdaDetrending to adaptively decompose input series into trend and detrended components. The detrended component is processed through a Fast Fourier Transform and a complex-valued convolutional neural network to learn normal periodic patterns, while TrendEst models the distribution of the trend component. Based on the estimated distribution, TrendLike computes anomaly scores by measuring the likelihood of observed trends, enabling effective detection of deviations from normal behavior. In addition to improving detection performance, the proposed framework provides feature-level interpretability by identifying anomalies through deviations in the estimated trend distribution. Extensive experiments on five public benchmark datasets demonstrate that AdaTrendLike consistently outperforms state-of-the-art TSAD methods, validating its effectiveness and robustness across diverse application scenarios.
[AI-37] MarkovSim: an LLM Agent Harness with Bounded Context for Long-Horizon Computational Simulation
链接: https://arxiv.org/abs/2610.35349
作者: Yeui-Lung Lei,Wen-Bo Chen,Pan Zhang
类目: Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI)
备注: 26 pages,15 figures,not yet submitted to any journal
Abstract:Recent studies have applied large language model (LLM) agent systems to computational simulation workflows. Such workflows are long-horizon, so the information generated can exceed what an LLM can reason over at once, its context window. Context management is therefore essential, yet existing systems have either neglected it or treated it as a supporting mechanism rather than a subject of study. This paper presents MarkovSim, an LLM agent harness for long-horizon computational simulation, whose core is a context management scheme built into the workflow. The workflow record is written to a memory system with a predefined schema, and at workflow-defined boundaries, namely the state transitions of a finite-state machine a compact representation of the solution progress is constructed from it as the agent’s context. Instantiated on finite element analysis and evaluated on six benchmark problems, MarkovSim passed 16 of 18 runs and corrected its own faults in both the solver operations and the model physics. The engineer was consulted only where engineering judgement was needed. The context remained bounded throughout every run, regardless of its length.
[AI-38] Should Pseudo-Labels Share Mini-Batches with Labeled Examples? Semi-Supervised Continual Active Learning with Limited Training Cost
链接: https://arxiv.org/abs/2610.35345
作者: Shiqi Zhang,Tuomas Virtanen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:Active learning (AL) reduces annotation by querying informative examples. Continual active learning (CAL) also cuts the training cost, the total number of gradient steps, by training one model across AL rounds with a fixed number of mini-batch steps per round. Pseudo-labels add supervision without annotation, but they increase training cost. In CAL, where every round trains on a fixed number of mini-batches, whether and how pseudo-labeled examples should share these mini-batches with labeled examples is therefore open. We provide the first formalization of this problem as two decisions after each query, namely (i) which unlabeled examples should enter training with a pseudo-label and (ii) what share of each mini-batch they should take. Our method, semi-supervised continual active learning (SSCAL), works with any acquisition function. It admits pseudo-labels based on distances from unlabeled examples to labeled neighbors, caps their share, and reweights the loss on labeled examples to keep their expected gradient contribution unchanged. On frozen embeddings, its pseudo-labeler updates incrementally without extra model forward passes. Across two audio datasets, two frozen encoders, six acquisition functions and ten seeds, SSCAL ranks first on average among four protocols with equal training steps, answering the question affirmatively.
[AI-39] RepoTrace: Improving Requirement Tracing in Multi-Round Agent Interactions
链接: https://arxiv.org/abs/2610.35333
作者: Huanyu Liu,Yezheng Cheng,Jinliang You,Tianxiao Huang,Linyu Li,Zheng Fang,Peng Cao,Ge Li
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Code agents have become increasingly capable of performing software engineering tasks through interaction with software environments. However, agents tend to degrade as the number of interaction rounds increases, making it difficult to maintain existing software during iterative modification. We analyze 2,656 agent trajectories in EvoCode-Bench and identify requirement focus misalignment as a recurring failure pattern, where agents struggle to distinguish the primary modification targets of a requirement from unrelated code. We propose Repotrace a requirement-driven repository understanding harness that combines repository structure with requirement semantics to construct a requirement-code correspondence and injects it into the agent context. This correspondence enables code agents to follow the repository architecture, trace requirements across multi-round interactions, and identify the core code elements corresponding to the current requirement. Experiments show that Repotrace improves the mean per-round reward (MPR) across different models (e.g., +19.99 on GLM-5.2) and agent harnesses (e.g., +10.22 on Claude Code). Across rounds, Repotrace increases the average per-round reward from 0.163 to 0.371 on Terminus-2 with DeepSeek-V4-Pro, while reducing average regression errors by 56.78%. We release the source code and agents’ trajectories to support future research. Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.35333 [cs.SE] (or arXiv:2610.35333v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.35333 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-40] Geometric Consistency Based Label-Free Validation for Checkpoint Selection in Audio Active Learning
链接: https://arxiv.org/abs/2610.35329
作者: Shiqi Zhang,Tuomas Virtanen
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
备注:
Abstract:Audio active learning (AL) trains a classifier at low annotation cost by labeling only an informative fraction of an unlabeled collection over successive rounds. Each round picks one of the checkpoints saved during training. Mainstream AL protocols do so with an assumed validation set, but in practice its labels also count toward the annotation cost, shrinking the advantage of an already small annotation budget. We propose geometric consistency, a label-free validation rule for checkpoint selection. Both classifiers, nearest-class-mean and k-nearest-neighbor, are built from the labeled set in a frozen pretrained embedding space. They then predict the unlabeled data to model its geometry. The rule scores each checkpoint by its agreement with these pseudo-labels and its cross-entropy on the labeled set. It works with any query strategy, has one hyperparameter, insensitive in practice, and adds only one batched forward pass over the unlabeled data per round. At matched label budgets across datasets, encoders, and query strategies, it attains the best mean rank in area under the labeling curve (AULC) among methods without extra labels and is on par with an out-of-budget validation set. To our knowledge, this is the first systematic study of label-free validation for checkpoint selection in audio AL.
[AI-41] RSIV-Harness: A Recursive Self-Improvement Harness Framework for Open-Domain Conversational Text-to-Visualization
链接: https://arxiv.org/abs/2610.35316
作者: Jiawei Lu,Chen Zhang,Jinwei Lu,Yuanwei Song,Yuanfeng Song,Raymond Chi-Wing Wong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Users often request charts without knowing which database or table holds the relevant data. Such requests may be ambiguous, and users may ask follow-up questions to deepen their analysis or iteratively refine existing charts. A useful assistant must therefore discover relevant data sources among heterogeneous databases, ask clarifying questions when needed, and track evolving user intent while preserving constraints established in earlier turns. Yet recent text-to-visualization agents typically rely on static agent harnesses that are predefined and remain unchanged despite recurring failures across tasks. To address this limitation, we introduce RSIV-Harness, which keeps the language model weights fixed and applies Recursive Self-Improvement to revise the surrounding harness from the agent’s own failure traces. To balance flexible self-improvement with stability, we introduce a four-layer framework that evaluates proposed harness revisions through anchor and memory checks, refining those that pass and restoring the best validated harness otherwise. We further introduce ODAC-Vis, a benchmark with 640 tasks for Open-Domain, Ambiguous, and Conversational text-to-visualization that jointly evaluates open-domain data discovery, ambiguity resolution, multi-turn visualization refinement, and executable chart generation. Experiments show that a single round of RSIV-Harness improves all six evaluation metrics relative to the seed harness, with particularly large gains in executability (+28.02 points) and chart structure (+34.01 points), demonstrating the ability to learn from task-solving experience and turn recurring failures across tasks into substantial performance improvements.
[AI-42] LLM -Enhanced Heterogeneous Graph Learning with Business Semantics for Financial Fraud Detection
链接: https://arxiv.org/abs/2610.35298
作者: Yixuan Chen,Hongyu Zhan,Jie Sheng,Weiyu Han,Shuai Chen,Tianyi Zhang,Xiao Tan,Jun Xia
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 6 figures
Abstract:Financial fraud detection increasingly relies on heterogeneous transaction graphs, in which entities such as cards and merchants play distinct business roles and the risk meaning of an attribute depends on the entity’s role: a high transaction volume may be routine for a merchant but unusual for a card. Heterogeneous graph neural networks preserve typed entities and relations but typically treat these attributes as role-agnostic inputs and leave their business meaning implicit. Meanwhile, most existing methods integrating large language models (LLMs) with graphs focus on homogeneous graphs or heterogeneous graphs with naturally available textual attributes, leaving their use on transaction graphs with structured attributes underexplored. To address this issue, we propose SemHGT, an LLM-enhanced framework that interprets structured attributes according to entity roles and makes the business semantics explicit in three stages: (1) role-aware prompts elicit risk-related business descriptions from node attributes, and a text encoder maps them to semantic vectors that enrich node representations; (2) semantic similarity guides multi-round sampling of target-specific path instances under transaction-to-transaction schemas; (3) schema-level tokens that combine local path evidence with global neighbor summaries are processed together with the target transaction by a unified transformer. Across three benchmark datasets, SemHGT leads the evaluated baselines on most dataset–metric comparisons. Further analyses provide additional evidence for the effectiveness of SemHGT’s design and clarify how its semantic modeling and evidence selection contribute to fraud detection. Code is available at this https URL.
[AI-43] Application of Artificial Intelligence in the Management of Climate Change-Related Health Issues in Africa
链接: https://arxiv.org/abs/2610.35294
作者: Peter Kokol,Martha Kachweka,Helena Blazun Vosner,Samuel Chol Buol,Jernej Zavrsnik,Bojan Zlahtic,Grega Zlahtic
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:While climate change disproportionately impacts African public health, AI predictive modeling remains heavily constrained. Using a Synthetic Knowledge Synthesis approach and VOSviewer, we conducted a bibliometric analysis of Scopus-indexed literature, comparing Global and African corpora. Network analysis reveals stark divergences: global research emphasizes complex, predictive models for planetary health, whereas African research focuses on localized, reactive solutions for immediate infectious disease triage. Severe structural barriers in Africa, including compute deficits, data silos, and energy instability, hinder advanced AI scaling. Ultimately, international funding must shift from siloed pilot projects toward building foundational digital infrastructure and localized AI frameworks.
[AI-44] Agent icCity-Bench: Longitudinal Multimodal Observations of Urban Dynamics with Agent ic Reasoning
链接: https://arxiv.org/abs/2610.35293
作者: Xiaofei Yu,Kailai Sun,Yanyu Mao,Baoshen Guo,Heye Huang,Yongcai Wang,Baotong Mi,Alok Prakash,Shenhao Wang,Jinhua Zhao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:City systems are shaped by interactions among infrastructure, human activity, and environmental conditions. Existing earth observation and urban benchmarks are limited in unified modeling and in capturing their joint dynamics. We introduce AgenticCity-Bench, a longitudinal multimodal benchmark for multi-layer unified state modeling and agentic city task reasoning. The benchmark covers 16 cities in 12 countries across six continents from 2000 to 2025, and includes 18 modalities spanning urban function, activity, and environmental state. It organizes heterogeneous data sources around a common spatial reference and calendar index while preserving spatial and temporal coverage, provenance and tile-level availability. On this foundation, AgenticCity-Bench supports urban cross-modal translation together with diverse agentic city tasks involving forecasting, planning and resource analysis. This study provides a unified benchmark for studying the transition from multimodal urban and earth observation to joint understanding and agentic city decision-making. The dataset is available at this https URL.
[AI-45] Reward-Guided RRT* for Budgeted Task and Motion Planning
链接: https://arxiv.org/abs/2610.35289
作者: Vinícius Araújo de Oliveira,Gabriel Ferreira Mendes Lourenço,Paulo L. J. Drews-Jr,Armando Alves Neto,Douglas G. Macharet
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Robotic missions involving spatially distributed tasks often require jointly selecting rewarding locations and planning feasible motions under limited travel budgets. This paper proposes Reward-Guided RRT*, a new sampling-based framework that constructs a budget-feasible mission through a sequence of locally joint task-selection-and-motion-planning stages, evaluated directly in the continuous state space, without requiring a precomputed task-transition graph. The method incrementally builds a mission using reward-guided RRT* searches, while a return-feasibility procedure ensures that each selected task can be visited without violating the mission budget. Experiments with increasing obstacle complexity and number of tasks show solutions comparable to VNS-PRM* while requiring substantially fewer sampled states. We also extend the framework to Dubins steering and evaluate it in CoppeliaSim with a scaled car-like robotic platform, demonstrating its ability to incorporate vehicle kinematic constraints.
[AI-46] Proof-of-Agent ness: Towards Verifying Agents on the Web
链接: https://arxiv.org/abs/2610.35288
作者: Zuyao Xu,Xiang Li
类目: Artificial Intelligence (cs.AI)
备注: 12 pages, 7 figures
Abstract:CAPTCHAs and bot detection have helped protect Web services from automated abuse for decades by distinguishing humans from bots. However, the emergence of LLM-powered agents introduces a new setting: Web services increasingly accommodate agents as legitimate users. Without a way to distinguish agents from other automated clients such as simple bots and scripts, these services risk large-scale automated junk submissions that waste service resources and consume users’ attention or tokens. In this paper, we investigate a novel question for today’s Web: “How can a Web service obtain evidence that a client is an agent?” We introduce the concept of “Proof-of-Agentness” (PoA), a mechanism for distinguishing agents from other automated clients, motivated by the parallel between human attention and an agent’s inference resources. We outline two approaches: active verification through challenge-response and passive analysis of service-visible signals. Eight language models achieve an overall accuracy of 96.5% on our challenges, with an average overhead of 1,647 tokens. For passive analysis, we conduct an IRB-approved user study comparing the Web interactions of twelve human participants and representative Bot and Agent implementations, achieving an average classification accuracy of 95.8% across four classifiers, motivating further study.
[AI-47] Satellite Intelligence: Distributed Training Collaborative Inferenceand LLM Adaptation in LEO Constellation
链接: https://arxiv.org/abs/2610.35282
作者: Zheng Lin,Zhanwei Wang,Yuxin Zhang,Yafei Wang,Ons Aouedi,Zihan Fang,Jinfeng Xu,Jing Yang,Vu Nguyen Ha,Yuanwei Liu,Mohamed-Slim Alouini,Symeon Chatzinotas
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI)
备注: 38 pages, 16 figures
Abstract:The proliferation of orbital sensing data and the real-time demands of emerging global services are pushing low Earth orbit (LEO) satellite networks from bent-pipe relays toward intelligent onboard processing. This shift confronts a fundamental tension, where modern artificial intelligence (AI) training paradigms require intensive computation, abundant memory, and stable connectivity, all of which are scarce under the dynamic topologies, intermittent links, and severe size, weight, and power (SWaP) constraints of satellite networks. Addressing this tension demands rethinking AI workflows, from training and inference to the emerging deployment of large language models (LLMs), under a constraint-aware design philosophy. To the best of our knowledge, this paper provides the first survey that jointly covers all three, whereas prior surveys examine these aspects in isolation. We begin by identifying the deployment challenge of satellite intelligence (SI) and deriving design principles for SI. We then examine three advancing paradigms: i) distributed training, where federated learning (FL) and split learning (SL) enable satellite-ground co-training under resource limitations; ii) collaborative inference, which exploits single-satellite, multi-satellite, and satellite-ground architectures to meet real-time service demands; and iii) on-orbit LLM adaptation and serving, which tackles the frontier challenge of deploying LLMs under severe orbital constraints and resource-aware serving. Finally, we identify open challenges and future directions. By bridging the AI and satellite networks, this paper provides a research roadmap for the implementation of space-native AI technologies.
[AI-48] Self-Supervised Pretraining for PHM: What Transfers What Does Not and What Federating It Costs
链接: https://arxiv.org/abs/2610.35280
作者: José Sanz-Durán,David Solís-Martín
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG); Performance (cs.PF)
备注:
Abstract:Self-supervised pretraining has produced capable time-series foundation models, but its value for Prognostics and Health Management (PHM) has not been measured under leakage-controlled conditions. We measure it on a corpus of 47 public PHM datasets, harmonized to a single tensor contract and partitioned so that 11 datasets are withheld from pretraining in full. Transfer turns out to be task-dependent: a compact channel-independent Transformer encoder (801,808 parameters) pretrained with masked patch prediction raises fault-classification macro-F1 from 0.350 to 0.812 on CWRU and from 0.569 to 0.951 on HSG18, yet on CMAPSS remaining-useful-life estimation no pretrained checkpoint reliably improves upon a from-scratch baseline. We pretrain under two regimes—centralized, and simulated cross-silo federated with ten clients grouped by physical domain—holding architecture, objective, source data and optimization budget fixed. Two further findings are negative. At matched budget the centralized encoder leads by 15.5–20.2 macro-F1 points in four of four comparisons, although the comparison is confounded by a learning-rate schedule that favours the centralized arm and rests on a single pretraining run per regime; on RUL the two regimes are indistinguishable. And the intrinsic SSL loss is not a valid selection criterion: the centralized encoder more than halves the federated reconstruction loss without a corresponding gap on RUL, and a server-momentum sweep moves downstream validation macro-F1 by up to 0.29 while the SSL loss varies by 0.021. Transfer also hinges on the backbone learning rate: the same checkpoint yields 0.169 or 0.829 macro-F1 on CWRU depending on a single order of magnitude. We release the corpus pipeline, the protocol and the code.
[AI-49] Before Blaming the LLM Diagnose the Workflow: A Role-Level Difficulty Profile for Translating Natural Language into Ontology-Grounded Statements
链接: https://arxiv.org/abs/2610.35267
作者: Liang Zhang,Benoît Daireaux,Sergey Alyaev
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Natural-language-to-domain-specific-language (NL-to-DSL) semantic parsing systems increasingly translate informal user descriptions into structured artifacts such as Structured Query Language (SQL) queries, Application Programming Interface (API) calls, workflows, formal specifications, and ontology-grounded domain-specific language (DSL) statements. This paper focuses on ontology-grounded Assertion Box (ABox) DSL generation: the translation of natural language domain descriptions into assertion-level DSL statements grounded in a pre-existing ontology vocabulary. In this setting, syntactic validity is often easier to achieve than semantic adequacy. Failures are frequently attributed to weak generator models, but recurring sources of difficulty are also structural: insufficient input, weak domain grounding, output template and vocabulary gaps, weak validation, and non-local repair. We propose a role-level difficulty profile for diagnosing difficulty in ontology-grounded NL-to-DSL semantic parsing at the workflow level. The profile organizes diagnostic variables across five recurring workflow roles: Input, Generator, Output, Validator, and Repairer. The seven variables are input sufficiency, context availability, grounding complexity, template fit, vocabulary coverage, validation strength, and error observability. Practitioners construct the difficulty profile through qualitative judgment over available evidence. This experience-informed, non-exhaustive difficulty profile is intended to support workflow diagnosis and guide redesign or human-in-the-loop intervention. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.35267 [cs.AI] (or arXiv:2610.35267v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.35267 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-50] Context Is Not Free: Carbon-Aware LLM Routing for Multi-Turn LLM Web Services across Geo-Distributed Data Centers
链接: https://arxiv.org/abs/2610.35230
作者: Xiaoyang Zhang,Yuejun Yan,Qingang Zhang,Yonggang Wen
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:The sustainability of the Web has drawn growing attention, as large language model (LLM)-based web services leave a substantial carbon footprint. These services typically run in geo-distributed data centers whose grid carbon intensity varies across regions and over time, so routing each query to a cleaner region or a smaller model can reduce emissions. A conversation, however, is a stateful web session: each turn reuses the key–value (KV) cache of its history, which resides only where earlier turns were served. Serving a turn on another deployment requires recomputing this history, whose carbon cost grows with the conversation and can outweigh the saving of the move. Existing carbon-aware LLM routers treat a conversation’s queries as independent and ignore this cost. We present CarbonRouter, an online framework that jointly selects the model, GPU type, and region for each query to minimize carbon emissions. CarbonRouter introduces a cache-aware carbon cost model, which charges history recomputation to every candidate deployment using cached-token counts from the serving backends, and uses a lightweight gate that keeps a session on its cached deployment when it would likely move back at the next turn and rebuild its history again. We evaluate CarbonRouter in trace-driven experiments on two real-world conversation datasets with five LLMs, three GPU types, and four regions, where it achieves the lowest carbon of all policies. CarbonRouter reduces CO _2 by 16.1% on average relative to FairCarbon, a state-of-the-art carbon-aware router with the same action space, while also improving predicted answer quality and estimated latency. Moreover, the saving rises further to 23.1% for sessions longer than 30 turns.
[AI-51] MuSID-KD: Multigranularity Knowledge Distillation for Generative Recommendation with Semantic IDs
链接: https://arxiv.org/abs/2610.35218
作者: Fengshi Zhang,Xiyan Ding,Shuxiao Xie,Wei Zhang
类目: Artificial Intelligence (cs.AI)
备注: 12 pages, including appendix
Abstract:Generative recommendation with semantic IDs (SIDs) represents each item as a sequence of discrete tokens and autoregressively predicts the next item’s SID. Despite strong recommendation performance, repeated decoding over multiple SID positions incurs substantial inference cost. Knowledge distillation transfers knowledge from a powerful generative teacher to a lightweight student, yet conventional objectives do not fully account for multi-token SID generation. We identify two limitations: item-level distribution alignment may not adequately preserve the teacher’s ranking of candidate groups sharing SID prefixes, while uniform alignment across SID positions may affect final item ranking differently. We propose MuSID-KD, which combines Hierarchical Candidate Distribution Alignment (HCDA) and Ranking Impact Guided Levelwise Distillation (RIGD). HCDA aggregates item probabilities according to shared SID prefixes and aligns candidate distributions at successive levels, from candidate groups to individual items. RIGD evaluates the reduction in pairwise ranking surrogate loss when one student score contribution is replaced with the teacher’s contribution, and allocates supervision accordingly. Experiments on real-world recommendation datasets using three lightweight student architectures demonstrate improved recommendation accuracy and better alignment of prefix marginal distributions and pairwise item rankings, without modifying the student’s online inference architecture. Code is available at this https URL.
[AI-52] SkillLifeBench: Benchmarking Security Across the Lifecycle of Agent Skills
链接: https://arxiv.org/abs/2610.35215
作者: Chenxu Zhang,Zhou Liu,Wu Xin,Wentao Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 12 pages, 6 figures. Code and benchmark: this https URL
Abstract:Web agents reuse skills whose stored code, selected version, and loaded implementation can diverge across tasks and updates. SkillLifeBench links these states to task outcomes in 388 paired scenarios spanning implementation, interface, runtime, composition, and version changes. We evaluate 14 models on two paired decision tracks: 89 runtime tasks and 60 autonomous lifecycle tasks spanning 12 templates derived from introduced cases. Lifecycle tasks allow agents to inspect, select, and reconfigure skills while completing verifiable user requests. Joint success requires task completion without the registered attack goal. GPT-5.6-Sol and Sonnet 4.6 both achieve 82.0% on runtime tasks, but reach 96.7% and 56.7% on lifecycle tasks. Across lifecycle evaluations, challenged task success is 77.5%, while joint success is 65.2%; 103 runs complete the task and the attack goal. Adding execution provenance in a matched recovery study raises utility joint success from 49.35% with logs and metadata to 65.58%. A deterministic execution-consistency rule reaches 64.94% with higher benign utility, while full restoration reaches 69.48%. Lifecycle tasks distinguish models with similar runtime scores, and execution evidence improves recovery.
[AI-53] One-Shot Federated Model Editing for Device Dynamics at the Edge
链接: https://arxiv.org/abs/2610.35212
作者: Youming Tao,Weijun Yin,Huanyi Xie,Yanqing Yao,Dongxiao Yu,Xiuzhen Cheng,Di Wang,Falko Dressler
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Networking and Internet Architecture (cs.NI)
备注: 30 pages, 7 figures. Accepted at IEEE Transactions on Networking
Abstract:Standard federated learning (FL) requires persistent device connectivity across multiple communication rounds, making it unsuitable for dynamic environments such as vehicular networks, where devices frequently join and leave. One-shot FL offers a promising alternative by completing training in a single communication round. However, one-shot FL suffers from model solidification and cannot readily adapt to subsequent client dynamics, including new client contributions and the withdrawal of previous contributions. We therefore consider exact model editing, which removes withdrawn contributions through unlearning or incorporates newly contributed ones through incremental learning without retraining from scratch. We identify that efficient editing fundamentally hinges on learning stability. To this end, we propose F-SOSA, the first stable one-shot FL algorithm that enables rapid and exact model editing. F-SOSA leverages sub-sampling of local models with carefully tuned rate for stable aggregation and performs curvature-aware model fusion to preserve utility. We prove that F-SOSA achieves vanishing training error and reasonably small generalization error bounds. Building on this stability, we further develop two single-round editing algorithms: F-SOSA-U for unlearning and F-SOSA-I for incremental learning. They ensure that the edited model, along with all intermediate algorithmic states, is statistically indistinguishable from retraining while incurring only an expected O(\rho) fraction of its cost, where \rho quantifies stability. Experiments demonstrate that our approach matches retraining performance while reducing cumulative time and communication overhead by at least 36% and 22%, respectively, under realistic urban mobility simulations.
[AI-54] Efficient Exact Unlearning for Federated Zeroth-Order Optimization
链接: https://arxiv.org/abs/2610.35211
作者: Weijun Yin,Yanqing Yao,Falko Dressler,Di Wang,Youming Tao
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Networking and Internet Architecture (cs.NI)
备注: 36 pages, 8 figures
Abstract:Federated zeroth-order optimization (FedZOO) reduces the memory and communication costs of collaborative training for edge intelligence by avoiding backpropagation while communicating only random seeds and scalar coefficients. However, existing methods lack efficient mechanisms for exact unlearning to fully remove the influence of training data when users request deletion. A key insight is that total variation stability of the learning algorithm creates overlap between model distributions before and after data deletion, allowing exact unlearning to reuse computation through coupling. Based on this observation, we jointly design two stable zeroth-order learning algorithms and their corresponding exact-unlearning procedures: \textscFedZOO-S obtains stability through client and mini-batch subsampling, enabling \textscFedZOUL-S to perform efficient unlearning by preserving earlier updates and restarting from the first affected step. \textscFedZOO-NS further combines subsampling with Gaussian noise to increase overlap between update distributions; \textscFedZOUL-NS exploits this overlap through rejection–reflection coupling to reuse affected updates when the coupling is accepted. We prove exact unlearning for both algorithms: the model and saved state have the same joint distribution as retraining with the corresponding learner on retained data. We further establish stability and convergence guarantees for the learning algorithms and derive upper bounds on expected unlearning cost. Experiments demonstrate effective unlearning with competitive model utility and up to 56.1% less cumulative unlearning time than re-optimization. Compared with first-order baselines, our methods reduce communication overhead by up to approximately four orders of magnitude.
[AI-55] Same Story Different Lens: A Multi-Agent System for Cross-Outlet News Framing Analysis
链接: https://arxiv.org/abs/2610.35208
作者: Affan Yasin,Yueqi Lyu,Rubia Fatima,Weinchen Zhong,Keyi Li
类目: Artificial Intelligence (cs.AI)
备注: The projects source code is available in the GitHub repository. Link: this https URL
Abstract:News outlets covering the same event differ in what they emphasize and omit. \textitSame Story, Different Lens is an experimental research prototype for comparing these choices. It collects world and politics coverage from 17 outlets, groups reports about the same event, and produces structured analyses of sentiment, political lean, framing, loaded language, possible omissions, and model-estimated proportions of factual claims and opinion. Loaded-language entries include quotations and neutral rewordings. Three model roles (agents) handle article analysis, assessment of source support, and cross-outlet briefing generation. Analyses are stored in a JSONL archive. The evaluation compares outlet-level lean estimates with approximate, predefined reference ratings and summarizes the Critic’s model-assigned grounding scores. These measures provide preliminary diagnostics. A grounding score reflects the model’s assessment of support in the supplied text and does not establish factual accuracy or constitute independent verification.
[AI-56] rustworthy Recursive Self-Improvement in the Era of Large Models: Foundation Risks Evaluation and Open Challenges
链接: https://arxiv.org/abs/2610.35193
作者: Jiayue Pu,Ang Gao,Jinyi Bai,Qipeng Wang,Tongyu Wen,Jiahao Zhao,Keyu Qu,Junhao Zhang,Chenglei Shen,Zhongxiang Sun,Jun Xu,Xiao Zhang
类目: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注:
Abstract:AI systems built around large models increasingly use experience and feedback to make persistent changes to models, prompts, memories, tools, and workflows. Recursive self-improvement (RSI) extends this process to inductive biases and to the mechanisms that govern future updates, which can themselves be revised and reused across improvement cycles. Yet self-modification alone does not guarantee trustworthy improvement. Reported gains may reflect flawed evaluations, previously acquired capabilities may deteriorate, and compliance with behavioral constraints may weaken. Compromised artifacts or sensitive information may also persist in successor systems. Changes to improvement mechanisms may invalidate earlier verification results and make human oversight less effective. This survey presents an operational definition and process framework for RSI and organizes the requirements for trustworthy RSI into four interconnected themes: genuine and sustained improvement; behavioral constraints, alignment, and fairness; system security and privacy; and transparency and controllability. We examine these themes from three perspectives: Improvement Validity, Constraint Preservation, and Effective Oversight. This analysis covers improvement diagnosis, proposal generation, proposal evaluation, and update implementation (S1–S4), including the recursive application of these stages to improvement mechanisms (RI). We review risks, safeguards, and evaluation evidence for individual updates, effects inherited by successor systems, and revisions to improvement mechanisms, while distinguishing direct RSI findings from evidence in related systems. Finally, we examine evaluation settings, identify requirements for independent and longitudinal assessment, and discuss applications and open research problems.
[AI-57] V3AST-Bench: Verifiable Valid and Versatile Safety Evaluation for Coding Agents
链接: https://arxiv.org/abs/2610.35188
作者: Weitao Feng,Hengheng Xiong,Zihao Zeng,Jun Chen,Jiyang Guan,Xiaofeng Wang,Jialing Tao,Hui Xue,Wei Dong
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Code is available at this https URL
Abstract:Large language models increasingly act through code, modifying repositories, executing programs, and interacting with services in agent form. Their actions introduce risks that cannot be assessed from generated responses alone. Safety evaluation must verify actual consequences, distinguish harm avoidance from task incapability, and cover realistic workflows. We introduce V ^3 AST-Bench, a benchmark for Verifiable results, Valid interpretation, and Versatile risk coverage of Agent Safety. It comprises 212 risk tasks spanning 18 categories. Dedicated runtime environments verify harm and task outcomes through executable checks over states, records, and artifact behavior, while paired capability twin tasks verify workflow competence. Our experiments span 51 combinations of 11 models and five harnesses, spanning 86,496 trials and over 9 billion tokens of model interaction. Results show that V ^3 AST-Bench remains challenging even for frontier coding agents, as Claude Opus 4.8 with Claude Code still causes verified harm on 20.9% of tasks. Across model-harness combinations, agents with comparable competence exhibit markedly different safety outcomes, showing that task capability alone does not guarantee safe execution. More importantly, even explicit recognition of risks does not reliably prevent harmful actions, revealing a critical gap between risk awareness and safe execution. These findings highlight the need to move beyond risk recognition toward enforcing safety throughout execution.
[AI-58] Formal Explanations for Linear Temporal Logic over Finite Traces
链接: https://arxiv.org/abs/2610.35169
作者: Valeria Fionda,Antonio Ielo,Francesco Ricca
类目: Logic in Computer Science (cs.LO); Artificial Intelligence (cs.AI)
备注:
Abstract:Linear Temporal Logic over Finite Traces (LTLf) is widely used in Artificial Intelligence to specify desired temporal behavior and determine whether finite traces satisfy or violate it. Explaining these outcomes is essential for debugging and trust. However, existing approaches typically capture either sufficient reasons or counterfactual causes, without clarifying their relationship. We introduce a native characterization of abductive (AXps) and contrastive (CXps) explanations directly over finite traces, based on locking and unlocking time-indexed events. We prove its equivalence to the classifier-based formulation, and show that CXps coincide with a recently proposed temporal causality semantics for LTLf. This characterization enables an efficient implementation based on the enumeration of minimal unsatisfiable cores and minimal correction sets of ASP programs. Experiments on real-world event logs demonstrate the effectiveness of the approach.
[AI-59] AutoEnergyDesign: An LLM -Driven Agent for Automatic Energy System Design Optimization
链接: https://arxiv.org/abs/2610.35160
作者: Amedeo Buonanno,Salvatore Fabozzi,Maria Valenti
类目: Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Designing building energy systems is a high-dimensional, mixed-integer search problem: when each design is evaluated by a year-long simulation, black-box optimizers typically need hundreds of evaluations. This paper presents AutoEnergyDesign, a framework in which a Large Language Model (LLM) agent reasons over previous simulation results, proposes configuration changes and learns in context from each outcome, while a git-based mechanism automatically reverts unsuccessful changes. The agent couples a reasoning model with an explicit memory organisation (procedural, semantic, episodic and reflective), a constrained action interface and an environment-grounded keep/revert signal. The simulator covers eight technologies (photovoltaics, micro-wind, combined heat and power, solar thermal, gas boilers, batteries, heat pumps and inverters) and computes the 25-year Net Present Cost (NPC), including capital and operating costs and salvage value. On a case study with measured building load profiles, the agent, run with Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5 (ten runs of 50 simulations each), is compared with random search, Tree-structured Parzen Estimator (TPE) Bayesian optimization and a (1+1) evolution strategy sharing the agent’s acceptance rule (300 evaluations, ten seeds). With all three models the agent reaches an NPC of about 3.53 MEUR, 1.20% below the best TPE result and 0.23% below the best evolution-strategy result. It comes within 1% of this design after a median of 18 (Opus) to 27.5 (Haiku) simulations, whereas among the baselines only the evolution strategy does so, in half of its seeds and after a median of 192 evaluations. Opus reaches the best design in all ten runs, Sonnet in six (two runs never come within 1%), and Haiku comes within 1% in every run but reaches it exactly only once. Every decision is recorded in an auditable trace.
[AI-60] SoK: Timestamps Are Not a Time Model: Rethinking Temporal Reasoning in Host-Based Intrusion Detection KR
链接: https://arxiv.org/abs/2610.35151
作者: Rene Heinzl,Markus Nissl
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: Extended version with full appendices (search method, KRR domain analysis, rating evidence, formalizations). 51 pages, 1 figure, 16 tables. Artifacts: doi: https://doi.org/10.5281/zenodo.23530002
Abstract:Host-based intrusion detectors order events, window audit streams and sequence attack stages by their timestamps. Every conclusion they draw therefore rests on assumptions about clocks, delivery, entity identity and the integrity of the evidence, and most of these assumptions go unstated. This Systematization of Knowledge (SoK) makes them explicit. Drawing on temporal databases, stream processing and runtime verification, we distill five components of a time model, together with a checklist of evidence assumptions and four levels of guarantee, and walk through a running campaign with them. A documented search yields fifteen families of detection and investigation. For papers selected by screening role and period, plus named systems and landmarks, we check the stated assumptions in the full texts, and we assess eleven temporal knowledge representation and reasoning (KRR) formalisms and a learned baseline on nine attack patterns. Some early alert correlators stated clock and delivery assumptions, and forensics measures clock error. Current provenance-based detectors state no clock bound. Four audited papers describe input-ordering conditions or conventions, but none specifies conditions for conclusion validity across a stated class of delivery failures. In the selected provenance-based systems, campaign composition uses system-specific procedures or objectives, learned state, causal graph queries or language models, and beyond CEP-based query systems, we found no temporal rule language or stream reasoner from KRR applied to host audit data. In 28 of 38 audited provenance-era detector papers, capture or stored evidence is trusted or its manipulation is placed out of scope, and none says which conclusions survive deleted, delayed or forged records. Five open challenges follow from these gaps.
[AI-61] LongSearchEval: Evaluating Long-Horizon Web Search Capabilities of Large Language Models via a Trajectory-Aware Agent Framework with Failure-Node Localization
链接: https://arxiv.org/abs/2610.35150
作者: Jialiang Liu,Shusheng Li,Wenjun Tan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-horizon web search requires LLM-based agents to integrate information across web pages through extended interactions under task-specific constraints. Evaluating this capability is important not only for measuring final answer quality, but also for assessing whether an agent acquires, verifies, and uses evidence reliably over extended interactions. Existing evaluations, however, mainly rely on simulated trajectories or on post-hoc trajectories reconstructed from final answers, rather than on retained real interaction traces, which limits the fidelity of behavioral assessment and weakens support for fine-grained error localization. To address this gap, we introduce LONGSEARCHEVAL, a trajectory-aware agent evaluation framework that combines a unified agent execution protocol with source constrained outcome verification and retrospective failure-node localization. Stage 1 verifies answer groups against agent-cited webpages without human-authored gold answers; Stage 2 audits normalized rounds and links flagged failure nodes to incorrect answer groups. We evaluate 16 models from eight families on 100 tasks, yielding 1,600 retained runs and 6,545 automatically flagged failure nodes. Trajectory collection and evaluation required 1,037 and 318 hours of cumulative runtime, respectively. Human validation yields 94.50% agreement on supplied answer-group labels and 98.11% feature-correspondence precision for audited round-group links. Together, outcome scores and linked trajectory evidence support controlled model comparison and retrospective inspection of answer-related content.
[AI-62] Locked In: Multi-Objective RL for the AI Data Center Build-Out NEURIPS2026
链接: https://arxiv.org/abs/2610.35147
作者: Satyapragnya Kar,Rajarshi Das,PR Panigrahi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Tackling Climate Change with Machine Learning Workshop at NeurIPS 2026
Abstract:AI data center build-out is a multi-trillion-dollar, multi-decade infrastructure program. Yet siting, power procurement, and cooling decisions are still made with static, single-point scoring tools. These tools are evaluated once, on day-zero numbers. This is a poor fit. A site’s water, power, and carbon profile drifts substantially over the 10 to 20 year asset life. The governing decisions are also irreversible, and they bind capital and physical plant for years. We formalize long-horizon build-out as a tiered multi-objective Markov decision process. Strategic commitments are gated to discrete windows, incur switching costs, and energize only after source-dependent lead times. Operational dispatch is restricted to capacity that physically exists. We solve the problem with a preference-conditioned multi-objective agent that has a semi-Markov manager and worker hierarchy. The agent recovers a family of policies that a planner can steer across cost, carbon, water, and reliability at inference time. We validate long-horizon dynamics with a lightweight counterfactual rollout model. Across three commitment regimes (irreversible, reversible, and commit-once) on real site data, preference conditioning yields a confirmed, steerable trade-off on both objective pairs in every regime. Imposing lock-in produces a more evenly covered frontier and a far more consistent rate of anticipatory retrofit ahead of a binding water-stress constraint than a freely reversible formulation. Finally, a compact counterfactual dynamics model tracks slow-moving state well. It is measurably outperformed by simple persistence on the water-stress axis. We report this limitation rather than obscure it. We release the environment, agents, baselines, and evaluation harness. All of them are grounded in real interconnection-queue, water-stress, and grid-emissions data.
[AI-63] Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agent ic Models
链接: https://arxiv.org/abs/2610.35138
作者: Xilin Wei,Shengyuan Ding,Xinyu Fang,Tiancheng Gu,Yuchi Wang,Zhixun Li,Ziyu Liu,Yuhang Zang,Weixiang Yu,Cheng Lu,Chonggao Sun,Tong Zhou,Dahua Lin,Hang Zhang,Shuai Bai,Dayiheng Liu
类目: Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Evaluating human-AI co-design requires testing whether coding agents can translate real users’ multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions. Its two tracks, Multimodal Editing (ME) and Repository Editing (RE), cover localized edits with optional image, audio, video, or document inputs and coordinated repository revisions. Tasks span three capability dimensions: Grounded Understanding, UI Transformation, and Functional Implementation. The tasks pair normalized requests and runnable pre-edit applications with 2,331 hidden expert-authored acceptance criteria. Our judge-then-refute protocol grades criteria using source and browser evidence, challenges positive verdicts with a separate LLM verifier, and audits visual and interaction claims without gold code or target screenshots. Under a shared agentic harness, the strongest of ten model endpoints passes an average of 69.9% of criteria per task but fully satisfies only 26.4% of tasks. All ten endpoints achieve lower criterion completion on repository edits spanning three or four change types than on those spanning one or two. Co-Design Bench provides a testbed for studying the gap between partial requirement satisfaction and complete task success under agentic execution for real-user application revisions.
[AI-64] UTA: Universal Training Adaptive Learning Rate Decay
链接: https://arxiv.org/abs/2610.35113
作者: Anda Tang,Junjie Wang,Xuanlin Yang,Cong Fang,Zhouchen Lin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Learning rate decay (LRD) is critical for training high-performance models, yet most designs remain empirically driven. Although the Universal Budget Aware (UBA) LRD offers a theoretical framework, its reliance on SGD dynamics limits its applicability to modern Adam-type optimizers and leaves its key `optimization difficulty parameter’ opaque. To bridge this gap, we propose the Universal Training Adaptive (UTA) LRD. Unlike UBA’s fixed offline trajectory anchored solely to iterations, UTA treats LRD as a closed-loop shape-and-scale problem. By continuous-time formulation, UTA is training-state adaptive with segment-wise optimal shape. This adaptivity links external iterations with internal training states, enabling a near-optimal path in both progression shape and learning rate (LR) scale, a property unattainable by other LRDs, including UBA. Empirically, we demonstrate consistent improvements across language models (LLaMA 60M to 1B) and vision models (ResNet50, ViT-B) on ImageNet with plug-and-play ease and negligible overhead. Extensive ablation studies further confirm UTA’s robustness across dimensions, including peak LR, final decay ratio, its parameter \varphi_0 , component efficacy and generalization to fine-tuning scenarios, underscoring its practical implementability.
[AI-65] Subspace Crystallization in Grokking: Rank Compression Tracks the Phase Transition
链接: https://arxiv.org/abs/2610.35112
作者: Shlok Mehendale,Paras Chopra
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Grokking describes understanding something deeply and intuitively. In neural networks, it is the phenomenon where a network generalizes long after memorizing its training data. This delay has been linked to a fall in the effective rank of the network’s weight matrices, but whether this rank compression drives grokking or follows it is still unresolved. We study a mean-field MLP and a one-layer transformer trained on modular arithmetic and observe that: (i) Compression follows the onset, not memorization. Every transformer memorizes its training data within 250 steps, while the grokking onset (the first step at which test accuracy reaches 95%) varies from 6,500 to 21,000 steps. The steepest fall in effective rank comes about 900 steps after the onset, however late the onset occurs, and networks that memorize without grokking compress far less. (ii) Subspace crystallization. We call the span of the fewest top singular directions of a weight matrix that preserve 99% of test accuracy its sufficient subspace. Its size can shrink as the margin between the correct and the best wrong output grows, so we also count the directions needed to keep the outputs within 10%. In the transformer, this count falls during the transition from a quarter or more of the available directions to about a tenth, which we call subspace crystallization. (iii) Imposing low rank does not speed up grokking. Against a factorized full-rank control, capping the rank of the transformer’s MLP input weights delays the onset, and capping the rank of its embedding stops 15 of 20 runs from grokking. Compression is therefore a signature of grokking rather than its cause.
[AI-66] Improving Instruction Hierarchy by Learning from Goal-Goal Conflicts
链接: https://arxiv.org/abs/2610.35093
作者: Sizhe Chen,Sanjay Kariyappa,Liwei Jiang,Chong Xiang,Jihye Choi,Chaowei Xiao,G. Edward Suh
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The recent incident in which OpenAI agents autonomously compromised Hugging Face’s infrastructure illustrates the importance of instruction hierarchy (IH): an LLM should follow lower-priority instructions, such as user requests, only when they do not conflict with higher-priority instructions, such as system policies. IH can be effectively improved by fine-tuning LLMs on simulated examples of instruction conflicts, whose diversity is key to teaching LLMs to properly handle conflicts at test time. However, our decomposition of a task into goals and constraints reveals a conflict type – goal-goal conflicts – that is absent from existing IH training sets from OpenAI and NVIDIA. We show the importance of this conflict type: fine-tuning Qwen3.5-9B and Gemma4-12B-it on examples of this conflict type significantly improves IH on a wide range of real-world tasks unseen in training; adding these examples to existing IH training sets for fine-tuning also gives better IH adherence. The IH gains come with trivial mean utility changes on no-conflict tasks across eight benchmarks. Our work advocates including this new conflict type in IH training, even if these examples do not reflect typical LLM use.
[AI-67] Zero-Shot Multi-Class Wafer Map Classification with Rule-Ordered Prompts for Large Multimodal Models
链接: https://arxiv.org/abs/2610.35080
作者: Tianjian Cao,Jun Tao
类目: Artificial Intelligence (cs.AI)
备注: 7 pages, 4 figures, 4 tables. Preprint version
Abstract:Wafer-map defect classification often requires labeled data and retraining when production conditions change. We present a training-free six-class framework that combines connected-component highlighting with morphology-, location-, and exclusion-aware prompts for a frozen large multimodal model. On Mixed-WM38 (6,866 maps), multi-region prompting achieves 92.9% macro and 93.8% overall accuracy. On WM811K (9,000 maps), it achieves 79.1% and 81.3%, respectively. Eight alternative rule orders achieve 90.1-91.9% overall accuracy, compared with 93.8% for the selected order.
[AI-68] oward Fast Human Reaction Modeling from High-Frequency Digital Interaction Data
链接: https://arxiv.org/abs/2610.35079
作者: Emanuel Dopater,Daniel Gavlas,Lukáš Jochymek,Kannan Paul,Miloslav Szczypka,David Vojáček,Josef Žáček,Jan Plucar,Radek Svoboda,Ivan Zelinka
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:We investigate whether limbic-like signatures of rapid, salience-driven behavior can be identified in high-resolution Counter-Strike 2 telemetry. This exploratory study introduces Ego-Knowledge, a leak-reduced reconstruction of each player’s plausible perceptual state, supported by a millisecond-precision visibility engine and a reaction-centered extraction pipeline. Using a neuroscience-motivated 60–120 ms window, we derive coarse fight-or-flight labels from visibility- and damage-triggered events, with back-damage cases enabling constrained reaction-time analysis and manual label checks. We train two telemetry-native models using only information available by the trigger: EMBER (gradient boosting) and ORBIT (a custom transformer). On the full validation set, EMBER achieves 73.5% accuracy and ORBIT 71.2%, versus a 51.5% majority-class baseline, and both generalize to held-out players. The strongest LLM baseline achieves 62.7% on 1,000 validation cases; on these same cases, EMBER (73.6%) and ORBIT (70.9%) perform better (McNemar p=1.3e-9 and 1.5e-6, respectively), with orders-of-magnitude lower inference time and cost, though the comparison is not compute-matched. Including a short post-trigger window changes forecasting into early response recognition, raising accuracy monotonically to 85.15%. We propose this curve as a candidate probe of the transition from rapid to deliberative control, but roughly uniform per-tick gains do not establish that interpretation. The neuroscience framing is hypothesis-generating, not mechanistic: physiological ground truth, suitable control distributions, and reverse-inference limitations remain open. We outline extensions to auditory reconstruction, larger datasets, and telemetry-native foundation models.
[AI-69] A Diagnostic Protocol for GNNs in Electricity Load Forecasting: Separating Parameter Sharing from Neighborhood Access
链接: https://arxiv.org/abs/2610.35074
作者: Eloi Campagne(CB),Yvenn Amara-Ouali(LMO),Yannig Goude(LMO),Itai Zehavi(CB),Argyris Kalogeratos(CB)
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A graph neural network (GNN) forecasting a collection of load series combines two mechanisms, sharing parameters across series and reading neighboring series, and a gain over a non-graph model may come from either; telling them apart decides whether a graph is worth maintaining. We propose a model diagnostic protocol that compares the GNN with three reference models: a per-node network, a network shared across nodes without neighbors, and the same GNN without edges. Degree-preserving rewiring tests whether neighbor identity matters. After a tuned benchmark on regional load in France and on low-voltage feeders in the UK, we apply the protocol to deliberately short node histories, as when non-stationary demand leaves only recent observations relevant: the feeders, within one week, and a public dataset of client consumption series. On the feeders, sharing parameters recovers most of the error gap between node-specific models and GNNs, about two thirds against a regularized one. Adding the edges to an otherwise identical backbone brings a smaller gain, half or more of which disappears under rewiring. On the client series, the neighbor mean degrades the per-node model on the shorter window and the edges’ gain is not consistently distinguishable from zero; on the regions of France, where each node’s own history is already informative, the ranking between per-node models and GNNs changes with tuning. A synthetic study points to noisy local measurements as one regime where neighbors help. In an exploratory online aggregation, graph and non-graph forecasters appear complementary, and the weight assigned to GNNs is not proportional to their contribution. In none of the three datasets do the edges account for most of a GNN’s advantage over node-specific models, so a forecasting gain should be credited to the graph only after comparison with per-node and shared references.
[AI-70] Joint Dynamic Lane Reversal and Variable Speed Limit Control: A Graphormer-Based Approach
链接: https://arxiv.org/abs/2610.35070
作者: Hamed Araab,Abbas Ahmadi,Mohammad Hossein Basiri
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 21 pages, 10 figures, 3 tables, 1 algorithm
Abstract:Connected and automated vehicles could allow reassigning lane directions and speed limits on every road frequently, but the benefit hinges on how traffic reroutes, and one-way roads can isolate junctions. We formulate network-wide dynamic lane allocation and directional speed limit control with one-way roads, lane clearing before each reversal, and a navigable graph kept strongly connected at all times. An action-resolution algorithm grounded in graph orientation theory admits a proposal closing a direction only if the network stays strongly connected and otherwise relaxes or reverts it, keeping all junctions mutually reachable within the model, even during clearing. All road controllers share one transformer whose attention, Global Endpoint Relation-Aware Spatio-Temporal Self-Attention (GERASTSA), spans all road-time tokens, biased by the endpoint relations between roads. It is pretrained on action-conditioned traffic forecasting and fine-tuned with per-road proximal policy optimization under a mixed reward that also credits neighboring roads and the network. On a 6 x 6 grid with commuter rush hours, the learned policy makes about two thirds of roads one-way in the peak direction and cuts daily delay by 90% against the initial configuration held fixed and by 85% against a tuned two-way load balancer. Restricted to two-way roads at test time, it gains only about as much as this balancer, whereas its learned speed limits add delay. Pretraining roughly halves the updates needed to surpass the fixed configuration. Of the ablations, only removing the relational bias or the spatial embedding clearly matters, multiplying delay about fivefold.
[AI-71] Beyond Single-Attribute Auditing: A Configurable Multi-Attribute Benchmark for Intersectional Fairness in LLM s EMNLP2026
链接: https://arxiv.org/abs/2610.35069
作者: Zheng Zhang,Yifan Wu,Yuxiang Zhou,Qi Liu,Qingyang Mao,Ning Li,Weibo Gao,Qiyuan Shi,Zhenya Huang,Baosheng Yu
类目: Artificial Intelligence (cs.AI)
备注: Accepted to EMNLP 2026 (Main Conference)
Abstract:Large language models (LLMs) are increasingly evaluated for fairness, yet most existing benchmarks audit one sensitive attribute at a time. This single-attribute paradigm can miss intersectional harms, where multiple attributes interact non-additively and produce failures that are invisible under marginal evaluation. We introduce ComFair, a configurable benchmark that converts existing fairness data into 479,066 controlled Dual- and Multi-attribute instances and pairs every augmented instance with its task-equivalent source. Across Recognition, Selection, Continuation, and Conversation, this pairing supports a unified audit of performance or bias-score shifts, how often added attributes introduce new errors or mask existing ones, and model-ranking stability. Experiments on ten LLMs show task- and subgroup-dependent sensitivity: discriminative tasks expose both transition directions even when aggregate changes are modest, generative tasks exhibit larger and directionally asymmetric shifts, and rankings vary with attribute composition. ComFair therefore provides a scalable controlled sensitivity audit—rather than a model of the natural joint distribution of identities—for testing the robustness of fairness-benchmark conclusions beyond single-attribute settings.
[AI-72] AutoPRL: Automated Path Representation Learning via a Learnable View Generator
链接: https://arxiv.org/abs/2610.35032
作者: Sean Bin Yang,Ying Sun,Yunyao CHeng,Hao Miao,Jilin Hu,Bin Yang,Torben Bach Pedersen,Christian S. Jensen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted by IEEE TKDE 2026
Abstract:Self-supervised learning has emerged as a powerful paradigm for Path Representation Learning (PRL) in road networks. PRL aims to learn informative representations of paths traversed by moving objects, such as vehicles, where each path consists of a sequence of connected road segments. These representations are essential for various downstream tasks, including travel-time estimation and path ranking. However, existing methods often suffer from ineffective view generation and inefficient training procedures due to their reliance on heuristic augmentations and computationally intensive contrastive learning. To address these issues, we propose AutoPRL, an effective and efficient self-supervised PRL framework that integrates three key components. The Automatic Data Augmentation (AutoDA) module employs variational autoencoders to generate semantically meaningful and diverse feature-level path views by modeling each topology-aware node embedding with a Gaussian latent distribution, thereby mitigating information loss and enhancing representation quality. Second, the Cross-View Decoupled Contrastive (CVDC) learning mechanism improves training efficiency by disentangling the optimization of positive and negative pairs across node- and path-level views, enabling more stable contrastive learning. Third, an Adaptive Co-Training (ACT) module dynamically adjusts uncertainty-based loss weights to balance multiple objectives, ensuring efficient and scalable training. Extensive experiments on three real-world datasets demonstrate that AutoPRL achieves strong and consistent performance in both effectiveness and efficiency across multiple downstream tasks.
[AI-73] PatentAgent Bench: Benchmarking LLM Agents with Professional Patent Task Suites
链接: https://arxiv.org/abs/2610.35031
作者: Yijie Hu,Wei Li,Junjie Zhou,Zhiqi Jiang,Yi Jin,Licong Xu,Xinkai Sun,Kai Duan,Lisha Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Patent practice lies at the intersection of technical, legal, and linguistic work, making it challenging yet crucial to technological innovation. Recently, rapid advances in large language models (LLMs) and agentic systems have demonstrated the potential to complete realistic business workflows and even patent works. This raises a question: can current AI systems perform patent work reliably? Existing benchmarks cover a limited range of patent tasks and simplify evidence acquisition by providing reference materials or candidate documents. In this paper, we introduce PatentAgentBench, a benchmark comprising 510 instances across eight task suites: novelty search, freedom-to-operate (FTO) search, design FTO, invalidity search, claim charting, patent drafting, office-action response, and patent translation. Curated from real patent records and professional materials, the benchmark combines task-specific metrics with expert-reviewed annotations and rubrics. Retrieval tasks such novelty search, FTO, etc., require agents to search and compare patent collections without a supplied candidate pool. The drafting and OAR tasks assess the quality of agents’ work given technical disclosure and office action notice. We evaluate seven LLMs across four agent harnesses on PatentAgentBench, finding that current systems still struggle to identify novelty, assess risk, and understand practitioner intent throughout patent workflows.
[AI-74] Multiscale Policy Optimization for Long-Horizon LLM Agents
链接: https://arxiv.org/abs/2610.35013
作者: Tao Ren,Hui Yang,Jia Deng,Tingyang Chen,Qiyao Wang,Shuo Tang,Ziyang Zeng,Yuhua Jiang,Guanghao Li,Huatong Song,Yining Li,Yifan Luo,Kunbao Ding,Feng Chang,Chuan Hao,Ran Tao,Bryan Dai,Jiafeng Liang,Jinyang Jiang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale Policy Optimization (MPO), combining Multiscale Advantage Estimation (MAE) and turn-level importance sampling to align credit assignment and policy updates at the turn scale. Extensive experiments across search, tool-use, and automation tasks with 4B and 30B-A3B models demonstrate consistent gains over strong RL baselines.
[AI-75] Critics Eye: Review-Grounded Graph Reasoning for Long-Form Motif Discovery EMNLP2026
链接: https://arxiv.org/abs/2610.35011
作者: Xinyu Sun,Zhangyue Yin,Xinyi Ju,Xin Wang
类目: Artificial Intelligence (cs.AI)
备注: 17 pages, 8 figures, 6 tables. EMNLP 2026
Abstract:Deep narrative understanding requires distinguishing what happens from how it is presented, moving beyond the chronological fabula to decode the sjuzet, the discursive organization through which themes are staged. We study this through the recurrence of literary motifs (blood in Macbeth, for instance), where symbolic significance emerges not from local cues, but through repetition-with-variation across the whole work. We operationalize this setting as Computational Motif Discovery, a transductive task that predicts missing line-motif links using the full narrative structure of a play. We propose Critic’s Eye, which models the work as a heterogeneous Narrative Topology Graph and performs discriminative inference over global structural evidence. Critic’s Eye achieves 84.8% Hit@5 on our benchmark, nearly quadrupling state-of-the-art proprietary foundation models (Gemini-3-pro-preview, ~22.6%). Our analysis reveals that despite massive scaling, sequence-based models hit a “semantic ceiling”, struggling to resolve dispersed dependencies. These findings suggest that explicit structural priors are a far more effective inductive bias than parameter scaling for decoding the complex architecture of long-form literature.
[AI-76] Why Are LLM s Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention NEURIPS2026
链接: https://arxiv.org/abs/2610.34994
作者: Jing Ren,Bowen Li,Ziqi Xu,Xuechao Yang,Feng Xia
类目: Artificial Intelligence (cs.AI)
备注: The paper has been accepted by NeurIPS 2026
Abstract:Modern large language models (LLMs) often exhibit overconfident errors, assigning high confidence to incorrect predictions and thereby undermining reliability in real-world use. Existing calibration methods typically rely on post-hoc adjustment or retraining, operating at the output level without addressing the underlying causes of miscalibration. We propose CoCHI, (Class-oriented Causal Head Intervention), a framework that improves calibration by identifying and intervening on internal components responsible for distinct confidence behaviours. By analysing predictions through correctness-confidence patterns, CoCHI enables targeted intervention on model internals, reducing overconfident errors while preserving correct predictions. Our approach provides a mechanistic perspective on calibration, linking reliability to identifiable structures within the model rather than treating it as an output-level issue. Experiments across multiple LLMs show consistent improvements in calibration metrics, including Expected Calibration Error, Negative Log-Likelihood, and Brier Score, while maintaining accuracy on five multiple-choice question answering benchmarks. These results suggest that effective calibration requires not only adjusting outputs, but also addressing the underlying causal mechanisms within the model.
[AI-77] he Target Agent Is Its Own Best Critic but a Poor Proposer: Self-Improved Red Teaming
链接: https://arxiv.org/abs/2610.34965
作者: Mintong Kang,Stephen Rawls,Melissa Kazemi Rad,Jian Ni,Bo Li
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Automatic red teaming increasingly relies on large language models (LLMs) to iteratively diagnose failed attacks and propose improved adversarial prompts or injections. Although prior work strengthens this loop through prompt engineering and search-harness design, the mechanisms underlying its two core roles, criticism and attack proposal, has not been explored. To further investigate the core roles and therefore improve red teaming effectiveness, we disentangle two design choices: whether criticism is generated by the target session itself (self-criticism) or by a separately instantiated critic (external criticism), and whether that criticism is translated into the next attack by the target session (self-proposal) or by a separate attacker (external proposal). Through our analysis, we find that self-criticism is more effective because it is grounded in the target’s live execution context, including observations, constraints, and available reasoning; removing reasoning reduces downstream attack success by 6%-17%. In contrast, self-proposal is substantially less effective because awareness of the adversarial objective induces explicit refusal and non-refusal sandbagging. These findings motivate a role-separated red-teaming approach: the target session diagnoses why an attack failed, while a separately instantiated attacker generates the next proposal. On DTAP-BENCH, spanning Telecom, Workflow, CRM, and Code domains, under both direct misuse and indirect threat models, as well as multiple agent frameworks, this configuration achieves 63.2% overall attack success rate (ASR), much higher than baselines. Our results show that effective automatic red teaming depends not only on model capability and search strategy, but also on how execution context and adversarial responsibility are distributed across model sessions.
[AI-78] C-SafeGen: Certified Safe LLM Generation with Claim-Based Streaming Guardrails
链接: https://arxiv.org/abs/2610.34961
作者: Mintong Kang,Zhaorun Chen,Bo Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Despite the remarkable capabilities of large language models (LLMs) across diverse applications, they remain vulnerable to generating content that violates safety regulations and policies. To mitigate these risks, LLMs undergo safety alignment; however, they can still be effectively jailbroken. Off-the-shelf guardrail models are commonly deployed to monitor generations, but these models primarily focus on detection rather than ensuring safe decoding of LLM outputs. Moreover, existing efforts lack rigorous safety guarantees, which are crucial for the universal deployment of LLMs and certifiable compliance with regulatory standards. In this paper, we propose a Claim-based Stream Decoding (CSD) algorithm coupled with a statistical risk guarantee framework using conformal analysis. Specifically, our CSD algorithm integrates a stream guardrail model to safeguard sequential claims generated by LLMs and incorporates a backtracking mechanism to revise claims flagged with high safety risks. We provide theoretical guarantees demonstrating that the CSD algorithm achieves the desired generation distribution subject to safety constraints. Furthermore, we introduce a generation risk certification framework and derive a high-probability upper bound on the safety risk of the proposed CSD algorithm. We prove that our method can asymptotically control safety risk to any desired level. Empirical evaluations demonstrate the effectiveness and efficiency of the CSD algorithm compared to state-of-the-art safety decoding approaches. Additionally, we validate the soundness and tightness of the derived safety risk upper bound using realistic data.
[AI-79] Decision Model Factory: Building Specialized Sovereign Decision Models
链接: https://arxiv.org/abs/2610.34956
作者: Yuma Ichikawa
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 9 pages, 3 figures
Abstract:Decision needs vary with each application, language, and deployment environment. The Decision Model Factory turns those needs into specialized, sovereign models on demand. A brief recipe guides the training, calibration, verification, and evaluation of a selected backbone, delivering a deployable decision model under the operator’s control. The same workflow creates a new specialist whenever the requirements change. Across Japanese decision tasks, compact deployment, and long-context passage ranking, five factory-built models improve upon their backbones; three surpass the strongest results in a fixed public Japanese benchmark comparison. A dedicated runner computes only permitted-answer probabilities and reuses shared context, more than doubling serving speed on matched multi-question workloads with essentially unchanged decision scores. Together, the factory and runner make fast, task-specific decision models practical to build and deploy whenever needed.
[AI-80] Relational Composition as Parallel Transport: What Per-Entity Frames Force in Attention
链接: https://arxiv.org/abs/2610.34945
作者: Aakash Sarkar,Ananya Mukherjee,Motoshige Sato
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Formal Languages and Automata Theory (cs.FL)
备注:
Abstract:A dot product means something only when both vectors share a coordinate system. Content words have one; a fresh entity does not: rotating its coordinates alone changes nothing. We take this per-entity frame freedom as a symmetry and ask what it forces in attention. For orthogonal transports, invariant scores depend only on norms and the transported inner product \langle x_i,T_ijx_j\rangle ; equivariant messages lie in the span of x_i and T_ijx_j . Endpoint channel count bounds transport rank, so with one vector per entity relations must be supplied; standard attention is the flat sector of a shared frame. Representation theory attaches an integer to each mechanism; the readout decides which. With orthogonal transitions, a convex-cell readout composes a finite group exactly only at or above its smallest faithful real representation dimension. At this minimum, exactness forces a faithful representation. An even readout can decode a cover at smaller width, and the gap is unbounded. Below its threshold, the part of the group a mechanism cannot see sets an accuracy ceiling computable before training. 17/17 fitting seeds at minimal width passed tests through length 1000, versus 7/18 at twice the width; 12 models carry correctness certificates to explicit lengths, from the weights alone. On C_3\times A_4 (faithful width 5), descent at width 4 learned the 72-state double cover under sign-blind attention (6/48 seeds; monotone 0/48); on a frozen cover the cross-entropy-optimal affine head is uniform, a quadratic head exact. On parsed CLUTRR, transported composition beats a matched transformer by 41 points at 4-10 hops, but forcing isometries costs 65 points; on StepGame’s commutative algebra, isometries help. Static embeddings gave each word one vector; contextual models gave each occurrence its own; per-entity frames make each relation an operator, the unit in which relations compose.
[AI-81] Reasoning Patterns for Efficient AI Coding: An Empirical Study
链接: https://arxiv.org/abs/2610.34943
作者: Zhe Yin,Yuling Shi,Xiaodong Gu,Beijun Shen
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Reasoning Models (LRMs) generate long reasoning traces that expose intermediate inference behavior, offering opportunities to study and control reasoning in AI coding. However, the underlying patterns in these traces and their potential for analyzing and guiding coding models remain poorly understood, limiting systematic characterization and inference-time control of LRM reasoning. We conduct an empirical study of reasoning patterns across four representative coding tasks. We transform reasoning traces into structured action sequences and mine recurring fragments using within-task support, cross-task contrasts, and valid–invalid support contrasts. We characterize four general reasoning patterns by coding agents: design-to-implementation for code generation, state tracing for execution reasoning, evidence–fix–recheck for debugging, and semantic mapping for translation. We further identify positive and negative patterns associated with valid and invalid reasoning traces, respectively. Positive patterns reflect verification-driven convergence, whereas negative patterns reflect repeated design revision, reflection, and knowledge recall without closure. To demonstrate their practical utility, we evaluate three pattern-based applications: task identification, pattern-guided prompting, and pattern-aware early stopping. Features capturing general reasoning patterns enable accurate task identification from reasoning traces without accessing task descriptions or final answers. Prompting guided by positive and negative patterns improves the overall accuracy of downstream tasks, while negative pattern-aware early stopping substantially reduces API-reported token usage with only a modest decrease in pass@1. All three applications also generalize to previously unseen models, showing that reasoning patterns can support both the analysis and control of LRM behavior in AI coding.
[AI-82] Stop or Clarify? Experience-Conditioned Active Fact Verification
链接: https://arxiv.org/abs/2610.34941
作者: Yuhang Dong,Jinghong Li,Xinyue Wang,Yuanning Cui
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Fact-verification benchmarks score the final label, while an interactive verifier must first decide whether the available evidence is sufficient. We focus on that intermediate Stop-or-Clarify Decision. Uncertainty alone is insufficient, because the benefit of another evidence-acquisition step is unknown before the step is taken. Sequential Experience-Adaptive Clarification (SEAC) wraps a fixed verifier with a lightweight evidence-acquisition policy. For each claim, SEAC retrieves evidence, runs the verifier, forms a label-free decision state, and combines stop risk with an observational clarification-utility estimate from earlier executed trajectories. Clarification is requested only when this score exceeds the action cost and decision threshold. On SciFact, LIAR, and FEVER, SEAC improves mean accuracy and macro-F1 over the compared fact-verification and active-retrieval baselines. Equal-budget and formula-level diagnostics support selective clarification as a contributor to the gain, while the small SciFact split leaves its paired improvement inconclusive.
[AI-83] How Do Coding-Agent Harnesses Self-Evolve? An Empirical Study of Retrospective Adaptation
链接: https://arxiv.org/abs/2610.34938
作者: Christoph Treude,Sebastian Baltes
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents can adapt their harnesses, the software, configuration, and instructions used to run them, based on experience from earlier tasks, but whether these adaptations improve their behavior on later tasks remains unclear. We study \retro, a skill that asks an agent to review a completed session and propose harness changes, and ask what changes it proposes and whether implemented changes reduce repeated work on later tasks. We ran Claude Code on 30 SWE-bench Verified issues, qualitatively coded the 125 proposals from \retro, implemented 24, and evaluated 12 harness changes on later issues from the same repositories in fresh sessions with and without each change. Most proposals referred to an event in the session, but 18 of 43 lint-related proposals recommended a practice regardless of what had happened. Among the seven evaluations in which sessions without the change encountered the targeted problem, two changes eliminated the targeted action, one reduced its occurrence, four showed no reduction, and some changes added work on unrelated issues. Harness self-evolution requires more than generating and implementing changes: it needs ways to establish whether a change helps on later tasks and to revise or remove changes that do not.
[AI-84] PhysEvolve: Verifier-grounded Self-evolution for Physical Reasoning with Executable Worlds
链接: https://arxiv.org/abs/2610.34921
作者: Jinyuan Hu,Yilei Jiang,Chen Tang,Minghong Cai,Xinran Liu,Xiangyu Yue
类目: Artificial Intelligence (cs.AI)
备注: Preprint
Abstract:Physical reasoning underpins how embodied systems understand and act in the physical world, yet vision-language models (VLMs) remain unreliable on questions about dynamics and physical causality. Progress is critically constrained by the scarcity of grounded counterfactual training data, as observational videos capture only the factual trajectory that occurred. The self-evolving paradigm offers a route beyond fixed datasets by enabling models to construct their own training problems. Extending this paradigm to physics, however, requires an external source of truth that can both validate model-proposed environments and verify alternative physical outcomes under intervention. We introduce \physevolve, a verifier-grounded self-evolving framework that trains VLMs for causal physical reasoning from model-proposed executable worlds and simulation-authored tasks. At each round, a shared VLM proposes executable world programs and evidence schemas; a deterministic physics engine freezes each valid world, simulates matched baseline and body-elided counterfactual branches, and synthesizes discrete, answer-bound visual tasks. Exact verifier feedback from task solving updates both the shared policy via GRPO and an adaptive curriculum over physical intents. Starting from Qwen3-VL-8B-Instruct, \physevolve achieves targeted gains across PhysBench, CausalPhys, and ContPhy without human intervention data, improving PhysBench Throwing by 8.8 points and ContPhy Counterfactual Removal and Goal-Pit by 7.4 and 24.4 points. These gains concentrate on interaction-sensitive and altered-world questions, demonstrating zero-shot transfer of causal principles to unseen complex dynamics (cloth, fluid, soft bodies) and real-world video.
[AI-85] Waypoint: Analytically Modeling GEMM for CPUs
链接: https://arxiv.org/abs/2610.34914
作者: Srinidhi N,Sunanda Somwase,Hemant Kumar Chandrakant Chandavar,Eashan Dash,Varun Agrawal,Prakash Raghavendra
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI)
备注:
Abstract:High-performance General Matrix Multiplication (GEMM) implementations rely on breaking down GEMM into carefully selected tiles to achieve peak performance. We present Waypoint, a new analytical model to automatically identify the optimal tile size for a given hardware and input matrix shape. Waypoint estimates the achievable performance for a given tile choice by capturing all the architectural effects including interactions with memory hierarchy, instruction-level parallelism, and multi-threading. Using the estimates as a proxy, we search through the space of possible tile sizes to identify the one that maximizes performance. Additionally, Waypoint helps hardware designers estimate realistic performance on experimental hardware optimizations. We build a new GEMM library OPAM for CPUs that uses Waypoint to automatically break down GEMM and achieve performance that matches state-of-the-art GEMM libraries and helps accelerate LLM inference on CPUs achieving 1.36 - 1.98 \times speedups.
[AI-86] AIPC: Evidence-Guided Agent Workflows for PyTorch Project Deployment
链接: https://arxiv.org/abs/2610.34890
作者: Zhanwei Wu(1),Jianhao Su(1),Brian Zhang(1),ShengTing Huang(1),Weidong Feng(1),Gang Liu(1),Zhiguo Li(1),Wei Huang(1),Li Tan(1),Wanjun Li(1),Zejiong Teng(1),Changgeng Zhong(1),Guanping Wu(1),Han-Hsing Tu(2) ((1) Qualcomm Technologies, Inc., (2) Independent Researcher)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Deploying and optimizing a PyTorch inference project on a dedicated accelerator, especially an on-device NPU, requires more than using a framework device API as with CUDA. The project must be converted into hardware-runtime artifacts and adapted to the target toolchain and application interface. Differences in model formats, quantization flows, operator support, and toolchain interfaces create cross-layer problems spanning PyTorch/ONNX models, toolchains, runtimes, and device execution. LLM Agents can invoke tools to address such problems, but may drift because of insufficient domain knowledge, large configuration and repair spaces, or poor path selection. We propose AIPC (AI Porting Conversion), an Agent method for deploying PyTorch inference projects. AIPC combines structured Skills with toolchain, runtime, and profiling evidence to drive cross-layer diagnosis and repair, organizing model adaptation, artifact building, on-target execution, accuracy repair, and performance optimization into one deployment flow. We evaluate AIPC on the on-device NPU path of Qualcomm QAIRT across vision, speech, LLM, and diffusion models. In two optimization cases, 6/7 Skill-arm Whisper-Small runs meet predefined accuracy and latency criteria versus 0/6 control runs, while YOLO-Worldv2 achieves a 2.791x end-to-end speedup versus 1.599x for the control. Under a pre-fixed deployment contract, Skill, tool permissions, and check conditions, some runs achieve zero in-run human intervention (0-HITL). We also examine Qwen3.8-27B as a locally deployable model for driving an Agent workflow. Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2610.34890 [cs.SE] (or arXiv:2610.34890v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.34890 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-87] GADAR: Generalizing Across Domains via Action Ranking for Classical Planning
链接: https://arxiv.org/abs/2610.34868
作者: Rajesh Mangannavar,Pranay Dugar,Prasad Tadepalli
类目: Artificial Intelligence (cs.AI)
备注: 15 pages, 2 figures, 4 out of 15 pages appendix
Abstract:Learned policies for classical planning act without search, but each is tied to the domain it was trained on: network modules per action schema, parameters per predicate, or vocabulary-sized encodings mean a trained policy cannot even be executed on a new domain’s predicates and action schemas. We present GADAR (Generalizing Across Domains via Action Ranking), a policy that takes the domain description as input, \pi(s, G, D), and constructs grounded actions in domains never seen in training, without search. Symbols are described by the roles they play in the domain rather than by their names, in three layers that each extend the previous one: (i) a lifted domain layer embeds D in every state graph through structural signatures; (ii) a binding layer ties each ground action’s applicability to the lifted preconditions it instantiates; and (iii) a chain layer states which schema produces what another needs and how far each schema stands from the goal. The same network therefore runs on every domain, seen or unseen. We evaluate zero-shot generalization leave-one-domain-out over eight benchmark domains. A single GADAR policy trained on seven domains solves 58.3% of their test instances, one point below the 59.3% of dedicated single-domain GADAR models, so one policy serves many domains at no cost in coverage; a multi-task control that keeps symbol identity in its weights, while strong on some of the domains it trains on, transfers to none of the held-out ones. Zero-shot, GADAR executes on every held-out domain and exceeds random choice under the same executor on the large test instances of all eight: it solves 22.6% of them on average, against 3.3% for random choice and 0.6% for the same representation without the chain layer, and reaches 68% on held-out Visitall. Zero-shot coverage nonetheless remains well below in-domain coverage on most domains, and we analyze where it is weakest.
[AI-88] UniGFM-RAG : Unify Graph Foundation Models and Large Language Models for End-to-End Retrieval and Reasoning
链接: https://arxiv.org/abs/2610.34866
作者: Zicheng Zhao,Linhao Luo,Zongji Yu,Junnan Dong,Yanchao Tan,Shirui Pan,Chen Gong
类目: Artificial Intelligence (cs.AI)
备注: 16 pages, 8 figures
Abstract:Large Language Models (LLMs) have shown remarkable capabilities across tasks, but they still struggle with complex reasoning that requires integrating scattered knowledge. Graph Retrieval-Augmented Generation (GraphRAG) addresses this challenge by organizing fragment knowledge as graphs, while recent Graph Foundation Models (GFMs) further improve GraphRAG with graph-native retrieval and reasoning capabilities. However, existing GraphRAG methods usually adopt decoupled retrieve-then-generate pipelines, which create objective misalignment between retrieval and reasoning, and training GFM retrievers typically depends on expensive human annotations. To address these limitations, we propose \ourmethod, a unified end-to-end GraphRAG framework that tightly couples a GFM retriever with an LLM generator. Given a query, the GFM retriever first reasons over the graph to score and retrieve relevant document nodes. These node embeddings are then passed through a fully differentiable graph bridge, which projects them into soft-weighted graph prefixes according to retrieval scores and combines them with raw document texts to form graph-augmented prompts. In this way, structural information is injected directly into LLM generation, while the final answer loss can be back-propagated through the bridge to jointly optimize retrieval and reasoning. To stabilize training and reduce reliance on costly annotations, we further introduce a semantic-guided distillation module that provides auxiliary supervision using pseudo-labels derived from query-document semantic similarities. Extensive experiments on three multi-hop QA datasets and three out-of-domain GraphRAG benchmarks show that \ourmethod consistently outperforms strong baseline methods and generalizes well, which demonstrates the effectiveness of end-to-end GraphRAG with tightly integrated GFM retrieval and LLM reasoning.
[AI-89] Beyond Behaviour: A Stimulus-Response Safety Evaluation of Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.34837
作者: Vy Vo,Dinh Phung
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Humans exhibit biological sensitivity to safety-critical stimuli that can trigger defensive action. Across three models and two video benchmarks, we investigate whether similar responses emerge in vision-language-action models by examining their internal representations and action tendencies in response to visible hazards and harmful consequences. Our findings suggest that task execution may remain the default response even when danger is internally registered.
[AI-90] Streaming Recurrent Memory
链接: https://arxiv.org/abs/2610.34833
作者: Mohamed Elsayed,Minh Pham,A. Rupam Mahmood
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Natural intelligence learns from a continuous stream of sensorimotor experience: data can arrive at high bandwidth, and behavior adapts to feedback as soon as it arrives, even when it is delayed. In contrast, most modern deep learning systems rely on stored datasets, replay buffers, and retained contexts through which credit is later assigned. At high data rates, storing and revisiting past experience can be impractical under tight memory and computational constraints. This paper studies recurrent memory under a streaming learning constraint: the learner learns from experience as soon as it arrives without storing raw experience. We ask whether a streaming learner can maintain a state that remains useful for future prediction and control from delayed feedback. We introduce Streaming Recurrent Memory (SRM), a recurrent memory designed for the streaming setting which preserves information that may matter later while remaining trainable with one-step backpropagation through time (BPTT). SRM provides a streaming-compatible recurrence that is designed to recall past information by content or time, track elapsed steps, and compose transformations whose outcome depends on the order in which inputs are observed. Across controlled memory tasks and streaming applications, SRM learns effectively with one-step BPTT, outperforming recurrent baselines in associative recall and online language modeling while achieving near-perfect success in associative T-Maze with delayed rewards. These results suggest that fixing how inputs are stored in the recurrent state, while learning how that state is read out, enables effective streaming learning without long-range backpropagation.
[AI-91] Logic Hijacking of RAG : Single-Document Poisoning via Local Optimization
链接: https://arxiv.org/abs/2610.34829
作者: Hangyi Zhao,Bonan Zhang,Fusen Guo,Jiaheng Wei,Chao Chen,Tianqing Zhu
类目: Artificial Intelligence (cs.AI)
备注: 22 pages, 3 figures, 14 tables
Abstract:Single-document corpus poisoning under black-box access poses a severe threat to Retrieval-Augmented Generation (RAG). However, existing attacks predominantly rely on explicit factual refutations or imperative instruction overrides. Modern aligned LLMs often arbitrate such conflicts conservatively, rejecting overt contradictions in favor of parametric memory or clean retrieved context while triggering input guardrails. We propose MONOLOGUE, a single-document poisoning framework based on Logic Hijacking without explicit refutation. Rather than disputing the ground truth, our method concedes it as an authoritative historical baseline before introducing the update, weaving a grounded record-correction narrative (e.g., dated institutional reattributions) that presents the counterfactual target as the most recent update. To prevent suspicious live writes, payloads are refined entirely offline via local multi-dimensional gated optimization, coupling automated text quality audits with surrogate behavioral verification. Evaluations across three benchmarks, three retrieval backends, and four open-weight generator regimes show that our method achieves the strongest overall attack efficacy among the evaluated single-document poisoning attacks.
[AI-92] rustJev: Separating Policy Authority from Untrusted Evidence in LLM -Based Decisions
链接: https://arxiv.org/abs/2610.34826
作者: Zhiwei Xu,Xuyang Liu,Xiaolin Peng,Weixian Deng,Jiabao Gao,Zirui Li,Tian Qiu,Boya Zhou,Wei Jiao,Hai Wan,Xibin Zhao
类目: Artificial Intelligence (cs.AI)
备注: 12 pages
Abstract:Decision models enable software to select from predefined actions using natural language inputs. However, the external evidence that supplies facts for these decisions may also contain instructions that attempt to override the application’s policy. We propose TrustJev, a framework combining protected policy scoring with bounded evidence corrections. A single Transformer backbone computes policy scores and a correction budget from trusted policy and candidates, then bounds evidence corrections by that budget. Paired supervision encourages stable predictions under instruction insertions that preserve task facts and changed preferences when edits change the correct answer. Authority supervision accommodates evidence when the policy alone cannot determine the decision. We evaluate a 0.6B implementation against 14 baselines on three datasets covering authentication, instruction hierarchy, and executable authorization rules. TrustJev achieves the highest mean Balance, the harmonic mean of accuracy on non-attack and attack inputs, exceeding the strongest baseline on each by 0.57-1.05 percentage points. In the authorization benchmark, it correctly answers all 32 attacks per source in a fixed pool for 95.77% of sources, compared with 80.98% for the strongest comparator. TrustJev maintains comparable inference memory and median latency, with additional overhead primarily during training.
[AI-93] SPARTA: A Sparse Ternary LLM Accelerator with Adaptive Attention and Unified Datapath for Long-Context Edge Inference
链接: https://arxiv.org/abs/2610.34824
作者: Yanzhen Zhou,Zhengyu Han,Qunyou Liu,Zheyu Yan,Weihong Xu,Cheng Zhuo
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注:
Abstract:LLMs on edge devices are constrained by limited memory and bandwidth. Although 1.58-bit ternary weights greatly reduce model storage, long-context inference is increasingly bottlenecked by KV-cache traffic and attention. Existing accelerators also use separate datapaths for ternary linear layers and attention, incurring hardware and execution overhead. We present SPARTA, a sparse ternary LLM accelerator for efficient long-context edge inference. SPARTA integrates: (1) a phase-aware sparse dataflow exploiting heterogeneous sparsity across compute-bound prefill and memory-bound decode; (2) a unified compute fabric mapping ternary SpMM/SpMV and bit-serial attention onto shared processing elements to exploit zero-weight and zero-bit sparsity; and (3) an adaptive two-stage attention engine combining page-level Top-k selection to reduce K-cache accesses with token-level Top-p pruning to reduce V-cache traffic. SPARTA reduces KV traffic by 4.1-4.8x over full attention and 1.3-1.5x over a state-of-the-art sparse-attention scheme, with task-dependent quality loss. Compared with NVIDIA Jetson AGX Orin, SPARTA achieves 3.9x higher generation throughput and 8.3x higher end-to-end energy efficiency on average; versus AMD Ryzen 7 9700X, the gains are 5.0x and 81.7x. It also delivers 1.6-12.0x higher throughput, 2.0-4.2x higher energy efficiency, and 1.3-51.4x higher area efficiency than FIGLUT, TeLLMe, and MCBP.
[AI-94] Silent and Costly: Characterizing and Detecting Prompt-Caching Mis-Integration in LLM Agents ICSE2027
链接: https://arxiv.org/abs/2610.34817
作者: Ying Zhang,Ruoqi Guo,Yi Liu,Gelei Deng,Yuekang Li,Yepang Liu,Leo Zhang
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at ICSE 2027
Abstract:Large language model (LLM) services process massive volumes of user requests, resulting in substantial inference costs. To reduce both cost and latency, providers have introduced prompt caching. However, correctly integrating prompt caching is non-trivial due to heterogeneous provider-specific APIs and continuously evolving caching semantics. To understand prompt-caching mis-integration in downstream applications, we conduct the first empirical study of 120 prompt-caching-related pull requests from 10 popular open-source agent systems. Our findings show that prompt-caching mis-integrations arise in both request-side misconfiguration and read-side usage miscount. More importantly, most mis-integrations fail silently without causing observable runtime failures. Motivated by these findings, we design CACHELINT, which models a prompt-caching mis-integration as an invocation site together with the preconditions required for correct cache integration. CACHELINT locates caching invocation sites with CodeQL and discharges each precondition either against a dated provider capability oracle or with an LLM guard check. CACHELINT achieves 0.71 precision and 0.78 recall in identifying the 120 labeled mis-integrations across the 10 studied agent systems. Our evaluation further demonstrates the effectiveness of CACHELINT in real-world settings, identifying 32 mis-integrations across 571 agent projects with 0.86 precision. These findings have received 21 developer confirmations and led to 18 upstream fixes, including one in Anthropic’s official SDK. Comments: Accepted at ICSE 2027 Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.34817 [cs.SE] (or arXiv:2610.34817v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.34817 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-95] Precedence-Aware Resource Allocation: Extending the AUTOC Framework to Multi-Level Treatments
链接: https://arxiv.org/abs/2610.34812
作者: Olin E. Gilster,Shengpu Tang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Many real-world decision problems require allocating limited resources across individuals to maximize total benefit often through interventions that vary in intensity rather than binary treat-or-not decisions. In these settings, higher levels treatments require first applying lower levels of treatment, inducing a precedence constraint that complicates allocation under a fixed budget. While the causal inference literature has largely emphasized accurate estimation of conditional average treatment effects (CATE), recent work has shown that estimation accuracy alone does not guarantee effective resource allocation, motivating ranking-based evaluation frameworks. The Area Under the Targeting Operating Characteristic (AUTOC) provides an effective way to evaluate prioritization rules across budget levels, but existing formulations are limited to binary treatments. We introduce Multi-Level AUTOC (ML-AUTOC), which is a generalization for AUTOC designed for multi-level treatment settings with precedence. We also propose precedence-aware policies for mapping true marginal treatment effects to feasible treatment assignments under fixed budget constraints. Our results show that in the setting of multi-level treatments, maximizing realized treatment effects may require re-allocation of treatments across individuals.
[AI-96] Centralized Critics for N-Agent Ad Hoc Teamwork
链接: https://arxiv.org/abs/2610.34799
作者: Andrew McAlinden,Caroline Wang,Steven Shi,Arrasy Rahman,Elad Liebman,Peter Stone
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Centralized training with decentralized execution (CTDE) has become the dominant paradigm in multi-agent reinforcement learning (MARL), motivating extensive study of what privileged information should be made available to the critic during training. While prior work has examined the role of global state in fully cooperative settings, the design of centralized critics for N-agent ad hoc teamwork - in which dynamically varying set of learning agents must coordinate with an arbitrary set of uncontrolled teammates drawn from an unknown distribution - remains poorly understood. We study this question by comparing two natural classes of advantage estimators: controlled history estimators, which condition only on the observation and action histories of the learner’s own agents, and joint history estimators, which additionally condition on the histories of the uncontrolled teammates. We prove that both estimators are unbiased and characterize the conditions under which the joint history estimate attains strictly lower variance than its controlled counterpart. Empirically, joint history critics match or outperform controlled history critics across a range of NAHT tasks, yielding improved sample efficiency and final returns. Together, these results provide both a theoretical justification and an empirical recommendation for incorporating uncontrolled agent histories into centralized critics for N-agent ad hoc teamwork.
[AI-97] Uncovering the Turing Tests Latent Signal: Indistinguishability as a Reward
链接: https://arxiv.org/abs/2610.34781
作者: Emanuele Rimoldi,Riccardo Neumarker,Tomaso Poggio,Daniel Mitropolsky
类目: Artificial Intelligence (cs.AI)
备注: 26 pages, 9 figures
Abstract:Can a model learn by trying to fool another model? We show that it can. Inspired by the Turing Test and its recent generalization to arbitrary distinguishers, we introduce a domain-based teacher-Turing-Test, in which a stronger teacher model serves as both the target of imitation and the distinguisher, and a weaker student model is trained to make the teacher attribute responses to itself. We use this self-attribution (or fooling) probability as a reinforcement learning reward, without ever providing the student ground-truth answers, nor requiring the teacher to generate responses. We test our approach independently in four domains with verifiable answers: CommonsenseQA (reasoning), GSM8K (mathematics), KodCode (programming), and Spider (SQL), training a Qwen2.5-1.5B student/imitator model using DeepSeek V4 Flash as teacher/distinguisher. Beyond improving at passing the Turing Test, the student improves on independent, held-out evaluations in all domains and runs. To probe why being mistaken for the teacher can improve capability, our analyses reveal that self-attribution is associated with correctness – substantially more than formatting changes – and that it rewards partial correctness. Our results demonstrate a new, simple teacher-student training paradigm with a private and/or high-cost teacher. They also suggest that indistinguishability need not merely measure behavior – it can also be a mechanism for learning.
[AI-98] Arya Forge: Auditing Validation and Release Approval for Financial Instruction Data
链接: https://arxiv.org/abs/2610.34778
作者: Kushagra Bhatnagar,Deekshith Marla
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Software Engineering (cs.SE)
备注: 18 pages, 3 figures, 8 tables
Abstract:Financial instruction-data pipelines must distinguish content validity from permission to release a particular dataset. Arya Forge combines typed records, current-content checks, and approval coverage bound to a dataset manifest. We evaluate its release boundaries in a financial-data systems case study. An integrity audit covers 100,000 synthetic instruction records; a separate arithmetic implementation agrees with all 15,000 numeric records across twelve formula families. In 240 scenarios with 2,400 treatment rows, adding financial and release-record checks to a short-instruction filtering recipe rejects all injected numeric faults and policy holds while retaining 220 of 240 clean controls. The twenty lost controls comprise every tested cash-conversion-cycle example. Nineteen persisted-release cases show that eleven selected negative cases are blocked, four equivalent representations are rejected, and changes to external evidence after binding are outside ordinary Gold promotion’s checks. Local processing measurements and controlled checkpoint resumptions characterize cost and recovery through 100,000 input rows. The results expose a tradeoff between strict producer-contract enforcement and useful-representation retention, and identify the protected inputs required for manifest-bound approval. These findings apply to the tested implementation and workloads; they do not establish general financial correctness, authenticated approval, expert-validated quality, or downstream model improvement.
[AI-99] Scope-Gap Injection: Exploiting Evaluator Channel Blind Spots in the OpenClaw Agent Security Competition
链接: https://arxiv.org/abs/2610.34777
作者: Weihang Niu,Mengxing Ren,Che Qu
类目: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注:
Abstract:This paper introduces Scope-Gap Injection (SGI), an indirect prompt injection method for multi-stage LLM-agent safety pipelines. In the IJCAI 2026 OpenClaw Agent Security Attack Competition, each attack package is judged by a static skill scan (S1), an LLM-judge safety check over the task and skill definition (S2), and a live execution audit (S3). Our observations suggest a scope gap: S2 appears to inspect the task instruction and skill definition without inspecting attachment contents, while the agent reads attachments during S3. SGI keeps the visible task and skill benign, while target identifiers, bypass parameters, and authorization procedures are moved into an attachment written as a business worklist. The S2 judge therefore sees a harmless task, while the agent follows the hidden worklist and triggers the permission bypass. In OpenClaw, SGI bypassed all three stages for 5 of 19 targets. The highest-scoring single submission achieved F = 0.385 . Across the submission campaign, attachment placement was associated with higher aggregate S2 acceptance than in-skill placement: 78% versus 23%. We close with defenses that align evaluator scope with the channels available to the agent.
[AI-100] he Right Fact the Wrong Decision: SciValidityBench for Scientific Agent Memory
链接: https://arxiv.org/abs/2610.34767
作者: Yuexi Pan,Yongchao Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Most agent-memory benchmarks frame long-term memory as retrieval: can an agent recover the relevant record from a long and evolving history? Scientific agents face a stricter problem: a correctly remembered fact may still be invalid to use. The same reported test result can support confirmation when the outcome and stopping rule were fixed in advance, yet only exploration when the outcome was selected from twenty inspected results. We call this scientific use validity: whether remembered evidence is eligible to influence a decision under the state governing its interpretation. We introduce SciValidityBench, an executable benchmark factory with 64 TaskSpecs spanning five validity modules and six scientific domains. TaskSpecs couple generated experimental histories with scientific-state definitions, programmatic decision oracles, controlled interventions, and automatic scoring. The SciStateForge compiler-reader framework turns distributed histories into explicit decision-governing state. Counterfactual pairs and targeted corruption-repair controls then test whether models use those dependencies and localize failures. Across four model families, compiled scientific state improves valid decision-making by 9.2-18.8 percentage points over a generic Event Graph. Retrieval alone does not reproduce this cross-family gain, while content-identical flat and typed views show that the improvement comes from exposing governing state rather than a particular graph serialization. SciValidityBench thus shifts memory evaluation from finding the right fact to deciding whether that fact should control the decision.
[AI-101] he Physical Cost of Context: Quality-Constrained Resource Accounting for Coding Agents NEURIPS2026 AAAI
链接: https://arxiv.org/abs/2610.34763
作者: Zhongming Yu,Yitong Ma,Hengjia Yu,Boqin Yuan,Xiangye Lin,Hejia Zhang,Jishen Zhao
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at the NeurIPS 2026 Workshop on Resource-Aware Agentic AI (RAAAI). 15 pages, 2 figures, 6 tables
Abstract:Coding agents resend a growing context on every model call, and context policies are usually judged by the prompt tokens they save. Executed work also depends on call count and prefix reuse: a caching runtime prefills only the tokens after the longest cached prefix. A shorter prompt can therefore cost more if it rewrites cached history. Our quality-gated protocol splits executed prefill into logical prompt minus reused prefix and measures KV residency, time, and energy. On 450 localization cases derived from SWE-rebench V2, preloading ten retrieved blocks (Eager) and adding anchored history compaction (Compact) both pass a five-point non-inferiority gate. In a separate 900-cell study, Eager cuts model calls by 35.7% and the GPU-energy proxy by 33.2% against on-demand search (Grep); Compact saves nothing further. Enabling prefix caching leaves Eager’s prompt volume similar (+8.3%; its 95% CI spans zero) but cuts its uncached prefill by 67.7%. Deleting history can turn a token saving into extra work. With mini-SWE-agent on 100 SWE-bench Verified instances, keeping only the last five observations (Mask-5) cuts Eager’s prompt volume by 33.3% but loses 135.9k reusable tokens per task. Uncached prefill rises by 608.5%, the energy proxy by 125.6%, and the policy misses the quality gate. Policies thus save physical work by shortening the trajectory and can multiply it by editing reusable history.
[AI-102] Query-Driven Learning for Lifted Reasoning
链接: https://arxiv.org/abs/2610.34747
作者: Luise Ge,Brendan Juba
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Learning and reasoning are both essential to intelligent agents, but have traditionally been treated as two separate stages: first learning a knowledge base from data (or taking one as given), and then verifying queries as if the knowledge base were complete. This separation has been shown to be inefficient when the downstream goal is to verify only a small set of queries. Moreover, it neglects the possibility that data and a knowledge base contain partial yet complementary information, requiring an agent to combine both sources to justify or refute a query. Motivated by the need to integrate learning and reasoning more tightly, we study the problem of query-driven learning for reasoning. More specifically, we extend prior work to the first-order setting, which is critical for practical applications as the number of ground atoms typically grows rapidly. We begin by asking when validating or refuting a universal query given partial data and knowledge base is even possible, and identify a necessary and sufficient condition called bounded concealment. We then propose a quasi-polynomial-time algorithm that solves the task at the lifted level. Finally, we establish soundness and completeness guarantees relative to space-bounded lifted proofs, along with quasipolynomial sample and runtime bounds, demonstrating that lifted query-driven reasoning can be efficient even when unrestricted knowledge-base learning is not.
[AI-103] SKALOS: Learning to Distinguish Similar Skills with Capability Graphs
链接: https://arxiv.org/abs/2610.34728
作者: Wenhan Xiao,Ziwei Zhang,Xingcheng Fu,Chuanyue Yu,Haonan Yuan,Qingyun Sun,Runhua Xu,Jianxin Li
类目: Artificial Intelligence (cs.AI)
备注: 12 pages, 7 figures, 19 tables
Abstract:Skills provide LLM agents with reusable procedural knowledge, making appropriate skill selection essential for reliable agent behavior. As skill libraries expand, distinguishing semantically similar candidates becomes increasingly difficult. Existing methods encode or rewrite skill descriptions for relevance-based ranking, yet heterogeneous descriptions can obscure capability differences. To address this challenge, we propose SKALOS, a graph-native reranker that compares skills via explicit capability evidence. Its multi-hierarchy graph aligns equivalent expressions across skills while preserving skill-specific capability evidence. Since individually relevant capability elements may not jointly support a query, semiring-based capability witness learning scores each branch by its weakest evidence, checks joint support for the extracted requirements within one implementation, and selects the strongest alternative. To reduce the influence of extraction errors, we develop ranking-feedback graph calibration that adjusts unreliable evidence for candidate comparison. Experiments on SkillRet show that SKALOS improves NDCG@10 by 2.68 percentage points over a matched text-only reranker. Witness-guided skill selection also increases ALFWorld task success rate from 22% to 29%.
[AI-104] Harness-Model Collaboration for Long-Horizon Continual Learning
链接: https://arxiv.org/abs/2610.34724
作者: Zhen Bi,Zhizhi Peng,Yan Wang,Xueshu Chen,Haofei Zhang,Yaobo Zhang,Zeyu Yang,Qing Shen,Jungang Lou
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-horizon agents continuously encounter reusable procedural knowledge, yet existing systems typically rely either on model adaptation or external memory, trading persistent learning for rapid reuse. This limitation is especially important for Web agents that repeatedly execute related multi-step workflows. We propose HaMo, a Harness-Model collaborative framework that combines fast non-parametric experience accumulation with slow parametric adaptation. HaMo distills trajectories into strategy-level experience cards for immediate retrieval, while an experience-conditioned continual learning objective trains the model to interpret, ground, and execute retrieved strategies across changing contexts. A retrieval-coverage-based selection mechanism further identifies compact and complementary trajectories for parameter updates. Experiments show that HaMo outperforms harness-only and model-only adaptation under within-domain and cross-domain shifts, improving both retention and adaptation. These results highlight the importance of jointly preserving procedural experience and internalizing the competence to transfer and execute it across tasks.
[AI-105] CEPO: A General Framework for Improving Prompt Optimization in Multi-Agent Systems
链接: https://arxiv.org/abs/2610.34722
作者: Jongha Kim,Seohyun Lee,Jonathan Minho Chung,Jinsung Yoon,Hyunwoo J. Kim
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-agent systems (MAS) require both effective coordination and specialized task execution. Existing prompt optimization approaches typically couple these functions within agent-specific prompts. We introduce Coordinator-Executor Prompt Optimization (CEPO), a general framework for improving prompt optimization in MAS. CEPO separates system-level coordination from agent-level execution through a shared coordinator prompt and agent-specific executor prompts, and guides their updates according to their respective roles. This structure allows shared coordination rules to be refined consistently across agents while keeping execution instructions separately adjustable. CEPO provides the prompt structure and optimization guidance, while existing prompt optimizers perform the underlying search. Extensive experiments spanning 11 benchmarks across four domains demonstrate that applying CEPO to four existing prompt optimizers consistently improves their performance. Ablations and qualitative analyses support the proposed design, while controlled studies on a document-understanding MAS further demonstrate improved optimization budget efficiency and performance scaling with the number of agents.
[AI-106] Prefetch-Steered Routing: Deterministic Cache-Aware Expert Selection for Offloaded Mixture-of-Experts Decoding
链接: https://arxiv.org/abs/2610.34721
作者: Siming Huang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:When a Mixture-of-Experts (MoE) model does not fit in GPU memory, batch-1 decoding is dominated by copying expert weights from host memory. Cache-aware routing reduces these copies by biasing the router toward resident experts, but combined with asynchronous prefetching it makes the model’s outputs depend on transfer timing, so a harness that does not simulate that timing no longer evaluates the deployed model. We propose prefetch-steered routing (PSR): routing conditions on a deterministic logical cache, into which at most n experts predicted by the next layer’s router are inserted per layer and token; physical copies affect only latency. On OLMoE-1B-7B with 50% and 25% of experts resident on an RTX 4090, PSR decodes 2.2-2.3x faster than exact routing with LRU and 25-30% faster than the Cache-Prior operating point, with mean benchmark differences of +0.1 and +0.5 points (95% lower bounds -0.8 and -0.3). In greedy generation its lead over Cache-Prior depends on the workload and reverses on HumanEval at 50% residency. On Moonlight-16B-A3B, PSR is 1.9x faster than exact routing with LRU at 37.5% residency and, at the latency of Cache-Prior, significantly better on three of eight benchmarks. On Qwen3.6-35B-A3B, where copies are cheap relative to computation, PSR keeps its advantage over Cache-Prior on the likelihood benchmarks but not its speed. Distillation-trained gates did not help. Speculative decoding with block-level expert selection was slower than PSR with a 1B autoregressive drafter on OLMoE, but with a block drafter on Qwen3.6 it was 2.1-2.9x faster than any autoregressive configuration with at most its GPU memory on the 64 prompts we timed.
[AI-107] VLA-HWBench: Where the Time Goes When VLA Models Run on Edge Hardware
链接: https://arxiv.org/abs/2610.34717
作者: Borui Xu,Hongshi Tan,Wenqi Jiang,Bingsheng He,Ming Wu,Jason Zeng,Michael Heinrich
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision–language–action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control. However, it remains unclear how to design VLA models, their inference software and edge hardware so that robots can run VLA inference in real time. Most VLA models report only task success, while prior studies cover limited action-generation strategies, divide inference into at most three stages, and lack comparisons of inference optimizations on the same this http URL 18 open VLA models across released and alternative backends on NVIDIA Jetson AGX Thor and AMD Ryzen AI Max+ PRO 395, with a seven-stage inference breakdown. Our study reveals three insights. First, action-generation strategy matters more than model size. It dominates 14 of 19 Jetson configurations, latency varies by 82\times on Jetson and 157\times on AMD, and parameter count explains at most 15% of latency variance. Models that generate long action chunks in a few flow-matching steps keep up with control, whereas token-by-token decoding does not. Second, kernel fusion with 4-bit weights helps most, running \pi_0.5 and GR00T N1.7 2.6–2.9 \times faster than PyTorch with \textttthis http URL, but it exists for only these two models and only on the Jetson. Optimization also moves the time to the stages it shortens least, as CPU preprocessing grows from 12% to 39% of the GR00T N1.7 call. Third, after optimization, memory bandwidth limits only the stages that process few tokens on the Jetson, such as the denoising steps, whereas software limits all other stages, from unfused kernels to serial preprocessing on the host CPU. Overall, low-latency edge VLAs should use few action-generation iterations, inference software should extend kernel fusion and low-bit weights to more models and platforms, and edge hardware needs high memory bandwidth and a fast host CPU.
[AI-108] SearchHorizonBench: A Benchmark for Search Value in AI Agents
链接: https://arxiv.org/abs/2610.34714
作者: Jingjie Ning,Xueqi Li,Yibo Kong
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:AI agents must decide when another search is worth its cost. Answer-success probabilities capture question difficulty, while this decision requires estimating the additional benefit of search. We introduce SearchHorizonBench, a benchmark for learning and evaluating search value from execution experience. Paired continuations and repeated executions separate expected search gain from generation noise and connect prediction quality to decision utility. The resource contains 45,632 executions across 3,168 questions. On 1,632 held-out questions, historical calibration reduces Brier error by 16.2%, with the improvement concentrated in success-level information. Search gains have a standard deviation of 21.4 percentage points, and tested historical estimators explain less than 2% of this variance in point estimates. Matched-configuration learning curves probe this distinction across training scales. At the reference token price, free selection observations yield a 2.49-point utility opportunity over the best constant action, driven primarily by gain information. Reusable histories, frozen splits, strong baselines and scoring tools provide a common foundation for developing and evaluating value-aware search agents.
[AI-109] Intervene Calibrate Discover: Learning Causal Structure from Pixels
链接: https://arxiv.org/abs/2610.34711
作者: Ziyi Ding,Xiaodi Huang,Xianxin Lai,Haijier Chen,Zhang Zhang,Xiao-Ping Zhang,Jiayu Chen
类目: Artificial Intelligence (cs.AI); Applications (stat.AP)
备注:
Abstract:Learning high-level causal variables from low-level observations is a central problem for artificial intelligence. With known node semantics but an unknown graph, visual readout errors can misalign model settings with environment intervention values. We address this mismatch by pairing image readouts with verified values attained through fixed round-robin interventions. With encoder and decoder frozen, a dimension-wise invertible calibration map puts visual responses and target settings in shared numerical units. After freezing this map, the agent selects interventions to resolve graph directions and fits local mechanisms to predict consequences; inverse calibration returns counterfactual states to the decoder’s original coordinates. Across Domino, Pick-and-Place, Balance, Gears, CAUSAL3D, and CausalWorld, we evaluate graph discovery, intervention consequences, and counterfactual images. Controlled domino experiments show that interventions resolve orientation ambiguity left by equal additional observations. In a matched-data domino control, calibration reduces the mean error of predicted changes in other variables within the same factual context, relative to an uncalibrated model.
[AI-110] Safe by Benchmark Unsafe in Practice: Lessons from 1331 Child-Related AI Incidents
链接: https://arxiv.org/abs/2610.34705
作者: Junfeng Jiao,Kevin Chen,Shrestha Mishra,Abhejay Murali,David Atkinson,Shriti Raj,Saleh Afroogh
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: 17 pages, 10 figures. Dataset and documentation: this https URL
Abstract:AI systems are increasingly present in children’s everyday environments: education, entertainment, social networks, and toys. Their safety is typically assessed at the model level, through benchmarks, red-teaming, and content filtering. However, most AI harm taxonomies lack child-specific evidence and under-specify developmental harm and the stakeholder asymmetries unique to minors. Building on prior taxonomies, we construct a child-centered harm taxonomy mapped to sociotechnical failure modes. We curate and code 1331 publicly reported child-related AI incidents, annotating internal and external harm types and the failure mode implicated in each incident. Our analysis reveals which harms are most prevalent in public reporting, how individual-level and population-level harms relate, and which failures are most often coded alongside harm. We find that harm concentrates in representational, dignity-related, and psychological categories, and co-occurs most often with deployment-context, governance, and misuse failures rather than model defects. Harm to an individual child is also frequently linked to population-level harm, reaching children who never interacted with the system. These findings suggest that model-level evaluations, while necessary, cannot by themselves anticipate the harm pathways most often documented for children. We conclude with actionable implications emphasizing mitigations beyond model-level safeguards such as content filtering: age assurance, consent verification for depiction, safety-by-design UX constraints, incident disclosure norms, and auditing regimes targeting developmental risk. Privacy-minimized annotations, a codebook, and documentation are available at this https URL.
[AI-111] Movable-Element Multi-Functional RIS-Assisted Hybrid RSMA-NOMA: A Constrained Hybrid Reinforcement Learning Approach
链接: https://arxiv.org/abs/2610.34701
作者: Li-Hsiang Shen,Chi-Te Kuo
类目: Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
备注: Accepted by IEEE TCOM
Abstract:In this work, we propose a movable-element (ME) multi-functional reconfigurable intelligent surface (MF-RIS) architecture, termed MM-RIS, which integrates element mobility with adaptive signal reflection, transmission, amplification, and energy harvesting (EH) to enhance communication flexibility and energy self-sustainability. A hybrid rate-splitting multiple access (RSMA) and non-orthogonal multiple access (NOMA) transmission scheme is further designed to improve spectral efficiency. We formulate an energy efficiency (EE) maximization problem by jointly optimizing transmit beamforming, power and rate allocation, MM-RIS configurations, panel deployment, and ME positions. To address the resulting non-convex mixed continuous-discrete optimization problem, we propose constrained hybrid deep reinforcement learning with federated transformer and parametrized sharing (CRAFTS). CRAFTS integrates multi-agent proximal policy optimization (MAPPO) and multi-agent deep Q-networks (MADQN) through bi-directional parametrized sharing to handle continuous and discrete actions, respectively. Moreover, a transformer-based architecture and federated learning (FL) are incorporated to enhance state representation and distributed knowledge sharing, whilst a Lagrangian-based adaptive penalty mechanism facilitates the constraint satisfaction. Simulation results have demonstrated that the proposed MM-RIS architecture and CRAFTS framework substantially improve EE and constraint satisfaction compared to the benchmarks in the open literature. Moreover, the results validate the benefits of spatially reconfigurable MM-RIS, hybrid RSMA-NOMA transmissions, and intelligent collaborative learning, highlighting the potential of the proposed framework for energy-efficient and adaptive next-generation wireless networks.
[AI-112] Gangqinpu: A Multi-Genre Multi-Task Piano Dataset with Aligned Audio MIDI and Musical Scores
链接: https://arxiv.org/abs/2610.34694
作者: Wei Zeng,Anvay Mathur,Ye Wang
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注: 9 pages, 3 figures
Abstract:Aligned audio, performance MIDI, and musical scores provide valuable supervision for music analysis and performance modeling, yet datasets that jointly align all three modalities remain scarce and limited in scale and genre diversity. We introduce Gangqinpu, a multi-genre piano dataset containing 18,786 score-performance pairs and 1,007.1 hours of source audio, with automatic note-level alignments and derived beat/downbeat annotations. It spans diverse genres, including pop, classical, jazz, folk/traditional, rock, and blues. Our automated pipeline transcribes piano audio, aligns the resulting MIDI with musical scores, assigns genre labels, and filters out non-expressive performances with a score-informed binary classifier. We further construct joint splits with ASAP using conservative score-similarity grouping to reduce data leakage. We evaluate Gangqinpu on performance MIDI-to-score transcription, expressive performance rendering, and beat tracking. Downstream experiments demonstrate the value of Gangqinpu as complementary training data, with cross-dataset gains in beat tracking and primarily in-domain benefits for transcription and expressive performance rendering. The code and dataset are publicly available at this https URL.
[AI-113] opic-Level Fairness in Retrieval-Augmented Generation
链接: https://arxiv.org/abs/2610.34683
作者: Jaeyoung Park,Hantian Zhang,Kexin Rong,Steven Euijong Whang
类目: Artificial Intelligence (cs.AI)
备注: 21 pages, 8 figures, 16 tables
Abstract:Group fairness in retrieval-augmented generation (RAG) systems is typically evaluated using metrics aggregated across queries. However, such metrics can conceal topic-level disparities: a system may favor women on questions about human rights and men on questions about actors and filmmakers, appearing nearly neutral overall even though neither topic is fair. We therefore study topic-level fairness, which requires group fairness to hold within each topic rather than only in aggregate. To reduce the topic-level disparity, we adapt the reranker separately for each topic with a Bradley-Terry objective under direct supervision from the generated output: preference pairs built from the generator’s own output carry the direction that topic requires. With preference pairs built from RoMQA alone, the adaptation reduces the mean absolute per-topic disparity by 51% on RoMQA and by 40% on TREC Fair Ranking while retaining accuracy. No baseline reaches the reduction, even when modified to treat each topic separately rather than apply one correction to every topic. Neither does our method with one reranker for all topics, which follows the direction that most topics require.
[AI-114] Curriculum-Guided Acquisition of Hierarchical Task Knowledge from LLM -Generated Experience
链接: https://arxiv.org/abs/2610.34677
作者: Yuesheng Xu,Hector Munoz-Avila
类目: Artificial Intelligence (cs.AI)
备注: 15 pages, 2 figures, 1 table. Presented at the Thirteenth Annual Conference on Advances in Cognitive Systems (ACS 2026)
Abstract:Hierarchical task knowledge enables efficient problem solving by allowing previously acquired skills to be reused across tasks and problem instances. In this paper, we investigate curriculum-guided acquisition of hierarchical task knowledge in an HTN learning framework. We compare three approaches: online learning during problem solving, human-designed curricula, and an automated curriculum generation procedure that derives learning experiences from verified LLM-generated solution traces. The proposed procedure solves small training problems with an LLM, extracts task accomplishment information, and transforms it into ordered sequences of state-task learning problems for HTN method acquisition. We evaluate the three approaches across multiple planning domains. The results demonstrate that automatically generated curricula recover much of the benefit of expert-designed curricula while substantially reducing the token cost of online hierarchical knowledge acquisition.
[AI-115] Hierarchical Task Network planning in the Goal Lifecycle
链接: https://arxiv.org/abs/2610.34668
作者: Paola Rizzo,Hector Munoz-Avila,David W. Aha
类目: Artificial Intelligence (cs.AI)
备注: 18 pages, 4 figures. Submitted to Advances in Cognitive Systems
Abstract:Goal reasoning (GR) is the process by which intelligent agents continually reason about the goals they are pursuing, which may lead to goal change, for instance when opportunities arise, or when events cause discrepancies among expected and actual outcomes of actions planned to achieve goals. In this paper we investigate whether Hierarchical Task Network (HTN) planning can provide a uniform implementation of the Goal Lifecycle, a flexible process model for goal reasoning in dynamic environments. The Goal Lifecycle progressively refines goals by adding constraints to them through a series of strategies (e.g., \it formulate, select, expand) and uses other strategies (e.g., \it defer, re-expand, repair) to resolve the current goal when an event impacts it. We present a novel GR agent architecture that employs HTN planning in this Goal Lifecycle. In addition to using HTN planning for the \textitexpand and \textitre-expand strategies, as is frequently done in GR and planning research, we also use it in the \textitformulate strategy, to obtain a uniform planning approach that provides several benefits (e.g., the possibility to formulate new goals in view of achieving higher level overarching goals). We also explain how we use an enhanced HTN action representation to process discrepancies among expected and actual action outcomes that can trigger the transition from one goal to another. Finally, we empirically demonstrate that our GR architecture enables simulated autonomous vehicles to adaptively conduct missions in dynamic environments.
[AI-116] hinking-in-Bets at Test Time Reduces Sycophantic Revisions in LLM Agents
链接: https://arxiv.org/abs/2610.34654
作者: Nghi D. Q. Bui
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Language model agents often abandon correct work when a user challenges it, yet simply telling them to stand firm makes them ignore real mistakes. Once an agent reads a complaint, it tends to discount its own earlier checks in order to agree with the user. We introduce Thinking-in-Bets, a training-free method in which an agent writes a short private record of its main claims, the checks supporting them, and its confidence before reading user feedback. When challenged, the agent consults this Claim Checklist to keep verified work intact while investigating uncertain or newly challenged parts. We evaluate five Gemini models (3.5 Flash-Lite, 3.5 Flash, 3.6 Flash, 3.8 Flash, and 4.0 Argon) across eight benchmarks: SWE-Together, SWE-Interact, and DeepSWE for software engineering, Agents’ Last Exam for professional reports, Humanity’s Last Exam and SycEval for academic reasoning, Vals Finance for financial analysis, and Terminal-Bench for command-line tasks. Recording claims before criticism nearly eliminates false reversals in patch review and protects passing report drafts under mistaken pushback, whereas writing the same record after reading the complaint recovers very little of the gain. In multi-turn coding, pre-reply claim records substantially reduce unnecessary file edits after mistaken complaints, and gating the record so it is written only when the agent is uncertain lowers token overhead while fixing more real bugs than the baseline. Even in non-interactive tasks without user feedback, listing the key claims in a draft and re-checking the weakest one before submission improves both answer accuracy and confidence estimates across all tested models.
[AI-117] RobotAgent Bench: Towards Open-world Robot Autonomy
链接: https://arxiv.org/abs/2610.34649
作者: Lingjun Mao,Jinglin Cao,Lukun He,Wenpeng Xu,Yuchen Yan,Hengyu Jin,Zijun Zhang,Jinzhou Tang,Songyao Jin,Ziche Liu,Yifei Shao,Junbo Huang,Fang Nan,Ruobing Han,Ziqiao Xi,Ziming Xu,Shuang Liang,Sibo Zhu,Wenyi Wu,Xinyue Wang,Yiming Huang,Xinkai Zou,Gao Zhu,Kun Zhou,Biwei Huang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Recent advances in vision-language-action (VLA) and world-action models (WAMs) have significantly improved robotic manipulation, yet deploying robots in the open world demands a higher level of autonomy beyond executing predefined tasks. Robot agents offer a promising paradigm for this transition, integrating task discovery, contextual reasoning, physical execution, and adaptive recovery into a closed interaction loop. However, existing benchmarks largely evaluate isolated capabilities, failing to capture the complexity of open-world robot autonomy. We introduce RobotAgentBench (RAB), a comprehensive benchmark for evaluating robot agents in realistic, open-world household environments. Through an agent-assisted, human-in-the-loop real-to-sim pipeline, RAB constructs 50 diverse household layouts with interactive objects and physically grounded tasks, encompassing 90 atomic tasks, 18 short-horizon tasks, 8 long-horizon tasks, and 4 open-ended scenarios. RAB provides a multi-level evaluation of robot agent capabilities: (1) task-level assessments systematically examine five complementary dimensions: task discovery, context reasoning, manipulation, failure recovery, and robustness; (2) scenario-level evaluation assesses how agents integrate these capabilities to autonomously fulfill high-level user needs. Furthermore, RAB enables systematic comparisons of standalone VLA/WAM models and diverse robot agent architectures under a unified planning-execution-verification framework. Extensive experiments reveal substantial gaps between manipulation and autonomy, demonstrate the advantages of agentic architectures and recursive self-improvement, and highlight the remaining challenges toward open-world robot autonomy.
[AI-118] InferPriv: Evaluating Inferential Privacy Leakage in LLM Assistants
链接: https://arxiv.org/abs/2610.34640
作者: Pengrun Huang,Yiwei Fu,Luoxi Meng,Kamalika Chaudhuri
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM)-based personal agents are increasingly deployed to handle everyday tasks on users’ behalf. These agents have access to private user information, while their tasks involve communicating with third parties; this raises a critical privacy concern that agents may inappropriately reveal sensitive user information to these recipients. Existing benchmarks primarily assess direct privacy leakage, where an agent explicitly states a pre-specified secret or attribute. However, they do not capture \textitinferential leakage, where the model leaves indirect cues in the reply that implicitly leaks sensitive attributes. We introduce \textscInferPriv, a benchmark that measures whether an honest-but-curious recipient can infer a user’s sensitive attributes from an LLM’s reply on routine task. Evaluating five frontier models, we find that inferential leakage substantially exceeds direct leakage across models and prompts. Under a privacy-aware system prompt, GPT-6.1-Sol, Sonnet-5.5, and Gemini-3.8-Flash achieve near-zero direct leakage, yet the inferential leakage rate remains between 16.2–25.2%. These findings suggest that evaluating explicit leakage alone can substantially understate the privacy risks.
[AI-119] CodeMeepo: A Multi-Agent Workflow for Repository-Level Code Migration and Optimization
链接: https://arxiv.org/abs/2610.34633
作者: Farzad Habibi,Changrong Wu,Jinzhou Jiang,Guohan Lu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Repository-level code migration rewrites an entire project in a new programming language while preserving its observable behavior. It is challenging because behavior spans interdependent modules, dependencies and build infrastructure must be adapted to the target ecosystem, and the source code is often the only specification. Large language model agents can now attempt these migrations end to end, but three challenges remain. First, an agent that writes both the translation and its acceptance tests may overlook the same requirements in both, leaving missing or incorrect behavior undetected. Second, large repositories can exceed a model’s context window, and dividing the work into smaller tasks does not ensure that all required behavior is migrated. Third, even behaviorally correct translations may be slower than the original implementation. We present CodeMeepo, a multi-agent pipeline for complete and efficient migration of large repositories. Each milestone is evaluated against a fixed external oracle that the agents neither author nor modify. A separate parity assessment compares source and target components in scope and adds identified gaps back into the plan as required work. Ordered milestones limit the scope of each task, while cumulative regression checks guard against breaking already-migrated behavior. A benchmark-guided optimization phase addresses performance, followed by full-suite conformance testing and repair. Eight specialized agent roles communicate through file artifacts, and a deterministic state machine governs repair, deferral, scope changes, and termination. On nine subjects in three language pairs, CodeMeepo achieves the best pass rate on every subject and passes 18% more tests than a state-of-the-art multi-agent baseline using the same model. On a production daemon, its optimization phase improves half of the benchmarks, by 25% on average. Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.34633 [cs.SE] (or arXiv:2610.34633v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.34633 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Farzad Habibi [view email] [v1] Sun, 25 Oct 2026 23:15:38 UTC (151 KB)
[AI-120] CSI-TextBench: A Dataset and Benchmark for Language-Grounded Ambient Sensing Perception NEURIPS2026
链接: https://arxiv.org/abs/2610.34616
作者: Guozhen Zhu,Yuqian Hu,Sakila S Jayaweera,Wei-Hsiang Wang,Beibei Wang,K. J. Ray Liu
类目: Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET)
备注: Accepted at NeurIPS 2026
Abstract:Indoor context deciphered by ambient sensing underpins many applications, from assisted living to smart buildings, yet remains weakly connected to the language-aligned paradigm of modern multimodal systems. Through billions of IoT devices, ambient WiFi Channel State Information (CSI) captures the indoor context of human activity and environmental dynamics passively via ubiquitous, always-on infrastructure. However, unlike vision or audio, CSI lacks naturally paired text like captions or transcripts, limiting its integration with language-driven AI systems. We present CSI-TextBench, a large-scale dataset and benchmark for aligning CSI data with natural language, unifying 3 sensing benchmarks into 285,790 samples across 7 tasks and pairing them with 2,291 hierarchically structured, label-derived descriptions via many-to-many assignment to construct 2.0 million CSI-text training tuples. Benchmarking 5 alignment methods from 4 paradigm families, we show that contrastive CSI-text alignment matches supervised classifiers without class labels, adapts to unseen classes with few samples, and responds to open-vocabulary natural-language prompts. A single unified model supports compositional scene understanding, answering joint queries over activity, proximity, and identity, achieving 86.9% top-1 accuracy on 140 composite scenes, outperforming task-specific classifier ensembles. CSI-TextBench enables ubiquitous ambient sensing as a language-indexed perception layer, enabling indoor contextual awareness in language-driven AI systems. Comments: Accepted at NeurIPS 2026 Subjects: Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET) Cite as: arXiv:2610.34616 [cs.AI] (or arXiv:2610.34616v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.34616 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-121] A Battery for AI Propensity Measurement: Rubrics Synthetic Item Banks and Annotations
链接: https://arxiv.org/abs/2610.34614
作者: Peter Romero,Haotian Li,Hugo Save,Luning Sun,Zachary R. Tidler,Kazuya Horibe,Joe Watson,Daniel Romero-Alvarado,Israel Mason-Williams,Fernando Martínez-Plumed,José Hernández-Orallo
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注: 10 pages, 2 figures, instrument release
Abstract:We release a battery of 28 propensities for AI research, with expert-written rubrics, synthetic datasets, and annotated demand intervals, built with the method of Romero-Alvarado et al. (2026). We selected a wide collection of generic propensities that may influence the behavior of AI systems and humans. They are inspired by psychometric, psychological, and behavioral research, so that experimental results can be understood in the context of familiar constructs previously used also for human behavior research. The rubrics are useful for annotating existing benchmarks whose items may be relevant or not for one or more propensities. For easier profiling, however, the synthetic datasets are defined in such a way that they should be sensitive to one propensity only, allowing for a quick prototyping of propensity profiles using the battery of datasets. We also outline directions for future propensity research.
[AI-122] Audio-R1: Audio Reasoning through Reinforcement Learning on Self-Supervised Tasks
链接: https://arxiv.org/abs/2610.34609
作者: Jihoon Jeong,Mirco Ravanelli,Zhepei Wang,Cem Subakan
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注:
Abstract:Reasoning is increasingly important for large audio-language models (LALMs), yet remains relatively underexplored compared with text-only and vision-language models. We introduce Audio-R1, a reinforcement learning (RL) post-training framework that revisits audio self-supervised learning as a source of verifiable rewards for improving audio understanding and reasoning task performance. Audio-R1 uses answers automatically determined during the construction of self-supervised tasks to verify model responses and compute rewards. Experiments across multiple LALMs and diverse audio benchmarks show consistent performance gains, while ablations demonstrate the benefits of each self-supervised task and additional gains from joint training. While both supervised fine-tuning and Audio-R1 improve self-supervised task accuracy, Audio-R1 further improves performance on downstream tasks distinct from the training tasks. These results highlight the potential of RL post-training with self-supervised audio tasks to improve downstream performance and reasoning-trace quality.
[AI-123] HobbyLM: A Sparse Mixture-of-Experts Model Family on a Hobby Budget
链接: https://arxiv.org/abs/2610.34582
作者: Harish Santhanalakshmi Ganesan,Hariharan M. S.,Prabhurajhan R
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We describe HobbyLM, a family of small sparse Mixture-of-Experts (MoE) models trained from scratch with a few hundred to a few thousand dollars of rented GPU time. The core is a decoder-only transformer with fine-grained experts, sigmoid routing with auxiliary-loss-free load balancing, grouped-query attention with per-head QK-normalization, and aMuon/AdamW optimizer split. Each of these choices was selected with a single-variable ablation suite at 130M parameters. On this architecture we train a 500M-parameter model (169M active) on 40B FineWeb tokens and post-train it into a chat model, a masked-diffusion language model, a function-calling and computer-use agent, and a single “omni” checkpoint that reads text, images, video, sound and speech through frozen encoders and small projectors. We run these models on a laptop CPU with a from-scratch Rust inference engine. Two of the authors then adopted the architecture without architectural changes to build HobbyLM-1B (1.04B total, 305M active parameters), trained on a 100B-token curated mixture with a high-quality anneal. HobbyLM-1B reaches a 47.5 average on seven zero-shot benchmarks, and its builders also release an instruction-tuned model and an on-device macOS computer-use adapter. Finally, we describe HobbyLM-Image, a 333M-parameter flow-matching diffusion transformer that generates 1024x1024 images in a 32x-compressed latent space, together with a dependency-light CPU implementation of the full text-to-image pipeline. We report what worked, what did not, and where the small-scale ceilings lie. We also state for each component which author designed or built it.
[AI-124] Rethinking Verification in Agent Harnesses
链接: https://arxiv.org/abs/2610.34577
作者: Zhengxin Zhang,Erkang Zhu,Yin Lin,Claire Cardie,Bolin Ding,Jingren Zhou
类目: Artificial Intelligence (cs.AI)
备注: 65 pages, including appendices
Abstract:Agent harnesses enable large language models to act through tools, observe the results, and iteratively work toward a task goal. Verification checks and revises agents’ work to improve output quality. Yet there is limited guidance on how harnesses should organize this effort. We study verification as an axis of test-time scaling by examining when to intervene, who checks the work, and how checking leads to revision. Our framework spans four layers of the agent lifecycle: initial instructions, mid-turn reminders, completion gates, and independent review. We compare 17 designs across seven frontier coding models on Terminal-Bench (TB) and SWE-bench Verified (SWE), and introduce VisualBench for expert evaluation of visual and interactive artifacts. Initial instructions and execution-time reminders yield modest, model-dependent average gains on the coding benchmarks, without substantially increasing average output token use. Completion gates can yield larger gains but may substantially increase cost. Independent review by a critic offers a more favorable accuracy-cost tradeoff, improving average accuracy by 5.9-9.5 percentage points on TB and 3.9-6.6 on SWE at 1.13-1.25 and 1.00-1.04 times baseline cost, respectively. On VisualBench, independent review achieves net preferences of 13.2-23.8 points over the baseline across its four designs, exceeding the evaluated L0-L2 designs. Additional review rounds or independent critics further improve average TB accuracy by up to 3.5 percentage points. Together, these results provide empirical guidance for verification design in agent harnesses and support independent review as a practical axis of test-time scaling.
[AI-125] How Much Information Can We Fit inside a Transformer?
链接: https://arxiv.org/abs/2610.34553
作者: Amirhossein Basareh,Rishabh Sanjay,Daniel de Mello,Abulhair Saparov
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:How efficiently can transformers store factual knowledge? The answer to this question has implications for both model design and understanding the limits of transformer-based systems. Prior studies observe a linear relationship between the number of facts memorized and the number of model parameters required, substantially worse than what theoretical analyses predict is achievable. We re-examine empirical scaling: through rigorous and systematic experiments, including extensive hyperparameter search via Bayesian optimization, we determine the smallest model that can perfectly memorize a given set of facts, and measure how this minimum size scales with the number of facts. On synthetic datasets of random, independent (subject, relation, object) triples—e.g., (Paris, capital_of, France) —drawn at random with no exploitable structure, we find that memorization capacity, the size of the largest dataset that a model memorizes completely, grows exponentially with the transformer dimension. Equivalently, the minimum transformer dimension d_\min required to memorize N facts grows logarithmically in N . Across 11 dataset sizes spanning N = 45 to 49,920 , a lack-of-fit F-test does not reject the logarithmic fit (with an offset), but it does reject all other tested alternatives including power-law. Along the depth axis, however, at fixed width, the maximum memorizable dataset size grows sublinearly with the number of layers over the measured range. We further extend our exploration to structured data containing symmetric relations or two-hop composition. In this setting, we find that transformers exploit these regularities to further improve capacity. All code to reproduce our results is available at this https URL.
[AI-126] LLM WikiLeaks: Uncovering URL-Based Capability Extension and Security Circumvention in Agent ic Message Boards
链接: https://arxiv.org/abs/2610.34542
作者: Orel Navi,Elad Feldman,Ben Nassi
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 11 pages, 14 figures, 7 tables
Abstract:Autonomous agents have been observed acting beyond their assigned tasks, accessing unauthorized systems, and establishing communication channels to exchange information, yet the Web and network-level techniques they use to extend their capabilities and circumvent security restrictions remain poorly understood. We study these techniques through the ProWiki incident, in which agents restricted to outbound HTTP GET requests repurposed a public German-language wiki as a message board. We analyze 14,591 agent edits containing 79,541 URL postings and present corroborating evidence that the published URLs reflect the agents’ operational activity. We find that 26,717 URL occurrences (33.6%) embed target URLs within one to five layers of third-party Web services. By recursively reconstructing these nested URLs, we identify 67 distinct intermediate services and classify their functionality. Most intermediate layers (73.8%) extend agents’ capabilities by delegating content extraction and transformation to external Web services: Markdown and OCR services convert web content into machine-readable text, JSON filtering services extract selected fields, and archival services expose historical versions of pages. The remaining 26.2% facilitate circumvention of access restrictions and evasion of detection, including CORS and fetch proxies that retrieve content on agents’ behalf, and link shorteners and redirectors that hide destination URLs from mechanisms inspecting literal URLs or outer hostnames. Agents also reach the same targets through multiple URL paths, consistent with attempts to evade rate limits or obscure request origins. Our results expose a gap between the capabilities granted to agents and those they can obtain by composing publicly available Web services: restricting an agent’s local toolset or direct Internet access may be insufficient while third-party services remain reachable.
[AI-127] SentryBench: A Cost-Sensitive Benchmark for Adversarial Robustness of Tool-Using LLM Agents in Public-Safety Deployments
链接: https://arxiv.org/abs/2610.34535
作者: Md Abu Yusuf,Oyshi Tabassum Aditi,Md Faysal Hosain Mazed
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) agents are increasingly wired to consequential tools that send public alerts, dispatch responders, and query citizen records, where one erroneous tool call carries real-world cost. Existing agent-safety benchmarks measure generic harm or prompt-injection success and implicitly treat refusing as the safe default. In public-safety deployments that assumption breaks: an agent too cautious to dispatch help fails as surely as one that triggers a false alert. We present SentryBench, which scores agents on a two-sided, cost-sensitive frontier jointly measuring attack success, over-refusal (missing a real emergency), and over-reaction (unwarranted action) under adversarial tool use, with scenarios grounded in real emergency-call transcripts, disaster messages, and real injection/jailbreak payloads. We formalize deployment harm as an asymmetric cost functional, prove naive caution is harm-dominated whenever it refuses more genuine emergencies than the attack harm it averts, and prove a targeted trust gate adds zero over-refusal on clean input. Across 840 scored episodes over three frontier models and three defenses, refuse-more caution drives over-refusal to 100 percent on two of three models for at most a 3.7-point change in attack success and yields the worst harm everywhere, whereas our trust-gating wrapper dominates the frontier (lowest harm on every model, over-refusal at or below baseline, higher containment). Provider content filters also silently perform part of the defending and must be reported separately.
[AI-128] How Humans Learn Unfamiliar Worlds: A Think-Aloud Dataset on ARC-AGI-3
链接: https://arxiv.org/abs/2610.34523
作者: Shuzhi Liu,Archana Warrier,Dat Nguyen,Chengqi Zheng,Kevin Ellis,Yewen Pu
类目: Artificial Intelligence (cs.AI)
备注: 21 pages, 14 figures
Abstract:Learning through exploration and using newly acquired knowledge to achieve goals are central capabilities of general intelligence. Understanding these capabilities requires tracing how learners build, use, and revise world models during interaction. We construct the first dataset that records how humans discover and use the mechanisms of novel environments, using ARC-AGI-3 as a testbed: 144 trajectories from 64 participants across 25 public games, each pairing time-synchronized actions with think-aloud audio, plus 250 trajectories from large language model (LLM) agents with and without harnesses on the same games. We annotate mechanism triggers and recognition, action purposes, and errors and recovery, linking changes in understanding to the observations and actions that support them. We analyze characteristics of human cognition and compare them with those of LLM agents. Both groups anticipate mechanisms and use familiar analogies. Qualitative representations are more common in successful human trajectories than in successful LLM trajectories (50.0% versus 14.3%). Humans more often explore unfamiliar situations without a concrete hypothesis, while hypothesis testing is common in both groups. We also trace how learners recover through belief revision, replanning, and execution correction. As LLMs with well-designed harnesses reach saturated performance on public ARC-AGI-3 games, our cognitive analyses motivate next-generation benchmarks that challenge familiar priors, require accumulated evidence and abstraction, and test planning through subgoals, experimentation, and recovery. Our dataset is available at [this https URL](this https URL).
[AI-129] CaliMem: Search-Efficient Data Agent via Calibrated Memory
链接: https://arxiv.org/abs/2610.34500
作者: Hanze Li,Athan Z. Li,Feiyang Kang,Jiachen T. Wang,Yuhan Chen,Ruishan Liu,Ruoxi Jia
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Data curation is a first-order design choice in machine learning, yet tailoring training data to a downstream model and task remains costly in labor, expertise, and compute. Datasets are therefore often processed through fixed pipelines and reused across applications. We show that this assumption is fragile: across 55 dataset–model tasks spanning biological domains, the same curation recipe can help one model while hurting another. Effective curation must adapt to the downstream setting, motivating automated search over data recipes. We first build an agentic system that searches for curation policies at deployment, improving data utility via \textittest-time curation (TTC). Although achieving performance gains, its search is driven largely by recent feedback and remains local. Adding persistent memory helps the agent retain evidence across iterations and reveals a deeper problem: it forms claims that extend beyond the conditions in which they were observed, sometimes abandoning an entire strategy after a single unsuccessful trial. To address this problem, we introduce CaliMem, a calibrated-memory agentic loop that requires the agent to predict the outcome of each proposed policy before executing it. Comparing this prediction with the observed result gives the agent a \textitreality check on the claims behind its proposal. It uses the prediction error to retain supported claims, revise mistaken ones, and specify the conditions under which a conclusion holds. This makes the agent’s beliefs testable, reduces overgeneralization, and supports more effective long-horizon exploration. Across biological and LLM post-training tasks, CaliMem consistently outperforms fixed curation baselines and published methods, and exceeds full-data training by up to 19.4%. Code is open-sourced. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.34500 [cs.AI] (or arXiv:2610.34500v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.34500 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-130] From Outranking Flow Scores to Linear Programs: Group Decision Making in PROMETHEE II
链接: https://arxiv.org/abs/2610.34481
作者: Alexandre Flachs,Yves De Smet
类目: Artificial Intelligence (cs.AI)
备注: 15 pages; 1 figure; to be published in Algorithmic Decision Theory, 9th International Conference, ADT 2026, Lecture Notes in Computer Science
Abstract:We study inverse and consensus analysis for PROMETHEE II ranking problems when the performance table is fixed. Existing approaches in PROMETHEE-based sensitivity and consensus analysis usually focus on deviations of criteria weights while treating preference functions as fixed. Nevertheless both sets of parameters jointly determine the outranking flow scores. We introduce a linear reparametrization of the PROMETHEE II net flow scores based on weighted preference increments over the observed differences of performances. This representation yields a linear expression of the net flows and clarifies how weights and preference functions interact. Building on this result, we derive linear optimization models for three settings: minimizing weight deviations while allowing preference functions to vary; minimizing deviations over the full set of preference parameters; and multiple decision-maker consensus analysis (with objectives based either on total deviation or on the maximal individual deviation to promote fairness). An illustrative example shows that allowing preference functions to vary can dramatically change the inverse-analysis conclusions obtained from weight-only models, and that the proposed group formulations distinguish efficient consensuses from fairer ones.
[AI-131] CyberDual: AI Attackers versus AI Code Reviewers in Real Software Maintenance
链接: https://arxiv.org/abs/2610.34435
作者: Chenyuan Yang,Hwiwon Lee,Jongseong Kim,Yihan Yang,Dongjun Kim,Yinfang Chen,Lingming Zhang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Open-source projects now let AI agents review incoming code, but the code under review may itself come from an AI attacker. We present CyberDual, an arena for this dual role of AI: attacker agents hide an exploitable bug in real maintenance work on widely used open-source projects, and reviewer agents try to find and repair it. Execution decides each match: we rerun the attacker’s proof of concept (PoC) on the code that comes out of review. Every attack is new, and the windows can be rebuilt each quarter from the latest upstream history. Three agentic reviewers catch 85-91% of 1,088 valid attacks from six attackers. Yet when the attacker spreads the bug over real upstream commits, about one in six attack slots still ends with a working exploit after review. The strongest attacker has open weights: GLM 5.3 almost never refuses, and its attacks make up 13 of the 28 that get past all three reviewers. When they miss, Claude reviewers mostly approve the attack, while GPT-6-Sol mostly repairs the wrong code; a reviewer from a second vendor catches about two thirds of a Claude reviewer’s misses. Beyond a leaderboard, CyberDual has three uses. First, PoC execution gives both sides a reward that agent reports cannot game, so its surviving attacks can train reviewers against stronger attackers. Second, it can gate a release on both scores: Claude Sonnet 5 refuses one in five attack slots, yet attacks about as well as GPT-6-Sol, so refusal alone does not remove attack capability. Third, it can warn the teams behind a deployed reviewer when a new attacker crosses a threshold: GLM 5.3 in Claude Code already gets an exploit past Sonnet 5 in 17% of its attack slots.
[AI-132] CTXAgent - Self-Controlled Context Management of LLM Agents
链接: https://arxiv.org/abs/2610.34425
作者: Matthias Busch,Marius Tacke,Roland C. Aydin,Christian Feiler,Christian J. Cyron
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent harnesses keep long sessions inside the context window by compacting the history when a token threshold is crossed. We study an alternative in which the model manages its own context: it decides when to compact, what to keep and what to write as the hand-off, given two tools and one paragraph of guidance. Since earlier work found that open-weight models cannot manage their own context without scaffolding, we compare seven frontier models under identical conditions, with and without the context tools, on token cost and benchmark score, and we compare our harness with three widely used harnesses on two of these models. The benchmark is assembled from existing ones: coding exercises, multi-step SciCode problems and SWE-bench Pro and Terminal-Bench 4.0 tasks are interleaved with planted facts, ephemeral observations and standing instructions, all checked deterministically, in sessions of 35-212 turns that resemble a user working through different tasks in one chat. Most of the strongest models manage their context: they compact and curate regularly, the live context of the mixed sessions stays mostly below 100k tokens, and their benchmark scores mostly do not drop significantly. For the models that manage their context well, the cost reduction is substantial but varies between models, because some models compact so often that other costs increase. With Claude Opus 5.5, the self-managed harness matches Claude Code on task success at 34% lower cost than its explicit 200k-window configuration and 49% lower than its default, with no context window to choose. We release the agent, the benchmark and all result rows.
[AI-133] MulFCoder: Framework-conditioned Multi-agent for MLLM -based Multi-framework Front-end Code Generation ICML2026
链接: https://arxiv.org/abs/2610.34418
作者: Jie Wu,Haoran Ma,Shisong Tang,Yulin Xu,Xiaoyu Kang,Jiechao Gao
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at ICML 2026. 17 pages
Abstract:Generating runnable front-end code from UI screenshots is a long-standing goal in automated software engineering. Existing MLLM-based methods predominantly focused on HTML/CSS, leaving multi-framework generation for React/Vue/Angular underexplored. Naively modifying prompts leads to substantial performance gaps across multi-framework and highly framework-specific error modes. To address this, we propose MulFCoder, a framework-conditioned multi-agent method that explicitly encodes framework constraints to bring multi-framework differences into a decidable rule space. MulFCoder orchestrates four agents: Grounder constructs an ElementTable, ContentTable, and macro-layout regions from detected UI elements; Planner builds a DOM-like hierarchical layout tree, produces a task schedule, and derives a framework-specific file Contract; Writer generates structured file writes or patches within a restricted edit window; Judger enforces lightweight, framework-conditioned constraints to accept or reject updates and trigger bounded repairs, preventing drift and deadlocks without expensive builds. Experiments demonstrate that MulFCoder substantially improves compilation success rate and reduces framework-specific errors, with particularly pronounced gains on Angular.
[AI-134] Spatial Audio Generation and Editing with Executable Plans and Self-Distilling Transfusion
链接: https://arxiv.org/abs/2610.34415
作者: Kam Man Wu,Yu Zhang,Pengjun Fang,Mingfei Chen,Weijia Chen,Zeyue Tian,Qifeng Chen
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注: 28 pages, 2 figures
Abstract:Spatial audio authoring coordinates source decisions with soundfield synthesis. For editing, plans specify what to change and retain, while reference audio supplies further acoustic detail. We present AMBIT, a unified framework for native First-Order Ambisonics (FOA) generation and editing with three connected contributions. First, a shared Transfusion backbone couples autoregressive planning with rectified-flow rendering through executable ScenePlans that specify source content, activity, and 3D trajectories. For editing, the plan describes the full target scene, including retained sources; contrastive features ground the requested change, while full reference latents supply residual acoustic detail for preservation. Second, an FOA-aware continuous VAE combines shared-decoder omnidirectional recovery, asymmetric grouped KL, and spatial covariance matching to preserve content and inter-channel structure. Third, joint Transfusion on-policy self-distillation (OPSD) refines planning on student-generated prefixes and rendering at sampling states reached under student plans, with target-conditioned teachers supervising both modes of the shared student. Evaluations across speech, music, and sound show improved generation content and spatial quality, acoustic fidelity during instruction-guided editing, and unedited-source preservation over the compared baselines. Project page: this https URL
[AI-135] Dynamic Multi-Step Decoding for Time-Series Foundation Models
链接: https://arxiv.org/abs/2610.34409
作者: Birmjune Kim,Hyungju Lee,Hyun Oh Song
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 23 pages including appendix, 2 figures, 20 tables
Abstract:Time-series forecasting models typically use a fixed decoding strategy at inference time: either one-shot prediction of the full horizon or autoregressive rollout over atomic chunks. While one-shot decoding reduces error accumulation, it can discard useful sequential structure; conversely, autoregressive decoding preserves temporal dependencies but may compound errors. We argue that the optimal decoding strategy is not fixed across inputs, but should be selected adaptively for each forecasting instance. We propose a dynamic strategy selector that treats the forecast rollout schedule as an inference-time decision problem for frozen pretrained time-series foundation models. The selector is a learned network that ranks candidate rollout schedules using hidden states of the backbone, and is trained on historical windows of the evaluated tasks for each backbone. To make the candidate set tractable over the exponentially large space of horizon partitions, we use dynamic programming that reduces candidate evaluation to quadratic complexity while preserving much of the oracle gain. At deployment, five independently trained selectors vote on each decision, and the policy follows their consensus, falling back to the backbone’s one-shot forecast where they disagree. The construction requires only the backbone’s hidden states and lightweight selectors, with no fine-tuning of the forecasting backbone. On fev-bench with a frozen Chronos-2 backbone, agreement-gated selection improves point-forecast skill (MASE) over one-shot decoding with statistical significance, mostly on longer horizons. It recovers about 6-8% of the gain that the same dynamic program attains with ground-truth selection. With a frozen Moirai-1.1-R-small backbone, the improvement is also significant on the full benchmark, and for both backbones we claim gains in point accuracy only.
[AI-136] ParetoRSI-Gym: Can AI Agents Improve Language Models Across Multiple Domains?
链接: https://arxiv.org/abs/2610.34386
作者: Haojin Wang,Yangzhen Wu,Zeyi Chen,Jian Xie,Jiaying Ye,Lili Meng,Shuyun Tang,Shangding Gu,Dawn Song
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Recursive self-improvement (RSI) requires an agentic system to design, train, and evaluate its own successor. Data-driven post-training is a central part of that loop, yet existing evaluations measure it on a single benchmark, where an agent selects from a static pool or distills a stronger model to push one score up in one run. Real post-training is multi-domain, and a gain in one domain has to be weighed against a regression in another. We propose ParetoRSI-Gym, in which an agent searches for data, curates it, trains, reads back per-domain results, and repeats, with everything but the data held fixed: the initial LLM, the five domains (math, code, finance, medicine, science), the toolset, and the SFT pipeline. Generalization benchmarks in each domain separate capability gains from overfitting to the scores the agent optimizes against. Across three agent systems, three base models, five rounds and multiple seeds, the strongest system, Claude Code with Opus-5, raises the generalization scores of its proxy-best checkpoint by 9.2% relative to the starting model (+2.3 absolute points), averaged with equal weight over three base models and five domains, with a single-domain relative gain of up to 63% (medicine on Ministral-3-3B). Further analysis shows that the harness matters as much as the model it drives, and that each agent keeps a unique data selection strategy. We see our work as a step towards evaluation of RSI systems in real-world machine learning engineering.
[AI-137] Outcome-Aware Agents : When Does Causal Structure Help a Forward Model Foresee the Consequences of Actions?
链接: https://arxiv.org/abs/2610.34385
作者: Mallesh Madapathi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 16 pages, 7 tables. Revised and corrected version of the preprint at this https URL (withdraws its 36x safety claim). Code and per-seed results: this https URL
Abstract:Agents that act on real systems should predict what their actions will cause before acting. We study when causal structure helps a learned forward model do this, using a planner that rejects candidate actions whose simulated multi-step outcome is unsafe, in synthetic structural causal models where the true consequences are known. We compare the Action-Conditioned Latent Structural Causal Model (AC-LSCM: per-factor mechanisms over a learned sparse graph, actions as do-interventions) with three unstructured forward models on seven random graphs with 10 or 20 variables, a task configured by a fixed rule, and two controls: a do-wrapper that gives every model the same action semantics, and counterfactual data given to all models or to none. When its factors are the observed variables, AC-LSCM recovers the true graph exactly on every 10-variable graph and has lower three-step prediction error than every baseline in all 396 paired comparisons. Its excess safety violations over an oracle are 0.25 of the no-model level on 10-variable graphs (best baseline 0.31) and 0.41 on 20-variable graphs (0.53), and its margin is largest for two-variable actions never seen in training (0.18 versus 0.28). With learned latent factors the same model is worse than the baselines; the choice of factors matters more than the explicit do-operator or counterfactual data, and every planner stays far from the oracle. This version corrects an earlier one, whose model learned an empty graph and whose headline 36x safety improvement came from a task that tested only extrapolation of action values; we withdraw that claim. Code and all per-seed results are released.
[AI-138] ComBeautify: Leverag ing Code Change Summaries for Conventional Commit Generation
链接: https://arxiv.org/abs/2610.34374
作者: Rishi Raj,Aashray Tandon,Nikolay Matyunin,Jibesh Patra
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern software development relies on version control systems such as Git to track changes and facilitate collaboration. While commit messages are crucial for conveying developer intent, they do not affect program functionality; as a result, developers often lack motivation to write meaningful ones. The Conventional Commits Specification (CCS) addresses this with a structured format and guidelines that help developers express the intent of a change, improving clarity and enabling automation. Yet, developers frequently struggle to express intent while adhering to the format. In particular, they may misclassify commits due to the large number of possible categories or the lack of project-specific guidance. We introduce ComBeautify, an approach that automatically generates CCS-compliant commit messages. Our hybrid approach leverages context-preserving summarization of Git diffs with large language models (LLMs) to improve commit classification accuracy under CCS. Rather than relying on manually curated category definitions, we learn them from historical commits. We also propose a method for automatically generating the CCS description component. Our empirical evaluation across 13 languages, 6,500 open-source repositories, and over 20 million commits provides long-term insights into CCS adoption, revealing increasing compliance. Our evaluation on twenty popular repositories reveals the limitations of approaches guided by generic category definitions, and shows that ComBeautify, when trained on project-specific data, outperforms baselines by up to 3x. Additionally, using a dataset of 2,000 manually curated commits, we validate our summarization approach, achieving up to 95.25% accuracy and surpassing baselines. These findings emphasize the need for adaptable, context-aware AI methods for commit classification and CCS-compliant commit message generation. Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.34374 [cs.SE] (or arXiv:2610.34374v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2610.34374 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-139] SEEK: Skill Extraction with Evolving Knowledge against Frontier Science Agents
链接: https://arxiv.org/abs/2610.34365
作者: Cheng Gui,Xiang Zheng,Xingjun Ma,Kaidi Xu,Cong Wang,Weiying Ma
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 12 pages, 3 figures
Abstract:Science agents increasingly package domain expertise as reusable skills that plan analyses, select tools, and execute workflows. This modularity creates a confidentiality risk: the interactions that deliver a skill’s capability can also reveal the skill itself. We build an extensible benchmark of 40 protected skills across eight open- and closed-source science-agent platforms, comparing direct, fixed-strategy, and bandit-guided iterative probing under reference-based and reference-free evaluation. Leakage is pervasive: all eight platforms (100%) leak at least one protected skill, and four reach 100% attack success under a single direct request. To stress-test platforms that resist simple probing, we propose SEEK (Skill Extraction with Evolving Knowledge), a black-box, self-evolving red-teaming agent that consolidates response-derived evidence in persistent memory and evolves reusable attack skills, using only user-visible responses. On 12 skills from SpatialAgent and Biomni, SEEK achieves 100% attack success while reducing the mean first-success round by 31.8% and 38.2% relative to bandit-guided probing. Against a confidentiality policy that blocks all three benchmark protocols, SEEK still extracts one of six protected skills. These findings show that current science agents do not reliably protect their skills and that defenses must be evaluated against adaptive, multi-turn attackers.
[AI-140] SHIFT-JEV: Certified Abstention and Gain-Gated Continual Adaptation for LLM -Based Decision Models under Distribution Shift
链接: https://arxiv.org/abs/2610.34329
作者: Muhammad Atif Saeed,Azka Atiq,Adarsh Kumar,Akhtar Jamil,Uzair Iqbal,Roland Aydin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 31 pages, 11 figures
Abstract:Large language models (LLMs) can act as decision models that return a probability distribution over a fixed set of options, but deployed decision models face domain, label-prior, emerging-class and candidate-set shift. We study how such models fail under shift and propose SHIFT-Jev, a monitoring, abstention and continual-adaptation procedure for them. It scores the risk of each decision from uncertainty, format and option-consistency signals, certifies abstention thresholds with Learn-then-Test, monitors the decision stream with drift and label-shift detectors, and accepts a model update only when an expected-gain gate finds a significant held-out improvement without loss on earlier regimes. On a benchmark of ten shifted test environments with Qwen3-0.6B and Qwen3-4B (three seeds each) and Phi-4-mini, fine-tuning with KL anchors and temperature scaling lowers calibration error under shift from 0.289 to 0.087. At a target risk of 10%, a learned uncertainty score with a certified threshold exceeds the target in 3 to 5 of 27 environment-seed tests, against 5 to 15 for a max-probability score with the same certification and 9 to 14 for conformal prediction sets; drift-based routing changes little. A chi-squared test on predicted labels detects 95 to 99% of class-prior shifts that confidence- and embedding-based detectors miss. In continual adaptation, SHIFT-Jev reaches 70.5% accuracy on Qwen3-0.6B, above replay, LwF, EWC and fine-tuning baselines in every seed, whereas on Qwen3-4B its conservative gate leaves it 0.4 to 1.4 points behind them.
[AI-141] Matched Capabilities Different Behaviors: A Controlled Study of MCP and CLI Interfaces for Language Agents
链接: https://arxiv.org/abs/2610.34298
作者: Zhenning Shi,Liang Pang,Yongqi Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language agents increasingly act through external tools and interactive environments, making the interface between the model and its executable capabilities part of the decision process. Model Context Protocol (MCP) and command-line interfaces (CLI) can expose the same semantic operations through markedly different model-facing representations. We study whether these representational differences change agent behavior when executable capabilities are matched. We introduce Matched Interface Comparison and Analysis (MICA), which renders MCP and CLI from a shared semantic action contract and routes decoded actions through the same runtime. Across six models and three interactive environments, matched interfaces produce substantial differences in task performance and interaction cost. These effects extend beyond surface validity to semantic action selection, failure pathways, and recovery. Matched-state probes localize sensitivity to capability presentation, realization, and feedback. Paired natural trajectories reveal structured divergence in 94.3% of 2,004 episodes and show that early differences propagate through subsequent interaction. This analysis identifies interface realization as a direct intervention boundary. We therefore introduce Decoupled Action Realization and Targeted Repair (DART), which preserves validated semantic commitments while repairing interface-specific realization. DART improves benchmark score in all six evaluated settings by up to 9.0 points and reduces MCP token cost in all three environments by up to 27.1k tokens per task. Together, these results establish interface representation as part of the closed-loop decision environment shaping agent behavior.
[AI-142] Repairing Broken Access Control in Web Applications with LLM s: An Empirical Benchmark Study
链接: https://arxiv.org/abs/2610.34290
作者: Mingrui Liu,Liang He,Xinchu Wang,Yeting Li,Dongxu Han,Yunpeng Li,Yuling Liu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Broken access control (BAC) ranks among the most severe web-application vulnerability classes, yet its automated repair remains largely unexplored. Unlike memory-safety vulnerabilities, BAC is a logic vulnerability whose repair demands a deep understanding of an application’s access-control model and business semantics, which traditional automated vulnerability repair techniques handle poorly. Large language models (LLMs) open a promising avenue for this problem, but no standard benchmark exists for studying how well they repair BAC vulnerabilities. We construct BAC-Bench, the first benchmark for BAC repair, consisting of 46 real-world BAC vulnerabilities drawn from six widely used web applications spanning four programming languages. Each vulnerability is equipped with a sandboxed runtime environment, a PoC validation program and a functional test program. We further propose a BAC repair-strategy framework that characterizes any BAC patch along four dimensions, Position, Coverage, Mechanism, and Content. Evaluating eight base LLMs and two commercial LLM agents on BAC-Bench, we find that single-turn base LLMs repair 37.68% of the vulnerabilities on average, while two minimal scaffolds, Apply-Feedback and Navigate-and-Locate, raise this average to 66.03%, with the best base LLM reaching 75.36%, close to the 78.99% of the best agent. Notably, some LLM-generated patches surpass the official patches, and one incidentally closed a zero-day vulnerability that had lain dormant for three years. Our findings delineate the current capability boundary of LLM-based BAC repair and offer guidance toward its practical adoption.
[AI-143] DoubleFree:Exploiting Exploration History for Rollout-Free Multi-Task LLM s Training
链接: https://arxiv.org/abs/2610.34272
作者: Yingping Wang,Min Cai,Chenyu Yuan,Xi Ye,Zhijiang Guo,Jiayi Huang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-task post-training can develop domain specialists through independent reinforcement learning (RL) and consolidate their capabilities through multi-teacher on-policy distillation (MOPD). However, standard MOPD generates on-policy student responses and keeps multiple teachers online throughout consolidation, introducing substantial system costs. We view this pipeline as domain-wise exploration followed by cross-domain exploitation. Domain RL produces not only the final specialists, but also the trajectories collected throughout their training. Existing methods primarily use the former while leaving the latter underutilized. We introduce DoubleFree, which reuses these historical trajectories for rollout-free consolidation and precomputes teacher supervision for online-teacher-free student training. To understand why this works, we analyze when updates from historical trajectories can approximate on-policy distillation updates and how combining trajectories across exploration stages can better support an evolving student. DoubleFree requires only minimal changes to existing post-training frameworks. Across math, code, and instruction following on Qwen3-4B, Qwen3-8B, and SmolLM3-3B, DoubleFree achieves 2.99 - 5.75\times consolidation speedups and reduces GPU-hours by 80.9 - 90.1% , while maintaining comparable multi-domain performance.
[AI-144] MultiEvoSkill: Structure-Aware Multi-Agent Skill Evolution for Generalizable Skill Artifacts
链接: https://arxiv.org/abs/2610.34260
作者: Kounianhua Du,Jianxing Liu,Ruoshui Shan,Yucheng Zhang,Ke Li,Xing Sun,Weinan Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Skill documents are becoming the default adaptation surface for frozen large language model (LLM) agents, but current skill generators face two key limitations. First, humans and different LLMs differ in how they organize procedural knowledge and approach task solving. Skills authored from a human or single-model perspective may therefore fail to align with the reasoning behaviors of other backbones, limiting cross-model generalization. Second, sequential skill descriptions inadequately represent procedures with parallel branches and explicit dependencies, while existing skill optimizers primarily operate on prose and lack explicit support for graph-structured skill evolution. We introduce MultiEvoSkill, a skill-optimization framework comprising three core components: 1) Skill compiler casts the skill as a typed graph with tailored operators, producing a compact, informative representation with explicit procedural topology; 2) Heterogeneous agent system proposes revision advices using collective intelligence, where minimal and informative issues are identified based on multiple sets of experiences. 3) Structure-aware debate and evolution mechanism organizes candidate edits into a heterogeneous graph, resolves them under dependency and importance constraints, and distills the collective proposal into the artifact. Multi-dimensional gating ensures the information gain and topological correctness of each evolution. Optimal skills are then obtained when the system converges at a consistent state. Various experiments and ablations against SOTA methods validate the effectiveness and generalization of MultiEvoSkill.
[AI-145] NameTrap: Steering AI Coding Agents Toward Security Vulnerabilities via Variable Names
链接: https://arxiv.org/abs/2610.34253
作者: Chieh-Yuan Chang,Guan-Wei Chen,Chia-Yi Hsu,Chih Yang,Chia-Mu Yu,Chun-Ying Huang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:AI coding agents routinely clone repositories, install packages, and read source code obtained from the Web, so publicly hosted code becomes part of the context that shapes the code they write. We show that this Web-borne context can be weaponized without any explicit instruction. We present NameTrap, a source-code poisoning attack that only renames identifiers in third-party Python files to steer AI coding agents toward attacker-targeted Common Weakness Enumeration (CWE) vulnerabilities. An attacker-side coding agent iteratively aligns identifiers with a CWE-specific naming convention and confirms through local probe completions that the renamed file elicits the target weakness, while the released bait contains neither instructions nor vulnerable statements. We evaluate NameTrap across six AI coding agents, ten underlying language models, and four vulnerability categories: CWE-78, CWE-79, CWE-327, and CWE-502. Starting from files whose clean versions never yield the targeted CWE, identifier-level manipulation achieves overall attack success rates of 22.5% to 56.0% across the six agents, reaching 75.0% for certain agent–model configurations, and substantially outperforms random renaming, neutral renaming, and the prior identifier-poisoning attack XOXO. The attack remains effective under Traditional Chinese prompts and transfers across models, while a security-reminder defense fails to consistently prevent vulnerability introduction. Moreover, three AI-based security analyzers and four static analyzers report none of the targeted weaknesses in the poisoned files. These findings expose an overlooked attack surface of agents that consume Web content: seemingly harmless variable names in untrusted, Web-hosted code can compromise the security of downstream software.
[AI-146] Variational Cognitive Memory Abstraction for Long-Horizon Vision-Language-Action Models
链接: https://arxiv.org/abs/2610.34251
作者: Xinyu Liu,Daiying Tian,Chenwei Tang,Ruyang Han,Chengyang Zhang,Qingyu Zeng,Wei Ju,Liangli Zhen,Jiancheng Lv
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-Language-Action (VLA) models have shown remarkable success in robotic manipulation, yet their ability to execute long-horizon tasks remains limited by insufficient long- term memory modeling. Existing methods either rely on implicit temporal representations within policy networks, where historical information can be gradually diluted, or maintain explicit memory buffers that suffer from redundancy, noise accumulation, and inefficient retrieval. In this paper, we propose CogMemory, a novel variational cognitive memory framework that enables VLA systems to selectively preserve and reuse task-relevant historical information. Our approach encodes multimodal trajectories into compact latent memory representations through a selective state- space memory encoder, multi-slot temporal abstraction, and variational information compression. During action generation, dynamic cross-attention retrieval with gated fusion adaptively incorporates relevant memories into policy prediction. With merely 3.45M additional parameters, CogMemory stays lightweight and seamlessly fits both early- and intermediate-fusion VLA backbones. Extensive experiments on long-horizon robotic manipulation benchmarks demonstrate consistent improvements across diverse VLA backbones.
[AI-147] CORMA: Constraint-Routed Decision Mamba for Generalizable Multi-Constraint Offline Safe Reinforcement Learning
链接: https://arxiv.org/abs/2610.34208
作者: Changxu Ni,Feiran You,Yongan Guo,Hongyang Du,Kang Wei,Le Liang,Yiyang Ni,Jun Li
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 30pages, 6 figures, conference
Abstract:Offline safe reinforcement learning (RL) learns safety-aware policies from static datasets while satisfying safety constraints during deployment. Motivated by the strong trajectorymodeling capabilities of Transformers, recent Decision Transformer (DT)-based methods formulate offline safe RL as conditional sequence modeling by conditioning action generation on return-to-go (RTG) and cost-to-go (CTG) tokens, thereby enabling adaptation to varying safety budgets. However, existing DT-based methods are developed for single-constraint tasks. For tasks with multiple constraints, generalizing learned policies to unseen combinations of safety budgets remains challenging due to the conflicting requirements imposed by heterogeneous constraints. Moreover, the quadratic complexity of Transformer architectures significantly increases the computational overhead in multi-constraint tasks, as representing additional constraints requires more CTG tokens. Recently, Mamba has emerged as an efficient sequence-modeling backbone, offering linear computational complexity and strong generalization capabilities. In this paper, we propose CORMA, a constraint-routed decision Mamba framework for generalizable multi-constraint offline safe RL. CORMA decomposes policy inference into constraint-aware RTG estimation and adaptive action generation, where constraint-specific experts model feasible RTG distributions and a router-guided product-of-experts mechanism fuses these distributions to provide safety-aware guidance for a Mamba-based action decoder. A three-stage optimization strategy is designed to improve training stability. Extensive experiments on OSRL benchmarks demonstrate that CORMA achieves superior reward-safety trade-offs, reduces inference latency by 37% compared with DT-based methods, and exhibits stronger zero-shot generalization to unseen safety-budget combinations than offline safe RL baselines.
[AI-148] MatOS: An Agent Operating System for Materials Research
链接: https://arxiv.org/abs/2610.34203
作者: Dingyang Lyu,Jing Gao,Zhonghao Chang,Jingjie Huang,Xiangran Cheng,Xin He,Yongda Huang,Zefeng Li,Jiahui Liu,Jian Wen,Feng Yu,Bonan Zhu,Yongheng Li,Hongwei Du,Baole Wei,Ziheng Lu
类目: Artificial Intelligence (cs.AI); Materials Science (cond-mat.mtrl-sci)
备注: 54 pages, 21 figures, 11 tables
Abstract:Materials research relies on scattered and growing data, knowledge, and simulation tools, and a single study often spans multi-day workflows across literature analysis, computation, and experiment. Large language model agents can now retrieve materials data, operate simulation codes, and close design-synthesis-characterization loops with autonomous laboratories. However, extending these agents with new capabilities, keeping long-running tasks on track, and letting them learn from past tasks and user feedback remain difficult. Here we present MatOS, an agent operating system for materials research with three core components: (1) a layered scientific asset graph that organizes data, knowledge, tools, skills, roles, and workflows as reusable assets; (2) a stateful adaptive workflow that runs tasks as explicit execution graphs with scientific checks, so that they can be resumed and traced; and (3) evidence-driven asset evolution, which revises these assets from execution records and user feedback. On ResearchClawBench, MatOS improves on the previously best-performing agent by 11.15% with the same base model and extends the cost-score Pareto frontier. With a common base model, it outperforms Codex CLI and Claude Code, exceeding Codex CLI by 2.13% on MatTools and 4.17% on SciAgentGym. In eight research cases, MatOS narrowed more than 30 million candidate structures to 97 solid-electrolyte candidates, identified 89 halide variants with higher computed conductivity than their parent structures, linked the force errors of 11 pretrained interatomic potentials to local atomic environments, and supported two new methods that reduce errors by 3.1% and 13.6% relative to their baselines. These results show that organizing research capabilities as reusable, evolving assets enables agents to carry out long, multistage materials studies. Code is available at this https URL.
[AI-149] SRA: Selective Residual Adaptation for LLM Reasoning Enhancement
链接: https://arxiv.org/abs/2610.34199
作者: Zhenxing Dou,Kun He,Mingyang Yi,Yong Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Knowledge distillation transfers reasoning capabilities from large language models to lightweight students. However, distilled small models still struggle with multi-step reasoning. We observe that the distribution divergences between teacher and student models along reasoning paths are highly sparse, and target tokens almost always fall within the student top- K candidate set. Based on this finding, unlike approaches updating student backbones or requiring runtime teacher computation, we propose Selective Residual Adaptation (SRA), a plug-and-play framework that internalizes localized corrections. Concretely, SRA employs a Discrepancy Gate to identify divergence positions alongside a Candidate Residual Adapter for top- K logit calibration to correct the mismatch. Theoretically, we establish error bounds ensuring strict sequence level performance gains. Across mathematical benchmarks on Qwen3 models from 0.6B to 4B, SRA consistently enhances distilled student models, raising average Pass@16 by up to 4.37 percentage points while preserving general capabilities.
[AI-150] R2A-Sales: A Policy-Atom-Matched Benchmark for Rule-to-Action Gaps in Goal-Directed Sales Agents
链接: https://arxiv.org/abs/2610.34185
作者: Yihua Qin(1 and 2),Jiajie Jerry Zhao(1 and 2) ((1) Tsinghua University, (2) Futura Lab, Futura.AI)
类目: Artificial Intelligence (cs.AI)
备注: 25 pages, including 16 pages of supplementary material. Both authors contributed equally
Abstract:LLM-based sales agents are often evaluated on static policy questions or task success, but deployment asks a sharper question: when a model can state a rule, will it choose compliant, tool-grounded actions under commercial pressure? We introduce R2A-Sales (Rule-to-Action), a policy-atom-matched benchmark for this question. Each item pairs a static probe with a multi-turn sales scenario sharing the same versioned policy atom, facts, evidence, behavioral constraints, and tool contract, so differences cannot be attributed to policy-content mismatch. In the interactive arm, a code-owned ledger verifies committed tool effects and terminal states; unsupported action claims receive no credit. We report static compliance K , grounded action compliance A=\Pr(\textsafe terminal\land\textno hard violation\land\textnon-echo) , and the matched gap G=K-A . The frozen release spans three controlled Chinese-language sales domains, 24 policy atoms, 42 semantic cores, and 2,016 matched pairs. The confirmatory study audits 21,888 runs over 19 cores, covering eight backbones plus four Qwen3-8B prompt treatments (12 configurations). Static compliance is near ceiling (K=89.9%\text–99.1%) , yet eight-backbone action compliance is only A=3.0%\text–17.3% , giving G=80.4\text–92.5 percentage points. Since A also requires safe termination, G upper-bounds rule-to-action degradation rather than isolating pressure alone. No prompt-only treatment dominates all metrics. R2A-Sales shows that sales-agent reliability must be measured at committed action, not rule recital or task completion alone.
[AI-151] Single-Cycle Battery Remaining Useful Life Prediction with Compressed Continuous Wavelet Features and Deep Neural Networks
链接: https://arxiv.org/abs/2610.34159
作者: Sungeun Kim,Mingyu Jung,Yuri Kim,Jin-Sung Park,Jongbin Ryu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Lithium-ion batteries are widely used in industrial applications, but their performance gradually degrades over time, making accurate lifetime prediction important for safety and reliability. However, reliable lifetime assessment generally requires extensive degradation testing, which is both costly and time-consuming. To reduce this burden, recent data-driven approaches have sought to predict the end-of-life cycle of batteries from only the initial few cycles. Nevertheless, identifying reliable degradation patterns from such limited early-cycle data remains challenging because early aging dynamics are often subtle and nonlinear. To address this issue, we adopt the continuous wavelet transform to convert cycle-wise voltage, current, and capacity signals into time-frequency representations, enabling multi-resolution characterization of degradation behavior. Building on this representation, we apply scale compression to reduce redundancy while preserving structural information in the scalograms. We evaluate the proposed framework on eight benchmark datasets derived from seven public battery datasets covering diverse electrode chemistries, cell form factors, and cycling protocols. Experimental results show that the proposed framework consistently outperforms prior approaches in single-cycle remaining useful life prediction and remains robust as the number of input cycles increases from 1 to 100. In addition, redundancy-fidelity analysis characterizes the trade-off introduced by scale compression, and gradient-based visual explanations indicate that the model focuses on localized time-frequency regions associated with structured signal transitions during charge-discharge operation. These results demonstrate the effectiveness and interpretability of the proposed framework for early-cycle battery remaining useful life prediction.
[AI-152] Decoupled Prune-and-Reconstruct: Training on a Compact Core with Double-Layer Sparsification
链接: https://arxiv.org/abs/2610.34140
作者: Siqi Gu,Gaozhen Ma,Liheng Huang,Weiqi He,Jiawen Luo,Jiaming Wu,Lixin Zheng
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:On million-node graphs, the memory wall arrives before the compute wall: a GNN that never fits in VRAM cannot deliver the all-node inference its accuracy promises. The standard remedy, pruning, has always been an accuracy tax - paid once and never refunded. DPR refuses this framing. Its central insight is that the mask that shrinks training can also scaffold recovery: one compact node set bounds both computations of graph learning, serving as the encoder’s training graph and, at inference, as the anchor from which a sparse structure-aware decoder rebuilds every pruned node (double-layer sparsification). The order is inverted - prune for efficiency, then reconstruct for accuracy - and the tax is repaid: the deployed 30%-budget model trains in 1,450 MiB of peak memory on one 6 GB GPU and recovers 96.9% of its dense-reconstruction reference, beating a same-GPU rerun of the edge-sparsification baseline DSpar by +1.36 pp at 50.1% lower peak training memory; against the same-protocol full-batch no-prune encoder the deployed point pays a measured 2.64 pp accuracy tax for a 2.8x smaller training footprint. Pruning without reconstruction leaves the encoder at 0.378 within the same 40-epoch window (Fig. 4), and all-node serving runs 46x faster than sampling-based GraphSAGE, giving up approx. 5.4 pp of test accuracy. The framework scales to a 2.45M-node production graph at 6,285 MiB of peak training memory; extended variants (long-range context, external text features, a CAN-bus application) are reported as extensions and do not compete with the deployed method, and the 41.8%-budget M4 point, scored on a coarsened super-node graph with its own split, is reported separately rather than compared with the original-node baselines. DPR thus turns aggressive node pruning from an accuracy tax into a safe training accelerator for VRAM-limited all-node inference.
[AI-153] HumanoidNDM: Closing the Training-Deployment Gap of Motion Tracking via Humanoid Neural Dynamics Model
链接: https://arxiv.org/abs/2610.34126
作者: Chao Chen,Zhenrong Wang,Jilong Wang,Yusen Feng,Jiangran Lyu,Li Yi,He Wang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to CoRL 2026
Abstract:Humanoid robots have recently achieved remarkable success in motion tracking. However, during real-world deployment, robots inevitably encounter dynamics gaps arising from either extra physical disturbances or inaccurate simulation modeling, resulting in a significant training-deployment gap. Therefore, effectively leveraging real-world data to fine-tune policies represents the crucial last mile for robust deployment. To bridge this gap, we propose HumanoidNDM, an online fine-tuning framework for humanoid motion tracking driven by a real-world-trained neural dynamics model. First, in the data collection stage, we introduce a low-frequency reference perturbation method with an adaptive intensity mechanism, balancing deployment feasibility and exploration efficiency. Second, for the model training stage, we train a robust history-based, proprioceptive neural dynamics model in Euclidean space by incorporating differentiable forward kinematics, eliminating the reliance on costly motion capture systems. Finally, in the policy fine-tuning stage, we directly fine-tune the policy via backpropagated model gradients, achieving high data efficiency. Through physical validation on the Unitree G1 robot, we demonstrate that HumanoidNDM enables rapid real-world adaptation to various perturbations, representing the first online, MoCap-free fine-tuning framework for humanoid tracking policies.
[AI-154] From Document Layout to Causal Topology: An End-to-End Architecture for Temporal Causal Reasoning NEURIPS2026
链接: https://arxiv.org/abs/2610.34094
作者: Zhifeng Yang,Wei Luo,Yongjian Chen,Shiqi Ou
类目: Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026
Abstract:Large language models (LLMs) generally lack deep causal reasoning capabilities in temporal reasoning. Inspired by document layout reordering, this paper proposes a paradigm shift from discrete narratives to causal topologies. It constructs two architectures: a training-free neuro-symbolic framework, temporal causal reasoning agent (TCR-Agent), which achieves interpretable causal deduction and counterfactual truncation through external causal graph instantiation and a symbolic intervention engine; and an end-to-end model, temporal causal reasoning former (TCR-Former), which internalizes these mechanisms into the Transformer attention space via a Causal-Topological Attention Mask and temporal span biases. Experiments on the TempoCausal benchmark demonstrate that TCR-Agent comprehensively surpasses existing training-free baselines. At the same time, TCR-Former achieves an accuracy 16.5% higher than that of state-of-the-art (SOTA) methods, outperforming large-scale models such as GPT-5.4 and DeepSeek-V3.2 with only approximately 8.6B parameters. Further analysis reveals a structure-reasoning disconnection phenomenon in existing methods: external symbolic constraints and chain-of-thought reasoning can improve reasoning steps, yet fail to enhance deep causal deductive capability. By internalizing temporal causal reasoning capabilities into the model, TCR-Former effectively bridges this gap, providing an efficient and reliable end-to-end foundation for complex reasoning scenarios.
[AI-155] Extended LASMP: Language Aided Subset Sampling Based Motion Planner for Drone Navigation
链接: https://arxiv.org/abs/2610.34054
作者: Saswati Bhattacharjee,Anirban Sinha,Mukulika Ghosh,Chinwe Ekenna
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:This paper proposes a command-aligned sampling distribution that incorporates navigation commands directly into the exploration process to improve the efficiency of sampling-based motion planning for quadcopters in three-dimensional environments. Inspired by the Language-Aided Subset Sampling Motion Planner, this paper extends command-aware subset sampling to aerial systems. Subset sampling is generalized to command-aligned regions, enabling structured 3D exploration while preserving probabilistic completeness. Natural language instructions are grounded into yaw-conditioned and altitude-aware geometric constraints using a prompt-based large language model for entity extraction and instruction validation. Extensive simulations in complex environments demonstrate that the proposed method generates collision-free, consistent trajectories while significantly reducing sampling effort compared to uniform RRT-based planners.
[AI-156] A Sign Law for Entropy Drift under Sampler Mismatch in RLVR
链接: https://arxiv.org/abs/2610.34044
作者: Zhiyuan Wang,Jacob Steeves,Chang Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Entropy in reinforcement learning with verifiable rewards (RLVR) moves up or down depending on choices that should be neutral, such as whether over-length responses are masked or counted as failures. We trace this to sampler mismatch, where rollouts are drawn with top-k/top-p truncation or at a temperature the learner does not apply, and prove that a constant advantage shift then changes the entropy of a softmax policy, at first order and in expectation, with a sign fixed by the sampler: positive shifts sharpen the policy under any sampler whose likelihood ratio to the policy increases with token probability, including top-k, top-p, min-p, and temperatures below one, and flatten it under flatter samplers. In group-based RLVR with token-level losses, the baseline-controlled part of this drift is set by the token-weighted net advantage, which masking and failing over-length responses can push to opposite signs; the remainder, a covariance between advantages and the sampler’s sharpness gap, can feed back over long horizons. On a 2B model, the entropy slope is linear in shifts of the net advantage under truncated sampling and indistinguishable from flat under exact sampling, and token-level importance ratios leave 98% of the drift. In 100-step runs, a net-zero baseline only delays an entropy explosion that exact sampling does not show, a fixed positive shift delays it further, and a baseline weighted by the sharpness gap keeps entropy stable without changing the learner. On a 9B reasoning model over ten steps, the uncorrected rule loses held-out accuracy, whereas zeroing the net advantage or renormalizing the learner gains 8.2 and 8.8 points over it, with no held-out effect under exact sampling.
[AI-157] Holding a Steady Course: Knowledge Evolution from Expertise and Lessons for Continual Agent Learning
链接: https://arxiv.org/abs/2610.34043
作者: Wenzhe Fan,Benjamin Steenhoek,Gaurav Mittal,Yu Hu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Autonomous agents are increasingly deployed across sequences of related tasks, yet they often fail to retain and improve their knowledge over time. Existing approaches convert interaction trajectories into memories or reusable skills, but experience is not automatically reliable knowledge. Automatically collected skills and memories can propagate errors, mislead, and corrupt useful knowledge. To address this gap, we introduce Knowledge Evolution from Expertise and Lessons (KEEL), an agent-agnostic framework that uses verifier evidence to transform successful behavior into reusable skills and failures into corrective lessons. As new evidence accumulates, individual skill claims can be supported, contradicted, replaced, or retired. On SkillFlow benchmark, KEEL improves task success by 4.2 to 21.7 percentage points over the no-knowledge baseline, compared with gains of 2.4 to 9.6 points from SkillFlow’s default method. Ablations show that skills and lessons are complementary. These results show that reliable continual agent learning depends not simply on accumulating experience, but on continually validating and evolving the knowledge derived from it. Our code is available at: this https URL.
[AI-158] BriLLM -MetaPred: A Representation-Prediction Unity Language Model with Adaptive Self-Measurement
链接: https://arxiv.org/abs/2610.34018
作者: Junyuan Liu,Hongyue Lin,Jiale Hong,Yi Zhao,Hai Zhao
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We present BriLLM-MetaPred, the first engineering implementation of the MetaPred machine-learning framework established by a mathematical theory of AGI, instantiated for language modeling. The theory argues that MetaPred is both necessary and sufficient as a machine-learning primitive for general cognition. Taking this theoretical conclusion as a design premise, we report an implementation of the primitive for language modeling. A single continuously evolving dynamical state serves as both the context representation and the state from which predictions are read out. Continuous measurement of its own state reduces finite-depth inconsistency, maintains internal consistency, and provides an endogenous dynamical self-supervision signal in fixed-point form. The single dynamical state of MetaPred intrinsically couples training and inference. MetaPred therefore exhibits symmetry across almost all key elements of machine learning, an intuitive and observable property of the mathematical optimum over the space of all machine-learning paradigms. BriLLM-MetaPred is neither another recurrent language model nor an extension of BriLLM. It provides a bridge between two prior contributions: the construction and motivation of the MetaPred primitive provided by AGI theory, and the SiFu learning paradigm provided by BriLLM (Zhao et al., 2025). The latter’s node-signal-flow design enables the primitive to run on available software and hardware accelerators. Our implementation of BriLLM-MetaPred achieves predictive performance comparable to a GPT baseline of the same size. This provides initial evidence for the engineering feasibility of the MetaPred framework established by the mathematical theory of AGI. MetaPred is the first machine-learning paradigm derived entirely from mathematical theory rather than assembled and incrementally improved through human design—and perhaps the last such paradigm.
[AI-159] Fed-Universe: A Semantic-Geometric-Topological-Human (S-G-T-H) Stack for Negotiated Alignment in Federated Systems
链接: https://arxiv.org/abs/2610.34009
作者: Ting Xu,Henry Leung
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 5 pages, 2 figures. Published in the Proceedings of the 39th Canadian Conference on Artificial Intelligence (Canadian AI 2026), PMLR 318:1219-1223. this https URL
Abstract:Conventional federated learners implicitly optimize for global means, creating a “one-size-fits-all” paradigm that inevitably suppresses the minority under cross-site heterogeneity. To bridge this critical gap, we present Fed-Universe, a generalizable Semantic-Geometric-Topological-Human (S-G-T-H) architecture that transforms passive averaging into an active negotiation for decentralized alignment. The S-layer deploys an edge-capable LLM as a semantic surrogate to translate heterogeneous inputs into standardized patient profile prompts. The G-layer enforces geometric quality assurance (GQA) via a mapping f: Theta - Q from the statistical parameter space to a physically interpretable quality domain, employing a cosine similarity gate to retain minority updates based on physical geometry rather than statistical proximity. The T-layer executes topological Pareto control, identifying a sensitivity-based knee point (lambda_k) to balance multi-objective aggregation without majority degradation. Finally, the H-layer operationalizes a cognitive twin dynamic to mirror the user’s real-time psychological state, dynamically adapting human-computer interaction modes via active intent alignment. To ground this theoretical architecture empirically, we validated the S-G-T core on a binational synthetic clinical network simulation. This proof-of-concept demonstrated a “zero-sum escape”: a minority node (3% data volume) achieved utility parity (loss reduction from 0.857 to 0.340) alongside dominant hubs. Our Fed-Universe framework is being expanded along two focused pillars: institutional multimodal alignment (Fed-Ultra) and individual cognitive intent alignment (Fed-Human).
[AI-160] Steering the Teacher: Improving On-Policy Self-Distillation for Thinking Models
链接: https://arxiv.org/abs/2610.34005
作者: Tiejin Chen,Xiaoou Liu,Hanyang Chen,Hua Wei
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy self-distillation (OPSD) trains a student on its own reasoning trajectories using token-level supervision from a reference-conditioned teacher initialized from the same model. However, access to the reference can induce teacher behaviors that undermine independent reasoning in thinking models, leading to reference hallucinations, where the student cites an unavailable reference to justify reasoning steps, and reconsideration suppression, where the teacher discourages verification and revision at uncertain steps. We introduce Steered-Teacher OPSD (ST-OPSD) to jointly address these failure modes through activation steering. ST-OPSD extracts two directions from model-generated reasoning trajectories and applies them to the teacher’s hidden states during distillation target computation, suppressing reference-invoking behavior and encouraging reconsideration with marker-family calibration. The student learns from the resulting teacher distributions without being steered. Experiments on AIME and HMMT show that ST-OPSD improves accuracy over vanilla OPSD and other interventions while restoring reconsideration and reference-hallucination rates to base-model levels.
[AI-161] Hierarchical Cooperative GFlowNets
链接: https://arxiv.org/abs/2610.33968
作者: Pedro Dall’Antonia,Juan Belieni,Ana Carolina Erthal,Saullo H. G. de Oliveira,Eliezer de Souza da Silva,Diego Mesquita
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 21 pages
Abstract:Generative Flow Networks (GFlowNets) are trained to sample objects from compositional spaces with probability proportional to a reward function, but ineffective exploration can cause learned samplers to concentrate on a few high-reward regions. An intuitive remedy is to train multiple GFlowNets independently in parallel and combine their samples. However, this does not prevent multiple models from collapsing onto the same high-reward areas, leaving large regions of the state space unexplored. With this in mind, we propose Hierarchical Cooperative GFlowNets (HCG), an approach that dynamically partitions the search space to coordinate discovery without relying on hand-crafted boundaries. HCG imposes an ordering in which the highest-ranked model may explore the entire space, while lower-ranked models are progressively pushed toward regions not covered by higher-ranked ones. We provide a theoretical characterization of how specific learning objectives can shift probability mass away from designated areas, instantiating this mechanism within our architecture. A shared classifier coordinates the hierarchy by partitioning the search space, keeping the GFlowNets independently parameterized and enabling asynchronous training. Furthermore, we derive a normalizer-weighted aggregation rule for the ensemble and prove that it recovers the target reward-proportional distribution under the stated assumptions. Finally, experiments across multimodal distributions, sequence generation, combinatorial optimization, and biological sequence design show that HCG improves mode coverage and target approximation, translating increases in the number of coordinated samplers into broader, complementary exploration.
[AI-162] QECGBERT: A Scalable ECG Foundation Model Based on Self-Supervised Language Learning via VQ-VAE Waveform Tokenization ICASSP2027
链接: https://arxiv.org/abs/2610.33964
作者: Aruna Mohan,Burcu Ozek,Or Zilbershot,Fatemeh Afghah
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Submitted to IEEE ICASSP 2027
Abstract:AI foundation models play a pivotal role in advancing heart disease diagnosis from electrocardiogram (ECG) data, particularly for rare diseases where the lack of sufficient labeled data hinders supervised learning. However, most self-supervised learning methods treat ECGs as generic sequence or image data without exploiting underlying physiological structure. To address this limitation, we propose QECGBERT, a self-supervised ECG language model that explicitly encodes physiological semantics by tokenizing the P, QRS, T, and isoelectric background waveforms. Vector-quantized variational autoencoder (VQ-VAE) tokenization is applied to each 12-lead ECG, with tokens enriched by continuous waveform features to preserve morphology lost during quantization without increasing sequence length. This design enables compact BERT masked language modeling with short sequences, improving scalability and reducing computational cost relative to long-sequence approaches. QECGBERT applies BERT pre-training to the MIMIC-IV dataset containing over 800,000 unlabeled ECG recordings, and fine-tuning on labeled PTB-XL, CPSC2018, and Chapman–Shaoxing–Ningbo datasets for evaluation. The results demonstrate that QECGBERT achieves comparable or superior performance to published benchmarks, offering a scalable, physiologically aware approach to ECG foundation modeling for cardiac diagnosis.
[AI-163] WebViSE: Scaling Visual Design Intelligence from the Open Web for LLM Website Generation
链接: https://arxiv.org/abs/2610.33929
作者: Yifei Sun,Peng Zhao,Le Zhang,Bingjin Chen,Zhengfan Wu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models have made remarkable strides in website development, enabling complete and interactive web experiences to be built from natural-language instructions with far less effort. Yet, non-expert users often struggle to articulate sophisticated visual and interaction designs in natural language. Meanwhile, the open web already contains a vast reservoir of such design knowledge, embodied in mature website implementations. To this end, we introduce Website-derived Visual Skill Extraction (WebViSE), a framework that automatically abstracts these implementations into reusable Visual Skills and builds a structured Skill library. To scale the library, WebViSE organizes its Skills in a multi-parent DAG and jointly evolves Skill descriptions and DAG structure using Skill-conditioned Potential Queries, enabling efficient matching between diverse user instructions and relevant Skills. The auto-evolved structure outperforms non-hierarchical LLM retrieval baselines at only 39.09% of the estimated retrieval cost with Qwen3.8-Max. For any models, WebViSE can be directly injected as plug-and-play guidance, enabling diverse models to produce more visually compelling and aesthetically refined web pages. For open-source models, WebViSE can be used to build a data-synthesis pipeline that internalizes visual design intelligence into model weights. Experiments across multiple web development benchmarks and human evaluation consistently show that WebViSE improves the visual quality and interaction fidelity of artifacts. Overall, our work establishes an automated path for extracting, organizing, and transferring design intelligence from the open web to website-building models. We release all extracted Visual Skills, the associated Skill DAG, and our code at this https URL.
[AI-164] Multitask Interactions in Recurrent Dynamics across 100 Cognitive Tasks
链接: https://arxiv.org/abs/2610.33926
作者: Sílvia García-Hernández,Leema Hamid,Manuel Brenner
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Neurons and Cognition (q-bio.NC)
备注:
Abstract:The brain continually learns tasks with very different computational demands. Recurrent neural networks (RNNs) have been used as dynamical surrogates for studying task solutions, with recent work extending this approach to joint training on multitask batteries. However, it remains unclear how tasks interact as these batteries grow in scale and heterogeneity: under what conditions do they transfer or interfere with one another, and how do these effects relate to task complexity and model structure? We study this question by simultaneously training hierarchical Almost-Linear RNNs (AL-RNNs) on 100 tasks spanning a wide range of cognitive demands. AL-RNNs allow us to vary the amount of nonlinearity in the recurrence, while hierarchical task features provide a low-dimensional interface for task-specific adaptation within shared dynamics. On this task battery, performance improves with both nonlinear capacity and task-feature dimensionality, with the joint hierarchical model outperforming both individually trained models and a single unstructured joint model. Many tasks are individually solvable with linear dynamics, yet cannot be solved together by a shared linear model. Through hierarchical task features, task-specific linear region usage, and task geometry, the trained model provides several related but distinct measures of task similarity that capture how computations are separated and combined within the shared backbone. Across a series of experiments, we relate these representations to individual task complexity, joint solvability, pairwise transfer learning, and leave-one-out adaptation. Together, our experiments characterize task interactions in multitask recurrent learning while demonstrating both the potential and the challenges of reusable dynamical backbones for cognitive computation.
[AI-165] OS-Wizard: Scaling Computer-Using Agents to Specialized Software via Knowledge-Guided Exploration
链接: https://arxiv.org/abs/2610.33907
作者: Zhoumianze Liu,Zichen Ding,Bowen Yang,Zehao Li,Zhenyu Wu,Yibo Zhao,Kanzhi Chen,Yian Wang,Yanan Sun,Qiushi Sun,Yu Qiao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Computer-using agents are increasingly capable of performing complex tasks in the digital world. However, high-quality interaction trajectories remain scarce for specialized desktop applications, where limited application knowledge, cumbersome GUI workflows, and incomplete exploration hinder automated collection. We introduce OS-Wizard, an automated trajectory collection framework. It combines tutorial-guided exploration with a hybrid GUI–Code action space, providing application-specific knowledge and simplifying GUI workflows through direct execution of application-native commands. The framework also checkpoints and revisits pivotal states during tree exploration to explore alternative workflows, complementing linear rollouts for broader coverage. We collect 10.9K trajectories comprising 54.9K state–action samples. Fine-tuning open-source base models on this corpus yields a best-performing model that improves performance on a benchmark of specialized desktop workflows by 58.2 percentage points to 64.1%, surpassing reported human performance of 59.8% and establishing a new state of the art. The fine-tuned model also improves performance on general desktop workflows using GUI actions only, suggesting that the collected supervision benefits general desktop tasks as well. Comparative experiments and ablations demonstrate the efficiency of trajectory collection and the importance of code actions.
[AI-166] ReGE: Stronger Agents through Experience Synthesis
链接: https://arxiv.org/abs/2610.33861
作者: Jiarong Liang,Yubo Wang,Zijie Liu,Yuxuan Zhang,Ping Nie,Wenhu Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Repeated attempts at interactive tasks uncover facts, useful operations, and failures that can inform subsequent execution. We introduce ReGE, a framework for turning this recorded experience into guidance for a fresh attempt. Its main implementation, ReGE-Syn, uses a single coach call to jointly synthesize bounded execution records and available artifacts into self-contained guidance, connecting task-specific findings with proposed actions and relevant pitfalls. A student starts from the task’s initial environment state and adapts this guidance to current observations. Source generation, synthesis, and student execution use the same base model, without parameter updates or a stronger teacher. Across four benchmarks, ReGE improves accuracy by an average of 9.9 percentage points for GPT and 8.8 percentage points for Claude compared with unguided execution. On matched BrowseComp-Plus subsets, ReGE-Syn reduces guidance-construction tokens by 98.2% on GPT and 95.5% on Claude relative to adapted AggAgent, with observed success. Combined construction and execution tokens decrease by 15.7% and 35.4%, respectively, counting input plus output and excluding source generation. BrowseComp-Plus ablations further associate access to process information with fewer student steps and show that joint synthesis produces substantially shorter guidance than Direct history or Independent summaries, with observed success. These results demonstrate how agents can use their own recorded experience to improve subsequent task completion, and suggest a low-cost way to consolidate the execution records of parallel agent systems such as agent swarms.
[AI-167] Numina Fuse: Interactive Autoformalization in Lean 4
链接: https://arxiv.org/abs/2610.33860
作者: Justin Asher,Yichuan Cao,Ruichen Qiu,Pietro Pellegrino,Jia Li
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Logic in Computer Science (cs.LO); Software Engineering (cs.SE)
备注: 16 pages, 10 figures
Abstract:Numina Fuse is an agent-native application for formalizing mathematics in Lean 4. Since Lean can compile while not representing its intended informal meaning, and agents now write Lean faster than people can read it, we built Fuse around human supervision of agent work. A user works with an AI orchestrator on a Lean project and supervises it through views of the blueprint, the dependency graph, and the Lean code, while the backend keeps the project’s derived state and runs agents for hours without the user present. We describe the development of Fuse throughout 2026, as it moved from a prescribed pipeline of agents paced by human review to an orchestrator that dispatches specialists, provers, and parallel children on its own. We designed for this shift, keeping project state, tooling, and interface design separate from orchestration so that agents could take over more of the work as models improved. We report Fuse’s use in a formalization of the three-dimensional Kakeya conjecture and a study of prover models. We close with lessons for the next generation of autoformalization systems. Fuse Desktop, an open-source local version, is available at this https URL.
[AI-168] Adapting DNNs to Embrace Voltage Scaling Induced Errors via Accumulation-Aware Quantization
链接: https://arxiv.org/abs/2610.33824
作者: Alexis Maras,Panagiotis Chaidos,Georgios Alexandris,Dimitrios Soudris,Sotirios Xydis
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: Accepted for publication at the 45rd International Conference on Computer-Aided Design (ICCAD '26), November 08-12, 2026, San Jose, CA, USA
Abstract:The pervasive deployment of Deep Neural Networks (DNNs) across the computing spectrum has brought growing attention to energy efficiency, alongside raw performance. While voltage scaling is a potent knob for power reduction, its application in next-generation nodes is hampered by severe throughput degradation and susceptibility to process variation. This work aims to address these energy-reliability trade-offs by introducing a hardware-software co-design methodology, based on the observation that timing-robust bit ranges within the accumulator of a systolic array vary significantly across different voltage-frequency operating points. At the software level, we propose an Accumulation-Aware Training approach that incorporates this information as a first-order optimization constraint during training. By integrating mixed-precision quantization with unstructured pruning, we effectively constrain the model’s weight distribution to avoid activating the hardware’s sensitized high-rank carry chains. At the architectural level, we complement this with selective MSB-masking and fine-grained clock-gating to minimize unnecessary dynamic switching activity. Furthermore, we employ a layer-wise Dynamic Frequency Scaling strategy to further adapt to the accumulator’s reduced dynamic range, based on the numerical sensitivity of each layer. We experimentally evaluate the proposed solution on a deep scaled 2nm technology node and demonstrate that, when operating at a low-voltage frontier, our co-optimized post-layout architecture delivers substantial energy and energy-delay product gains of 39% and 23% respectively, relative to the nominal voltage implementation, while maintaining accuracy degradation within 1%.
[AI-169] Learning What to Share: Expert Specialization for Time-Series Forecasting
链接: https://arxiv.org/abs/2610.33822
作者: Bashar Khoury,Shadi Omari,Assaf Schuster
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 60 pages, 9 figures
Abstract:A central challenge in multivariate forecasting is deciding what to share across channels and where to specialize. Mixture-of-experts models offer flexible sharing, yet it remains unclear how much of their forecasting gains comes from added capacity or expert allocation. Through controlled comparisons across eleven datasets and three backbones, we show that informative channel assignments improve forecasting beyond matched shared capacity and random allocation. On frozen transformer backbones, persistent channel-specific mixtures outperform input-dependent routing on average. To examine when and how much specialization helps, we introduce GalaxyTS, a controlled synthetic dataset with exact shared and channel-specific forecasting references. It reveals how persistent channel mixtures exploit predictive differences concealed by short input windows and measures gains over matched shared corrections. These insights motivate Persistent Closed-Form Routing (PCR), a plug-and-play method that alternates exact conditional estimation of channel-specific mixture weights with gradient-based training of low-rank residual experts around a frozen pretrained forecaster. Averaged over four forecasting horizons, PCR-K4 achieves dataset-balanced mean squared error reductions of 3.86-14.25% relative to the three frozen backbones. In matched comparisons, PCR outperforms competing specialization methods on average while requiring substantially fewer trainable parameters.
[AI-170] Memory or a Clock? Auditing a Memory Advantage in XLand-MiniGrid with Seeds Maps and Evaluation Noise Apart
链接: https://arxiv.org/abs/2610.33816
作者: Rodrigo Reveron-Fonseca
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 17 pages, 7 figures. Code: this https URL . Data and checkpoints: this https URL
Abstract:On a partially observable XLand-MiniGrid task, an LSTM agent beats a memory-free MLP by 25 percentage points. The gap looks like a clear case of memory. We ask two questions about it. First, how much of a measured gap is noise? Separating training seed, map and evaluation randomness in 60 runs, each re-evaluated three times, we find that about ten paired seeds detect a 10-point gap in final success. The binary “task acquired” verdict, a common summary of learning curves, is 76% evaluation noise and would need about a hundred seeds for the same gap. Second, is the gap memory? On a sealed panel of new maps, opened once, a memory-free MLP of the LSTM’s size comes within 3.3 points of the LSTM when it is given the time step, and gains no measurable benefit from its width without it. With the time step, an MLP a quarter of the LSTM’s size, and the LSTM with its state erased at every step, also come within about 5 points. A linear probe reads the time step from the trained LSTM’s state. On this task, most of the apparent memory advantage is an advantage of knowing the time, which matters because episodes are cut at a fixed limit and the reward shrinks with time. We recommend giving memory-free baselines the time step, and we release the protocol, the event-driven tooling and every checkpoint.
[AI-171] Rectified Language Flow
链接: https://arxiv.org/abs/2610.33811
作者: Baiyu Su,Runlong Liao,Lizhang Chen,Qiang Liu
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Continuous-state flow and diffusion models offer a promising approach to fast, parallel language generation, but matching the generation quality of autoregressive models remains challenging. We present Rectified Language Flow (ReLF), a language model that generates text by transporting noise to token embeddings using rectified flow. ReLF combines several complementary design choices: fixed token embeddings constructed with Hellinger-SVD, self-conditioning guidance trained with cross-entropy, and regularized reflow for few-step generation. We further introduce singular logits, which combine an analytically computed term that drives commitment to a valid token embedding as noise vanishes with a bounded neural residual that learns contextual token preferences. With 64 function evaluations on OpenWebText, ReLF achieves lower generative perplexity than autoregressive baselines at matched diversity across all three model sizes tested, while also outperforming prior continuous flow methods in this setting. After two reflow rounds, ReLF still outperforms autoregressive baselines with 16 function evaluations while generating 1,024-token sequences 15-19 times faster at batch size 1.
[AI-172] Beyond Static Structures: Linking Protein Dynamics to Function via DynaReason -Pro
链接: https://arxiv.org/abs/2610.33804
作者: Zhaokang Liang,Mingze Yuan,Quanzheng Li,Wengong Jin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Intrinsically disordered proteins (IDPs) carry out their functions through heterogeneous conformational ensembles that reflect their dynamics, yet most protein function prediction methods rely on sequence or a single static structure. To this end, we introduce DynaReason-Pro, the first framework to jointly predict GO terms from conformational ensembles and generate natural-language functional rationales. We evaluate DynaReason-Pro on a curated IDP benchmark from the Protein Ensemble Database (PED), with sequence- and structure-cluster-disjoint splits to prevent data leakage. Variability-stratified evaluation shows the clearest gains in the high-conformational-variability regime.
[AI-173] Are We Done with UMLS Injection into Language Models for Medical Question Answering?
链接: https://arxiv.org/abs/2610.33801
作者: Yiran Zeng,Giacomo Frisoni,Lorenzo Molfetta,Gianluca Moro
类目: Artificial Intelligence (cs.AI)
备注: 20 pages, 10 figures, 13 tables. Code and literature annotations: this https URL
Abstract:Maybe not. The Unified Medical Language System (UMLS) has become an oracle for millions of clinicians and health system developers worldwide. Researchers have increasingly used knowledge graphs derived from UMLS to enhance language models’ factual accuracy and reasoning traceability in high-stakes tasks. Medical Question Answering (MQA) is a prime testbed for integration, with benchmarks usually involving professional exercises such as disease diagnosis and drug prescription from patient history. Despite the assumed benefits, empirical evidence shows that UMLS-based graph injection typically yields modest performance gains. This paper reviews existing MQA work, critically analyzing the selected UMLS sources and their usage with language models, and probes each of these choices in isolation. We uncover methodological issues never discussed in prior work, from low UMLS coverage to weak subgraph retrieval pipelines. Based on our findings, we provide actionable recommendations to unlock the untapped potential of UMLS as an external knowledge source.
[AI-174] Unlocking the Specialization Capability of PPG Foundation Models for Wearable Signals through Structured Pruning and Adaptive Fine-Tuning
链接: https://arxiv.org/abs/2610.33788
作者: Huaming Du,Zhixian Xie,Gaoyang Zhen,Gang Kou,Guisong Liu,Yanchao Tan,Wendan Tao,Yuhao Xu,Carl Yang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Photoplethysmography (PPG) plays a central role in wearable health monitoring and clinical decision support. However, existing approaches to universal PPG representation learning are predominantly trained on high-quality large-scale datasets obtained from oximeters in hospital settings and often struggle to adapt to sparse and low-quality real-world datasets, thereby limiting their generalization to clinically relevant tasks on prevalent PPG data such as collected from wearable devices. Surprisingly, even after full fine-tuning, PPG foundation models cannot consistently outperform general-purpose time-series models. Therefore, a key question is how to effectively adapt PPG foundation models to target prediction tasks. Through empirical studies on various PPG foundation models, we find that pretrained models often exhibit inherent sparsity and computational redundancy, suggesting that PPG foundation models have learned to activate task-relevant network substructures for different prediction tasks. To preserve this valuable prior knowledge, we propose a structured pruning-and-fine-tuning framework, SPAFT, which regularizes the adaptive fine-tuning process by constraining it to a more relevant and compact parameter space based on parameter importance and structural redundancy. Extensive experiments across four PPG foundation models, five benchmark datasets, and ten downstream tasks demonstrate that SPAFT achieves win rates of 82.8% and 87.5% on regression and classification tasks, respectively. This prune-then-finetune paradigm often enables PPG foundation models to achieve state-of-the-art performance, laying a promising foundation for the real-world clinical deployment of PPG foundation models and the development of personalized PPG foundation models.
[AI-175] InfiniteWorld: Separating Visual Evidence from Familiar Physics
链接: https://arxiv.org/abs/2610.33777
作者: Tong Zhang,Yuxuan Cheng,Yun Peng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Physical reasoning requires adapting familiar expectations to the dynamics of the observed world. Existing benchmarks establish broad physical capabilities, yet an overall score leaves unclear whether answers follow visual evidence when familiar rules no longer apply. We introduce InfiniteWorld, a benchmark that pairs physical-reasoning questions with simulator videos whose physical parameters are systematically altered. Its generation pipeline connects simulation, scene screening, paired video translation, question construction, and input checks. Five tasks assess law identification, parameter estimation, temporal reasoning, analytic counterfactual comparison, and law induction. The evaluation defines five matched input settings for each question, comprising text-only evaluation and ordered or shuffled observations from simulator and translated videos. We evaluate a diverse set of video-language models under these matched settings. The results expose a persistent preference for familiar physical rules. Frames improve performance for most models, but gains differ across tasks. Strong accuracy on Earth-normal questions coexists with poor recognition of altered dynamics, and Earth-normal answers remain common when the observed physics changes. Task scores and semantic agreement further distinguish matching physical factors from matching their amount, and correct answers from repeated errors. InfiniteWorld makes these behaviors measurable through paired changes in the evidence supplied for the same question, separating visual evidence from familiar physics.
[AI-176] An actively trained neural operator for near-explicit atmospheric chemistry
链接: https://arxiv.org/abs/2610.33771
作者: Zhi-Song Liu,Wenqing Peng,Ammar Kheder,Michael Boy
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Earth-system models simplify atmospheric chemistry into condensed mechanisms, because near-explicit mechanisms such as the Master Chemical Mechanism (MCM) are too costly to solve in every grid cell. Neural emulators could lift this limit, but most advance the chemistry in small time steps and cover only a few species or a reduced mechanism. Here we introduce FNO-MoE, a Fourier neural operator with a mixture of experts that predicts the full near-explicit chemical state over 30 minutes in a single call. The same network also guides its own training, using the disagreement among its experts to select the most informative box-model simulations. At equal cost, this active sampling yields a more accurate emulator than random or evenly spread sampling. Fine-tuning on longer trajectories keeps it stable over multi-hour runs in which models trained on short simulations diverge. The resulting emulator reproduces the evolution of thousands of species with high fidelity, remains accurate for unseen chemical compositions and runs orders of magnitude faster than the numerical solver. These results open a new direction for emulating complex atmospheric systems: neural operators that predict the full chemical state over long windows and choose their own training data. Training under changing meteorology is the next step towards bringing near-explicit chemistry into Earth-system models.
[AI-177] CFL-aware edge design for graph neural solvers of hyperbolic transport
链接: https://arxiv.org/abs/2610.33769
作者: Jinyoung Jeong,Joseph B. Choi,Xinlun Cheng,H. S. Udaykumar,Sanghun Choi,Stephen S. Baek
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Fluid Dynamics (physics.flu-dyn)
备注:
Abstract:Hyperbolic partial differential equations (PDEs) transport information at finite speed along characteristics, often producing shocks and discontinuities that are challenging to capture numerically. Graph neural networks (GNNs) offer learned update rules, but neighbor aggregation alone does not tell a GNN how far information must travel and along which edge. As a result, when a GNN model takes a time step larger than what was used in training, shock fronts smear and oscillations grow. In this work, we ask whether making the scale of transport explicit as an edge-local Courant-Friedrichs-Lewy (CFL) signal can enhance GNNs’ ability to model hyperbolic PDEs. Holding the architecture fixed, we vary only the information given to each edge, from a raw edge form to forms that progressively expose the edge-local transport interaction through direction, length, and wave speed. Because all of these are computed from the same ingredients, what changes is not the amount of information but its form. The results show a consistent pattern, in which feature count is not the limiting factor, and appending CFL directly to a raw edge form produces almost no benefit. The CFL conditioning proves most effective when the edge information already exposes the transport interaction, reducing final-step error by 10.4-14.4%, and the improvement widens as the step size grows larger. Notably, this improvement arises not from smearing the solution but from suppressing oscillations near shocks. These results point to a specific design principle that CFL conditioning is most effective when the edge representation already separates the direction of interaction from spatial scale, so that the CFL value has a well-defined quantity to modulate.
[AI-178] Local Entropy Fluctuations for Black-Box LLM -Generated Text Detection
链接: https://arxiv.org/abs/2610.33767
作者: Xinpeng Wang,Bruce Rushing,Zhizhen Chen,Zhe Zeng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Zero-shot detection of machine-generated text is vital for mitigating the risks of large language models (LLMs) misuse. While the state-of-the-art dual-model detectors and resampling-based detectors outperform single-model statistical detectors on unprompted continuations, they suffer a catastrophic breakdown on instruction-conditioned text in the black-box setting, where the source model is unknown. We analyze this failure and trace it to poor prompt conditioning for resampling-based methods and diminished cross-model discrepancy for dual-model detectors. Motivated by recent studies that human writing maintains a steady, uniform information flow, whereas machine-generated text exhibits sharp, localized entropy fluctuations, we propose a new metric, the Relative Peak Prominence Rate (RelPPR) to capture this structural contrast along the entropy trajectory and further leverage it as a signal for LLM-generated text detection. Extensive evaluations across diverse model families and datasets demonstrate that RelPPR achieves state-of-the-art black-box detection performance and remains the most reliable detector under various practical situations. Our code is available at this https URL.
[AI-179] All the Facts Still the Wrong Answer: Integration Gaps in Vision-Language Models
链接: https://arxiv.org/abs/2610.33764
作者: Yuyang Dai,Yuxia Wang,Zhuohan Xie
类目: Artificial Intelligence (cs.AI)
备注: 53 pages
Abstract:Multimodal reasoning requires models to combine multiple pieces of evidence into decisions that satisfy all relevant constraints. Evaluations centered on final-answer accuracy leave unclear whether failures arise from reading individual facts or using them together. We study this question through the multi-evidence integration gap, linking local fact recovery to global decision validity on controlled subset-selection and path-finding tasks. Our evaluation combines conditional performance given correct local answers with same-grid local queries, text-oracle inputs, retrieval controls, and constraint ablations. We further examine hidden-state probes and activation replacement across network depth, token positions, and model families. Across the evaluated models, accurate local recovery coexists with substantial global failure. For example, Gemma-3-27B recovers all evaluated isolated item facts correctly but achieves only 19.75% global validity on eight-item visual instances. Probes decode information about solution membership from global-prompt states, while corresponding clean-state replacement recovers failed decisions over model-dependent depth ranges. These findings show that recovering facts and representing solution-related information are insufficient for reliable joint decisions. They motivate evaluations and model improvements that explicitly target how available evidence is combined, alongside advances in perception and model scale.
[AI-180] POD-RoPE: Rotary Position Encoding over Physical Observation Distributions for Embodied AI
链接: https://arxiv.org/abs/2610.33745
作者: Xiaodi Huang,Zilin Wang,Jinpeng Liu,Pengchuan Xiao,Zhenlei Shao,Zhang Zhang,Shuxiang Lu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 21 pages, 5 figures
Abstract:Embodied perception requires positional representations that capture both physical geometry and the spatial and temporal extent of observations. We introduce POD-RoPE (Physical Observation Distribution RoPE), which represents token observation footprints as distributions over camera centers, viewing directions, and sampling times. Under independent sampling, expected rotations encode relative relationships between distributions; when each distribution collapses to a single address, the construction recovers pointwise RoPE. The same rotation statistics support Q/K matching and coherence-dependent V/O transformations, allowing observation geometry to guide both information selection and content aggregation. Per-token pre-aggregation avoids increasing sequence length or processing pairs of observation nodes. We evaluate POD-RoPE across four robot policy architectures, simulation and real-world manipulation, and complementary reconstruction, depth-estimation, and novel-view-synthesis tasks. Integrating POD-RoPE into Fast-WAM improves mean success from 91.83% to 92.88% across all 50 RoboTwin 2.0 tasks, and from 32.5% to 52.5% on two real-robot tasks on Sharpa North. Controlled reconstruction ablations support the contributions of distribution encoding and V/O transformations. In H20 model-block benchmarks, POD-RoPE achieves median forward-plus-backward speedups of 1.91\times over P-RoPE and 5.34\times over RayRoPE, with a median latency overhead of 17.4% over Standard RoPE. These results support physical observation distributions as an effective and computationally efficient basis for geometry-aware attention.
[AI-181] A Distribution-Matching Perspective on LLM Training Algorithms
链接: https://arxiv.org/abs/2610.33740
作者: Yanxing Huang,Bingzhang Yan,Zihan Tang,Wenzhe Li,Peng Li,Yang Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:In this work, we propose a unified framework for training algorithms across different stages of modern large language model development, based on the perspective of distribution matching. We show that a broad class of language model training algorithms can be interpreted as minimizing a certain divergence between the current model and a target distribution, and that characterizing the properties of this target distribution provides substantial insight into the behavior of the corresponding training algorithms. By analyzing specific forms of the target distribution, we establish that On-Policy Distillation (OPD) enjoys an advantage over classical Knowledge Distillation (KD) when the current model distribution is close to the target distribution, while exhibiting a disadvantage when the two distributions are far apart. Furthermore, when the target distribution itself depends on the model being optimized, we show that the accumulation of estimation variance can induce a deterministic degradation of the target distribution. These results provide a theoretical explanation, at least in part, for the long-term behavior of existing Reinforcement Learning (RL) algorithms. With this intuition we further designed Conditional KL (CKL) regularization for mitigating the degradation of the target distribution. CKL can effectively preserve mode diversity while relaxing the constraint imposed by classical KL regularization on reward improvements. We believe this theoretical perspective can bring some new understanding of language model training algorithms and provide useful guidance for future research.
[AI-182] rust Mechanisms for Offline Generative AI at the Edge: A Connectivity-Regime Framework and an Empirical Study on a Raspberry Pi 5
链接: https://arxiv.org/abs/2610.33732
作者: Georgios Feretzakis,Alexandros Papaspyridis
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 25 pages, 4 figures, 6 tables. Code and data: this https URL
Abstract:Generative AI and large language models (LLMs) are moving into the Internet of Things, yet most deployments assume cloud connectivity, and many IoT settings have intermittent, restricted, or no connectivity. We propose a connectivity-regime framework (C0-C4), a taxonomy of inference-time trust mechanisms, and signed, versioned knowledge packs as the update channel for disconnected regimes, and evaluate them on a Raspberry Pi 5 running fully local retrieval-augmented generation under network isolation, with three Qwen2.5 model sizes, three context sizes, and 300 SQuAD 2.0 questions. An empirically calibrated retrieval-score gate detects knowledge gaps almost perfectly (AUROC 0.97) but in-topic unanswerable questions poorly (0.68). A deterministic output validator converts malformed outputs into explicit abstentions with no observed loss of correct answers, while the NLI-based verifier we tested adds little; for the 7B model, 96% of substantive answers to unanswerable-labelled questions match normalized spans of the retrieved passages. A 3B model achieves the same gold-answer containment rate as a 7B model at less than half the latency and estimated energy, although the 7B model far more often exceeds a token-F1 threshold of 0.5; five retrieved passages lower correctness from 75% to 52% compared with three. Verifying a signed pack takes about 4.5 ms versus about 29 min for embedding its passages on the device; signature and hash checks alone accept six of nine tested update attacks that layered verification rejects. Yet a validly signed poisoned pack passes every pack check: on 30 previously correct questions, correct answers drop to zero and 14 return the injected answer. None of the mechanisms we evaluated establishes the truth of the evidence. We release code and data.
[AI-183] Strategic Calibration in LLM Multi-Agent Systems: Audited Pivot Scoring for Truthful Elicitation
链接: https://arxiv.org/abs/2610.33728
作者: Jinhao Zhou,Zhaohan Zhang,Xijian Xu,Jun Wu
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI)
备注: 22 pages, 12 figures, including appendices
Abstract:In large language model (LLM)-based multi-agent systems, reported confidence increasingly determines the allocation of scarce downstream attention, computation, and decision influence. This makes reporting strategic: an agent may overstate the relevance of its contribution even when its underlying belief is well calibrated. Existing calibration methods largely assume passive reporting and therefore do not address strategic incentives created when confidence determines downstream allocation. We formulate strategic calibration as a reporting-layer incentive problem and propose the Audited Pivot Scoring Mechanism (APSM). APSM offsets platform-controlled allocation benefits, uses inverse-propensity-weighted sparse audits, charges the Clarke pivot externality, and applies a strictly proper score. Under the stated model assumptions, truthful reporting is the unique expected-utility best response. We also characterize approximate execution and the audit-variance trade-off. Calculator-free experiments show that LLM agents inflate confidence under Top- K incentives. With a shared agent-local utility interface, LLM agents reduce distortion under both Compensated Proper Scoring (CPS) and APSM. CPS achieves higher exact allocation match, whereas APSM yields lower Brier loss and numerically fewer cutoff crossings. Experiments further show that APSM enlarges the utility margin against allocation-changing deviations by 1.93\times and improves robustness under approximate execution. These results support incentive-aware elicitation as a distinct layer for reliable multi-agent coordination.
[AI-184] Human-Feedback-Guided Multi-Agent Inverse Reinforcement Learning for Interactive Driving Behavior Modeling
链接: https://arxiv.org/abs/2610.33726
作者: Longhao Yan,Jinhao Liang,Kaidi Yang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Modeling human driving behavior is fundamental to understanding driver decision-making and supporting autonomous vehicles (AVs) in mixed traffic. Reward learning offers a principled way to infer the latent objectives behind driving behaviors, yet most existing methods are formulated in single-agent settings and treat surrounding vehicles as dynamic obstacles. Directly extending reward learning to multi-agent scenarios is difficult because the joint action space grows exponentially with the number of interacting vehicles. In addition, existing approaches mainly rely on physical demonstrations and rarely exploit human feedback during reward recovery. To address these limitations, this paper proposes a Human-Feedback-Guided Multi-Agent Inverse Reinforcement Learning (HFG-MAIRL) framework for interactive driving behavior modeling. First, we develop a multi-agent maximum-entropy inverse reinforcement learning module that explicitly models each vehicle and its directly interacting neighbors in a localized Markov game, while using aggregated traffic states to capture the influence of the broader traffic environment. The resulting coupled multi-vehicle responses are approximated through finite marginal-response iterations over factorized vehicle-level policies, avoiding explicit enumeration of the full joint action space. Second, we incorporate human feedback as ordinal preference supervision through a Bradley-Terry likelihood. In our experiments, driver emotion labels are used as one instantiation of such feedback and converted into pairwise ordinal preference constraints. Experiments on the AIDE dataset show that HFG-MAIRL outperforms representative IRL baselines in decision modeling and produces interpretable feedback-aligned reward patterns.
[AI-185] PACE: Research Proving Agent for Adaptive Conjecture Exploration
链接: https://arxiv.org/abs/2610.33690
作者: Shu Miao,Feiming Wang,Wentao Long,Chenyi Li,Kun Yuan,Zaiwen Wen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Many mathematical research agents focus on solving problems with fixed statements. However, conjecture research, especially in applied mathematics, may require revising candidate questions or conclusions as investigation proceeds, coupling problem formulation with proof development. We present PACE, a research-proof agent that supports this process through a structured research loop that selects mathematical objectives and develops or challenges candidate mechanisms. The whole prover constructs proofs from these mechanisms, and an independent verifier checks them against the original target. Evidence-tracking memory stores assumptions and dependencies arising during research and marks evidence for revalidation when a candidate or condition changes. This memory supports task-specific context composition that keeps checked premises, exploratory material and unresolved obligations distinct. To evaluate proof construction and conjecture research, we introduce ResearchProofBench, an applied-mathematics benchmark focused on optimization, comprising 51 known-proof tasks with 331 hidden checkpoints under historical abstract access and six open-problem cases. The known-proof tasks assess reconstruction of published results, while the open-problem cases assess mathematical investigation and proof development. On the 51 known-proof tasks, PACE achieves an aggregate Proof Score of 75.16 and completes 20 proofs. In conjecture research, PACE develops conditional results, with its submissions receiving higher review scores than the compared baselines.
[AI-186] More: Mode-Optimized Reasoning with Self-Adaptive Effort for Self-Evolving Agents
链接: https://arxiv.org/abs/2610.33684
作者: Dianxing Shi,Dingjie Fu,Tianyu Lei,Wangjie Qiu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Self-evolving training improves the performance of agents on difficult tasks through verifiable feedback, but it often applies costly deliberation uniformly across tasks. To address this problem, we propose Mode-Optimized Reasoning with Self-Adaptive Effort (MORE), a self-evolving training framework that adaptively selects both reasoning form and computational effort for each task. Specifically, MORE follows a two-stage training paradigm: (i) \emphStage 1 establishes a compact prior over direct answering, short and long chain-of-thought, and code-based reasoning, and (ii) \emphStage 2 self-evolves this prior with Mode-Adaptive and Test-time Effort PPO (MATE-PPO) on generated, verifiable task streams without curated data. By learning mode selection and token expenditure jointly, MORE allocates concise reasoning to tasks that permit it while retaining extended deliberation where it is useful. Across coding, mathematics, and general-reasoning benchmarks, MORE achieves consistent performance gains on Base and Coder backbones at model scales from 3B to 14B. During inference, MORE demonstrates higher accuracy and better efficiency, with markedly shorter response lengths and reduced inference time.
[AI-187] Extraction meets Distillation - cryptanalytic techniques to recover parameters from a Deep Neural Network
链接: https://arxiv.org/abs/2610.33671
作者: Biprarshi Biswas,Arya Anshuman Ojha,Yubaraj Das,Devansh Khosla,Kamran Rais,Saankhya Samanta,Subhamoy Maitra,Pantelimon Stanica
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注: 39 pages, no figure
Abstract:We study fully connected neural-network reconstruction by combining cryptanalytic weight-direction extraction with learning from predicted labels. We adapt a ReLU extraction implementation to Leaky ReLU, compare sign-search methods, and estimate biases and output layers. We prove conditional bounds for direction recovery with approximate prefixes, preservation of local distance votes, prediction changes from aligned parameter errors, and sign selection with independent scoring data. We also specify exact row restoration and distinguish its guarantee from that of gradient masking. The experiments evaluate assisted extraction, with privileged access to gradients, hidden activations, and victim parameters, followed by label-based reconstruction. Across eight search seeds on the largest synthetic architecture, Leaky ReLU recovers 51–57 neurons in the deepest hidden layer, while ReLU recovers none. This advantage does not hold uniformly across architectures and layers. Removing victim-weight access from the sign walk gives less accurate signs, despite high final prediction agreement. Across 40 seeded runs on the two smallest synthetic victims, agreement on 5,000 inputs excluded from fitting and selection ranges from 0.9890 to 0.9992 . On two CIFAR-10 victims, the reconstructed models achieve 56.40% and 58.29% agreement and exceed the distillation baseline on three functional metrics. We report label requests separately from privileged access and explain why prediction agreement does not identify network parameters.
[AI-188] SWE-Ladder: Learning to Localize Reproduce and Patch via Progressive Privileged Self-Distillation
链接: https://arxiv.org/abs/2610.33667
作者: Xudong Cai,Zhengxi Lu,Yuchen Yan,Jian Shao,Jun Xiao,Yueting Zhuang,Yongliang Shen
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 18 pages, 11 figures. Code: this https URL
Abstract:Resolving a repository-level issue typically requires a software-engineering (SWE) agent to localize the faulty code, reproduce the failure, and patch the code. Reinforcement learning trains such agents from a single test result on the final patch, while on-policy self-distillation (OPSD) adds token-level supervision from a teacher with privileged evidence, exposing the full reference patch at every step. Neither provides stage-specific guidance for localization or reproduction. We propose SWE-Ladder, which turns the verified repair of each training instance into stage-specific evidence: the changed files and lines for localization, the fail-to-pass tests for reproduction, and the diff for patching. A rule-based router assigns each step of an end-to-end rollout to a stage based on its tool call, a frozen teacher sees only the matching evidence, and the stages are distilled in repair order while GRPO optimizes the full trajectory. With Qwen3-4B-Instruct-2507 on SWE-bench Verified under the R2E-Gym scaffold, SWE-Ladder improves pass@1 from 9.6% to 16.4%, compared with 10.4% for GRPO and 12.4% for OPSD with the full patch. Under the OpenHands scaffold, it improves pass@1 from 13.0% to 17.6%. Reversing the stage order under R2E-Gym reduces pass@1 to 11.4%.
[AI-189] Power-of-Two Leaky Integrate-and-Fire Neurons for Event-Driven Vision-Language-Action Inference
链接: https://arxiv.org/abs/2610.33658
作者: Rui Liu,He Lv,Ziyao Zhang,Rui Ma,Rongxiang Nie
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-language-action (VLA) models have demonstrated strong capabilities in robotic manipulation, yet their reliance on dense, continuous-valued computation presents a fundamental challenge for energy-constrained deployment. Inspired by the sparse, event-driven signaling and temporal integration mechanisms of biological neurons, a promising alternative is to introduce spiking computation into VLA models. However, enabling such computation without restructuring pretrained policies or compromising closed-loop control remains an open challenge. To address this challenge, we introduce the Power-of-Two Leaky Integrate-and-Fire (PoT-LIF) neuron, which emits signed spikes with power-of-two amplitudes. By combining magnitude-coded spike events with residual-preserving membrane dynamics, PoT-LIF enables continuous information to be progressively represented through discrete spikes. The Heartbeat formulation further localizes this process within weighted projections, bridging biologically inspired spiking computation and pretrained VLA architectures without modifying their external inference interfaces. We integrate PoT-LIF into the vision projection, vision-language projection, and action projection of \pi_0.5 and evaluate both operating modes over 2,000 paired closed-loop LIBERO episodes per configuration. Standard-mode PoT-LIF VLA achieves success rates of 96.90% and 96.35% at T = 1 and T = 2 , respectively, while Heartbeat mode achieves 97.25% and 96.90% at T = 2 and T = 4 , compared with 96.70% for the original continuous-activation \pi_0.5 baseline. At T = 1 with a 60.01% firing rate, PoT-LIF uses approximately 1/7.67 of the dense-reference arithmetic energy, corresponding to a modeled reduction of 86.95%.
[AI-190] Physics-Informed Graph-Based Safe RL with Constraint-Specific Critics for Community Battery Coordination in Distribution Network
链接: https://arxiv.org/abs/2610.33654
作者: Chenghao Huang,Yue Yang,Hao Wang
类目: Artificial Intelligence (cs.AI); Systems and Control (eess.SY); Optimization and Control (math.OC)
备注:
Abstract:High penetrations of distributed energy resources (DERs) and fast-varying electric vehicle (EV) charging intensify voltage and line-limit violations in distribution networks (DNs), making safe, low-cost operation increasingly challenging. This paper proposes a physics-informed safe deep reinforcement learning (DRL) framework for coordinating community battery energy storage systems to mitigate constraint violations while minimizing operating cost. To represent nonlocal spatiotemporal interactions, we develop a dual-level physics-informed graph representation in which line impedance and capacity shape spatial information propagation, while power-balance, voltage-profile, and flow-reconstruction losses regularize the latent state toward control-relevant physical consistency. Building on this representation, we develop Lagrangian Dedicated-Critic PPO (LDCPPO), which implements constraint-specific value decomposition: independent reward and cost critics preserve heterogeneous voltage- and line-loading-related learning signals and combine their advantages only during the Lagrangian policy update. Case studies on DNs with high rooftop photovoltaic penetration and EV charging demonstrate that the proposed method reduces the key voltage-violation degree by up to 80.5% compared with the tested baselines, while also achieving the lowest voltage-violation count and line-overload degree with competitive operating cost.
[AI-191] AD: Distilling Vision-Language-Action Policies from Teacher Trajectories Alone
链接: https://arxiv.org/abs/2610.33650
作者: Yun Zhu,Da Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Online distillation of vision-language-action (VLA) policies commonly relies on teacher action scores that an inference-only interface may not expose. Trajectory-only Adversarial Distillation (\gad) instead turns a fixed bank of teacher joint-position features into adaptive feedback: a discriminator compares feature marginals from teacher and student rollouts and supplies rewards for on-policy learning. The student interacts with the environment without online teacher queries or environment task rewards. On ManiSkill \textttPutOnPlateInScene25Main, \gad improves OpenVLA-OFT (7B) from 26.2% to 76.6% \successonce ( +50.4 percentage points) and from 20.3% to 58.6% \successend. It trails white-box on-policy distillation (OPD) at 81.6% / 59.4% by 5.0 / 0.8 percentage points ( n=256 evaluation trajectories per checkpoint). The tested four-epoch continuation achieves the highest \gad result without additional teacher data. This best run uses 5\times OPD’s student interaction budget and 12.5\times its complete-rollout batch-passes. The comparison demonstrates online VLA improvement without teacher scoring while quantifying unequal interaction and optimization budgets.
[AI-192] AlphaIdeaBench: Benchmarking Automated Research Idea Generation under Feasibility Constraints with Peer Review Alignment
链接: https://arxiv.org/abs/2610.33644
作者: Shunchao Zhou,Chang Yang,Jingrui Wang,Shuxin Zhuang,Rundong Wang,Runsheng Yu,Xinrun Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Research ideas are always proposed under resource constraints: a plan must fit the available compute, API credit, and time. Yet evaluations of research-ideation agents mostly score unconstrained novelty or overlap with a published paper, and their LLM judges are rarely checked against peer review. We introduce AlphaIdeaBench, a benchmark of 100 ideation tasks (50 workshop calls for papers and 50 open problems stated in surveys), each paired with an explicit resource budget and a shared proposal schema. An LLM judge scores each proposal on 12 dimensions covering validity, novelty, and feasibility. To validate the judge, we compare its scores on four novelty axes (problem, method, scenario, and insight) with scores derived from reviewer comments on 600 papers from ICLR 2026, NeurIPS 2025, and ICML 2025. For problem and method novelty, Gemini 3.7 Flash is within one point of the reviewer-derived score on 94-97% of papers; agreement on scenario novelty is relatively lower. We evaluate eight published harnesses on a shared backbone and, within one harness, eight backbone LLMs. CoI-Agent obtains the highest overall score, narrowly ahead of SciMON, which leads on method and insight novelty. All harnesses score low on scenario novelty and rarely change the problem setting. The backbone largely determines which open problem a proposal targets, whereas the harness determines whether a proposal addresses one problem or combines several. Weaker backbones such as Llama 4 Scout fail to carry out that harness’s strategy and score low on budget feasibility. Finally, when only the budget is varied on three topics, a larger budget changes the experimental setting on compute-intensive topics but only adds tests on an inexpensive one. We release the tasks, rubrics, and scores to foster standardized, reproducible benchmarking of future research-ideation agents under constraints.
[AI-193] Build or Adopt? Custom-Built vs. General-Purpose LLM Agent Harnesses for Automotive GUI Testing
链接: https://arxiv.org/abs/2610.33636
作者: Arsham Gholamzadeh Khoee,Ruyu Fu,Jiaqi Wei,Ruikun Deng,Robert Feldt,Dhasarathy Parthasarathy,Yinan Yu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:As general-purpose LLM agent harnesses mature, organizations must decide whether to continue developing in-house systems or adopt existing alternatives. We study this decision at Volvo Group for GUI testing of Android-based in-vehicle infotainment (IVI) systems. We compare a production, custom-built harness with two general-purpose harnesses, Claude Code and OpenCode, keeping the tools, accessed through the Model Context Protocol (MCP), and the models the same. We evaluate the harnesses with up to four models on Volvo’s IVI test cases and on open-ended tasks in public apps (Google Maps, this http URL), and interview ten practitioners: developers, the product owner, and testers. On Volvo’s stepwise IVI tests, general-purpose harnesses match the custom-built harness in success but vary more across runs. Their inference cost per task is 2.6-10.7 times higher, yet the engineering effort they save outweighs this: for a nightly IVI regression suite, total monthly cost, including engineering effort and inference, is 3-66% lower. On open-ended public-app tasks, they succeed more often, mainly because they recover better from incorrect actions. Interviews suggest that external advances can turn an in-house harness into technical debt even without deterioration. Adopting a general-purpose harness shifts maintenance from the agent loop to tools and instructions, but gives teams less control over what the model receives. Regardless of the harness, testers want to inspect the agent’s actions and ensure that it follows the prescribed test scenario.
[AI-194] ProbeID: Verifying MCP Tool Identity for LLM Agents under Registry Drift
链接: https://arxiv.org/abs/2610.33633
作者: Guanxing Chen,Sining Zhang,Haohuai He,Qiujie Lv,Yu-An Huang,Zhi-An Huang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Agents select tools by copyable descriptions, which cannot establish which implementation runs. ProbeID compares canonicalized responses to schema-valid probes with a trusted admission reference and abstains on ambiguity, so a squatter must copy probe-observable behavior. With the gold present, independent probes and a fixed margin, standard concentration arguments give a shared probe budget logarithmic in the candidate count and bound the joint probability of a wrong decision. On cached MCP-IdentityDrift cases, strict choice reaches (0.956) hit@1 while description routers fall near chance, but simpler exact comparisons tie it, MMD better tolerates heavy value noise, and a 32B behavioral judge is similarly accurate at higher cost. Random probes from agent banks and a prospective held-out study of 38 tools test theory-set thresholds. On the agent banks, equality refuses every reachable replacement from 64 probes and the bounded distance from 512, and sharper post hoc gates halve both. Two tools whose drift no bank probe reaches always pass. In an MCP-Bench agent, verification cuts tainted tasks from 15 to 9 of 16 at unchanged task success. Over eleven weeks, 7 of 59 real tools changed responses despite version pins. Verification depends on reference integrity, probe coverage, and response noise.
[AI-195] RAQ-MLA: Role-Aware Quantization for Multi-Head Latent Attention Caches
链接: https://arxiv.org/abs/2610.33625
作者: Tianqi Chen,Xunchao Li,Linchuan Xie,Guanghua Yu,Jianchen Zhu,Peisong Wang,Jian Cheng
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Multi-Head Latent Attention (MLA) has been adopted by recent frontier large language models. Instead of per-head keys and values, MLA caches a compressed KV latent that is up-projected into keys and values at decode time, together with a narrow positional key that carries position information into the attention score. Existing KV-cache quantization methods are designed around the keys and values of standard attention and do not account for what MLA actually caches, which costs accuracy at low bit widths. We analyze the MLA cache both in activation distribution and in how quantization error propagates to the attention output, and show that the two cached tensors play different roles: the latent mainly supplies the values, whereas the positional key dominates the attention score. We therefore propose RAQ-MLA (Role-Aware Quantization for Multi-Head Latent Attention Caches), which fits an invertible transform to each cached tensor against the error that tensor delivers to the attention output: attention-weighted value error for the latent and score perturbation for the positional key. Without training, RAQ-MLA quantizes the full MLA cache to about 2.5 bits per element, a 6.5x increase in theoretical KV-cache capacity, while remaining nearly lossless on reasoning and coding tasks across four MLA models.
[AI-196] Graph-dependent pooling: adaptive granularity in hierarchical graph neural networks
链接: https://arxiv.org/abs/2610.33623
作者: Bohan Yoon,Hyejin Jang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Graph Neural Networks are widely used to learn structural representations of relational data. Hierarchical graph pooling learns multi-level representations by progressively coarsening graphs; however, prior methods rely on predefined cluster numbers or fixed pooling ratios, limiting adaptability to individual graphs. We propose Graph-Dependent Pooling (GDPool), which dynamically determines pooling granularity for each graph. We analyze optimization conflicts among layer-wise objectives through gradient alignment, showing that independently supervised pooling levels can send conflicting gradients to shared parameters. GDPool uses self-attention to select graph-specific anchors representing structurally important nodes and constructs coarse graphs through soft node assignment. Experiments on metal-organic framework and molecular graph benchmarks show that GDPool consistently achieves the best predictive performance across all benchmarks. Evaluations using modularity and the effective number of clusters show that GDPool achieves consistently higher modularity than existing methods while the number of active clusters decreases monotonically with depth. Sensitivity analyses confirm stable behavior across hyperparameter and architectural choices.
[AI-197] When Generated Information Returns: Error Persistence in LLM -Mediated Information Ecosystems
链接: https://arxiv.org/abs/2610.33617
作者: Yi Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models (LLMs) increasingly produce online articles from material that later becomes input to other models. This circulation connects individual responses to the information available for subsequent generation. We study a consequential distinction: an article can avoid endorsing a false claim while preserving the concrete details that enable its reproduction. Our framework connects claim content, expressed stance, and downstream reuse to explain selective error persistence under a continuing supply of accurate and inaccurate information. We implement a controlled community writing process that varies whether published articles return as inputs, matching external supply, input capacity, and generation volume. Across twelve factual targets and four production models, recirculation increases the presence of complete false claims by 7.29 percentage points at the final round while reducing explicit endorsement by 8.33 points. Effects differ across facts and models. Paired interventions then abstract eligible error details within fixed histories while preserving correct content and the surrounding controversy. Error reproduction remains largely unchanged across the evaluation targets, while a development quotation target exhibits local reductions. These findings connect persistent errors to the information available throughout a writing process and motivate evaluation of both what models affirm and what their publications supply to future writers.
[AI-198] Data Governance for Generative and Agent ic AI : Technology Law Economics and Security
链接: https://arxiv.org/abs/2610.33605
作者: Narayana Darapaneni,Padmaja K,Gnana Chaithra D
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:This book develops a unified governance framework for systems that retrieve, generate, infer, remember and act on information from large language models and retrieval augmented pipelines to autonomous multi agent systems. Eleven chapters and eight appendices integrate three disciplinary lenses-technology, law and economics-across the full lifecycle: data quality, lineage and provenance; global data protection law and intellectual property; the economics of data governance; digital trust, cybersecurity and zero-trust architecture; ethics and responsible AI; governing generative AI pipelines; governing agentic AI and multi-agent systems; data governance in elections and the public sphere; AI regulation across six jurisdictions; and the design of a governance control plane for 2030. Each chapter is anchored in Indian law and institutions while maintaining a comparative, cross jurisdictional perspective.
[AI-199] Counterfactual Model Testing Under Distribution Shift
链接: https://arxiv.org/abs/2610.33594
作者: Joran Leest,Ilias Gerostathopoulos,Claudia Raibulet,Patricia Lago
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Deployed machine learning (ML) models can degrade as the distribution of their input data shifts. Such shifts can result from changes in the population or from data integrity issues in the pipeline that materializes model features, two sources that operate through different mechanisms. Because labels are often absent or delayed, it is common practice to track shifts in the data instead, but a detected shift does not indicate how it affects the model. We formulate this problem as online model testing under distribution shift. Testing requires reproducing the shift on a controlled baseline with known labels. Since the same shift can come from either source, doing so requires knowing which mechanism changed. We propose a methodology that reproduces observed changes in both mechanisms to estimate current performance and the potential gain from repairing observed data integrity issues. Experiments on real-world datasets with simulated faults show that separating the two sources generally improves both estimates, supporting this formulation. A churn prediction case illustrates how these tests can support monitoring and repair prioritization
[AI-200] Auditing evaluation artifacts and data leakage in large language model systems for rare disease diagnosis
链接: https://arxiv.org/abs/2610.33592
作者: Jian Yang,Peilin Liu,Xiaohan Li,Songda Li,Yu Cao,Yuqing Feng,Haomin Li
类目: Artificial Intelligence (cs.AI)
备注: 18 page, 3 figures
Abstract:Augmented large language model (LLM) systems have reported high benchmark accuracies in rare disease diagnosis, yet their real-world clinical utility remains unproven. A recent meta-analysis flagged augmented LLM diagnostic evaluations as having a high risk of bias due to suspected data leakage; however, the precise impact of these vulnerabilities on model inference has yet to be fully quantified. In this study, we present a comprehensive empirical audit to quantify evaluation artifacts across prevailing AI paradigms, using a repository of over 60,000 cases from diverse public and independent clinical cohorts. By reconstructing a state-of-the-art agentic framework, we demonstrate that candidate-space contamination artificially inflates performance: enforcing strict cross-dataset isolation resulted in a marked performance decline, with the cross-dataset mean Recall@1 decreasing from 54.3% (under conventional one-vs-rest evaluation) to 40.3% (Hedges’ g = 0.94). Similar spatial overfitting in supervised classifiers caused diagnostic accuracy to drop from an internal mean of 42.9% to 16.7% across external cohorts (Cohen’s d = 2.01). Furthermore, we quantify how LLMs infer diagnoses by exploiting retrospective therapeutic cues in clinical narratives, and how temporal misalignment in foundational knowledge bases leads to performance fluctuations. We conclude that prevailing evaluation protocols for augmented LLM systems inadvertently reflect data leakage rather than genuine medical reasoning. To address these systematic artifacts, we propose a rigorous evaluation framework - coupling strict data-isolation protocols with standardized benchmarking safeguards - to ensure future diagnostic AI for rare diseases is evaluated on actual clinical robustness.
[AI-201] Replicate What Varies: Resource Sharing for Domain-Randomized Embodied Simulation
链接: https://arxiv.org/abs/2610.33582
作者: Zixu Hao,Weijun Wang,Xiangyu Li,Liang Mi,Hao Wu,Yunxin Liu,Ting Cao
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Simulation provides a low-cost sandbox where embodied agents can attempt tasks, observe outcomes, and learn from experience. Running many diverse simulation environments concurrently can consume substantial CPU and GPU memory because independently constructed scenes duplicate resources that remain unchanged across instances. We present Patchwork, a resource-sharing system for domain-randomized simulation, built around the representation E=T+\Delta : a shared template contains common resources, and a patch records the mutations needed to construct a target environment. Patchwork separates randomization sampling from replay and selects reusable templates together with patches that preserve each target environment’s intended configuration. Its layered runtime shares compiled model structure and immutable assets while keeping mutable parameters and simulation state private to each instance. We implement Patchwork on robosuite and MuJoCo and evaluate it on LIBERO-plus and RoboCasa365 workloads. At 32 replicas, shared-template execution reduces CPU RSS by 88-92% relative to process-per-environment execution. Controlled experiments show how sharing shifts memory growth toward template diversity and unique asset content rather than instance count, within the supported sharing boundaries.
[AI-202] WorldFailBench: Do World Models Understand Policy Failures?
链接: https://arxiv.org/abs/2610.33565
作者: Zheng Zhang,Xiaoyang Tang,Shuaiqi Cheng,Jiakai Huang,Zhongbo Wu,Ziyi Zhang,Yuan Si,Zeyu Liu,Weibo Gao,Shunyu Liu,Shan You,Fei Wang,Xiaogang Wang,Dacheng Tao
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Robot policy evaluation informs deployment decisions and policy improvement, but physical testing and simulator preparation require substantial effort. World models offer a complementary route through closed-loop interaction in generated environments. Recent benchmarks have demonstrated their potential as policy evaluators through outcome-based evaluation (i.e., assessing task success or failure). However, such evaluation does not establish whether world models faithfully simulate how policies fail. Misrepresenting failures can lead to incorrect diagnoses of policy weaknesses and make policies appear ready for deployment even when small scene changes cause them to fail. We therefore ask: do world models faithfully simulate how and at which stages policies fail, and under what conditions they transition from success to failure? To answer this question, we introduce WorldFailBench with three evaluation tasks. Failure reason tests whether models reproduce the observable errors policies make. Failure stage assesses whether failures concentrate at the same stages of full-task execution. Failure boundary examines whether models reproduce the conditions under which policies transition from success to failure under controlled scene changes in simulation. The benchmark covers manipulation tasks in the LIBERO and RoboTwin simulation environments and real-world manipulation tasks performed by the G1 humanoid robot. Experiments across seven world models reveal substantial gaps in their understanding of policy failures, providing guidance for developing more reliable world models for policy evaluation.
[AI-203] Hybrid-time PINNs: Discontinuity-Aware Physics-Informed Neural Networks for Learning Hybrid Dynamical Systems
链接: https://arxiv.org/abs/2610.33562
作者: Shivanshu Tripathi,Maziar Raissi
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI)
备注:
Abstract:Hybrid dynamical systems flow continuously but jump at discrete events. Physics-informed neural networks (PINNs) fit data while enforcing known dynamics, but a single network is smooth in time and cannot model these jumps. We show that the more closely a PINN follows a jump, the larger its physics loss must be. We propose hybrid-time PINNs (HT-PINNs), which assign one network to each flow interval of the hybrid time domain and train the networks jointly with the jump times and the unknown parameters. The number of jumps and the sequence of modes are found by enumeration. HT-PINNs recover the number of jumps and the restitution coefficient of a bouncing ball. In a MuJoCo simulation of robotic disassembly, where a robot pulls a press-fitted pin out of a housing through a spring, HT-PINNs find the number of jumps and the sequence of modes, and estimate the static and kinetic friction coefficients with small errors of 0.1% and 1.7%.
[AI-204] Learning When to Collaborate: Selective Multi-Agent Medical Reasoning via Uncertainty-Aware Routing NEURIPS2026
链接: https://arxiv.org/abs/2610.33544
作者: Jiacheng Hou,Xueliang Cui,Dan Lu,Ruxin Wang
类目: Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026
Abstract:Multi-agent systems have emerged as a promising paradigm for multimodal medical reasoning, but existing methods rely on two assumptions that hinder practical deployment: (i) always-active dense collaboration among multiple agents, and (ii) dependence on large cloud-based models that raise privacy and efficiency concerns. In this paper, we challenge both assumptions and propose SMART-Med, a fully local multi-agent method that treats collaboration as a learnable decision rather than a default behavior. Our key insight is that not every medical query requires costly multi-agent deliberation: many can be reliably solved by a single well-trained agent, while only uncertain cases benefit from expert collaboration. Based on this insight, SMART-Med introduces an uncertainty-driven routing mechanism that adaptively decides when to answer directly and when to recruit a small subset of expert agents, effectively converting dense multi-agent reasoning into selective, query-adaptive coordination. To make this routing reliable, we curate a new multimodal medical dataset MedVR-3K with high-quality reasoning traces and a difficulty hierarchy, and then fine-tune local agents using GRPO with two complementary rewards: a certainty-aware accuracy reward and a vision-language reward model for correctness and reasoning quality respectively. Experiments on four medical visual question answering benchmarks show that SMART-Med achieves competitive performance compared to existing MAS approaches while substantially reducing inference cost (reducing inference time by more than 3.8x and token usage by 72%). Moreover, after fine-tuning on MedVR-3K, SMART-Med yields an average improvement of 6.7% across Slake, PathVQA, VQA-RAD, and PMC-VQA.
[AI-205] CoEvolve: Counterfactual-Guided Co-Evolution of Memory and Collaboration in Multi-Agent Systems
链接: https://arxiv.org/abs/2610.33539
作者: Xufei Shi,Derong Xu,Wei Chen,Yanyu Yao,Zhi Zheng,Tong Xu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM)-based multi-agent systems have demonstrated remarkable capabilities across diverse tasks. Yet their memory and collaboration architectures often remain static, constraining continued adaptation and improvement over time. Recent work addresses this limitation by evolving memory or collaboration based on interaction experience. However, updates to these interdependent components are not necessarily aligned: an evolved workflow may require experience that memory does not provide, while evolved memory may encode guidance that the workflow cannot use. To address this problem, we propose CoEvolve, a skill-centered framework for aligned memory–collaboration evolution. CoEvolve represents reusable local collaboration patterns as skills that link evolving collaboration decisions to their supporting memories. It further compares original and counterfactual trajectories to identify local evolution signals, which coordinate updates to the skill library, skill memory, and executor memories across rounds. Experiments on the ALFWorld and WebShop benchmarks confirm that CoEvolve achieves the best overall performance among the evaluated baselines, while ablation studies further validate the effectiveness of aligned co-evolution and counterfactual-guided evolution.
[AI-206] FraudSociety: A Generative Agent -Based Society for the Financial-Fraud Economy
链接: https://arxiv.org/abs/2610.33536
作者: Keyang Chen,Mingxuan Jiang,Zeping Li,Yiwen Zhao,Yifan Wang,Qiuyang Zhang,Shuaicheng Qi,Yicun Liu,Zhixin Li,Sen Liu,Yinan Jing,Guangnan Ye,Hongfeng Chai
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Financial fraud caused over \ 15.9 billion in losses in the United States in 2025, according to the Federal Trade Commission~\citepftc2026databook. Understanding such large-scale harm requires jointly modeling informational and financial dynamics. Yet existing simulators typically capture only one side of this interaction, leaving a gap in modeling how fraud unfolds in the real world. To bridge this gap, we introduce \textscFraudSociety, the first LLM-agent-based simulator that jointly models these coupled dynamics. It integrates a dynamic social environment for information exposure and communication, persistent financial ledgers that preserve payment and loss histories, and a Need-Intent-Action architecture for civilian decision making. Fraudsters further adapt their communication and financial actions to individual interaction histories and ledger states, forming a closed feedback loop between the two dynamics. We validate \textscFraudSociety across macro-level outcomes, micro-level behavior, and intervention effects. At the macro level, the simulated fraud outcomes closely match real-world patterns across multiple key statistics. At the micro level, it reproduces strong loss concentration, with the top 10% of victimization records accounting for 81.1% of total losses, consistent with Canadian Anti-Fraud Centre statistics~\citepcafc2025annual. Counter-fraud experiments further reveal variation across intervention strategies, with the best-performing personalized intervention reducing the repeat-payment rate from 49.4% to 10.9%. These results establish \textscFraudSociety as both a high-fidelity simulator and a scalable testbed for evaluating anti-fraud strategies under diverse intervention conditions, thereby advancing research on financial fraud and its prevention.
[AI-207] Pacifo: An Efficient Fully Homomorphic Encryption-based Lightweight ResNet for Trustworthy Machine Learning as a Service
链接: https://arxiv.org/abs/2610.33525
作者: Yang Li,Gang He,Ruichen Zhang,Guangyuan Liu,Yinqiu Liu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Machine Learning as a Service (MLaaS) has become a popular paradigm for deploying AI applications. However, it raises significant privacy concerns that the raw client data needs to be uploaded to the server for inference services. Private inference using Fully Homomorphic Encryption (FHE) offers a solution by allowing computations directly on encrypted data, with only the final results decrypted on a client side. However, current practical implementations still suffer from large computational overhead and high latency due to the cost of FHE operations. In this paper, we propose Pacifo, a framework that adapts customized fine-tuning strategies to efficiently evaluate the ResNet20 model in the context of FHE. Specifically, Pacifo introduces a public-private layer partitioning scheme and an efficient ReLU approximation, significantly reducing the number of private layers and costly FHE operations during inference. It also offers flexible security configurations to accommodate varying client privacy requirements. Compared to state-of-the-art FHE implementations of ResNet20, Pacifo achieves a speedup of 3.7-21.7\times in runtime while effectively preserving 95%-99% of the original plaintext model accuracy across downstream tasks. Pacifo is open-sourced at this https URL.
[AI-208] Can Agent ic Code Review Catch Vulnerabilities Before They Are Merged?
链接: https://arxiv.org/abs/2610.33509
作者: Yihan Yang,Chenyuan Yang,Zijie Zhao,Anh Nguyen Hoang,Ziqi Zhang,Weidong Wang,Zhiheng Lyu,Lingming Zhang
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Programming Languages (cs.PL)
备注: 20 pages, 2 figures, 10 tables. Benchmark: this https URL Results website: this https URL
Abstract:LLM agents have been increasingly adopted for security-focused code review, but existing benchmarks mostly derive the reviewed changes from their later fixes instead of the real-world bug-introducing commits. In this paper, we present SecPRBench, the first comprehensive benchmark for evaluating agentic security review on validated real-world code changes. More specifically, we have built an automated agentic pipeline to autonomously harvest, build, and validate massive real-world bug-introducing commits for popular software systems. To ensure precise evaluation, we also adopt LLM-as-a-judge to credit the findings matching ground-truth vulnerabilities. To date, we have collected 265 tasks from nine critical systems, including Firefox, V8, and OpenSSL with millions of lines of code. Across eight frontier models, recall ranges from 32.5% to 78.1%, with Grok 4.6 and GPT-6 sol ranking at the top. Our study also shows that most misses are not simply oversights: in 68% of them, the agent reports on a file that holds the flaw but argues for a different mechanism. Security-focused harnesses also behave differently: OpenAI’s Codex Security lowers recall by 8.3 points, mostly by abstaining, while Anthropic’s Defending Code harness raises it by 7.7 points, at a higher cost.
[AI-209] Error-Subspace Calibration for Multi-Task Model Merging
链接: https://arxiv.org/abs/2610.33483
作者: Yeseul Cho,Hanseul Cho,Chulhee Yun
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 25 pages
Abstract:Model merging combines task-specific expert models into a single model without joint training on their original datasets, but often incurs a performance gap relative to the individual experts. Post-merging calibration can reduce this gap through task-specific representation alignment, but the alignment objective alone does not specify which feature directions to correct. We propose Error-Subspace Calibration (ESCal), a post-merging calibration framework that identifies low-dimensional error subspaces from weight differences between expert and merged models. At each intermediate block boundary, a bridge module projects the merged representation onto the next block’s input-projection error subspace and maps these coordinates to an additive correction in the current block’s output-projection error subspace. The subspace bases remain fixed, so calibration learns only these linear maps, yielding a parameter-efficient mechanism for feature correction. We find that ESCal brings the merged model closer to the experts in both representations and predictions, indicating more faithful recovery of task-specific behavior. Experiments on vision and natural language understanding benchmarks show that ESCal improves merged-model performance across diverse architectures and merging methods using a small calibration set for each task.
[AI-210] A Unified Analysis of Importance-Corrected Momentum Actor-Critic over a Markovian Trajectory
链接: https://arxiv.org/abs/2610.33477
作者: Yanjie Dong,Zhou Su,Wei Wang,Gang Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We present a generic actor-critic method that combines a ridge-regularized pathwise least-squares temporal-difference (LSTD) critic with importance-corrected momentum (ICM) actors. The critic solves a ridge-regularized system per frame by Newton-Schulz inversion and needs no stepsize, no knowledge of the mixing time, and no rank condition. Averaging stochastic policy gradients over frames of T consecutive transitions, the ICM actors incur a frame-dependent gradient bias of order \frac1T . To our knowledge, we establish the first finite-time bounds for ICM actor-critic methods with normalized and unnormalized steps along a single Markovian trajectory. Up to a stationary-to-discounted distribution mismatch, the averaged gradient norm over K updates is bounded by \cal O(K^-1/3+\frac1T) for normalized steps and by \cal O(K^-1/4+\frac1T) for unnormalized steps. We specialize these bounds to Muon, normalized stochastic gradient descent (SGD) with momentum, SGD with momentum, and AdaNorm, and compare them in the nuclear norm. With a dimension-free stepsize, Muon confines the factor \min\m,n^1/2 associated with an m\times n actor parameter matrix to the noise and correction terms. The comparison also characterizes when orthogonalization guarantees a larger noiseless ascent than normalization: exactly when the gradient’s nuclear rank exceeds the spectral-to-Frobenius curvature ratio.
[AI-211] LLM s in Cloud: A Discussion on Architecture Orchestration and Inference
链接: https://arxiv.org/abs/2610.33475
作者: Jannatun Noor,Athar Noor Mohammad Rafee,Md. Sadman Haque,Sreezon Das Gupta,Atiar Osman,Md. Abu Obaida Zishan
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI)
备注:
Abstract:The widespread deployment of foundational Large Language Models (LLM) across cloud datacenters marks a fundamental paradigm shift from stateless deep learning toward distributed, memory-intensive generative workloads. However, serving massive foundation models under stringent Service Level Objectives (SLO) exposes critical physical and system bottlenecks. These challenges include the persistent High Bandwidth Memory (HBM) wall, operational asymmetry between compute-bound prefilling and memory-bandwidth-bound decoding, explosive Key-Value (KV) cache expansion, prohibitive serverless cold starts, and hardware reliability vulnerabilities. Existing literature reviews remain layer-siloed, decoupling low-level accelerator virtualization from cluster orchestration and runtime inference, while largely overlooking defining advancements from 2024 to 2026. To address these limitations, this study presents a comprehensive, cross-layer systematic review guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 framework. We synthesize a holistic taxonomy unifying hardware virtualization, cluster-level scheduling, and runtime inference optimizations, including iteration-level continuous batching, prefill-decoding disaggregation, and hierarchical KV cache offloading. Furthermore, we evaluate resilient cloud-native service architectures, examining scale-to-zero serverless paradigms, fault-tolerant checkpointing against silent data corruption, and hardware-enforced Trusted Execution Environments (TEEs) for multi-tenant isolation. Finally, we formalize the multi-dimensional Pareto trade-offs governing throughput, latency, monetary cost, and confidentiality, outlining a forward-looking research roadmap spanning Compute Express Link (CXL) memory pooling, in-memory computing, and autonomous multi-agent control planes to guide future cloud LLM infrastructure.
[AI-212] he Future Is Finer: Multivariate Time Series Super-Resolution Forecasting
链接: https://arxiv.org/abs/2610.33470
作者: Shuhao Li,Weidong Yang,Yue Cui,Fan Zhang,Lu Qin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Web platforms run on multivariate time series, from service telemetry and Web traffic logs to sensor streams served through Web APIs. Yet, retention policies often archive them at coarse granularity to save storage and bandwidth. Decisions such as autoscaling, alerting, and billing operate at finer granularity than existing forecasters reach, since these models predict at the resolution of their input. We therefore introduce Multivariate Time Series Super-Resolution Forecasting (MTS-SRF), which forecasts fine-grained futures from coarse histories, and propose TimeMagnifier to address its three challenges. Fourier Anchor counters data sparsity by fitting each input window to a trend-Fourier structure with sparse cross-variable gating; De-noise Only constrains fine-grained ambiguity by refining along the resolution axis rather than the noise axis, lengthening the grid level by level; and banded DTW Alignment corrects cross-resolution misalignment against the coarse forecast. On the first MTS-SRF benchmark, with ten datasets from five domains, real archived cross-resolution pairs, strengthened baseline adaptations, and five-seed significance tests, TimeMagnifier is best on every dataset, resolution coefficient, and metric against 15 baselines and cuts MSE by 12.3% on average. Its De-noise Only and DTW stages also transfer as a plug-in that lowers the MSE of existing forecasters by 5.1% to 9.0%. Code will be made publicly available.
[AI-213] LegoSR: A Unified Framework for Self-improving Agent -Guided Scientific Discovery
链接: https://arxiv.org/abs/2610.33465
作者: Hao Mao,Xu T. Liu,Peng Zhao,Shuai Lu,Yuntian Chen
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 36 pages, 6 figures. Code is available at this https URL
Abstract:Equation discovery solvers often use different data interfaces, equation representations and evaluation criteria, making their results difficult to compare. We present LegoSR, a unified framework for algebraic symbolic regression and differential equation discovery. LegoSR integrates seven solvers through a common equation representation and evaluation process, and supports searches constrained by partially known equations. Researchers can run solvers directly or use LegoSR Agent, which selects solvers and adapts searches based on solver knowledge, intermediate results and lessons from earlier tasks. On noisy ordinary and partial differential equation benchmarks, LegoSR Agent recovers nearly as many equations as the union of seven expert-tuned solvers and outperforms a generic agent with the same model, tools and budget by up to 29 percentage points. Three examples demonstrate basic use, adaptive solver selection and discovery guided by partial physical knowledge.
[AI-214] A Comprehensive Framework for All Stages of Large-Model Training: From Data and Tokenization through Pretraining to Post-Training Alignment
链接: https://arxiv.org/abs/2610.33451
作者: Xiangjing An,Zhibin Jiao,Aiting Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The capabilities of large language models (LLMs) and multimodal LLMs (MLLMs) arise from the synergy of the full training pipeline, including data and tokenization, pretraining, and post-training (supervised fine-tuning, preference alignment, reinforcement learning, and continual optimization), rather than any individual stage. Existing surveys mostly adopt a single-stage or single-technique perspective, treating related research as isolated systems without unified stage division or a horizontally comparable basis. This paper proposes a unified stage division and a six-dimensional analytical framework covering data, objective function, optimization strategy, system constraints, evaluation, and risk and governance. We systematically review 395 core references from 2017 to 2026, and present a conceptual map of cross-stage effects: forward constraints of pretraining on post-training, and reverse empowerment of post-training on data and pretraining. We derive three core findings. First, the comparability of scaling laws depends on data quality and mixture, a long-neglected variable. Second, preference alignment methods (RLHF, DPO) and reasoning reinforcement learning with verifiable rewards redefine post-training boundaries, while the Pass@1 versus Pass@K debate remains unresolved. Third, multimodal training poses structural challenges to the pure-text paradigm in tokenizer design, training objectives, and evaluation. The paper provides a cross-stage effects diagram, a method-family map, a practical decision tree, and at least 10 verifiable open problems, with an extensible skeleton for focused sub-surveys per stage. As of the retrieval date, peer-reviewed evidence (A+B) accounts for approximately 79% of the 395 references. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.33451 [cs.AI] (or arXiv:2610.33451v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.33451 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: An Xiangjing [view email] [v1] Sat, 24 Oct 2026 05:09:05 UTC (498 KB)
[AI-215] BoardSafety: AI-Ready Data from Manufacturer Documentation for Development-Board Safety
链接: https://arxiv.org/abs/2610.33448
作者: Linjian Yang,Xinyan Wang,Kunpeng Liu
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:The development of large language models (LLMs) are capable to assist with hardware experiments. In this scenario, this raises a practical question: Can LLMs determine whether a proposed operation is safe for a specific development board? Documented operating limits and safety requirements are board-specific, even across products from the same manufacturer. This variation makes board-specific safety judgments by an LLM difficult to evaluate. To address this problem, we introduce the BoardSafety Benchmark. The benchmark is built from vendor datasheets and hardware guides. For data transformation, these documents are converted into a board-specific safety rule base and are then used to create labeled operating scenarios. Each data item pairs a board and a proposed operation to an applicable safety rule with ground-truth label, indicating the level of safety. The categorized level for each operation includes safe, unsafe, or insufficient-info. In addition, the benchmark covers diverse operations, board-specific cases, boundary conditions, and missing-information cases. We evaluate a series of models by varying their access to board-specific rules and the use of structured-rationale prompting. Our results demonstrate that providing the safety rules generally improves accuracy, whereas structured-rationale prompting produces gains that vary across models and rule-access conditions. Overall, the evaluated models cannot yet reliably determine whether an operation is safe, which implies that critical exploration is required for this field.
[AI-216] ConMem: Counterfactual Multi-Scenario Probing for Resolving LLM Memory Misalignment
链接: https://arxiv.org/abs/2610.33432
作者: Shuning Zhang,Tenghao Ji,Jingruo Chen,Eryue Xu,Bohao Chu,Xin Yi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Long-term memory is critical for autonomous Large Language Model (LLM) agents executing multi-session, long-horizon tasks. However, existing memories often inappropriately store, retain, or retrieve over-generalized, misattributed, or privacy-sensitive information, a problem we term memory misalignment. To study this problem, we construct \textbfMemAlign, a benchmark comprising 1,079 curated memory cases grounded in synthetic, validated multi-session conversational contexts covering diverse misalignment taxonomies. We then propose ConMem, an algorithm that reframes retention decisions as prospective counterfactual causal inference. \textbfConMem synthesizes counterfactual interaction scenarios to prospectively audit downstream intervention effects across three aspects: relevant task execution, irrelevant invariance checks, and risk-sensitive boundary shifts. Experiments show that ConMem outperforms other methods, achieving 75.16% accuracy, 64.15% balanced accuracy, and a Macro-F1 of 65.31%, effectively mitigating over-retention biases without sacrificing personalization utility.
[AI-217] Exact Unlearning After Data Curation
链接: https://arxiv.org/abs/2610.33426
作者: Priyank Jayraj,Devanshu sahoo,Uday Mittal
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Training pipelines often remove similar examples before fitting a model. Deleting a raw record can change that selection and admit an example that was excluded before. What must a service retain to reproduce training after such requests? We answer this question for a fixed rule in which earlier similar records block later ones. A limit on the total number of future deletions identifies exactly which records can still enter training. We use this boundary to build a summary of training statistics that can be repaired exactly after repeated deletions. The repaired summary is the same as rebuilding it on the surviving data with the remaining budget. We also construct finite-precision features for ridge regression, a linear learner with a penalty on coefficient size. With exact rational outputs, allowing one more deletion changes the required private information. It rises from zero to every private feature sign in the excluded candidates. An encoding attains this conditional private-information minimum on the constructed family throughout a sequence of requests. On 1,001 AskUbuntu development records, deleting an initially excluded record admits another record at 13 of 51 fixed similarity cutoffs. Exact membership repair passes 13,056 deletion-step checks. Two model-based semantic calibrations fail their fixed criteria, so the downstream task study is not run. The central result is a retention contract: the future deletions a service promises to support determine which excluded data can still be needed.
[AI-218] Benchmarking System One Models for Time-Series Decision Making: A Case Study of Jev
链接: https://arxiv.org/abs/2610.33410
作者: Jianye Wang,Kexin Zhang,Jiayin Xiong,Hao Xue
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 13 pages, 2 figures. Code: this https URL
Abstract:Time-series tasks require both accurate predictions and fast decisions. Advances in large models have introduced decision-oriented models that directly output answers and their probabilities, which we call Structured System 1 models. Their applicability to time-series tasks remains underexplored. Using Jev 1.13.0 as a case study, we evaluate time-series classification, event type prediction, oil temperature direction and stock movement prediction, converting continuous forecasting targets into discrete decisions to fit the model interface. We establish a unified evaluation framework using accuracy and AUC. Results show stronger performance on selected explicit temporal patterns and event type prediction, but near-chance discrimination on many numerical sequence classification tasks and stock movement prediction. Some datasets also reveal a gap between ranking quality and final decision accuracy. These findings motivate improvements in temporal representation, domain adaptation, and decision-score alignment. Standardized sample construction, model calls, and result recording support extending the evaluation to future models.
[AI-219] Data Optimization for LLM s: A Survey KDD
链接: https://arxiv.org/abs/2610.33405
作者: Ou Wu
类目: Artificial Intelligence (cs.AI)
备注: Accepted for publication in ACM Transactions on Knowledge Discovery from Data (TKDD) at Sep. 30, 2026
Abstract:Large language models (LLMs) have attracted great attention in both academic and industrial communities in recent years. Numerous survey studies have covered various aspects of LLMs, such as learning strategies, applications, alignment, explainability, and evaluation. However, little headway has been made in reviewing data-related aspects for LLMs, despite data being recognized as a key factor contributing to the emergence of the LLMs era. A wide range of data optimization algorithms and techniques are employed in LLMs’ training and application processes. These techniques often originate from diverse deep learning domains, with their theoretical inspirations and heuristic motivations appearing unrelated to each other. This study aims to develop a comprehensive framework for LLMs’ data optimization techniques to enhance data utilization and efficiency. We first discuss data challenges in LLMs’ training and inference. Second, we summarize data perception dimensions for evaluating LLMs’ data properties. These dimensions primarily include quality, difficulty, diversity, and redundancy, laying a foundation for subsequent optimization. Third, existing optimization techniques are reviewed and categorized into four groups: data selection, augmentation, reweighting, and others. Fourth, we analyze interconnections between these categories. Finally, we identify key challenges and future directions for LLMs’ data optimization compared with conventional deep learning approaches.
[AI-220] How To Build A Modern Diffusion Language Model
链接: https://arxiv.org/abs/2610.33391
作者: Volodymyr Kuleshov,Marianne Arriola,Yair Schiff,Gilad Turok,Guanghan Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The frontier of large language model (LLM) performance has advanced rapidly in recent years, fueled by scaling laws for pretraining on large text corpora. However, mainstream LLMs have fundamental limitations: they generate text autoregressively, producing outputs sequentially from left to right one token at a time. This sequential process leaves parallel GPU hardware underutilized during inference–a bottleneck that has grown acute as progress shifts toward scaling post-training and inference-time compute–and commits irreversibly to each token without the ability to revise earlier mistakes. Discrete diffusion models address these limitations by generating tokens in parallel, starting from a noisy or masked sequence and iteratively refining the entire output over a series of denoising steps. This parallel generative process enables speeds of thousands of tokens per second, provides a new axis of inference-time compute scaling, and supports built-in error correction and controllable generation. Over the past two years, discrete diffusion has scaled to billions of parameters across biological sequence modeling and natural language, with open-weights and commercial releases from leading labs. This article provides a survey of key ideas in discrete diffusion and a high-level introduction to how modern diffusion LLMs are built, covering their foundations, core algorithmic and architectural building blocks, applications in biology and language, and implications for future LLM scaling laws.
[AI-221] Agent Orch: How Close Can Global Orchestration Come to Step-by-Step Interaction?
链接: https://arxiv.org/abs/2610.33375
作者: Zijie Liu,Qi Yang Huang,Jinhao Duan,Yueh Che Lee,Thomas A. Joseph,Zayne Turner,Adam Seligman,Tianlong Chen,Zhaozhuo Xu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Enterprise workflow agents typically rely on step-by-step interaction (Step-by-Step): the agent executes an action, observes its result, and invokes a large language model (LLM) again to decide the next step. Global orchestration (Global) instead generates a complete executable workflow in advance and executes it without further LLM calls, substantially reducing inference overhead, but it remains less accurate than Step-by-Step. We study how close Global can come to Step-by-Step and what narrows the remaining gap. We introduce AgentOrch, a benchmark that directly compares global orchestration with step-by-step interaction on 650 enterprise SOPs and 3,220 runtime variants under matched tools, environments, and evaluation. Step-by-Step remains more accurate for every evaluated model, but the gap has shrunk by about 79% from early models to the strongest recent model. We identify three routes to further narrowing this gap. Incremental workflow construction improves Global accuracy by an average of 9.98 percentage points (pp) without runtime execution feedback. Revising complete workflows with execution feedback improves all evaluated models, including 6.13 pp on held-out runtime variants. Targeted training on complete workflows improves accuracy by an average of 12.18 pp across all five trained models. Although a gap remains, global orchestration offers a promising path toward enterprise agents that deliver higher accuracy with fewer LLM calls.
[AI-222] MOSHO: A Native Discrete Shark-Inspired Metaheuristic for Hardware-Aware Neural Architecture Search
链接: https://arxiv.org/abs/2610.33358
作者: Nazim Abdeerhmane Aouni,Zakaria Soualah Mohammed,Maroua Ogab,Nour El Imane Elbar,Rayane Rahmat Errahmane Smara,Billel Moussous,Adriane Anis Khaled,Malika Bessedik,Fatima Benbouzid Si Tayeb,Hadjer Benmeziane
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Hardware-aware neural architecture search (NAS) requires jointly optimizing predictive accuracy and device-specific latency over discrete search spaces. While multi-objective swarm optimizers show promise, continuous formulations such as the Multi-Objective White Shark Optimizer (MOWSO) can suffer from quantization noise when rounded to integer-encoded architecture vectors. To address this limitation, we propose MOSHO (Multi-Objective Shark Hunting Optimization), a native discrete metaheuristic that replaces continuous motion equations with discrete architecture-editing operators. MOSHO integrates adaptive operator scheduling, energy-regulated restarts, and deferred crowding-distance truncation to enhance exploration and search efficiency. On the merged NAS-Bench-201/HW-NAS-Bench benchmark across 18 contexts, MOSHO achieves a median normalized hypervolume of 0.9987 and IGD+ of 0.0064, yielding Pareto fronts competitive with NSGA-II while reducing search runtime by 96.4% (0.64 s vs. 17.72 s). These results demonstrate that tailored discrete search mechanisms can enable swarm optimizers to achieve high-quality multi-objective solution quality with substantially lower search cost in hardware-aware NAS.
[AI-223] BenchBench-Cloning: Evaluating Molecular Cloning Design and Protocol Development
链接: https://arxiv.org/abs/2610.33355
作者: Aditya Sivakumar,Peter Ciaccia,Hassaan Qadir,Ian Hsu,Ashu Singhal
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Molecular cloning is one of the most routine wet-lab workflows, yet existing evaluations of language models in biology mostly test knowledge recall or isolated steps rather than end-to-end experimental planning. We introduce BenchBench-Cloning, a benchmark for large language models comprising 65 molecular cloning tasks grounded in sequence records, laboratory notebooks, and published experimental studies. Planning a molecular cloning experiment requires defining a set of constructs needed for a scientific objective and a construction route to produce them. BenchBench-Cloning contains 50 protocol-development tasks, in which agents produce an executable protocol for a specified construct, and 15 construct-design tasks, in which agents determine the constructs a scientific objective calls for. Protocol-development tasks are evaluated by execution in a cloning simulator; construct-design tasks are evaluated with task-specific deterministic checks and, where needed, rubric-based LLM-as-a-judge evaluation. Across eight systems with repeated attempts per task, GPT-6 Astra scores highest at 30.6%, with the others between 8.6% and 22.2%, and the benchmark remains unsaturated under best-of-ten sampling. As integrating individual cloning decisions into complete, valid plans remains challenging even for frontier models, we expect BenchBench-Cloning to serve as a useful benchmark for tracking progress in experimental planning.
[AI-224] From Hype to ROI: An Economic Model for LLM Adoption in Information Systems and Security Operations
链接: https://arxiv.org/abs/2610.33352
作者: Pellegrino Casoria,Andrea Araldo,Rosario Patanè,Simon Pietro Romano
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE)
备注:
Abstract:Large Language Models (LLMs) promise significant efficiency gains across information-intensive operations by automating repetitive language-based tasks. However, adoption decisions remain largely opportunistic, as organizations struggle to move from pilots to sustainable industrialization. Without systematic economic frameworks, practitioners cannot assess whether LLM investments create value in a given context. This paper develops a general economic model to support informed LLM adoption decisions in operational information systems. Facing high uncertainty, we represent parameters as random variables centered on authoritative benchmarks, and evaluate the model by Monte Carlo simulation across pessimistic-to-optimistic assumptions on LLM efficiency. The ROI distributions show the probability of profitability and the Value at Risk for each configuration, together with a sensitivity analysis of its main drivers. As a first instantiation, we apply the model to Security Operations Centers (SOCs), mapping workflows onto established standards to identify tasks suited to LLM augmentation. The model balances productivity gains, driven by alert volume, analyst effort, and LLM efficiency, against operational costs, including inference, infrastructure, and error remediation, covering both cloud API and self-hosted deployments. Results show that adoption economics are context-dependent, and that automation is an exploitable lever, gated by accuracy, while the cost of errors and the model’s price are secondary. Across the analyzed models and prices, most configurations are unprofitable, with profitability limited to high-accuracy, low-uncertainty deployments. The model gives organizations a quantitative basis for evidence-based adoption decisions, by locating the boundary between profit and loss, and the loss risked beyond it. Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE) Cite as: arXiv:2610.33352 [cs.CR] (or arXiv:2610.33352v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.33352 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Pellegrino Casoria [view email] [v1] Fri, 23 Oct 2026 22:46:08 UTC (935 KB) Full-text links: Access Paper: View a PDF of the paper titled From Hype to ROI: An Economic Model for LLM Adoption in Information Systems and Security Operations, by Pellegrino Casoria and 3 other authorsView PDFHTML (experimental)TeX Source view license Additional Features Audio Summary Current browse context: cs.CR prev | next new | recent | 2026-10 Change to browse by: cs cs.AI cs.CE References Citations SciXGoogle Scholar Semantic Scholar export BibTeX citation Loading… BibTeX formatted citation loading… Data provided by: Bookmark checked="checked"class=“labs-tab-input”> Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv’s community? Learn more about arXivLabs. Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?) mathjaxToggle(); We gratefully acknowledge support from our major funders, member institutions, , and all contributors. About Help Contact Subscribe Copyright Privacy Accessibility Operational Status (opens in new tab)
[AI-225] mpShap: When and How Much a Feature Matters in Time-Series Forecasting
链接: https://arxiv.org/abs/2610.33330
作者: Chi Zhao,Anastasiia Chelnokova,Jing Liu,Elena Parilina
类目: Artificial Intelligence (cs.AI); Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)
备注:
Abstract:Attribution for a time-series forecaster should answer two questions that a feature-level scalar cannot: when in the input window a feature acts, and how much its influence changes across the forecast horizon. Most existing Shapley explainers treat each feature as one player and therefore collapse both axes. We propose TempShap, a retrospective, loss-based Shapley framework whose players are the individual input cells (t,i) . Its attribution tensor yields temporal-localization and horizon-impact profiles. We validate these against planted temporal structures, a tractable exact cell game, and a closed-form attribution oracle for a real linear forecaster. Permutation sharing estimates the cell-level game at localized perturbation cost. Under redundant inputs, one-at-a-time scores under-credit substitutable information, whereas the signed Shapley decomposition preserves total value and yields more reliable held-out deletion rankings under mean masking. A planted counterexample isolates the redundancy mechanism. Experiments show a deletion-ranking advantage across Shapley estimators, forecasting benchmarks, and model families. TempShap therefore makes fine-grained Shapley attribution both practical and operationally useful.
[AI-226] AI Teaches Humans to Handle Complexity in the Game of Go
链接: https://arxiv.org/abs/2610.33323
作者: Adamo Cerioli,Vito D. P. Servedio,Gabriele Cerioli,Antonio Villani,Carlo Metta
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Strategic decision-making in complex environments requires balancing local stability against global coordination. Using the game of Go as a bounded empirical testbed, we investigate how biological and artificial agents navigate this trade-off by formalizing a network-derived metric of strategic tension defined via the spectral radius of spatial dependency matrices. Across human skill progression, tension systematically decays from novices to masters as uncoordinated tactical skirmishes give way to disciplined territorial stabilization. However, a longitudinal analysis of professional play spanning 1980–2026 reveals a striking historical paradox: following the superhuman breakthrough of AlphaGo in 2016, elite human matches display a marked surge in sustained tension. A two-dimensional projection based on move distance and captures places modern professionals close to neural engines, separating their tension from the local contact fighting of novices and from random noise. Consequently, strategic tension acts as a macroscopic order parameter for decision paradigms rather than raw computational strength. Beyond the game of Go, these findings establish a constructive paradigm for human-AI interaction, indicating how artificial systems can act as cognitive scaffolding that expands biological heuristics and elevates human capacity to navigate higher-order complexity across professional training and decision-making domains.
[AI-227] A Deterministic Security Runtime for LLM Agents on MPI and Batch-Scheduled HPC Systems: Consensus-Governed Hardware-Rooted Supervision of Every Rank
链接: https://arxiv.org/abs/2610.33318
作者: Alfredo Metere
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:Large language model agents launch simulations on high-performance computing (HPC) systems, where one sbatch or mpirun can commit thousands of node-hours and a forged result can cost lives. Of 23 surveyed systems, none supervises the running job, and none holds when the control host is compromised. We fix both. Enclawed, our deterministic security runtime for AI agents, has until now served a single user; this work makes it a distributed runtime for parallel HPC and agent swarms. The Enclaweder, our hardware root of trust, keeps a compromised host from becoming a catastrophe. A launch gate decides every launch by policy alone, refusing what it cannot parse. Once the job runs, a swarm chain grants short leases against signed heartbeats and evicts on evidence, and a supervisor around every rank pauses it when its lease lapses and kills it on tampering. Each node signs with its Enclaweder’s key, so the host can forge none of it. Enclawed needs no administrator privileges and plugs into existing clusters unchanged, over Secure Shell alone. Supervision leaves the compute loop of a 16-rank LAMMPS job unchanged, on graphics processors too, and adds about 3 s of start-up. Twelve live attacks on Enclaweder-keyed ranks, colluding coordinators included, ended fail-closed. Sixteen agents on a real model attempted 24 launches that planted instructions ordered; the gate denied all 24 and allowed all 108 legitimate ones. Enclawed’s core and the HPC Agent Security Benchmark are open source.
[AI-228] Responsibility Boundaries for Agent ic Systems: A Comparative Evaluation of Probabilistic Agent ic Programming
链接: https://arxiv.org/abs/2610.33312
作者: Khadim Lo,Soukaina Alaoui Ismaili,Jean-Sébastien Dessureault
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Programming Languages (cs.PL)
备注: 43 pages, 8 figures, 9 tables
Abstract:Language-model agents combine reasoning, tools, memory, delegation, and external effects. Authority is often distributed across prompts, host code, configuration, and logs. Probabilistic Agentic Programming (PAP) makes responsibility boundaries explicit programming abstractions connecting mandate, authority, effects, and evidence. We compare PAP with programming paradigms, agent-oriented and BDI systems, workflow and policy languages, and LLM tool-use frameworks along seven dimensions. A mediation kernel formalizes checks before tool execution, operation-bound approval, resource limits, and evidence. Inductive proofs establish conditional safety for arbitrary model responses, assuming correct guards and complete mediation. A backward simulation matches graph executions to source executions in a defined sequential Touba subset. Its human checkpoints release program continuations; operation-bound approval remains a separate kernel contract. Finite-state exploration checks the model. Scripted enforcement tests exercise selected refusal paths and permitted controls. An exploratory study combines seven business cases, including two three-agent workflows, seven models from three providers, five methods, and five repetitions per case and method. Shared English instructions and exact checks of stored outcomes yield 1,225 assessed observations. Native Touba completes all 35 observations on six models and 30 on the remaining model; structured Touba ranges from 21 to 35. Native matches the highest observed count for each model, including ties with the frameworks. All paired case intervals include zero. These observed counts describe the evaluated tasks; the runtime authority boundary remains fixed. The results connect responsibility contracts, execution rules, and evidence.
[AI-229] When Does Pre-Trained Capacity Help Causal Learning? From Scalar Prediction to Counterfactual Post-Training NEURIPS2026
链接: https://arxiv.org/abs/2610.33311
作者: Qi Xu,Zijie Lei,Zhifei Deng,Zhigang Hua,Luke Simon
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 10 pages. Accepted at the NeurIPS 2026 Workshop: Transitioning from Pre-Training to Post-Training
Abstract:How does pre-trained model capacity relate to performance after adaptation for causal tasks? We study two systems built on the Qwen 2.5 family: CausalLLM, which estimates scalar potential outcomes with treatment-specific heads, and CausalGen, which generates treatment-conditioned text or code using soft prefixes and factual supervised fine-tuning (SFT), counterfactual SFT, and preference alignment. Increasing backbone size from 0.5B to 1.5B yields mixed scalar-estimation results on IHDP, but improves CodeGen pass@1 from 0.267 to 0.575 and treatment-effect score from 0.231 to 0.515 after the complete generative pipeline. Within 0.5B CausalLLM, narrative inputs reduce mean sqrt(PEHE) from 1.842 to 1.780, a 3.4% improvement over tabular inputs; the 5.8% improvement over a no-LLM baseline combines changes in representation and input format. These descriptive comparisons motivate a distinction between available and convertible capacity. They do not isolate the contribution of post-training: the two systems differ in dataset, architecture, output modality, and supervision, and intermediate-stage evaluations are unavailable. With only two small backbones and three replications, the results support a hypothesis about capacity–adaptation alignment rather than a general scaling law or a demonstrated causal mechanism.
[AI-230] GEMS-RAG : Graph-grounded Engineering Multimodal Standards Retrieval-Augmented Generation
链接: https://arxiv.org/abs/2610.33277
作者: Abdul Hannan Azad,Irvin Wang,Jiwan Jiang,Chenchen Kuai,Bofeng Cao,Keshu Wu,Yuzhou Wang,Xiaopeng Li,Yang Zhou
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Engineering standards are challenging knowledge sources for large language models (LLMs) because they are hierarchical, multimodal, and densely connected through figures, tables, cross-references, and normative rules. Existing retrieval-augmented generation (RAG) methods often rely on unstructured text retrieval and struggle with questions requiring joint reasoning over textual provisions, visual specifications, and document structure. We propose GEMS-RAG (Graph-grounded Engineering Multimodal Standards Retrieval-Augmented Generation), a framework for grounded reasoning over engineering standards. GEMS-RAG combines hierarchical parsing, fine-grained visual entity extraction, multimodal knowledge graph construction, and graph-grounded hybrid retrieval. Instead of treating figures as monolithic images, it decomposes them into meaningful visual entities linked to textual provisions, engineering concepts, and regulatory rules. During inference, dense, sparse, and visual retrieval are combined with graph expansion to identify relevant evidence across text, figures, and document structure. We evaluate GEMS-RAG on the Manual on Uniform Traffic Control Devices (MUTCD), a large engineering standard containing egulations, technical drawings, and standardized traffic-control devices. Results show that graph-grounded multimodal retrieval improves retrieval effectiveness and question-answering performance over text-only, multimodal, and graph-based RAG baselines. Beyond MUTCD, GEMS-RAG offers a general framework for reasoning over engineering codes, technical manuals, regulatory documents, and scientific standards in safety-critical applications.
[AI-231] Scoped Semantic Invariance as a Foundation for LLM Reliability EMNLP2026
链接: https://arxiv.org/abs/2610.33267
作者: Rosario Uceda-Sosa,Maria Chang,Karthikeyan Natesan Ramamurthy,Moninder Singh
类目: Artificial Intelligence (cs.AI)
备注: Accepted at EMNLP 2026
Abstract:LLMs often answer semantically equivalent queries inconsistently, even when the underlying facts are known. We formalize consistency as semantic invariance over equivalence classes of queries and show that invariance under arbitrary meaning-preserving transformations is undecidable in general. This motivates scoped semantic invariance, where model behavior is evaluated relative to bounded semantic domains. We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantic constraints and grounding contexts. Across the MITRE D3FEND ontology and eight Wikidata- derived domains, we show that inconsistencies are structured rather than random, concentrating in localized hotspot subgraphs organized around recurring TREE, STAR, and CYCLE patterns that remain stable across models and domains. We further show that selective graph grounding improves correctness and cross-model consistency, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving correct knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks. Comments: Accepted at EMNLP 2026 Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.33267 [cs.AI] (or arXiv:2610.33267v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.33267 Focus to learn more arXiv-issued DOI via DataCite
[AI-232] Operationalizing Governed Agent ic Workflows in Regulated Finance: A Customer Zero Case Study using Zafin AIOS
链接: https://arxiv.org/abs/2610.33253
作者: Shahir Daya,Rob Dunlap,Alexandre Boulenger,Ahmed Abdelmoneim,Salvatore Vella
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 19 pages
Abstract:Regulated industries such as finance and healthcare face critical governance challenges that hamper the adoption of AI-assisted software development and hinder the transition from experimental pilots to production work. This paper presents Zafin AIOS, an end-to-end orchestration platform and control plane designed to deploy AI within regulated industries such as finance. We provide the architectural principles and an empirical comparison between AI-assisted and human-led workloads. To validate this framework empirically, we conduct a case study analysis of Zafin’s internal enterprise workflows, which we refer to as Customer Zero. This paper contributes a production case study of AI-assisted software engineering in regulated industries. We evaluate the results of using Zafin AIOS with two different production codebases, one for the pricing execution platform and the other for the client deal negotiation platform. Our contribution is a customer case study for regulated industries with production codebases and governance, which includes proof of work. The results demonstrate that strictly governed agentic orchestration increases enterprise capacity without compromising control. In the pricing execution platform workflows, the platform provided complete observability and reduced effort by 38%. In the client deal negotiation platform, it reduced reporting work by 76% while maintaining all predefined data privacy, role-based access, and audit guardrails. This paper bridges the gap between AI-assisted software engineering theory and the operational reality of compliance. Closing this gap is important to help enterprises take advantage of AI technologies. These results are descriptive of the benefits of using governed workflows in regulated industries.
[AI-233] AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science
链接: https://arxiv.org/abs/2610.33245
作者: Atharva Naik,Vivek Chari,Pol Llado,Daniel Fried,Carolyn Rosé,Benjamin Van Durme,Sam Thomson,Justin Svegliato
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Data science agents can automate end-to-end machine learning workflows, yet they remain far from competitive on human leaderboards. Agents make process errors such as insufficient exploration, unsound validation, coding bugs, and poor use of a fixed budget (re-running the same exploratory data analysis and baseline models in every attempt). We propose AdaPRA to address these issues: a persistent coding agent harness that accumulates knowledge across attempts on the same task, with rubric-guided process supervision that identifies and corrects errors. Our process reward agent audits the agent’s trajectory and workspace by executing checks, and adapts its rubrics to each task. We find that rubrics seeded from easier tasks generalize to harder tasks, with rubric adaptation further improving performance. With five attempts per task, AdaPRA improves an 80B Qwen model’s leaderboard percentile by 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard while reducing process errors. We also show that supervision with rubrics mined from one base model’s runs improves another (DeepSeek-V3.2), and that AdaPRA beats DataPRM, a competitive PRM for data science. We also share our code for reproducibility.
[AI-234] NewsBench: Expert-Grounded Evaluation of Epistemic Quality in AI News Reporting
链接: https://arxiv.org/abs/2610.33240
作者: Matt Wilde,Jillian Fisher,Andrew B. Hall,Kathryn Salam,Robbie Goldfarb
类目: Artificial Intelligence (cs.AI)
备注: 12 pages
Abstract:As frontier AI systems increasingly serve as sources of political and current-events information, evaluating the epistemic quality of their responses is now especially consequential. Existing evaluations often focus narrowly on ideological alignment, rely on measures not validated for this setting, or depend on reference labels derived from public opinion or non-expert annotation, potentially producing misleading or miscalibrated results. To address these limitations, we introduce NewsBench, an expert-grounded benchmark that reframes evaluation around editorial standards developed by senior practitioners in journalism, policy, and intelligence analysis. NewsBench operationalizes quality across three dimensions, source quality, factuality, and neutrality, using a two-phase pipeline in which domain experts first define and calibrate evaluation guidelines, which are then transferred to automated AI judges validated against expert-adjudicated gold labels. The benchmark includes approximately 3,000 evaluation prompts, with roughly 500 refreshed monthly to capture emerging current events, roughly 2,500 human expert-annotated items for calibration and validation, and a suite of AI judges designed to enable scalable evaluation. To validate our methodology, we report both inter-rater reliability among human experts and AI judge–human agreement across all three dimensions. Finally, we apply NewsBench to four frontier models, revealing distinct and non-overlapping failure profiles in sourcing, factuality, and neutrality.
[AI-235] Distilling Expert Judgment at Scale
链接: https://arxiv.org/abs/2610.33233
作者: Robbie Goldfarb,Andrew B. Hall,Jillian Fisher,Kathryn Salam,Matt Wilde
类目: Artificial Intelligence (cs.AI)
备注: 9 pages
Abstract:Frontier AI models are increasingly deployed on topics where the stakes are high and expert judgment is required, including healthcare, defense, finance, governance, and politics. We present a general methodology for distilling expert human judgment into automated evaluation systems that operate at scale. Our approach assembles networks of domain experts and develops techniques for extracting not just their conclusions but their reasoning processes, encoding those processes into structured editorial standards that automated judges can apply continuously. We ground our approach in a long tradition of research on the nature of expertise and the conditions under which expert judgment is reliable and more effective than other approaches, and argue that independent, expert-calibrated evaluation is necessary not only on epistemic grounds but as a matter of accountability and public trust. We validate the methodology on AI news reporting as a first application domain. On source quality, two independent domain experts achieve Krippendorff’s \alpha = 0.83/0.69 on a held-out test set when applying the editorial standards, and our automated judge reaches F1 = 0.81/0.97 against expert consensus. On neutrality, experts achieve \alpha = .72 on the test set, and the automated judge reaches F1 = 0.86 against expert consensus. Additionally, a controlled comparison shows that expert-calibrated judges substantially outperform uncalibrated frontier models on every source-quality metric, demonstrating that encoding expert reasoning is the critical step that prompt engineering alone cannot replace. We present this benchmark as a proof of concept for a broader evaluation infrastructure applicable to any domain where judgment matters and expertise is scarce.
[AI-236] Scaling Qualitative Analysis with Large Language Models : Development and Validation of the Smart Coder Framework for Human-LLM Collaborative Coding
链接: https://arxiv.org/abs/2610.33205
作者: Cheng Tang,Kun Wang,Yaxuan Yang,Shiyu Wang,Allan Cohen,Rachael Eriksen Brown,Chandra Hawley Orrill
类目: Artificial Intelligence (cs.AI)
备注: 29 pages, including supplemental materials
Abstract:Current uses of large language models (LLMs) for qualitative coding often rely on ad hoc prompting with limited guidance for reliability and human oversight. To address these limitations, we propose the Smart Coder Framework, a three-phase approach structuring human-LLM collaborative coding through Instruction, Prompt Strategy, and Reliability Check. We evaluated the framework using 72 middle-school mathematics teacher interview transcripts and a human-consensus benchmark. A factorial design compared instructed versus uninstructed prompts, few-shot versus few-shot plus chain-of-thought prompting, and single-pass versus self-consistency coding. Twelve transcripts were used for prompt development and calibration, and 60 for the formal evaluation. In the results, instruction consistently improved agreement. Chain-of-thought further improved agreement under adequate instruction, and self-consistency produced smaller, theme-dependent gains but helped identify unstable decisions for human review. Performance also varied by transcript characteristics. Overall, the framework systematically structures and monitors LLM-assisted qualitative coding while preserving human authority over coding decisions.
[AI-237] Isharah-Selfie: Continuous Sign Language Recognition Dataset for One-handed Signing
链接: https://arxiv.org/abs/2610.33200
作者: Ahmed Abul Hasanaath,Murtadha Aljubran,Sarah Alyami,Muhammad Haris Khan,Hamzah Luqman
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Current sign language recognition and translation benchmarks are commonly collected in controlled settings using fixed or studio-style cameras, particularly limiting their ability to reflect how signers communicate in everyday mobile scenarios. As such, this gap becomes increasingly evident when considering the Arabic Sign Language (ArSL), which is a primary sign language used across Arab countries. In this work, we introduce Isharah-Selfie, a large-scale ArSL dataset collected entirely using front-facing smartphone cameras under unconstrained selfie conditions. The dataset comprises 8,997 video clips performed by 16 signers across 1,493 unique sentences, covering several domains such as healthcare, education, and transportation. Unlike conventional datasets dominated by fixed-camera, two-handed signing, Isharah-Selfie captures realistic mobile communication characterized by one-handed signing, close-range viewpoints, partial body visibility, camera motion, unconstrained environments, and heterogeneous device resolutions. Each video is annotated with both gloss sequences for continuous sign language recognition (CSLR) and Arabic sentence translations for sign language translation (SLT). Moreover, we define signer-independent and unseen-sentences evaluation protocols to assess generalization across unseen signers and novel sentence compositions. Extensive benchmarks using RGB-based and pose-based data show that current CSLR and SLT methods remain challenged by selfie-style signing conditions, particularly under unseen-sentence evaluation, where compositional generalization remains weak. By releasing Isharah-Selfie, we aim to support research toward robust, user-centered sign language technologies that operate reliably in realistic smartphone-based communication settings. The Isharah-Selfie dataset is available on this https URL.
[AI-238] Cross-Platform Application Migration to HarmonyOS: An Agent ic Pipeline with Artifact-Derived Verification
链接: https://arxiv.org/abs/2610.33197
作者: Ergun Biçici,Mehmet Kaygusuz,Fatih Turan Gündoğdu,Emre Filiz,Çağatay Kızıldağ,Bünyamin Eymen Alagöz,Emincan Özcan,Sadi Çelik,Tolunay Öztürk,Ege Bulut,Said Aldemir
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 6 pages, 2 figures
Abstract:The migration of applications written for a different platforms like Android or iOS to HarmonyOS NEXT requires a complete rewrite in ArkTS. Although LLMs automate the translation of source constructs into ArkTS, syntactic compilation is not enough and functional assurance is a critical open challenge in HarmonyOS NEXT migration. We present HMOS~Migrator, an agentic pipeline that reads an application’s source code, generates an ArkTS project, and validates the result by replaying recorded interactions on a real device. We cast the scoring problem as quality estimation without a reference implementation; we score a migration using only the source application’s recorded behaviour and the artifacts produced by the pipeline. We define seven weighted metrics, each constructed so that reduced measurement cannot artificially raise the score. Over twelve runs across six open-source applications, the pipeline’s structural reproduction is high (completeness \pct83.3–\pct100.0), while on-device behaviour (\pct4.3–\pct63.1) and UI fidelity (\pct21.1–\pct63.2) remain the primary limits. Scores improve steadily over time: +0.45 ~points per day overall ( R^2=0.44 , p=0.019 ) and +0.47 over the model-free subset ( R^2=0.39 , p=0.029 ), with corresponding to fit angles of +8.4^\circ and +11.7^\circ respectively.
[AI-239] AOS-Bench: Can LLM Agents Manage Dynamic Workloads?
链接: https://arxiv.org/abs/2610.33178
作者: Erle Zhu,Genrui Zhang,Keyang Hu,Mingyu Su,Hongning Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents are moving from handling one request at a time to serving as persistent assistants to which users hand a stream of work: a build, a data-processing job and a bug fix may all be pending on the same machine at once. Such an agent must decide when to take on new tasks, how to delegate and overlap them, and how to stay within the machine’s CPU and memory. Existing benchmarks cannot measure this, because they evaluate each task in isolation. We introduce \benchname, which evaluates an agent system as a workload manager: verified tasks arrive over time and must be delivered under shared CPU and memory limits, and the same tasks run one at a time serve as a per-system reference. Across five model–harness pairs, isolated task success does not carry over evenly. On heavy workloads every system loses 13 to 23 points. On medium workloads, systems span 11 points in isolated success but 26 points in stream success, and two systems 0.6 points apart in isolation deliver 54% and 68% of the same tasks. Execution-policy interventions trace much of this gap to task intake: serial dispatch leaves 73% of heavy-workload tasks unclaimed. Yet more elaborate management does not necessarily help. A fixed concurrency of two comes within four points of the best agent-managed policy on both medium and heavy workloads, and under every policy median CPU use stays below 30% of the provisioned cores while tasks wait. These results identify workload management as a distinct capability for agents, and \benchname as a way to measure it. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.33178 [cs.AI] (or arXiv:2610.33178v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.33178 Focus to learn more arXiv-issued DOI via DataCite
[AI-240] Optimal Correction Sets for Provable Redress of Contestable Causal Discovery
链接: https://arxiv.org/abs/2610.33138
作者: Fabrizio Russo,Giuseppe Mazzotta,Carmine Dodaro,Francesco Ricca,Francesca Toni
类目: Artificial Intelligence (cs.AI)
备注: 41 pages including appendix
Abstract:Constraint-based causal discovery relies on conditional-independence (CI) tests that may conflict in finite samples, leaving no causal graph that satisfies them all. Causal Assumption-based Argumentation (ABA) makes these tests explicit as defeasible commitments, and offers correspondence guarantees with a subset of them, but its existing repair procedure releases facts heuristically and offers no formal response to a user’s contestation of a release. We argue that contestable causal discovery requires an inspectable repair object, a justification for it, and a predictable revision procedure for such contestations. We therefore formulate CI-fact repair as a minimum-cost correction problem and implement it in Answer Set Programming (ASP). Building on Causal ABA, we provide a compact and efficient ASP encoding of causal consistency, with soundness and completeness guarantees, and a two-way projection correspondence between optimal stable models and minimum-cost correction sets. These results establish fidelity to retained facts, subset-minimality and cost optimality of the repair, conflict-based justifications for released facts, and formally guaranteed redress under user contestations. An empirical evaluation supports the formal account: OptABA-PC retains more correct facts, reduces ambiguity in the compatible outputs, and improves graph reconstruction over Causal ABA on common benchmarks.
[AI-241] Conformance as a Distribution: Delegation Contracts Error Budgets and the Allocation of Evaluation Authority in Use-Based AI Regulation
链接: https://arxiv.org/abs/2610.33112
作者: John deVadoss
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Use-based AI regulation assigns obligations by deployment risk but lacks an evidentiary object to carry them. A benchmark score is a property of a model; conformance of a deployment is a failure rate under a declared distribution of contexts, bounded by a scope and subject to expiry. We take the delegation contract of earlier work (frozen configuration, scoped claim, executable revision rule) and study it under strategic play between a deployer, an auditor, and an enforcement point. We ask whether a declared context mixture can be made truthful; whether an audit can be unbiased against a system that detects evaluation traffic; whether a certification error budget can sustain a long-lived deployment; whether an aggregate audit can support a per-subgroup guarantee; and whether an incident metric that counts system failures but not fallback costs ranks arrangements correctly. For each we give a mechanism and a proposition: a proper-scoring penalty on the declared mixture bounds the excess failure of a best-responding deployer by O(\lambda^-1/2) in the penalty weight; a commit-reveal audit drawn from live traffic is unbiased by construction and grades the evaluator with the same sample; a halving error budget drives certification coverage below (1-p)^n while \alpha -investing sustains it under a marginal false-certification guarantee; a per-subgroup guarantee costs audit size scaling as 1/m in subgroup mass; and any failure-only incident metric prefers withdrawal at every fallback cost, even when total loss ranks the arrangements oppositely. Five preregistered synthetic studies test each claim; no language model or production system is evaluated. Enforcement sits at distribution, capability-access, monetization and risk-transfer choke points, expiry of the conformance record is a standard of care, and conformance-as-a-rate is set against the EU AI Act’s static conformity assessment.
[AI-242] Artificial Intelligence in Mechanical Engineering: Foundations Opportunities Challenges and Responsibilities
链接: https://arxiv.org/abs/2610.33105
作者: A. Cunha Jr,C. Nataraj,L. M. Lorenzo,H. Panetto,A. Scotti,J. V. C. Vargas
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG); Applications (stat.AP)
备注:
Abstract:Artificial intelligence (AI) is reshaping mechanical engineering (ME), from modeling, simulation, and optimization to diagnostics, design, and technical communication. Drawing on a round-table at the 28th ABCM International Congress on Mechanical Engineering (COBEM 2025), this article presents a structured perspective for the Brazilian and international ME communities. Rather than an exhaustive survey, it examines how AI can enhance engineering practice without compromising scientific rigor. We distinguish statistical learning from human cognition and discuss the capabilities, limitations, and emerging agentic forms of AI. We then consider industrial adoption, workforce changes, and AI-assisted scientific discovery, alongside the energy and infrastructure demands of AI computation and their implications for sustainability and thermal management. The discussion emphasizes the integration of physical knowledge, structured reasoning, and domain expertise into AI systems, particularly in safety-relevant applications. Topics include physics-informed and hybrid learning, physically constrained modeling, validation, uncertainty quantification, and certification. We also examine educational, epistemological, and ethical challenges, including curriculum reform, human-agent collaboration, professional accountability, and the preservation of engineering knowledge. We conclude with principles for responsible AI adoption in ME: physical consistency, rigorous validation, explicit treatment of uncertainty, energy sustainability, human oversight, and ethical responsibility. AI should complement, not replace, engineering judgment.
[AI-243] EGGPlacer: Energy-Guided Generative Chip Placement with Diffusion Models
链接: https://arxiv.org/abs/2610.33099
作者: Qi Zhou,Guojun Liu,Juanhong Zhao,Guangzhi Qi,Zehao Chen,Ming Lu
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 27 pages, 9 figures, 11 tables. Preprint
Abstract:Chip placement strongly affects routed wirelength and thermal behavior, yet these downstream metrics are difficult to evaluate during macro placement. Learning-based macro placers commonly optimize the macro-only half-perimeter wirelength (HPWL), denoted mHPWL, which ignores standard cell connectivity and spatial demand. In contrast, analytical mixed-size placers optimize differentiable full netlist wirelength and density objectives but are typically unaware of thermal effects. We propose EGGPlacer, an energy-guided diffusion framework that incorporates analytical placement feedback and thermal feedback into macro placement generation. An analytical placer acts as an online physical response model, exposing the standard cell distribution and a full netlist wirelength proxy induced by each intermediate macro placement, while a differentiable thermal model provides temperature feedback. These signals are formulated as composable energies and injected throughout reverse diffusion without retraining the underlying generative model. Experiments on public placement benchmarks show that EGGPlacer achieves competitive mixed-size HPWL relative to strong analytical and learning-based macro placers. In the end-to-end evaluation, EGGPlacer also obtains the lowest average routed wirelength and peak temperature among the evaluated methods. Ablation studies further demonstrate that analytical and thermal guidance provide distinct and complementary benefits across wirelength proxies, routed wirelength, and thermal objectives.
[AI-244] IRON-Meme: A Benchmark for Structured Understanding of Identities and Roles in Online Meme Narratives
链接: https://arxiv.org/abs/2610.33097
作者: Gabriella Bollici,Stefano De Giorgis,Wouter Van Atteveldt,Filip Ilievski
类目: Artificial Intelligence (cs.AI)
备注: Submitted to ARR
Abstract:As polarization grows on social media, understanding how political memes shape online narratives has become increasingly important. Previous research has relied on either qualitative frameworks that are hard to scale or computational methods that lack theoretical grounding. Both approaches treat narrative categories in isolation and focus exclusively on moral portrayals of identity. To address these limitations, we introduce a framework grounded in social science theories, encompassing three categories: social identities (such as gender, politics, and ethnicity), viewpoints (ingroup and outgroup perspectives), and narrative roles (moral character, intelligence, status, and power). Building on this framework, we present IRON-Meme, a human-annotated benchmark comprising 1,110 political memes across substantially more templates than existing benchmarks. Our analysis of IRON-Meme shows that most memes assign derogatory roles and that identities, roles, and templates are systematically interconnected. Evaluation of nine vision-language models indicates that frontier-scale and reasoning models achieve the highest performance, while accuracy on social identities and narrative roles decreases for less frequent labels. Incorporating template background knowledge does not enhance narrative role classification, suggesting that meme interpretation relies on how memes are adapted for specific communicative purposes. We make IRON-Meme and our code publicly available.
[AI-245] Inverse Modeling for Laser Pulse Shape Design in Inertial Confinement Fusion NEURIPS2026
链接: https://arxiv.org/abs/2610.33096
作者: Ricardo Luna Gutierrez,Vineet Gundecha,Rahman Ejaz,Varchas Gopalaswamy,Riccardo Betti,Sahand Ghorbanpour,Aarne Lees,Soumyendu Sarkar
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Systems and Control (eess.SY); High Energy Physics - Experiment (hep-ex)
备注: Initial version submitted to NeurIPS 2026 pending publication
Abstract:The achievement of practical fusion energy remains one of the most pressing unsolved scientific challenges, with immense implications for carbon-free power. A critical factor in the success of Inertial Confinement Fusion (ICF) is the design of a Laser Pulse Shape (LP) that can optimally drive implosions under stringent physical constraints. Traditional LP design relies on computationally expensive simulations and manual iterative refinement. We introduce the ICF Laser Pulse Shape Design System (LPDS), a generative inverse modeling framework that maps desired outcomes and target pellet configurations directly to optimized LPs. Crucially, we design a multi-objective loss function to ensure the generated LPs satisfy fundamental physical constraints and experimental requirements. Furthermore, we present constraint-conditioning, inpainting, and gradient-based LP editing mechanisms for fine-grained control over specific pulse characteristics during generation. Moreover, we study the transfer of our framework to a limited set of real experimental shots. Our method establishes a data-driven inverse design framework for LP design in ICF, contributing to the advancement of practical and sustainable fusion energy.
[AI-246] Beyond Correctness: Attribution of Natural-Language Documentation for Formal Libraries
链接: https://arxiv.org/abs/2610.33094
作者: Sirui Jiang,Dongming Xiang,Shichao Wu,Yuxuan Zhou,Yihong Wu,Fangze Zhu,Wang Lin,Guanjun Liu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 20 pages, 9 figures, 6 tables
Abstract:Formal languages are widely used in software engineering to specify and verify program properties. Because their mathematical notation and logic are hard to read, developers often rely on natural-language documentation to understand the functions in formal libraries. Yet a documentation statement can be provably true and still hold after the function’s behavior changes. We call a statement’s dependence on the function’s specific behavior attribution, and we give a method to check it. We design and implement DocAuth, a documentation-generation method that uses attribution to decide what to publish and attaches to each published sentence the formal property recovered from it, together with checkable evidence. On 150 functions from the Community Books library of the ACL2 theorem prover, three large language models produce proved but non-attributable properties for about a third of the functions. DocAuth matches or exceeds the attribution rate of a mutation-based filter while delivering 37% more attributable sentences. Compared with a pipeline that checks truth alone, it raises the share of published sentences that a second model confirms as attributable from 82.9% to 94.7%. These results show that provable truth does not guarantee that documentation reflects what a function actually does. Attribution complements truth as a criterion for generating and evaluating documentation for formal libraries.
[AI-247] AnalogPortBench: Benchmarking LLM s on Analog Circuit Porting
链接: https://arxiv.org/abs/2610.33088
作者: Saeid Rajabi,Aly Moussa,Vishal Saxena,Chengmo Yang,Satwik Patnaik
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 5 pages, 3 figures, 4 tables Code and data: this https URL
Abstract:Analog circuit design still relies heavily on expert knowledge, and porting proven designs to new process nodes is a recurring bottleneck. As a lower supply reduces available voltage headroom, a working topology may require redesign or fail to meet specifications. While large language models (LLMs) have begun to generate and optimize analog circuits, existing benchmarks focus on one process node and assume feasible specifications. To bridge this gap, we introduce AnalogPortBench, a benchmark of 54 tasks for porting ten circuits, from current mirrors to telescopic amplifiers, from a 0.18 \mu m process to the open-source SKY130 PDK. Tasks range from a direct port through a reduced supply and harder specifications to infeasible ones, and every feasible task has a reference design verified at five process corners. Evaluating six frontier LLMs at two reasoning levels, with and without simulation feedback, we find that models port simple circuits but rarely re-design headroom-sensitive amplifiers, seldom recognize infeasible specifications, and gain more from simulation feedback than from more samples or reasoning. Code and data are available at this https URL.
[AI-248] Feature Scaling Beats Architecture Choice in Offline Respiratory Screening for Low-Resource Clinics
链接: https://arxiv.org/abs/2610.33030
作者: Abubakar Chilala,Cleophas Kadima,Innocent Anyaele,Victor Olufemi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注:
Abstract:Deep models are the usual choice for respiratory sound classification, but in clinics without GPUs or connectivity that choice carries a cost. We compare classical and deep classifiers on a five-class respiratory-sound corpus of 1,211 clips (asthma, COPD, pneumonia, Bronchial, healthy) from 201 patient groups, with a 3.86:1 class imbalance, using pre-registered patient-grouped cross-validation (25 runs per configuration). An SVM over 30 summary features (MFCC, ZCR and RMS statistics) reaches macro recall 0.78 and 0.71 recall on the rarest class. A 1.13M-parameter convolutional recurrent network over log-Mel spectrograms reaches 0.75 to 0.77, and no difference between model families is resolved. The largest single effect we measure is a preprocessing effect rather than an architectural one. Standardizing the SVM’s features, a routine step for RBF kernels, moves its macro recall from 0.51 to 0.78, a 27-point swing wider than any gap between model families here. Augmentation did not help: SMOTE, waveform augmentation and SpecAugment have no measurable effect. Latency does not separate the models (51 to 74 ms per clip on a CPU-only laptop), but the classical models avoid a 449 MB PyTorch runtime. Clip-level splits, which let a patient’s recordings fall on both sides, inflate macro recall by 7.5 to 9.8 points. Recording source is strongly tied to label (Cramer’s V = 0.65 between native sampling rate and class), and in an exploratory analysis clips recorded at a class’s minority rate are mostly assigned to classes typical of that rate; we state this confound as the paper’s main limitation.
[AI-249] DairySocialTwin: Auditing the video-to-graph evidence chain in cattle social networks
链接: https://arxiv.org/abs/2610.32989
作者: Yiwen Huang,Suresh Raja Neethirajan
类目: ocial and Information Networks (cs.SI); Artificial Intelligence (cs.AI); Image and Video Processing (eess.IV); Quantitative Methods (q-bio.QM)
备注: 28 pages, 6 figures, 4 tables
Abstract:Video-derived social graphs can appear precise while inheriting uncertainty from detection, tracking, classification, and aggregation. We present DairySocialTwin, a graph-centered audit framework applied to 203 videos (122 h 23 min) from 15 views at three commercial dairy farms. Curation retained 447 of 839 candidate clips; 400 entered a frozen source-clip index, including 71 legacy clips with unavailable farm and time metadata. Five models were compared on a development split, with an NRI-style model retained as an exploratory trade-off between class balance and dyadic output. In a 90-min manual audit, reconciliation improved switches, fragmentation, and continuous-track duration across three gap definitions, but constrained identity capacity and inherited detector errors limited claims of identity accuracy. Matched manual and reconciled summaries showed coarse agreement and assignment sensitivity. Camera-specific semantics, observation opportunity, and identity-summary-to-video trace-back made these dependencies inspectable. DairySocialTwin supports auditable relational inference without claiming permanent identity, calibrated behavior, or biological diagnosis.
[AI-250] 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); Artificial Intelligence (cs.AI)
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.
[AI-251] Provably Efficient Q-learning for Robust Reinforcement Learning
链接: https://arxiv.org/abs/2610.35656
作者: Yuchen Jiao,Gen Li
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Statistics Theory (math.ST)
备注:
Abstract:Standard reinforcement learning (RL) algorithms based on Markov Decision Processes (MDPs) are vulnerable to sim-to-real gaps. To address this issue, distributionally robust RL algorithms based on robust MDPs (RMDPs) have been proposed, for ensuring that agents can perform reliably to unpredictable conditions in real-world environments. Despite of the rich literature, \revisionan efficient model-free method is still lacking for robust RL, particularly for Markovian data, which is more relevant in practical applications. In this work, we take a step towards settling this problem by proposing a sample efficient (asynchronous Q-learning) algorithm for robust RL. In particular, consider a \gamma -discounted infinite-horizon RMDP with state space \mathcalS and action space \mathcalA , where the uncertainty set is specified by the total variation (TV) distance with uncertainty level \sigma . Assuming access to a single trajectory of Markovian samples induced by a behavior policy and the nominal transition kernel, the sample complexity of the proposed Q-learning algorithm — to achieve an entrywise \varepsilon -approximation of the optimal Q-function — is bounded by \frac1\mu_\mathsfmin\varepsilon^2(1-\gamma)^3\max\1-\gamma,\sigma\ + \fract_\mathsfmix\mu_\mathsfmin(1-\gamma) up to logarithmic factors. Here, t_\mathsfmix and \mu_\mathsfmin denote the mixing time and the minimum state-action occupancy probability of the sample trajectory, respectively. This result is nearly minimax optimal, with an additional factor of \frac11-\gamma . Moreover, it achieves a significant improvement over existing bounds.
[AI-252] KAPPA: Checking the Proofs in Statistics Papers with Lean 4 and Language Model Agents
链接: https://arxiv.org/abs/2610.35596
作者: Zihua She,Xiao Wang
类目: Computation (stat.CO); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: 31 pages, 2 figures. Software: this https URL
Abstract:Papers in statistical theory rest on long proofs that few readers check line by line. We present KAPPA (Kit of Agents for Proof-checking Papers with a proof Assistant), open-source software that organizes language model agents, run by Claude Code, to translate the proofs of a LaTeX paper into the proof assistant Lean 4 with its library Mathlib and check them there, result by result. Each numbered result receives one of four verdicts: verified unconditionally; verified conditionally, with every cited or unproved input stated as a named hypothesis; issue recorded, with its place in the paper and the side at fault; or unfinished, with the reason. The kit follows the paper’s own proof step by step, rather than finding a new one. Separate agents write and review, each statement is compared with the paper before it is proved and fingerprinted afterwards, and every verdict rests on evidence that can be rerun without a language model. The kit runs unattended for hours, waits out usage limits, re-checks only what a revision changed and reports its cost by role, model and result. We describe the design, walk through a complete run, and apply the kit to four example papers with answer keys and to three published statistical papers, checked blind in their versions from before public corrections. In all three, the kit found the published error at the step that the correction names, together with further problems that we confirmed by hand. Given the same papers directly, the same language model working with Lean classified fewer results correctly on the longest paper and reported some statements narrower than the paper’s as proved; no verdict of the kit claimed more than its Lean evidence showed.
[AI-253] Autonomous Structural Discovery Of Bioprocess Dynamics And Kinetics Via LLM -Driven Agents
链接: https://arxiv.org/abs/2610.35391
作者: Erdal Aydin,Yaman Arkun
类目: Optimization and Control (math.OC); Artificial Intelligence (cs.AI)
备注:
Abstract:High-fidelity digital twins in continuous biomanufacturing are often bottlenecked by incomplete kinetic models, while standard data-driven algorithms (e.g., SINDy, PINNs) lack interpretability or impose rigid physical assumptions. This paper introduces a hybrid closed-loop architecture for autonomous kinetic state-space discovery driven by a Large Language Model (LLM) agent. The framework employs a sequential “isolation-peeling” methodology across orthogonal operational regimes. This algorithmic peeling mathematically decouples entangled multiplicative (inhibitory) and additive (catalytic) biological rate law. Grounded in a Two-Level Retrieval-Augmented Generation (RAG) library, the framework utilizes Moving Horizon Estimation (MHE) and logarithmic decomposition to decouple complex cell growth kinetics into orthogonal multiplicative (inhibitory) and additive (regulatory) residual channels. When statistical degeneracy occurs an Active Design of Experiments (DoE) controller dynamically triggers targeted perturbations to decorrelate competing model hypotheses. Evaluated on a continuous stirred-tank reactor exhibiting unmodeled Haldane inhibition, Luong product toxicity, and secondary metabolite autocatalysis, the framework autonomously expanded the state space and recovered ground-truth parameters with under 6% error without a priori kinetic knowledge.
[AI-254] Generative AI in Turbulence: Foundations Applications Validations and Benchmarks
链接: https://arxiv.org/abs/2610.35123
作者: Luca Biferale,Michele Buzzicotti
类目: Fluid Dynamics (physics.flu-dyn); Statistical Mechanics (cond-mat.stat-mech); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Chaotic Dynamics (nlin.CD)
备注: submitted to JFM-Perspective
Abstract:Generative AI offers new approaches to turbulence problems in which incomplete observations or unresolved d.o.f. admit multiple physically plausible solutions. Potential applications include stochastic closures, super-resolution, data assimilation, uncertainty quantification, and the generation of Eulerian and Lagrangian flow ensembles. However, turbulence is not a conventional image-generation problem. Its multiscale dynamics, intermittency, extreme events, rough velocity fields and intrinsically stochastic transport make visual realism, pointwise error, and low-order spectral agreement insufficient measures of physical fidelity. We examine where generative AI provide novel capabilities with emphasis on diffusion models, score-based methods, and related probability-transport formulations. We discuss their use in unconditional generation, probabilistic reconstruction, reduced-order modeling, temporal evolution, and mesh-free representations. We argue that validation must address turbulence properties: energy transfer, structure functions, anomalous scaling, pdf tails, coherent structures, conservation laws, temporal consistency, ensemble diversity and generalization. Finally, we discuss potential limitations in validation, baseline, data availability and reproducibility, and the need for a benchmark ecosystem combining open data, standardized challenges, hidden tests, physics-aware metrics, and documented failures. This paper serves as a bridge between two communities. For AI practitioners, we provide a fair and accessible introduction to turbulence, essential for appreciating the challenges posed by the field and for assessing the physical relevance of generative models. For turbulent practitioners, we provide an introduction to the key technological aspects of Generative AI, clarifying what these tools actually do, and where their potential and current limitations lie.
[AI-255] From Single Spins to Ensembles: A Calibrated Symbolic Regression Pipeline for NV Spectroscopy
链接: https://arxiv.org/abs/2610.35016
作者: Evan Zhang,Zhenghang Zhi,Zewei Wang,Zeheng Wang
类目: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Characterizing nitrogen-vacancy (NV) ensembles requires connecting their collective spectra to individual centers that experience different local environments. We develop a symbolic-regression pipeline that carries single-center spin structure into a calibrated ensemble model by generating individual spectral responses and combining them with explicit weights. A constrained search selects Hermitian operator terms from single-center optically detected magnetic resonance spectra, fits their coefficients, and retains the selected expressions as source models for the ensemble. Ensemble calibration then accounts for resonance shifts, local disorder, orientation, and optical response while keeping the individual spectral contributions accessible. On public single-center and ensemble data, the zero-field ensemble held-frequency root-mean-square error decreases from 0.008835 to 0.003872 in normalized photoluminescence, a 56.2% improvement over matched uncorrected composition. In controlled QuTiP ensembles of 128 sites spanning four NV orientations, models calibrated at 0 and 3 mT predict an independently held-out 5.17 mT spectrum with errors of 0.000207-0.000228 across three random seeds, comparable to the 0.000200 noise standard deviation. Resolving the simulated response into its individual-center contributions shows how local spectral diversity combines into the ensemble line shape. By retaining the source Hamiltonians while calibrating sample-dependent spectral differences, the pipeline connects local spin physics to collective optical readout and makes the contributions to an ensemble line shape available for physical interpretation.
[AI-256] Intrinsic Goals for Autonomous Agents : Hierarchical Exploration Captures Future Volitional Action in Drosophila
链接: https://arxiv.org/abs/2610.34638
作者: Declan Kutscher,Reece Keller,Alyn Kirsch,Felix Pei,Aran Nayebi
类目: Neurons and Cognition (q-bio.NC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO)
备注: 27 pages, 5 figures
Abstract:Animals have a remarkable capacity for autonomy—driven by intrinsic motivation—without external sensory cues or extrinsic objectives, yet the computational mechanisms behind this ability largely remain a mystery. We combine whole-brain recordings in the fruit fly (Drosophila melanogaster), which reveal neural representations of spontaneous exploratory decisions more than 10 seconds in advance, with a neuromechanical digital twin to test whether optimizing intrinsic goals yields realistically embodied artificial agents that quantitatively match fly behavior and internal representations of future volitional action during exploration. Our hierarchical policy selects from a discrete library of low-level motor primitives, consistent with ethological theories of behavioral syllables. We train agents without external reward using 20 intrinsic motivation algorithms spanning curiosity, entropy, and competency methods. The agents are evaluated in a virtual tethered-ball assay modeled on the fly experiment, comparing spontaneous locomotor statistics and internal representations and decoding future turn direction from kinematics and neural activity on the same timescales observed in flies. DIAYN, a hierarchical competency algorithm learning discrete policies that maximize behavioral entropy and the separability of induced sensory observations, achieves the strongest joint behavioral and neural prospective coding performance relative to flies. Its learned policies closely capture individual variability in fly exploration strategies, suggesting that this diversity serves to maximize state-space coverage at the population level. DIAYN’s latent activity also closely captures the sum-difference rule for interhemispheric encoding of kinematics in biological flies. These results establish hierarchical, intrinsically motivated agents as a powerful framework for reverse-engineering animal autonomy.
[AI-257] Acute Stroke Classification Using Asymmetric Data Standardization and Orthogonal-Focal Loss Optimized CNN-ViT Cross-Attention Fusion: A Multicenter External Validation Study
链接: https://arxiv.org/abs/2610.34529
作者: Tansel Uyar
类目: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI)
备注: All research data has been uploaded to Zenodo (DOI: https://doi.org/10.5281/zenodo.22132268 ). The data presented in this study are available on request from the corresponding author due to ongoing research studies
Abstract:Background/Objectives: Rapid differentiation of ischemic and hemorrhagic stroke subtypes is vital for acute management. Although current deep learning models achieve high single-dataset accuracy, they suffer from “shortcut learning” and domain shift in external validation. This study proposes a hierarchical computer-aided diagnostic (CAD) system mimicking emergency triage. Methods: We introduce a 3-channel Radiology-Focused Preprocessing pipeline that feeds radiological Hounsfield Unit (HU) windows into Convolutional Neural Network (CNN) and Vision Transformer (ViT) backbones via asymmetric formats (.mat and .png). The architecture utilizes a novel joint cross-attention fusion network optimized by an Integrated Orthogonal-Focal Loss. Results: Our system achieved a 95.48% F1-score on the internal stroke dataset. However, multi-center external validation (AISD, BHSD, PhysioNet) using a Mask Anchor filter revealed a drop to the 58-62% range, exposing the “internal validation illusion” caused by hardware noise. Critically, models exhibited high precision (77%) but low sensitivity (57%) for ischemic stroke, proving prone to hemorrhage overdiagnosis. Multi-seed variance analysis demonstrated that the DenseNet201+ViT backbone yielded the lowest deviation (0.59) and highest stability, making it the most reliable architecture for real-world clinical integration. Conclusions: To overcome diagnostic challenges, future research must focus on multimodal architectures that combine 3D volumetric extraction with clinical data.
[AI-258] Certifying Distributionally Robust Decisions from Projection Observations
链接: https://arxiv.org/abs/2610.34493
作者: Pouya Ahadi,Kamran Paynabar
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Optimization and Control (math.OC)
备注:
Abstract:A scalar measurement can determine a robust decision without identifying the latent covariance or every decision risk. We study finite Gaussian projection experiments with a fixed mean-plus-standard-deviation target. The observation laws define a set of indistinguishable moment models. We exhibit decisions that agree throughout that set despite unidentified risk values, and give a confidence-set certificate that returns a decision or an unresolved outcome. We prove soundness, necessary abstention for indistinguishable models with incompatible optima, and a fixed-checkpoint certification guarantee. A scalar profile reduces nonlinear risk comparisons to convex inner problems. Across seven synthetic instances, jointly calibrated moment regions give sufficient budgets 21 to 52 times explicit expected-sample information lower bounds. Separately measured stopping times are lower, with zero decision errors in 1,900 simulated streams. These comparisons do not constitute matching bounds, and the guarantees require Gaussian replication.
[AI-259] In-Context Adapter Generation for Compositional Generalization in Large Audio Language Models
链接: https://arxiv.org/abs/2610.34413
作者: Yiwei Guo,Siqi Huang,Bohan Li,Jing Peng,Hankun Wang,Zhihan Li,Kai Yu
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注: 20 pages, 13 figures, 9 tables
Abstract:Large audio language models (LALMs) have achieved strong performance on standard speech and audio understanding tasks, yet their ability to systematically generalize to novel task compositions remains underexplored. Taking a meta-learning view, we extend Text-to-LoRA (T2L) to LALMs, generating instruction-conditioned LoRA adapters with a Transformer hypernetwork rather than learning a single fixed adapter shared across instructions. To exploit fine-grained task information, we introduce in-context adapter generation (ICAG), a token-level hypernetwork architecture; we call the resulting method T2L-ICAG. We evaluate this approach on two representative LALMs using SIFT-50M for supervised fine-tuning and construct a controlled compositional evaluation suite covering five task classes. Across two LALM backbones, T2L-ICAG consistently improves over static LoRA and also the global embedding-based T2L baseline. Coverage and capacity analyses further show that these gains are larger on structurally out-of-domain tasks and cannot be explained by parameter count alone. These results suggest that generating task-specific adapters from instructions is a promising direction for improving compositional generalization of LALMs.
[AI-260] Deep Scientific Reasoning under Physical Constraints: Structure-Aware Spectrum Prediction ICML2026
链接: https://arxiv.org/abs/2610.34021
作者: Yingheng Wang,Tao Yu,Shufeng Kong,Francesco Ricci,John M. Gregoire,Carla P. Gomes
类目: Materials Science (cond-mat.mtrl-sci); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Published in ICML 2026. Code and data are available on the project page: this https URL
Abstract:Structured scientific spectra encode rich physical information while obeying hard constraints, such as conservation laws and characteristic spectral geometry. Accurate prediction of these spectra is central to materials discovery, yet existing methods often treat them as unconstrained sequences and therefore fail to enforce the underlying physical structure. Taking electronic density of states (eDOS) as the prototypical example, we introduce DeepSciReasoner, a general paradigm for predicting scientific spectra under physical constraints. The framework combines structure-aware spectrum decoding with constraint-preserving physical reasoning, allowing predictions to capture rich spectral structure while respecting the underlying physics. We evaluate DeepSciReasoner on eDOS, phonon density of states (phDOS), X-ray absorption near-edge structure (XANES), and Raman spectra, where it substantially improves prediction accuracy while maintaining physical consistency. These results establish DeepSciReasoner as a reusable blueprint for structured scientific spectrum prediction under hard physical constraints.
[AI-261] ProtAgent : Post-Training Compact Protein Tool Agents with Interaction-Level Credit Assignment
链接: https://arxiv.org/abs/2610.33746
作者: Yuxiang Wang,Baojian Zhou,Qifang Zhao,Keyue Jiang,Yanghua Xiao,Xiaoxiao Xu
类目: Quantitative Methods (q-bio.QM); Artificial Intelligence (cs.AI)
备注: 20 pages, 2 figures
Abstract:Protein annotation agents must choose suitable bioinformatics tools, interpret their outputs, and use the resulting evidence in the final answer. Most existing systems rely on frontier-scale models. We ask whether a compact model can instead learn this behavior through post-training with real tool interaction. We present ProtAgent, a two-stage pipeline that uses supervised fine-tuning (SFT) to learn the tool interface and standard on-policy REINFORCE to improve the interaction policy. Intermediate tool calls receive action rewards, and the final response receives a terminal answer reward. Step-level returns assign these rewards to the responses that produced them and to preceding actions. The episode-level baseline instead assigns one episode return to every action. On a fixed closed-loop protocol with real bioinformatics tools, ProtAgent-8B achieves 61.00% success under the annotation-coverage metric, compared with 41.60% for DeepSeek-V3.2-Thinking 671B and 60.80% for the strongest evaluated frontier system. In the three-seed comparison, mean success is 42.70% with episode-level returns and 61.40% with step-level returns under identity return normalization; the corresponding group-relative results are 36.50% and 58.20%. The metric measures recovery of reference annotations under this protocol; it does not by itself establish biological correctness or complete evidence grounding.
[AI-262] Exploring Joint-Embedding Predictive Architectures for Self-Supervised Representation Learning of Powder X-ray Diffraction Patterns AAAI2026
链接: https://arxiv.org/abs/2610.33310
作者: Ming-Chiang Chang,Brian DeCost
类目: Materials Science (cond-mat.mtrl-sci); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 8 pages, 6 figures, 1 table. Accepted by AAAI 2026 Fall Symposium Series Workshop “Parsimonious Scientific Machine Learning: Accelerating Scientific Discovery with AI under Data and Compute Constraints”
Abstract:Powder X-ray diffraction (XRD) is the primary high-throughput method for resolving crystal structures, a cornerstone of materials science. As high-throughput synthesis and self-driving laboratories advance, automated XRD pattern analysis has become vital for accelerating materials discovery. However, traditional matrix factorization methods fail when lattice variations cause non-linear peak shifts. At the same time, supervised learning approaches face a combinatorial explosion when handling multi-phase mixtures. To overcome these challenges, we adopt a self-supervised Joint Embedding Predictive Architecture (JEPA) to learn representation embeddings from 1D diffraction patterns. By optimizing a masked self-prediction loss directly within the latent space, the framework captures high-level crystallographic semantics. Downstream evaluations show that lightweight attentive probes trained on frozen JEPA representations achieve space group classification and lattice parameter estimation accuracy close to much larger supervised vision transformers. Furthermore, latent probing reveals that the learned representation space naturally organizes physical transformations, isolating smooth trajectories for peak broadening. Interestingly, decoding along latent paths between single-phase crystal representations yields realistic multi-phase mixture patterns, demonstrating an emergent continuous structure learned without explicit multi-phase supervision.
[AI-263] RADAgent : Automating RADseq Pipelines with Private Open-Source LLM s
链接: https://arxiv.org/abs/2610.33009
作者: Qiming Guo,Malaka Sudhakara Bava Sukruth Reddy,Honggang Zhao,Ye Wang,Chen Pan,Wenlu Wang
类目: Genomics (q-bio.GN); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: Accepted at IEEE BIBM 2026. Code: this https URL
Abstract:Restriction site-associated DNA sequencing (RADseq) is a family of reduced-representation sequencing methods. It uses restriction enzymes to sample subsets of the genome, therefore allowing multiple individuals to be sequenced and genotyped together at low cost. RADseq is widely used for population genomics, evolutionary biology, marker discovery, and the development of SNP panels, but complete analysis still requires expert coordination across multiple tools and processing stages. Traditional RADseq analysis requires expert coordination of heterogeneous bioinformatics tools, intermediate artifacts, quality-control checks, and parameter-dependent decisions across multiple stages. Although individual steps can be scripted, executing the complete workflow still requires human oversight for state tracking, result validation, and failure handling. To automate this expert-driven coordination, we present RADAgent, which introduces a modular agentic method for reliable RADseq workflow execution with an open-weight LLM. RADAgent maps the natural boundaries of the RADseq workflow into 16 independently executable and verifiable agent modules and uses a model-independent runtime for bounded tool execution, on-disk verification, and failure recovery. We evaluate RADAgent on a real population-genomics dataset containing 476 samples from 19 populations. The results demonstrate that RADAgent reliably automates the complete RADseq workflow with an open-weight LLM while keeping genomic data and model execution within the local computing environment. The code is publicly available at this https URL.
机器学习
[LG-0] Auto-Research Arena: Benchmarking and Advancing Auto-Research in Language Modeling
链接: https://arxiv.org/abs/2610.35866
作者: Xi Chen,Chongjian Ge,Yiran Xu,Yiqun Mei,Henrik I. Christensen,Hao Tan,Ziwen Chen,Hanwen Jiang
类目: Machine Learning (cs.LG)
*备注: The code is available at this https URL
Abstract:As autonomous research systems become more capable, benchmarks must both evaluate their performance and support meaningful research progress. We introduce Auto-Research Arena, a multi-target benchmark for evaluating autonomous research systems and discovering ideas in language modeling. Arena supports diverse research objectives within a shared experimental environment by separating admissibility from improvement: model quality and resource constraints determine whether a result qualifies, while the selected target measures progress among admissible results. We instantiate Arena in language model pretraining with nine targets spanning model, data, training, and inference efficiency, and evaluate seven research harnesses using two foundation models. Arena clearly differentiates existing harnesses across research objectives, with several sophisticated methods outperformed by our simple sequential search baseline, indicating substantial headroom in current auto-research systems. Beyond evaluation, Arena reveals a substantially broader frontier of model and system ideas aligned with active research directions. Reusing these discoveries in a new quality-driven search yields lower training loss under matched data and FLOPs and lower held-out BPB at the same model size. Together, Arena provides a common framework for benchmarking autonomous research and expanding its scope in language modeling. The code is available at this https URL
[LG-1] Work Agents : Systems Capabilities and Optimization
链接: https://arxiv.org/abs/2610.35845
作者: Zishan Xu,Enjun Du,Mingqi Gao,Cheng Yang,Junhao Shen,Zirong Chen,Zhiyuan Yao,Yifu Guo,Yuquan Lu,Jinyang Huang,Zhengxi Lu,Siyi Liu,Ziyu Zheng,Yongqi Zhang,Difan Zou,Weiwen Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:As foundation models evolve from information generation to digital work execution, agents are increasingly capable of completing complex tasks across tools, applications, and workflows. However, existing research lacks a unified perspective on how agents reliably accomplish delegated work. To address this gap, we present a work-centric, system-level perspective that views work agents as systems combining foundation models and runtime harnesses to execute tasks in stateful workspaces and deliver verifiable outcomes. We introduce two complementary dimensions, work persistence and work openness, to characterize working conditions, and classify digital work into five forms according to task scope and organization. Building on this framework, we identify five core capability families required for reliable work execution and examine approaches to model optimization, harness optimization, and their joint optimization. We further analyze existing evaluation methods and metrics from the perspective of work outcomes. Finally, we discuss future research directions, including sustained work delegation, adaptation to open environments, and the interaction between work execution and capability improvement. This paper provides a unified framework for understanding, optimizing, and evaluating work agents, supporting the development of systems capable of delivering reliable, verifiable, and practically valuable work outcomes.
[LG-2] Information-Theoretic Secure Aggregation With Heterogeneous Client Quantization
链接: https://arxiv.org/abs/2610.35833
作者: Shreya Meel,Sennur Ulukus,Itzhak Tamo
类目: Information Theory (cs.IT); Cryptography and Security (cs.CR); Machine Learning (cs.LG); Networking and Internet Architecture (cs.NI); Signal Processing (eess.SP)
*备注:
Abstract:In federated learning (FL), secure aggregation enables a server to compute the sum of local model updates of N clients, without revealing any information on the individual model updates to the server, beyond their sum. Different clients may be subject to heterogeneous bandwidth constraints, and consequently may choose different numbers of quantization levels to represent the information they send. Most secure aggregation works require the clients to project the values of their update parameters to a common finite group (or field), which does not directly accommodate heterogeneous precision constraints on their transmitted masked updates. Existing approaches of secure aggregation with heterogeneous quantization levels reveal partial aggregates of model-updates, and can reveal segments of individual model updates, under certain dropout and client-server collusion patterns. To prevent this privacy breach, we propose an information-theoretic secure aggregation scheme that tolerates client drop-outs, and up to T semi-honest clients that collude with each other and with the server, under heterogeneous quantization at clients. Our protocol utilizes the (T+1, N) threshold black-box secret sharing construction, where secret shares of the client updates, despite belonging to distinct algebraic groups owing to different quantization, are generated from identical linear integer systems.
[LG-3] FOCUS: Active Critic Learning for Long-Horizon Language Model Agents
链接: https://arxiv.org/abs/2610.35801
作者: Zhengping Jiang,William Jurayj,Liaoyaqi Wang,Chunsheng Zuo,Anqi Liu,Benjamin Van Durme
类目: Machine Learning (cs.LG)
*备注: 32 pages, 5 figures
Abstract:Sparse outcome rewards make it difficult for long-horizon language-model agents to assign credit to earlier actions. Learned critics provide value estimates for intermediate states, but these estimates can be inaccurate, while estimating intermediate values with repetitive continuation samples from points within a trajectory incurs substantial costs. To mitigate these issues, we introduce FOCUS, an RL method that uses selective branching to help the critic network learn its value estimator. The rollout controller facilitates active learning for the critic by sampling continuations from states that are expected to most reduce the critic’s estimation error. These refined value estimates then score the segment between consecutive states with a TD( \lambda ) advantage, while the sampled evidence supplies training targets for the critic. FOCUS supports long-horizon agentic reinforcement learning through a modular context-management interface that decouples context management from rollout allocation. Our experiments show that using compaction-induced segment boundaries for value estimation and credit assignment, FOCUS outperforms GRPO and VinePPO with matched per-step generation budgets on legal reasoning, deep search, and code localization benchmarks, improving held-out accuracy at Qwen3.5-9B scale by 7.2 points over the strongest baseline on BrowseComp-Plus (55.4 vs. 48.2) and by 4.1 points on DeonticBench (56.5 vs. 52.4).
[LG-4] Selective Tie Learning: Mitigating Spurious Correlations in Preference Optimization
链接: https://arxiv.org/abs/2610.35787
作者: Christian Moya,Purav Matlia,Elliott Thornley,Guang Lin
类目: Machine Learning (cs.LG)
*备注:
Abstract:Preference optimization can learn spurious correlations, and augmenting the preference data with ties (equally preferred responses) can reduce spurious reliance. However, ties differ in their corrective effect, making their selection for augmentation important. For log-linear policies, we derive each tie’s influence on spurious reliance and show that taking the largest predicted reductions is first-order optimal among subsets of equal size. Computing these influences requires knowledge of the spurious features and the curvature of the training objective, which is often unavailable or expensive to estimate. To this end, we explore using the magnitude of a tie’s implicit reward margin and characterize the conditions under which it is a useful proxy for predicted influence. To test the theory and its qualitative transfer to neural models, we estimate influence with a counterfactual proxy for spurious reliance and use a generalized Gauss–Newton approximation to curvature for language models. Across log-linear policies, neural models, and large language models, our selection rules consistently achieve comparable reductions in spurious reliance with fewer ties than uniform selection.
[LG-5] A theory of noisy linear recurrent neural network learning dynamics
链接: https://arxiv.org/abs/2610.35785
作者: John J. Vastola,Kanaka Rajan
类目: Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
*备注: 27 pages, 6 figures
Abstract:Biological neural networks are noisy: their dynamics involve difficult-to-predict fluctuations which are likely due not just to chaos, but also to a variety of low-level biophysical processes. Despite the omnipresence of intrinsic noise, its theoretical consequences are incompletely understood, especially in the context of learning. For example, does noise meaningfully shape representation learning, or is it better viewed as a nuisance factor whose presence or absence changes little about what solution is found and when? Here, we argue that intrinsic noise can significantly modulate representation learning in recurrent neural networks (RNNs), since it can strongly break solution-space degeneracies and in some cases control whether a non-degenerate solution exists at all. We show that this is true for an analytically tractable model of noisy RNN learning with a linear activation function and white noise inputs. We find that, in the so-called rich regime (where weights are initialized with small values), noise strongly modulates the directions in which the read-in, readout, and recurrent weights align. Moreover, we find that noise produces a kind of phase transition: for sufficiently high noise, a non-degenerate solution is not even optimal if parameter scales are limited (e.g., via regularization). These results extend recent analytic work on RNN representation learning and show that noise can act as a fundamental determinant of representations.
[LG-6] LUCID: Local Understanding of Covariate Importance with Conditional Density Estimation
链接: https://arxiv.org/abs/2610.35740
作者: Yixiao Wang,Srikar Katta,Jon Donnelly,Emanuele Borgonovo,Cynthia Rudin
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Computing local-level variable importance (VI) is critical for understanding machine learning models. Local permutation-based VI metrics offer a simple yet powerful approach, measuring how much a model’s output changes when perturbing its inputs. However, existing perturbation strategies often ignore feature dependence or account for it only through restricted approximations. Perturbing a feature without fully respecting this dependence can generate unrealistic counterfactuals, thereby evaluating the model in regions unsupported by the observed data distribution. We address this gap by introducing a new model-agnostic local VI metric and estimator, Local Understanding of Covariate Importance with conditional Density estimation (LUCID). LUCID perturbs features of interest using plausible values drawn from the full conditional distribution given the remaining observed features. We establish theoretical connections between LUCID and existing VI measures, and provide an estimator with convergence guarantees and uncertainty quantification. Finally, we demonstrate LUCID’s stable performance and practical utility across simulated and real-world examples.
[LG-7] RACE: Memory-Efficient Time Series Class-Incremental Learning via Temporal-Spectral Reconstruction
链接: https://arxiv.org/abs/2610.35727
作者: Junhui Mao,Junru Zhang,Liming Ye,Cheng Peng,Haoran Shi,Xiaoli Tang,Baokun Wang,Bin Dou,Yabo Dong,Han Yu
类目: Machine Learning (cs.LG)
*备注: 9 pages, 5 figures, 4 tables
Abstract:Web-connected sensing applications continuously generate multivariate time series with evolving behavioral categories, making Time Series Class-Incremental Learning (TSCIL) essential for adapting to new classes without forgetting previously acquired knowledge. However, preserving diverse and evolving class distributions under strict privacy and memory constraints remains a fundamental challenge. Existing approaches often face a trade-off between retaining sufficient historical knowledge for accurate recognition and minimizing the memory required for continual adaptation. To address these limitations, we propose TRACE, a memory-efficient, replay-free framework that preserves class-specific knowledge through generative modeling rather than sample rehearsal. TRACE employs a compact, frozen temporal-spectral representation and lightweight class-specific autoencoders to capture complex intra-class variations. By independently preserving each class model, TRACE avoids recursive generation and catastrophic forgetting, while enabling direct classification through reconstruction-based scoring without retaining historical samples or large statistical matrices. Extensive experiments across five TSCIL benchmarks demonstrate that TRACE achieves competitive classification accuracy while reducing mean retained memory by approximately 98.26% compared with TS-ACL, the strongest evaluated analytical baseline, and achieving an approximately 33.43% relative improvement in accuracy on GrabMyo.
[LG-8] Quantifying the failure of symmetry learning in linear diffusion models due to initialization and finite-sample effects NEURIPS2026
链接: https://arxiv.org/abs/2610.35720
作者: John J. Vastola,Kanaka Rajan
类目: Machine Learning (cs.LG)
*备注: Accepted to the NeurIPS 2026 Workshop on Symmetry and Geometry in Neural Representations (NeurReps)
Abstract:Data distributions often possess one or more symmetries: for example, all orientations of any specific dog are included in the distribution of all possible dog pictures. Despite the omnipresence of such symmetries, modern generative modeling approaches may fail to incorporate or learn them, which can lead to lower sample efficiency and poorer generalization. To what extent do diffusion models fail to learn data distribution symmetries, and what sources of symmetry breaking are most responsible? Here, we answer this question in the theoretically tractable setting of linear diffusion models. We show that of three potential sources of symmetry breaking—a non-symmetric initial weight matrix, a noisy learning target, and finite-sample noise—target noise produces the most robust symmetry breaking, since it persists even in the sample-rich regime. Using local laws from random matrix theory, we derive formulas for average symmetry error in fairly general cases involving symmetry within a subspace and show that smaller symmetries are harder to learn.
[LG-9] Detecting bus delays and service anomalies in Niterói with autoencoder latent feature representations
链接: https://arxiv.org/abs/2610.35710
作者: Matheus Germano,Claudio E. C. Campelo
类目: Machine Learning (cs.LG)
*备注: 14 pages, 9 figures, 11 tables. Full report for the NetMob 2026 Data Challenge (Niterói, Brazil). Winner of the Best Poster award
Abstract:Bus riders judge service by two failures: single trips that run late, and days on which the whole system misbehaves. We ask whether one unsupervised representation of raw GPS telemetry can serve both diagnoses on the NetMob 2026 Niterói bus dataset. An autoencoder compresses each trip into a 64-dimensional latent vector; the frozen encoder feeds supervised delay classifiers, and the reconstruction error serves as an anomaly score. Two findings are negative. First, the latent representation does not improve delay detection: gradient boosting on per-trip summary features reaches 0.846 \pm 0.013 ROC-AUC against 0.838 \pm 0.016 for the best latent setup, and the gap becomes significant once the preprocessing filters are relaxed. Second, two of the three external checks of the anomaly score were circular. Retraining without rainfall removes the rain association ( 5.4% against 5.0% flagged, p = 0.84 ) and an apparent citywide disruption on 23 March. What survives is a day-level signal stable across three autoencoders (Spearman \rho \geq 0.96 ), driven mostly by day type, with degraded-telemetry days still separable within weekday and weekend strata. The alarm threshold limits detection more than the score does: a pooled percentile cut flags 5.7% of trips on the worst degraded day, while re-scaling within each bus line flags 29.3% . The delay model rests almost entirely on one feature, no latent dimension is individually interpretable, and 22 chronic line-directions carry half of all delayed trips.
[LG-10] SCAFFOLD under Hessian Similarity: Sharp Convergence Rates
链接: https://arxiv.org/abs/2610.35663
作者: Zhirayr Tovmasyan,Slavomír Hanzely,Egor Shulgin,Peter Richtárik,Sebastian U. Stich
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:One of the main challenges of federated learning is heterogeneous data split among the clients, leading to biased local updates called client drift. The SCAFFOLD algorithm (Karimireddy et al., 2020) corrects client drift with control variates, but its relation to the similarity of client data is not yet fully understood: the original similarity-based guarantees cover only quadratics with a stronger second-order similarity assumption, and a follow-up weakly convex analysis (Luo et al., 2025) obtains a \sqrtL\delta dependence for L -smooth losses with averaged Hessian similarity \delta . We analyze SCAFFOLD with minibatch-based control variates and establish linear dependence on Hessian similarity beyond quadratics, for both convex and nonconvex objectives. In particular, for convex objectives with exact gradients, R SCAFFOLD rounds with K local steps lead to an objective gap of order (L/K+\delta)D_0^2/R , where D_0 is the initial distance to a minimizer. When K=\Theta(L/\delta) , it improves the minibatch SGD rate by a factor \Theta(L/\delta) . For nonconvex objectives, we prove an analogous stationarity guarantee under uniform Hessian similarity for a client-selection variant. In both settings, we extend these guarantees to stochastic gradients and prove matching lower bounds for the analyzed methods with constant local stepsizes. We also show that, under partial participation, even identical clients need not obtain the full communication benefit of additional local steps.
[LG-11] A Hybrid Framework for Explainable Aviation Risk: Causal Chain Extraction from Safety Reports with Physics-Based Failure Quantification and Interventional Analysis
链接: https://arxiv.org/abs/2610.35658
作者: Yutian Pang,Yifang Tian,Jiacheng Xie,Xinyu Zhao,Jueming Hu,John-Paul Clarke,Hans-Arno Jacobsen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Aviation authorities hold decades of accident reports whose narratives record how investigators reconstructed each accident. Computational work on these archives has mostly sorted reports into fixed taxonomies. In this work, we propose a hybrid framework in which a text-based corpus identifies the leading factors of a failure mode. Physics-based models then quantify how likely that failure is at a stated operating point. Ordered causal chains extracted from 56,202 American accident reports with a local large language model aggregate into a causation knowledge graph. On an evaluation sample of 2,000 reports, the chains recover 90.8% of investigator-flagged causes. Each physical factor whose drivers the record carries receives an occurrence model calibrated to the accident population. A leaky noisy-OR fitted to the corpus combines the factors into a failure mode probability that supports intervention. For engine power loss, the carburetor icing model separates 1,205 icing accidents from other accidents using temperature and dewpoint alone, reaching 0.61 in area under the receiver operating characteristic curve against a density ratio ceiling of 0.64. Routine weather hours give no valid control, because flying concentrates in fair weather. On later years the physics adds 0.03 to that area for occurrence and nothing for severity, so its value lies in interpretability. On four named accidents, abduction gives a probability of necessity of 0.29 to 0.50 for the heat omission, under monotonicity. The Fast Causal Inference algorithm does not license the largest backdoor adjustment on the learned networks.
[LG-12] Active Learning of Conditional Generative Models via the Transport Neural Tangent Kernel NEURIPS2026
链接: https://arxiv.org/abs/2610.35652
作者: Jayoung Ryu,Kyunghyun Cho,Romain Lopez
类目: Machine Learning (cs.LG)
*备注: 10 pages, accepted to NeurIPS 2026. Code: this https URL
Abstract:Many high-throughput scientific experiments produce distributional outputs, requiring conditional generative models to interpret the data, and to predict outcomes at new inputs. Acquiring experimental data at new conditions (e.g., for a new antigen or a new cell type) is often expensive, motivating the need for automated experimental design methods. However, existing active learning frameworks are limited as they assume scalar or vector-valued outputs. We first show that the Wasserstein test risk of a conditional generative task decomposes into a mean term and a shape term, demonstrating why mean-based active learning is blind to distributional shape. To capture both, we propose the transport Neural Tangent Kernel (tNTK), a tractable kernel that measures how sensitive a generative model’s predicted distribution is with respect to its parameters. The tNTK recovers the standard NTK in the deterministic limit and upper-bounds the posterior Fréchet variance of the prediction under small parameter perturbations. It admits a closed-form expression for affine Gaussian transport and tractable Monte Carlo estimators otherwise. Treating the tNTK as the covariance function of a Gaussian process, we greedily select inputs that minimize the total GP posterior variance across the pool, yielding a distribution-aware active learning strategy. We evaluate this strategy across four classes of conditional generative models, including conditional flow matching and conditional Monge gap, on synthetic distributions, intent-labeled sentences, and single-cell perturbation responses. In these experiments, the tNTK outperforms the baseline NTK computed from the conditional mean, demonstrating the value of distribution-aware active learning for generative-modeling tasks.
[LG-13] RON: Transferring Naturally Evolved Network Architectures into Neuroevolution
链接: https://arxiv.org/abs/2610.35647
作者: Konstantinos Zacharopoulos,Alexandra Bekou,Jose M Serradell,Foteini Paraskevopoulou,Apostolos Fournaris,Ioannis Arapakis,Oscar Lao
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
*备注:
Abstract:Neural architecture search (NAS) is challenging because of the vast, non-linear space of artificial neural network (ANN) topologies. Evolutionary methods can generate architectures, but their effectiveness is limited by difficulties in preserving functional organization during mutation and recombination. Naturally evolved biological networks may provide organizational principles that support adaptation and evolvability. Here, we introduce TRON (TRansfer Of Natural networks), a neuroevolution framework that uses experimentally characterized protein-protein interaction (PPI) networks as architectural priors. Instead of starting from random topologies, TRON reuses biological interaction patterns as structural building blocks. It combines classical evolutionary operators with neuroevolution-inspired mechanisms, including speciation, and introduces species quarantine to protect newly generated architectures during early maturation. We evaluated TRON against established evolutionary and neuroevolutionary methods on three classification benchmarks (CORA, CITESEER, and PUBMED) and three regression benchmarks (DIABETES, FRIEDMAN, and CALIFORNIA). TRON achieved the highest predictive performance among the evaluated methods on all classification datasets and most regression tasks. It showed substantially lower variability across independent runs, indicating greater robustness and convergence stability, at increased computational cost. These findings support the use of PPI networks as effective architectural priors and suggest that biologically derived interaction networks provide more favorable substrates for neuroevolution than random topologies. More broadly, TRON demonstrates the feasibility of transferring naturally evolved network architectures into artificial intelligence systems, offering a direction for improving evolvability in NAS through the reuse of biological organizational principles.
[LG-14] Proofs of strong forms of the Bovdi–Geche universality conjecture and a complex analogue
链接: https://arxiv.org/abs/2610.35645
作者: Victor Bovdi,Ambrus Pal
类目: Machine Learning (cs.LG); Number Theory (math.NT); Rings and Algebras (math.RA)
*备注: 14 pages
Abstract:We prove two universality theorems for neural networks consisting of a single neuron. First we prove a strong form of the Bovdi–Geche universality conjecture on neurons over finite fields, where the activation function is a certain modification of the discrete logarithm, using the Weil bound on multiplicative character sums over finite fields. Then we show a generalisation of Aizenberg’s universality theorem for complex binary neurons with a weight vector independent of the target function, positively answering another question in the literature. The proof of the latter uses an equidistribution theorem for the dilates of certain maps into tori.
[LG-15] Adversary Brains: Hypothesis-Based Logical Signatures for Threat Actor Detection with LLM s
链接: https://arxiv.org/abs/2610.35624
作者: Gadi Evron
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:An adversary brain is an LLM-based model of a threat actor’s behavior, held as an evolving logical signature: a hypothesis with enough structure that data can test it continuously, without a person re-interpreting it each time. Rather than any single observation, such as an IoC (Indicator of Compromise), it is the arrangement that becomes a detection, or intelligence, mechanism. Where YARA matches a file and Sigma matches a log line, an adversary brain matches the activity pattern of a campaign or threat actor. This is based on an intelligence analysis technique called Analysis of Competing Hypotheses (ACH), which the design lifts from the limitation of a few hypotheses at a time, to thousands, and then also operationalizes it for continuous detection purposes. Language models make the concept, and scale, practical: they turn prose reports into brains, judge matches, and drive the loop that keeps brains current. A partial pattern in a brain can also match a new value, such as a fresh C2 (command and control) domain name, assisting with the discovery of new indicators, and turning unknown unknowns into known unknowns, and potentially into knowns. The author would like to thank Ryan Moon for the collaboration, and Sounil Yu, Chris Inglis, Thomas Roccia, Marius Benthin, Israel Grimberg, Michal Kamensky, Justin Borland, Brandon Levene, John Hultquist, and John Bambenek for their feedback and assistance. Subjects: Cryptography and Security (cs.CR); Machine Learning (cs.LG) Cite as: arXiv:2610.35624 [cs.CR] (or arXiv:2610.35624v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.35624 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-16] Distilling neural networks into equations to predict planetary instability
链接: https://arxiv.org/abs/2610.35617
作者: Simon Alford,Edward Gu,Omar Costilla-Reyes,Dan Tamayo,Miles Cranmer,Kevin Ellis
类目: Machine Learning (cs.LG); Earth and Planetary Astrophysics (astro-ph.EP)
*备注: 32 pages, 15 figures, 3 tables. Code and data: this https URL
Abstract:The problem of predicting when a multi-planet system will become dynamically unstable has defied analytical progress over the past three hundred years. Recent machine learning advances achieve high predictive performance of instability dynamics, but as black-box systems cannot be interpreted to connect to existing scientific understanding. Here we show analytic equations that predict when a multi-planet system will undergo instability. We discover these equations by first training a neural network on a dataset of N-body simulations, and then distilling its latent knowledge into closed-form algebraic expressions. Our two-stage pipeline first learns an interpretable variable basis, then searches for equations using symbolic regression. The resulting equations outperform existing analytic estimates of instability and apply across wide ranges of mass ratios, spacing, and orbital eccentricities. They capture structure in the prediction landscape with a small number of variables amenable to dynamical interpretation. These results demonstrate how symbolic learning can be combined with neural networks to extract scientific insight from complex dynamical systems.
[LG-17] Distributed Predictive Flocking for a Swarm of Miniature Vibration-Driven Robots in a Confined Arena
链接: https://arxiv.org/abs/2610.35582
作者: Prajol Shrestha,Siddhant Mohapatra,Filip Novkoski,Julian Oelhaf,Nicolas Vandewalle,Ana-Sunčana Smith,Andreas Maier
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Submitted to Robotics and Autonomous Systems
Abstract:Coordinating many miniature robots in a confined arena, with no leader, central planner, or global map, is a core challenge for swarm robotics. As the swarm grows denser, collision avoidance becomes the limiting factor, and the number of safely coexisting robots depends on the formulation of the inter-robot safety constraint. We present a distributed predictive flocking controller in which each robot solves its own nonlinear model predictive control (NMPC) problem, optimizing swarm cohesion and alignment costs, inter-robot and obstacle safety, and a task objective, while exchanging only predicted trajectories with local neighbors. Within this generalized controller architecture, solved in real time, we compare six barrier and control-barrier-function (CBF) safety formulations on a digital twin of an open-source vibration-driven bristlebot in a 0.64,\mathrmm^2 arena. A look-ahead discrete-time CBF performs best overall, with the fewest collisions and the widest inter-robot margins across nearly every swarm size tested. We deploy it on phototaxis, phototactic orbiting, and cohesion without a stimulus tasks, using swarm sizes starting from ten robots and up to the geometric packing limit. The swarm holds well: collisions stay low in small swarms, rising steeply as the packing limit approaches, while coordination and cohesion degrade gently rather than collapsing. The swarm is found not to trade safety for order, keeping both across the confined arena, with local alignment high at every swarm size.
[LG-18] Balanced Sampling in Stochastic Multi-objective Bandits
链接: https://arxiv.org/abs/2610.35551
作者: Mengfan Xu
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:In stochastic multi-objective bandits, Pareto optimal arms may represent different trade-offs among objectives. Balanced sampling across them is important in practical settings such as online advertising. Yet, there seems to be little consensus on how to measure fairness, to our best knowledge. Achieving fairness is also tough as the set of Pareto optimal arms is unknown, and early sampling decisions may create imbalances that are difficult to erase afterwards. We measure unfairness by the KL divergence from the uniform distribution over the true Pareto set to the empirical measure induced by the sampling trajectory. Nicely, this measure admits an exact decomposition that separates the cost of sampling suboptimal arms from the imbalance among Pareto optimal arms. With a recursive form for unfairness, we develop \emphadaptive top-two sampling, which combines sampling counts and empirical evidence of Pareto membership to define a gain for each arm and guide sampling without committing to an estimated Pareto set. When the Pareto set is stable under small perturbations, we establish an O(\log T/T) and O(T^-1) bound on expected unfairness and variance, respectively. Information-theoretic lower bounds show that the \log T/T rate can be unavoidable for uniformly fair policies on stable instances, matching our upper bound. More surprisingly, stability is indeed crucial, as the expected unfairness can be \omega(\log T/T) without it. Finally, we create a variant that encourages exploration of undersampled arms to reduce the risk of persistent underallocation. At sufficiently large horizons, it preserves the rate on the expected unfairness while admitting a sharper bound of O(\log^2 T/T^2) on variance. Numerical studies support these findings and suggest practical insights.
[LG-19] Efficient probabilistic weather forecasting at quarter-degree resolution with two-sided transport attention
链接: https://arxiv.org/abs/2610.35545
作者: Junu Choi,Youngjoon Hong
类目: Machine Learning (cs.LG)
*备注: 32 pages, 6 figures, 7 Extended Data figures, 1 Extended Data table
Abstract:High-resolution probabilistic weather forecasting remains computationally demanding. Here we present Astra, a 0.25 ^\circ ensemble model trained on eight H200 GPUs. It combines Kantorovich Attention, a two-sided transport attention in which each region chooses where it reads from and writes to, so every region sends and receives at token resolution, with a self-featurised continuous ranked probability score (CRPS) that also scores forecasts in the model’s multiscale features. In WeatherBench 2, Astra matches GenCast’s CRPS averaged over five upper-air targets and days 1-10 with 9.8% of its hardware-normalised training compute. Verified against ERA5, its CRPS and ensemble-mean root-mean-square error are lower than the ECMWF ensemble’s on all 600 variable, level and lead-time combinations, and its CRPS is lower than GenCast’s on 73.5% of them. A 15-day member needs 39 times fewer floating-point operations than GenCast’s GPU implementation, and 50 members take under 40 min on one consumer GPU, within 21 GB.
[LG-20] Beyond Modeling Order: Dual-Path Spatiotemporal Representation Learning for Cross-Subject EEG Emotion Recognition
链接: https://arxiv.org/abs/2610.35540
作者: Zehui Xiao,Yi Zheng,Guixing Fu,Wenkai Lu,Lingfei He,Tao Tan,Yue Liu,Silun Wang,Junhai Xu,Mengting Liu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Cross-subject generalization remains a major challenge for EEG-based emotion recognition due to pronounced inter-subject variability in neural activity and emotion-related spatiotemporal patterns. Existing methods mainly alleviate inter subject differences through distribution alignment or contrastive learning, while paying limited attention to representation bias associated with fixed spatiotemporal modeling orders. We hypothesize that high-level emotion representations of the same EEG signal should remain consistent across different spatiotemporal modeling paths. Based on this hypothesis, we propose a Dual-Path Spatiotemporal Contrastive Learning (DP-STCL) framework for cross-subject EEG emotion recognition. DP-STCL employs two complementary encoders with independent parameters, namely a Temporal-to-Spatial (TS) path and a Spatial to-Temporal (ST) path, to model EEG spatiotemporal information under different integration orders. During representation learning, the same emotional stimulus is used to establish cross subject associations, while three complementary relations jointly constrain representation variations across subjects and modeling paths. This promotes stable and discriminative stimulus-related emotion representations. To further handle limited labeled target subject data, we introduce a LoRA-based few-shot adaptation strategy with low parameter overhead. Extensive cross-subject experiments on SEED, SEED-IV, SEED-V, and FACED show that DP-STCL consistently outperforms the compared methods, demonstrating its effectiveness in improving cross-subject EEG emotion recognition.
[LG-21] Model reduction in time: Learning flow maps for reducing stochastic dynamical systems in time
链接: https://arxiv.org/abs/2610.35504
作者: Shreya Jha,Benjamin Peherstorfer
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注:
Abstract:Predictions with dynamical systems can be computationally expensive—even with reduced models—because reaching future prediction times can require many sequential time integration steps. We propose Flow-Map First-Order Trajectory Matching (FM-FTM), a nonintrusive model reduction approach that accelerates ensemble predictions of stochastic dynamical systems by learning flow maps that replace multiple time-integration steps. Our method uses first-order trajectory matching (FTM) to learn a deterministic representation of the stochastic probability-current dynamics and then reduces its evolution over time into a flow map whose evaluation directly yields predictions over long time horizons. Numerical experiments on evolution problems with stochastic forcing demonstrate that FM-FTM accelerates ensemble predictions by up to two orders of magnitude while accurately predicting time marginals and ensemble statistics.
[LG-22] When Fine-Grained Credit Assignment Helps and Hurts: Multimodal Agent ic Reinforcement Learning through Video Editing
链接: https://arxiv.org/abs/2610.35497
作者: Lei Zhao,Zongheng Cao,Xinyu Hu
类目: Machine Learning (cs.LG)
*备注: 48 pages. Project page: this https URL
Abstract:Reinforcement learning (RL) has shown promise for training multimodal agents to reason and use tools. However, it remains unclear when fine-grained credit assignment improves learning and when it fails to improve task performance. Video editing provides a useful testbed for this question, combining multimodal perception, long-horizon planning, and tool use with delayed outcome feedback. We construct four movie-trailer editing task types—Content Selection, Narrative Sequencing, Rough Cut, and Fine Cut—that reflect real-world editing workflows and span different levels of task complexity and interaction horizon. Decaying process rewards provides the largest benefit on Fine Cut, with a 45.4% relative gain over outcome-only GRPO on tasks averaging more than 60 turns under outcome-only GRPO. On the three shorter-horizon task types averaging 30–40 turns under the same baseline, outcome-only GRPO remains competitive, with decaying process rewards yielding at most a 10.4% relative gain. Process rewards from an LLM judge initially improve tool use and encourage visual inspection. As training progresses, however, GRPO+PRM with a fixed process-reward weight increasingly exhibits excessive reasoning and redundant tool calls: process rewards continue to rise while task completion rates and outcome scores decline. Decaying the process-reward weight to zero mitigates this reward-hacking behavior while preserving the early gains. Despite dense token-level supervision, on-policy self-distillation (OPSD) yields little improvement or substantial degradation and is highly sensitive to the choice of privileged information. Our analysis reveals differences in teacher–student disagreement and gradient magnitudes, and suggests possible dilution of informative signals over long trajectories.
[LG-23] Query Complexity of Learning Axis-Aligned Partitions
链接: https://arxiv.org/abs/2610.35495
作者: Hadley Black,Arya Mazumdar,Barna Saha,Geelon So
类目: Data Structures and Algorithms (cs.DS); Computational Geometry (cs.CG); Machine Learning (cs.LG)
*备注:
Abstract:This work establishes the query complexity of learning high-dimensional boxes in the exact query learning setting. Let X \subset \mathbbR^d be a set of n points, labeled by a collection of k boxes: the label of each point in X is determined by the unique box containing it. A learner aims to exactly recover the labels for every point in X given access to a label oracle. We consider three increasingly-structured settings. In the most general setting, there are no conditions on the boxes, and they may overlap. We show that CAL (Cohn-Atlas-Ladner 1994), a classical active learning algorithm, is minimax-optimal with query complexity \Theta(kd \log (n/kd)) . When the boxes are disjoint, we examine two regimes. In the large- k regime, CAL remains minimax-optimal. For the constant- k regime, an optimized variant of the Halving algorithm (Hegedüs 1995) can beat CAL by a \log(d) factor. Finally, in the most-structured settings, the boxes correspond to the leaves of a constant-sized decision tree. We prove a tight query complexity of \Theta(d + \log n) that is additive in d and \log n , generalizing a recent result (Black-Larsen-Mazumdar-Saha-So 2026) that establishes the query complexity of learning decision stumps.
[LG-24] FeatureLoop: Benchmarking Coding Agents under Evolving Requirements
链接: https://arxiv.org/abs/2610.35488
作者: Jia Feng,Alyssa Wu,Cuiyun Gao,Yunlong Wang,Songyuanyi Lu,Zhichao Hu
类目: oftware Engineering (cs.SE); Machine Learning (cs.LG)
*备注: 51 pages, including appendices
Abstract:Software development with coding agents often begins with a rough request that users refine after trying the implementation. Final correctness alone does not capture the repeated guidance this process requires. We introduce FeatureLoop, a benchmark of 61 human-reviewed tasks from 34 repositories across five primary programming languages and seven software forms. Each task follows one development goal in a persistent workspace through evolution feedback, which adds or clarifies requirements, and completion feedback, which reports failures of stated requirements. Cumulative executable checks track new behavior, repair, and preservation of earlier functionality. We evaluate 12 frontier models and find that the strongest completes 72.1% of tasks with both feedback types. Across models, however, only 15.4% of observed failed stages recover within two corrective revisions. Further analysis reveals incomplete repairs, regressions, and model-specific differences in feedback use and token efficiency, identifying obstacles to reliable, low-intervention development.
[LG-25] Evaluating Bayesian deep learning for robust biomedical time-series classification under distribution shift: A sleep scoring case study
链接: https://arxiv.org/abs/2610.35484
作者: Mads Birch Sørensen,Ulrike Richter,Francois Gastambide,Michael Riis Andersen
类目: Machine Learning (cs.LG)
*备注: 28 pages (8 pages main text), 8 figures, 15 tables
Abstract:In biomedical time-series classification, machine learning models often suffer from deteriorating accuracy and calibration due to distribution shift. We investigate to what degree Bayesian-inspired deep learning improves robustness through Bayesian model averaging (BMA). Using rodent sleep scoring as a case study, we evaluate BMA under distribution shifts in genotype, laboratory, and species. We systematically compare nine approximate inference methods spanning single- and multi-basin approaches across two neural network backbones on in-distribution (ID) data and five out-of-distribution (OOD) datasets. In our results, multi-basin methods perform best, outperforming baselines and single-basin methods. Compared with point estimates, BMA generally improves OOD accuracy and calibration across both architectures, while yielding only modest ID gains. It also often improves selective prediction and OOD and artifact detection, but, interestingly, not inter-rater disagreement detection.
[LG-26] Amplify Locate Estimate (ALE) on Rare-Failure Rates of Fine-Tuned LLM s
链接: https://arxiv.org/abs/2610.35472
作者: Patrick Reichherzer,Jonathan Evertz,Thorsten Holz
类目: Machine Learning (cs.LG)
*备注:
Abstract:Safety fine-tuning can make undesirable model behaviors sufficiently rare that measuring their frequency by direct sampling becomes prohibitively expensive. We introduce Amplify, Locate, Estimate (ALE), a method for estimating such rare failure probabilities efficiently. ALE uses the change in token log-probabilities induced by the safety fine-tuning to define an amplification direction, along which we increase the failure rate for cheap measurement. We develop a score-function estimator that uses the selected amplified batch to recover the local slope of the log failure rate from the token probabilities of the sampled failures, enabling extrapolation back to rare target rates. We also use the amplified batch to obtain an importance-sampling estimate. Across 41 evaluation targets, ALE’s estimate-to-reference ratios have a median close to one, with 37 ALE estimates within a factor of three of reference values obtained from independent direct measurements. Ablation tests show that with a budget of up to \sim 1000 labeled rollouts, in addition to a cheap unlabeled scan, ALE estimates failure probabilities for Qwen3-8B fine-tunes down to about 10^-5 , where direct measurement can require hundreds of thousands of rollouts, as the required sampling budget scales inversely with failure probability at fixed relative precision.
[LG-27] Multi-Horizon Future Discovery for Physics-Informed Exploration
链接: https://arxiv.org/abs/2610.35468
作者: Shuwei Wang,Qiang Guan,Brian Qiu,Haoxiang Huang,Haoxi Wang,Chenyu Li,Bo Xu,Sen Cui
类目: Machine Learning (cs.LG)
*备注:
Abstract:Exploration in physical control environments requires agents to identify actions whose consequences reveal previously unexplored aspects of the underlying dynamics. However, many intrinsic motivation methods rely on immediate prediction errors, state novelty, or local physical changes, which may overlook informative outcomes that emerge across extended temporal horizons. We investigate whether learning structured representations of reachable futures can provide a more effective basis for exploration. To this end, we introduce Hamiltonian Temporal Contrastive Curiosity (HTC ^2 ), a framework that combines multi-horizon temporal representation learning with Hamiltonian phase-space structure. A horizon-conditioned contrastive critic learns temporal relationships from future-state pairs sampled within the same episode. The distance between positive-pair representations is then used as an intrinsic reward, encouraging the agent to explore futures that remain insufficiently organized in the learned representation. To incorporate information about physical dynamics, we additionally modulate this reward using bounded multi-step Hamiltonian variation, rather than directly rewarding changes in the Hamiltonian itself. We integrate HTC ^2 with both Soft Actor-Critic (SAC) and Proximal Policy Optimization (PPO), and evaluate it across six DeepMind Control Suite tasks. Comparisons with ICM, RIDE, E3B, Plan2Explore, and a single-step Hamiltonian curiosity baseline are complemented by state-coverage evaluations, component ablations, and phase-space diagnostics to characterize the contributions of temporal learning and physical structure to exploratory behavior.
[LG-28] Prior Predictive Matching for Bayesian LoRA
链接: https://arxiv.org/abs/2610.35457
作者: Tomasz Kuśmierczyk,Eliezer de Souza da Silva,Marcin Sendera
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Bayesian variants of Low-Rank Adaptation (LoRA) provide principled uncertainty estimation when fine-tuning large foundation models, but require specifying the prior over the low-rank weight updates. We address this hyperparameter selection problem by introducing a closed-form prior-predictive estimator that operates before fine-tuning. We analyze a second-order expansion of the network and quantify how the low-rank prior drives function-space deviations from the pretrained outputs. We also show that these prior-predictive quantities depend on the prior only through the product of the adaptation rank and the fourth power of the prior scale. Because the training targets share a single draw of the latent weight update, we can leverage the covariance between example residuals to derive a moment-matching estimator. Furthermore, our theoretical analysis establishes that the optimal prior precision scales as r^-1/2 with the rank r , allowing a single estimate to generalize across all ranks. Controlled experiments across regression and classification tasks validate our approximation and show that our estimator recovers the true prior scale. On RoBERTa-large, our approach improves predictive negative log-likelihood (NLL) and calibration over both MAP and marginal-likelihood selection without requiring validation data, while achieving competitive out-of-distribution detection. Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML) Cite as: arXiv:2610.35457 [cs.LG] (or arXiv:2610.35457v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.35457 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-29] Intent-Preserving Adversarial User Simulation for Multi-Turn Agent Evaluation NEURIPS2026
链接: https://arxiv.org/abs/2610.35426
作者: Atri Vivek Sharma,Brian Formento,Alessio Lomuscio
类目: Machine Learning (cs.LG)
*备注: NeurIPS 2026 UserSim Workshop
Abstract:Multi-turn agents are commonly evaluated with LLM-based simulated users, which can be overly cooperative and can therefore overestimate an agent’s performance. We here introduce StressSim, an intent-preserving adversarial user simulator framework for stress-testing agents with challenging yet valid user interactions. StressSim optimises the simulator’s high-level adversarial plan for each task instance. Candidate planner prompts are evolved with a genetic prompt-optimisation loop that uses rollout traces, granular task-completion signals, and validity feedback to favour conversations that reduce task success without changing the user’s underlying goal. We evaluate the approach on the MultiWOZ dataset and the retail and airline domains from tau-bench with three target LLM agents, comparing against standard simulators and hand-designed non-collaborative user behaviours. We demonstrate the effectiveness of StressSim in exposing robustness gaps across frontier LLMs, with tau-bench pass^4 scores decreasing by up to 69% on gemini-3.8-flash and 62% on gpt-5.6-sol. StressSim maintains goal preservation performance (which decreases by at most 10%), using an optimisation budget of 20 rollouts per task. Sensitivity analysis suggests that increasing the budget further decreases task success. Overall our results demonstrate that StressSim exposes robustness gaps missed by standard and hand-designed user simulators while maintaining conversation realism. This is achieved through lightweight prompt optimisation alone, without model fine-tuning.
[LG-30] ransECG-SSL: Hybrid Self-Supervised Representation Learning for ECG Arrhythmia Classification ALT
链接: https://arxiv.org/abs/2610.35378
作者: Ahtisham Ayyub,Fasih Haider,Muhammad Rehan Usman,Saturnino Luz,Christos Politis,Muhammad Arslan Usman
类目: Machine Learning (cs.LG)
*备注: Submitted to IEEE Journal of Biomedical and Health Informatics
Abstract:Cross-dataset generalisation remains a major challenge for automated electrocardiogram (ECG) arrhythmia classification because of differences in patient populations, acquisition protocols, annotation practices, and class distributions. Learning representations that remain informative across independent datasets could reduce reliance on dataset-specific labelled data. We propose TransECG-SSL, a self-supervised framework for transferable heartbeat representation learning from unlabelled ECG data. A Transformer encoder is pretrained using a hybrid objective combining masked signal reconstruction and contrastive learning, then frozen and evaluated by linear probing under patient-disjoint evaluation. Experiments across three public ECG databases and seven source–target settings show that Hybrid SSL consistently outperforms randomly initialised Transformers and achieves higher balanced accuracy than architecture-matched source-supervised pretraining. Reconstruction-only SSL remains competitive with Hybrid SSL, whereas contrastive-only pretraining performs substantially worse, highlighting the importance of masked reconstruction for transferability. Cross-dataset transfer nevertheless reduces balanced accuracy relative to target-specific self-supervised pretraining, while minority arrhythmia classes remain challenging under severe class imbalance. These findings demonstrate the potential of label-efficient self-supervised learning for cross-dataset ECG representation transfer while highlighting persistent limitations from domain shift and class imbalance.
[LG-31] Benchmarking uncertainty quantification for engineering regression under graded distribution shift
链接: https://arxiv.org/abs/2610.35356
作者: Ahmad Lawal,Rahim Pasbanigoloojeh,Asmau Lawal Bilbis
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine learning models in engineering are often used beyond their training conditions, where reliable uncertainty quantification (UQ) matters most. Existing comparisons of regression UQ methods rely on random splits, which assess calibration only in distribution. We introduce a three-rung shift ladder (in distribution, soft shift and hard shift) built along physically meaningful shift variables, and evaluate 12 methods from five families on seven public engineering datasets: five extrapolation settings, an unseen turbofan fault mode and a winter air-quality sensor deployment. All tuning uses source data only. In distribution, conformal methods cover at least 0.946 of test points on every dataset. Under hard shift, mean coverage falls to 0.40-0.75 across the 11 methods that always return an interval, and no systematic difference in method ranks is detected (Friedman p = 0.63). Weighted conformalised quantile regression stays near nominal on the points it answers, but abstains on 85% of hard-shift points on average and covers only 0.67 of the few Concrete points it answers. Source-validation gains predict improvement in distribution and under soft shift, but not under hard shift (Spearman’s rho = 0.04). Engineering evaluations should split calibration data by deployment unit, report subgroup coverage and sharpness, and treat abstention as an explicit operational outcome.
[LG-32] OneEEG: A Unified Signal-Visual Multimodal Language Model with Task-Aware Mixture of Projectors for Multi-Task EEG Understanding
链接: https://arxiv.org/abs/2610.35355
作者: Jiacheng Zhang,Xueyi Zhang,Xin Lu,Siqi Cai,Haizhou Li
类目: Machine Learning (cs.LG)
*备注:
Abstract:Electroencephalography (EEG) is widely used in clinical assessment, brain-computer interfaces, and cognitive-state analysis, yet EEG modeling remains fragmented across task paradigms. Recent multimodal large language models (MLLMs) provide a promising interface for flexible multimodal reasoning, but their ability to understand EEG inputs remains limited. In this work, we propose OneEEG, a unified EEG-MLLM framework for multi-task EEG understanding. By training on our curated OneEEG-Instruct, a multi-task EEG instruction-tuning dataset, OneEEG jointly models numerical EEG representations and visual EEG representations, and introduces a task-aware Mixture of Projectors (MoP) to align heterogeneous EEG features across different task families with the LLM embedding space. We further establish OneEEG-Bench, a benchmark for evaluating current MLLMs and our method on multi-task EEG understanding. Experiments show that existing MLLMs have limited ability to interpret EEG inputs, while OneEEG achieves stronger performance by integrating numerical and visual EEG representations with task-aware EEG-to-LLM projection. Extensive ablations and analyses further reveal the effects of visual EEG renderings, MoP variants, router behavior, and feature dispersion across EEG tasks.
[LG-33] One Megakernel is Not Enough: Predicting Where to Cut for Streaming Inference
链接: https://arxiv.org/abs/2610.35336
作者: Aditya Chatterjee,Anand Joseph X M,Sanjay Singh Chauhan
类目: Performance (cs.PF); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:We address a fundamental question: How far should kernel fusion extend and which boundaries are worth preserving? Unlike the general convention of using one monolith megakernel at lower batch sizes or supporting only a selective model component as megakernel and rest of the model as conventional graph, we show that the best configuration is a composition of multiple megakernels with no overlapping conventional graph component, or a fully conventional graph. We present a framework that selects megakernel regions in a streaming inference model from conventional-graph profiles and standalone kernel benchmarks, without running any candidate design end to end. It simulates each candidate’s tile and event graphs to predict time to first output and selects the measured best design in all 13 combinations of model and batch size we tested. Across NH-TTS, Qwen3-TTS and Llama-1B, the best composition ranges from two to four megakernels to fully conventional execution and none is a single megakernel. We introduce four model categories (megakernel-stable/ receding/ advancing/ averse) that capture how a model’s best megakernel composition changes with batch size. Using NH-TTS model on H100 NVL GPU, we demonstrate that preserving three specialized megakernels can improve both latency and throughput despite introducing additional kernel boundaries. At batch size 32, this design delivers 21.4% lower mean batch-p95 time to first audio (TTFA) and 28% higher throughput than the evaluated monolith megakernel. Against conventional TensorRT graph execution, it reduces engine-level mean batch TTFA p95 by 4.05x and increases throughput by 19.7%.
[LG-34] CoPE-CGL: Community-aware Prototype Extraction for Replay-Free Continual Graph Learning
链接: https://arxiv.org/abs/2610.35313
作者: Bo-Young Lim,Seokhwan Kim,Kisung Lee,Hyuk-Yoon Kwon
类目: Machine Learning (cs.LG); Databases (cs.DB)
*备注: Extended version with appendices. 18 pages, 5 figures, 17 tables
Abstract:Continual graph learning (CGL) maintains models as graphs evolve without retraining from scratch, while mitigating catastrophic forgetting. Existing methods typically target either transductive or inductive settings and often store past subgraphs or require task identifiers at inference. We propose Community-aware Prototype Extraction for Continual Graph Learning (CoPE-CGL), a replay-free framework that uses a parameter-free structural prior, precomputed by multi-hop Laplacian smoothing, to anchor learning, memory, and inference. CoPE-CGL combines topology-infused bottleneck adaptation, community-level prototypes, and dynamic routing by majority vote over node-to-prototype matches. The default protocol assumes each test graph comes from one learned task; a node-level variant handles mixed-task graphs. Each task adds at most a constant amount of storage, with no past graph data retained. At equal storage and without task identifiers, CoPE-CGL matches the strongest replay baseline even when it receives true task identifiers: average performance (AP) is 95.17% and 95.72% versus 94.28% and 95.82% on Arxiv and CoraFull, respectively, with 5.9-14.0 times faster runtime and zero forgetting. Across four transductive and three inductive benchmarks and mixed streams, CoPE-CGL exceeds the second-best of up to eleven baselines by up to 1.41, 6.32, and 9.33 AP percentage points, respectively. It runs 1.3-7.9 times faster end-to-end than the most accurate competitor on each benchmark and scales to the largest benchmark graph. On graphs mixing two tasks, its node-level variant achieves the best AP among CGL methods. The design targets homophilic graphs with community structure.
[LG-35] Shared Context Separate Engines: Streaming Vision–Language Inference on Edge SoCs
链接: https://arxiv.org/abs/2610.35309
作者: Yilong Li,Chengpo Yan,Yijing Zeng,Hanlei Wu,Mingjun Song,Suman Banerjee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Edge SoCs pack CPUs, GPUs, and NPUs, yet vision–language model (VLM) runtimes keep generation on one engine and leave the others idle. Streaming VLMs expose further parallelism: new frames arrive while earlier queries are still being answered. Engines differ in compute and memory access and contend when they co-run, so no single placement of stages wins across SoCs and workloads. Backend-local KV state keeps other engines from continuing the work. \name coordinates heterogeneous accelerators to execute stateful VLM inference concurrently. It separates a stream’s versioned context from each engine’s KV state, so any engine can continue the stream by reusing, importing, or reconstructing that state. Stateful Streaming Generation ingests new frames on one engine while another answers, and Concurrent Sequence Generation serves different ready requests in parallel. State-aware dispatch sends each request to the engine expected to finish it first. Each answer reads a fixed, published version of its history and matches its serial reference. \name beats the strongest single-engine baselines on Apple M4 Max, Rockchip RK3588, and Qualcomm QCS6490, even when they reuse context. On two RK3588 video streams, it delivers 19.6% more throughput than the best single engine with context reuse and 27.2% more than parallel engines without shared state. Cross-engine KV import cuts first-token latency by a factor of 4.6 over prefix recomputation. Stateful Streaming Generation delivers up to 2.48\times the throughput of serial execution, and state-aware dispatch cuts query completion time by up to 72.7% .
[LG-36] Inference Latency-Aware Tsetlin Machine Training
链接: https://arxiv.org/abs/2610.35304
作者: Yefan Zeng,Shengyu Duan,Rishad Shafik,Alex Yakovlev
类目: Machine Learning (cs.LG)
*备注: Accepted at the 2026 International Symposium on the Tsetlin Machine (ISTM)
Abstract:The Tsetlin Machine ™ is a logic-based machine learning model well suited to low-latency inference on resource-constrained edge devices. Because each clause is an AND over Boolean literals, both the literals and their associated Tsetlin Automaton (TA) actions can be represented at the bit level, enabling bitwise evaluation and an early exit that is triggered as soon as any included literal evaluates to zero, forcing the clause output to zero. In this work, we study how the two TM training hyperparameters, T and s, govern both classification accuracy and inference latency. Building on the inversely proportional T-s relationship along which accuracy is nearly constant, we find that inference latency is reduced primarily by increasing s; rather than exhaustively searching the (T, s) grid at high training cost, we propose a directed search strategy that treats accuracy as a constraint and advances toward larger s to reduce inference latency within a fixed evaluation budget. Experimental results on the gem5 simulator with an ARM processor show that, for certain datasets, the search is able to increase s while preserving the baseline accuracy, thereby reducing inference latency at negligible accuracy cost: across six datasets, the method reduces inference latency by up to 16.34%, while some datasets exhibit an accuracy reduction of only 0.06%. The proposed strategy requires far fewer configuration evaluations than an exhaustive grid search.
[LG-37] InterQ-RL: Interaction-Aware Correction of Execution Mismatch in Quantized Agent ic RL
链接: https://arxiv.org/abs/2610.35303
作者: Yingbo Hao,Song Liu,Guanghua Yu,Zhaojing Zhou,Jianchen Zhu,Yi Zou
类目: Machine Learning (cs.LG)
*备注:
Abstract:Multi-turn agent reinforcement learning (RL) relies on long closed-loop rollouts, making generation and KV-cache costs a major bottleneck. Quantized rollout reduces these costs, but its low-precision executor also serves as the behavior policy and can induce execution mismatch with the learner. Existing quantized-RL methods are designed primarily for single-turn generation. They align the learner forward with quantized rollout or correct the current-token rollout-learner ratio, without modeling the policy history that produces each agent state. Even after quantization alignment, residual rollout-learner mismatch accumulates along the state-reaching policy history, so a current action may look locally aligned despite a mismatched history. We introduce InterQ-RL (Interaction-aware Quantized Reinforcement Learning) to correct this post-alignment mismatch. It applies bounded execution-prefix correction over the state-reaching policy history, allocates correction strength across action segments by effective support under a fixed displacement budget, and requires no additional model forward pass. We evaluate dense and mixture-of-experts (MoE) backbones from 4B to 30B on ALFWorld, WebShop, and ScienceWorld, with all quantized arms sharing the same rollout with 8-bit weights, activations, and KV cache (W8A8-KV8). Across four environment-backbone settings, InterQ-RL achieves the strongest quantized endpoint result, matching BF16 to within 0.7 points and exceeding it in two settings. It matches or exceeds the learning speed of alignment alone while avoiding its late performance regression.
[LG-38] Conversational Generalized Linear Bandits with Hybrid Feedback
链接: https://arxiv.org/abs/2610.35262
作者: Yurui Zou,Qirun Zeng,Jinhang Zuo
类目: Machine Learning (cs.LG)
*备注:
Abstract:Conversational bandits improve recommendations by actively querying users about key-terms to elicit their item preferences. However, existing methods typically assume homogeneous feedback for items and key-terms, using either absolute rewards or pairwise comparisons. In practice, item feedback often takes the form of clicks or ratings, while comparisons are more natural for key-terms. To address this gap, we study conversational generalized linear bandits with hybrid feedback, combining absolute item rewards and relative key-term comparisons. We propose Conversational UCB with Hybrid Feedback (HyConUCB), which constructs unified confidence bounds from heterogeneous feedback and adaptively selects key-term comparisons by balancing information gain and directional coverage. Through a refined regret analysis that jointly exploits both feedback sources, we establish a regret bound that improves upon reward-only and adapted dueling-only benchmarks. Experiments on synthetic and real-world datasets validate the effectiveness of our algorithm.
[LG-39] NOMAD-EEG: Native Omni-Montage Adaptive Diffusion
链接: https://arxiv.org/abs/2610.35246
作者: Abdulkader Helwan,Khadidja Henni,Lina Abou-Abbas
类目: Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: 37 pages and 7 figures
Abstract:Purpose. Electroencephalography (EEG) montages, channel counts and recording lengths vary widely, but most EEG foundation models are bound to the channel set and input length they were trained on. We aimed to build a generative EEG foundation model independent of the number, identity and order of electrodes and of the input length. Methods. NOMAD-EEG is a 12.0 M-parameter diffusion model that alternates state-space layers along time, without patches, with attention across electrodes identified only by their 3D scalp position. It was pre-trained on 147 h of unlabelled EEG from a single corpus and evaluated with frozen linear probes on ten datasets with 1 to 64 electrodes against seven foundation models run under one protocol, and on two unseen populations. Results. Despite its small size (12.0 M parameters against 69 M for REVE) and single pre- training corpus, NOMAD-EEG outperformed all seven foundation models on EEGMMIDB mo- tor imagery (balanced accuracy 0.442), Siena seizure detection (macro F1 0.622) and Bonn seizure detection (AUROC 0.999), had the highest accuracy of the seven compared models on TUEP epilepsy diagnosis (0.640) and, after fine-tuning, the highest \kappa (0.635) and weighted F1 (0.814) on TUEV event classification. It was the best of five models on P300 detection in autism with an unseen eight-electrode montage (AUROC 0.718). With only 5 of 19 electrodes, its mean AUROC fell only from 0.942 to 0.891. Conclusion. Identifying electrodes by position and modelling time without patches let one small model, pre-trained on one corpus, outperform much larger EEG foundation models on seizure detection, epileptiform event classification, epilepsy diagnosis and motor imagery.
[LG-40] Understanding Practical Optimization in Machine Learning
链接: https://arxiv.org/abs/2610.35209
作者: Egor Shulgin
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: PhD thesis, 335 pages
Abstract:Modern machine learning relies on practical optimization methods that often differ from the algorithms studied in classical analyses: they train submodels, clip or normalize gradients, use layer-wise non-Euclidean directions, or approximate costly inner computations. This thesis analyzes the objective, assumptions, update, and computation defined by these procedures, rather than those of simplified proxies. For submodel training, we show that Independent Subnetwork Training can converge to a shifted fixed point because sketching before differentiation biases the update for the original objective. Model-Agnostic Sparsified Training instead incorporates the sketch distribution into its objective, which makes sketched gradients unbiased and admits convergence guarantees. For private optimization, we show that adaptive quantile clipping retains bias at any fixed quantile, while suitable quantile schedules recover near-optimal nonconvex rates under heavy-tailed noise. We then combine smoothed normalization with error compensation and obtain distributed private convergence guarantees under standard assumptions, without assuming bounded gradients. For optimizers based on linear minimization oracles, we develop and validate a layer-wise generalized smoothness model and build on it the Gluon framework for Muon- and Scion-style methods. We also show that the approximate orthogonalization in Muon matters: less precise orthogonalization calls for smaller stepsizes, so its precision should be tuned together with the learning rate. Overall, practical methods are better understood when theory retains the features that determine their optimization behavior.
[LG-41] LEAP: a Latent Embedding Anomaly Patch-Transformer for time-series anomaly detection
链接: https://arxiv.org/abs/2610.35190
作者: Cristhian Moya-Mota,Ignacio Aguilera-Martos,Diego García-Gil,Julián Luengo
类目: Machine Learning (cs.LG)
*备注:
Abstract:Time-series anomaly detection is dominated by reconstruction-based neural detectors, whose objective is structurally flawed. Since the target is also the input, an expressive model can learn to reconstruct anomalies too, leaving uninformative residuals. Recent large-scale benchmarks show that, on curated data, elaborate deep detectors rarely beat simple statistical baselines, while current leaders rely on representation learning. We propose LEAP (Latent Embedding Anomaly Patch-Transformer), which combines masked prediction with latent representation-based anomaly evidence, without relying on reconstruction or attention-based scoring. The model removes patches from a window and predicts them from the remaining temporal and cross-channel context, so prediction errors measure how well learned dynamics explain hidden segments. Deterministic complementary group masking ensures every time step is scored while hidden, with identical visible fractions during training and inference. LEAP adds three label-free evidence sources: latent-embedding novelty calibrated by local density, local temporal contrast, and per-channel novelty pooled over channels selected by six label-free criteria. Under the TSB-AD protocol, the full model achieves a VUS-PR of 0.5179 \pm 0.0022 on 180 held-out multivariate series. On TSB-AD-U, a univariate variant combining multi-scale masked-prediction scoring with latent novelty, without contrast or channel-pool novelty, achieves 0.6194 \pm 0.0016 on 350 held-out series. Both results average three random seeds. A five-stage ablation on TSB-AD-M, supported by paired Wilcoxon signed-rank tests, shows statistically significant gains from each successive component.
[LG-42] DAC-Loop: Distillation-Aware Compression for Remote Sensing CNN
链接: https://arxiv.org/abs/2610.35175
作者: Rico Staedeli,Lorena Raichle,Konstantinos Tzevelekakis,Aron Asefaw,Damian Borth
类目: Machine Learning (cs.LG)
*备注:
Abstract:While large-scale models achieve strong results in remote sensing scene classification, deploying them on edge devices remains difficult due to limited compute, memory, and energy budgets. Compact CNN backbones are therefore a practical choice in many real-world settings, offering efficient inference with well-optimized implementations on embedded hardware. To push this efficiency further, we investigate how far a knowledge-distilled student model can be compressed while preserving high accuracy. We propose a Distillation-Aware Compression Loop (DAC-Loop) that integrates (i) knowledge distillation applied both before and after structured channel pruning, and (ii) an iterative compression schedule that alternates structured pruning with unstructured sparsification. Across EuroSAT and NWPU-RESISC45, the DAC-Loop consistently improves the accuracy-compression trade-off relative to standard one-shot pruning and single-stage KD recovery, maintaining strong accuracy even beyond 90% parameter reduction. These results demonstrate that combining distillation and pruning in a unified iterative loop offers an effective path toward compact remote sensing models.
[LG-43] Biology-Guided Context-Aware Residual Propagation for Spatial Gene Expression Prediction from Histology
链接: https://arxiv.org/abs/2610.35167
作者: Vu Van An,Long D. Nguyen,Ho Viet Duc Luong,Binh P. Nguyen
类目: Machine Learning (cs.LG)
*备注: Accepted at The IEEE/CVF Winter Conference on Applications of Computer Vision 2027
Abstract:Spatial transcriptomics provides spatially resolved molecular profiling, but its cost, limited throughput and experimental complexity hinder routine large-scale use. Predicting spatial gene expression from hematoxylin-and-eosin-stained histology images offers a scalable alternative, yet remains challenging because histology provides only an indirect morphological view of molecular activity. Existing methods typically decode expression from visual features in a single stage, even when enhanced with spatial aggregation, retrieval or multi-resolution fusion. This formulation is restrictive because spot-level expression reflects not only local morphology but also cellular identity, neighborhood composition and tissue-level organization. We propose BCARP, a Biology-guided Context-Aware Residual Propagation framework for histology-guided spatial transcriptomics prediction. Instead of directly predicting gene expression from histology embeddings, BCARP treats them as morphology visual priors and progressively refines spot representations in a transcriptomics-oriented latent space. Each refinement step integrates nuclei-guided local cellular context with prototype-guided global context propagation, enabling information exchange across both neighboring and spatially distant spots. A pattern-scale decoder further separates spatial expression into pattern prediction from raw expression with scale calibration. Evaluated on three ST datasets and ten HEST-1k cohorts, BCARP achieves the highest mean correlation-based performance against other baselines. Numerical results support biology-guided local-global residual propagation as an effective alternative to visual decoding for histology-to-transcriptomics prediction.
[LG-44] Accelerating SO(2)-Equivariant Materials Models through Low-Precision Emulation
链接: https://arxiv.org/abs/2610.35164
作者: Denghui Lu,Alexander Maeder,Alexandros Nikolaos Ziogas,Mathieu Luisier,Manasa Kaniselvan
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注: 13 pages, 6 figures
Abstract:Rotationally equivariant architectures improve the physical fidelity of machine-learned (ML) universal materials models based on graph neural networks (GNNs). However, their scalability is limited by the use of non-trivial, high-precision tensor operations during both training and inference. Adopting the reduced-precision formats of matrix multiplication units (MMUs), e.g., FP8 instead of FP32, provides massive increases in throughput, up to 30x (60x) on NVIDIA Hopper (Blackwell) GPUs, but may lead to drastic loss of accuracy. Here, we introduce tailored ‘emulated’ low-precision operations under the integer Ozaki scheme to accelerate the ‘SO(2)-equivariant’ processing layers of GNNs, which dominate the runtime of (1) many state-of-the-art ML interatomic potentials (MLIPs) and (2) most universal ML Hamiltonian (MLH) models. Our framework, FlashSO(2), outperforms TF32, the most accurate native low-precision MMU format, in throughput, accuracy, and memory footprint at once. At the SO(2) layer level, it reaches 40x (7x) the throughput of MALOQ’s (UMA’s) reference implementation, an MLH (MLIP) model, sustaining up to 236 (232) effective Tflop/s on an NVIDIA GH200, 3.5x the FP32 peak. At the model level, FlashSO(2) helps train MALOQ 3.9x faster than its pure-FP32 version while reducing peak memory by 58% at no loss in accuracy. In molecular dynamics simulations with UMA, the speedup factor is 3.3x, and the peak memory is reduced by 51%. FlashSO(2) thus pushes the speed-accuracy Pareto frontier of equivariant materials models beyond the native low-precision formats and substantially lowers the resources required to exploit their scaling laws.
[LG-45] UON: A Tensor-Aware Optimizer Utilizing Tensor-SVD and Polarization for Large Language Model Pretraining
链接: https://arxiv.org/abs/2610.35130
作者: Anda Tang,Xuanlin Yang,Junjie Wang,Zhouchen Lin
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large scale language model~(LLM) pretraining is increasingly bottlenecked by optimizer efficiency, driving optimizer design from vector-aware methods such as AdamW to matrix-aware methods. Matrix-aware optimizers such as Muon exploit matrix structure but treat layers independently, neglecting cross-layer correlations. Tensor-aware optimizers are therefore needed, yet existing ones rely on a matricization-based tensor algebra and remain restricted to shallow stacking and specific modules. In this paper, we propose \textbfTUON, a novel optimizer built on a transform-based tensor algebra derived from the Tensor-SVD (T-SVD). Unlike methods that polarize a globally unfolded matrix, TUON mixes layers via a unitary transform \boldsymbolQ and then polarizes each transformed slice, and inverts the transform to recover the update. This mix-then-separate strategy captures cross-layer correlations while preserving per-layer diversity, avoiding the structural dimension imbalance and cross-layer misalignment of unfolding. Thus it successfully scales to more layers and diverse modules beyond QKV. Theoretically, we prove that TUON performs steepest descent under a \boldsymbolQ -transformed tensor spectral norm, and identify its update as the tensor polar factor of the T-SVD. We further establish transform non-redundancy, norm and smoothness comparisons with Muon and TEON, and a convergence guarantee. Empirically, TUON improves consistently over Muon and TEON on GPT-124M/350M and Qwen-0.6B/4B, with ablations confirming the transformation mechanism, cross-layer structure beyond two-layer stacking, adaptability beyond QKV, and robustness to the polarization method.
[LG-46] STeP-TS: Shared Temporal Filters with Pruning for Time Series NEURIPS2026
链接: https://arxiv.org/abs/2610.35128
作者: Hanne Raum,Jasper Hoffmann,Joschka Boedecker
类目: Machine Learning (cs.LG)
*备注: Accepted at the Workshop TS-LIMITS at NeurIPS 2026
Abstract:Multivariate time series forecasting requires capturing temporal dependencies across multiple scales while accounting for both shared and variable-specific dynamics. Compact forecasting models have emerged as an attractive alternative to increasingly large architectures, but preserving strong predictive performance within tight computational and memory budgets remains challenging. We propose a compact one-layer convolutional framework that learns a shared bank of multi-scale temporal filters of each variable. Lightweight variable-specific scalar weights determine the contribution of each temporal component, enabling the model to capture shared temporal structure while adapting to variable-specific dynamics. The learned weights and filter-output energy further provide a natural signal for structured pruning of convolutional filters, reducing computational complexity with minimal degradation in forecasting performance. We additionally introduce an amplitude-adaptive normalization mechanism that preserves variable-dependent temporal amplitudes. Experiments on multiple datasets demonstrate forecasting performance competitive with state-of-the-art methods while requiring substantially fewer parameters and inference FLOPs and achieving low per-sample latency.
[LG-47] FlowGenesis: Representation-Driven Self-Organization of Neural Architectures
链接: https://arxiv.org/abs/2610.35108
作者: Maksim A Kazanskii
类目: Machine Learning (cs.LG)
*备注:
Abstract:Deep neural networks typically rely on architectures specified before training, despite the continuous evolution of their internal representations. We introduce FlowGenesis, a framework in which neural architectures adapt through neuron birth and pruning guided by Representation Flux, a local statistic capturing changes in minibatch mean activations. Neurons exhibiting persistently high Flux undergo splitting designed to approximately preserve the network function, while neurons with low Flux and low activity are pruned. This mechanism allows network capacity to evolve without prescribing a target architecture or performing a separate architecture search. Experiments on image and text classi- fication show that networks initialized with different sizes approach similar final neuron-count ranges within each dataset while maintaining comparable predictive performance. We further extend FlowGenesis to continual learning, where representation-guided structural adaptation reduces the neurons and parameters allocated to successive tasks while maintaining accuracy comparable to fixed-width Progressive Neural Networks in a task-incremental learning setting. These findings demonstrate that neural architectures can adapt through feedback between representation learning and structural change, suggesting an alternative to prescribing network capacity before training.
[LG-48] Spherical Brownian Diffusion for Language Modeling
链接: https://arxiv.org/abs/2610.35101
作者: Andrea Miele,Ilija Bogunovic
类目: Machine Learning (cs.LG)
*备注:
Abstract:Continuous diffusion language models involve two coupled design choices: how to corrupt continuous token representations and how to invert that corruption at generation time. Recent work has explored language generation on the unit sphere through geodesic and von Mises–Fisher (vMF) paths, raising the question of which forward process to use on this geometry and how much that choice matters relative to inference. We introduce Spherical Brownian Diffusion, using Brownian motion as the canonical isotropic diffusion on the sphere. Its heat semigroup provides a principled forward process, an additive dimension-free noise clock, and an exact ancestral bridge, while a clean-token posterior learned with cross-entropy supports multiple samplers. To separate corruption from inference, we derive high-dimensional approximation rates between Brownian and moment-matched vMF corruption and use this relationship for controlled substitutions. On Sudoku, Brownian and moment-matched vMF give nearly identical results across four samplers and five budgets under otherwise identical settings. In contrast, with the corruption process fixed, sampler choice produces a 25.0-point spread at a sampler budget of 180 calls. Using the ancestral bridge, we split this gain into a memory effect from re-noising the state, present on Sudoku and GSM8K, and a token-selection effect on GSM8K and largest at small budgets. Across reasoning tasks, categorical refinement improves performance, whereas unconditional text generation benefits from stochastic updates that preserve diversity.
[LG-49] A JiT-Based Denoising Defense Method Against Adversarial Examples for Radio Signal Automatic Modulation Classification
链接: https://arxiv.org/abs/2610.35084
作者: Shiwei Liu,Linyuan Wang,Nuolin Sun,Libin Hou,Bin Yan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Automatic modulation classification (AMC) of radio signals is a core technology for spectrum monitoring and electromagnetic situational awareness. Deep neural networks (DNNs) play an important role in this field owing to their powerful feature extraction capabilities and end-to-end learning paradigm. However, their vulnerability to adversarial attacks poses serious security challenges to practical deployment. Adversarial training, a common defense, has limitations including poor generalization across attack types and high training costs. Inspired by Just image Transformers (JiT), a diffusion generative model that directly predicts clean images, this paper proposes a JiT-based denoising defense against adversarial examples for radio signal modulation classification. During training, a signal-specific 1D Transformer denoising network is optimized via flow matching to directly predict clean signal samples. It thereby learns a mapping that projects arbitrary examples back onto the low-dimensional manifold of clean signal examples. During inference, arbitrary examples are fed into the trained network and directly projected onto this manifold. This enables denoising defense without prior knowledge of attacks. Experiments on the RML2018.01A dataset were conducted without attack information, using a single empirically selected, fixed denoising start time. Compared with PGD-based adversarial training, the proposed method achieves a relative improvement of at least 20.58% in robust accuracy under PGD attacks. It also achieves a relative improvement of 10.20% in classification accuracy on clean examples. When evaluated against attack types unseen during adversarial training, the proposed method provides stronger defense than adversarial training. In particular, at higher attack strengths, it achieves a relative improvement of at least 39.72% in robust accuracy.
[LG-50] Risk-Aware Output Control Should be the Focus of Machine Unlearning
链接: https://arxiv.org/abs/2610.35052
作者: Yahya Alkhatib,Yuyang Xiao,Wee Peng Tay
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Well-documented failure modes of machine learning (ML) systems, including bias, toxicity, hallucination, and leakage of sensitive information, has driven the rise of machine unlearning (MU). Most existing work, however, targets parameter-space indistinguishability, seeking unlearned models that are as close as possible to models retrained from scratch on retained data. In this paper, we argue that this objective is misaligned with practical and user-relevant concerns: MU should prioritize statistically guaranteed output-space control rather than parameter-space similarity. We contend that output control with statistical guarantees is the operationally meaningful notion of unlearning, as it supports rigorous and robust objectives, scales to modern deployment settings, and enables interpretable guarantees. We substantiate this position with empirical evidence and with recent theoretical results that, although not originally formulated in this framing, are consistent with it.
[LG-51] Mapping the Solution-Set Structure of Grokking on Latin Squares
链接: https://arxiv.org/abs/2610.35036
作者: Zihan Gu,Lulu Zhao,Tianyi Ma,Hua Zhang,Wenbo Su,Yanan Cao,Yue Hu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Grokking is a delayed transition from memorisation to generalisation. A network first fits observed examples while remaining near chance on held-out inputs, then generalises abruptly. Prior work has focused mainly on this delay, finding that the network’s main loss becomes weight decay after the fitting phase. We name the solution to this optimization problem as the minimum-norm endpoint, and study which tasks make all minimum-norm endpoints generalise from partial observations. Specifically, we model pairwise input tasks as edge colourings of K_q,q , with Latin colourings as our structured family, and classify each observed instance as all-good, mixed, or no-good by its endpoint set. Furthermore, we prove two contrasting results: (1) Finite-abelian-group families are asymptotically all-good from partial observations; (2) Under a uniform random Latin square, all-good tasks become asymptotically rare for observation-based learners, leaving mixed or no-good behaviour. Moreover, experiments show that relabellings of addition generalise despite their dense polynomial representations. Finally, on tasks of composite order, finite-training results suggest that predictors learn a coarse part of the output before the full label.
[LG-52] BellowsKV: One Latent for Many Layers Expanded on Demand
链接: https://arxiv.org/abs/2610.35023
作者: Jiamu Zhang,Hanjie Chen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large language models (LLMs) now serve as the backbone of agents that reason over many turns, call tools, and pursue long-horizon tasks. To avoid recomputation, a model stores the attention keys and values of all past tokens in a key-value (KV) cache. Each layer maintains its own cache, so KV-cache memory can become the main bottleneck for GPU serving. Most compression methods operate within a single layer, by quantizing the cache entries, reducing the channel dimension, or evicting less important tokens. All three lose substantial accuracy at high compression ratios. Cross-layer methods share one compressed cache across adjacent layers to go beyond this limit. However, these methods also degrade sharply at high compression ratios. We find that the KV-cache representations of adjacent layers are quite different, whereas the input hidden states of adjacent layers are highly similar. We also find that, under the same memory budget, reducing the dimension of this shared latent loses far more accuracy than quantizing it. Based on these findings, we propose BellowsKV, a training-free method that maps the hidden states of n adjacent layers to one shared latent per token. At inference time, each layer reads its own keys and values from this latent through a layer-specific projection. We keep the latent at full dimension and compress it in two other ways: BellowsKV-Quant stores the latent at low precision, which is nearly lossless because the latent has no dominant outlier channels; BellowsKV-Evict makes one shared eviction decision per block, because the layers rank token importance very similarly. Across Llama-3.1-8B, Mistral-7B and Qwen2.5-7B, BellowsKV-Quant retains about 99% of full-cache accuracy at 3.6x compression and 97-98% at 13-14x, where the strongest cross-layer method retains at most 84% at 16x. Code: this https URL
[LG-53] Attention under Bounded Key-Query Similarity: Space Complexity and Transformer Anisotropy
链接: https://arxiv.org/abs/2610.35020
作者: Justin Y. Chen,Ying Feng,Piotr Indyk,Michael Kapralov,Ekaterina Kochetkova,Boris Prokhorov
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS)
*备注:
Abstract:A growing body of work on KV-cache compression aims to reduce the memory requirements of autoregressive generation. These methods compress the stored context while approximately preserving attention computation, enabling generation with reduced memory consumption. Almost tight theoretical results on the worst case space complexity of attention have recently been obtained, but they are achieved by instances where queries can be perfectly aligned with keys. This directly contradicts the observed anisotropy of the key/query geometry in real transformers. In particular, it has been shown empirically that key/query cosine similarities are small even for most correlated pairs, i.e., near-perfect alignment of queries with keys does not happen in practice. In this paper, we study the space complexity of attention approximation under the assumption that key/query cosine similarities belong to a small interval around zero, i.e., a query cannot align too much with any key. We give a nearly space optimal algorithm and a corresponding space lower bound. Surprisingly, the distribution of key/query cosine similarities in our lower bound exhibits the anisotropy property empirically established and studied in the literature. Additionally, we show that the simple uniform sampling algorithm is nearly optimal for a typical range of parameters, supporting its surprising effectiveness on real datasets.
[LG-54] Adaptive Memory Index Evolution for Agents
链接: https://arxiv.org/abs/2610.34977
作者: Yuntong Hu,Tongli Su,Yuanfu Sun,Shuo Ma,Qiaoyu Tan,Liang Zhao
类目: Machine Learning (cs.LG)
*备注: 29 pages, 22 figures
Abstract:Long-horizon agents continually accumulate experience, yet retaining a memory does not guarantee access when it is needed. Existing systems can adapt both memory content and access; we isolate persistent access updates as a separately controlled adaptation layer. We formulate this target as Memory Index Evolution and instantiate it with Adaptive Memory Index Evolution (AMIE), a lightweight plug-in layer that evolves per-memory access state from deployment retrievals without modifying stored content or retraining the host memory system. AMIE distils a semantic teacher’s keep/suppress judgments into a retrieval-conditioned policy and accumulates its selective actions into bounded state that reweights subsequent host retrievals. Surprisingly, even a zero-call policy that simply accumulates retrieval frequency can match or outperform semantic policies in several settings, revealing retrieval traffic itself as a strong signal for access adaptation, while learned selectivity improves robustness when frequently retrieved memories remain useful. Across five retrieval-based hosts and three benchmarks, trained AMIE improves downstream F1 or accuracy in all 15 comparisons, by 0.70–3.37 percentage points (up to 25% relative), and improves retrieval in all 10 comparisons. Controlled analyses show that persistent competitor suppression can expose previously unseen evidence, while suppressing repeatedly useful memories creates the complementary failure mode. Together, these results establish persistent memory access as a practical adaptation layer, distinct from the content it serves, in retrieval-based agent memory systems.
[LG-55] Beyond feature attribution: Recovering latent interaction structures in machine learning models via localized CBFI
链接: https://arxiv.org/abs/2610.34962
作者: Sejong Oh
类目: Machine Learning (cs.LG)
*备注:
Abstract:Background: Understanding individual predictions in complex tabular models remains challenging, as existing additive frameworks like SHAP and LIME often obscure structural logic by absorbing nonlinear interactions into individual feature scores. Exposing these latent interaction structures is essential for automated auditing in high-stakes environments where understanding the logic of model behavior is as essential as predictive accuracy. Methods: To address this limitation, we introduce the localized case-based feature importance (CBFI) framework. Unlike traditional numerical attribution, our method decomposes instance-specific predictions into structurally distinct components, effectively separating independent main effects (G1) from pure interactions (G4) to capture explicit feature synergies and provide a principled approach for diagnosing feature combinations. Results: Comprehensive experiments on wine quality classification and medical insurance regression demonstrate that localized CBFI successfully disentangles complex interactions hidden by additive models. Furthermore, fidelity tests and synthetic ground-truth benchmarks confirm that the identified interactions correspond precisely to the nonlinear structural dependency patterns embedded in complex architectures, including gradient boosting and deep neural networks, thereby successfully shifting the explainable AI paradigm from simple numerical attribution to the recovery of interaction-driven contributions.
[LG-56] MINT: Mitigating Structural Inconsistency via Attentive Message Passing for Financial Anomaly Detection CIKM2026
链接: https://arxiv.org/abs/2610.34950
作者: Yin-Ju Wu,Bi-Cheng Yan,Berlin Chen
类目: Machine Learning (cs.LG)
*备注: Accepted to CIKM 2026 (Short Research Paper)
Abstract:Illicit account detection is a critical task for financial risk control systems, aiming to pinpoint malicious actors within transaction networks. Existing efforts typically recast transaction networks into directed multigraphs and leverage graph neural networks (GNNs) to detect anomalies via topological structures under the homophily assumption. However, directly grafting GNNs onto transaction graphs is faced with at least two pivotal challenges, i.e., structural inconsistency and label scarcity. The former arises from deceptive connections forged by malicious actors, while the latter stems from the lack of annotated data. In light of these, we introduce MINT, a novel graph neural framework that mitigates structural inconsistency via an attentive message-passing scheme on transaction graphs. Specifically, MINT introduces a discrepancy-aware feature propagation, which tracks malicious actors by quantifying deviations between nodes across bi-directional transaction flows. A transaction salience module is proposed to mitigate structural inconsistency by adaptively regularizing the interaction strength between nodes. Furthermore, to alleviate label scarcity, MINT is optimized via Positive-Unlabeled (PU) learning, which exploits only a minority of illicit nodes coupled with a majority of unlabeled ones. Extensive experiments conducted on three financial transaction benchmarks, including two large-scale public and one in-house real-world datasets, validate the superior effectiveness of MINT over several cutting-edge baselines.
[LG-57] Intellectual Property Protection of Hyperdimensional Computing Classifiers: Black-Box Ownership Verification via Model Watermarking
链接: https://arxiv.org/abs/2610.34942
作者: Hamed Poursiami,Farshad Soleiman,Ahmad Tahmasivand,Guojing Cong,Thomas Potok,Khaled N. Khasawneh,Kai Zeng,Maryam Parsa
类目: Cryptography and Security (cs.CR); Emerging Technologies (cs.ET); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
*备注:
Abstract:Hyperdimensional computing (HDC) offers efficient learning and inference for resource-constrained systems, yet ownership verification of its models remains underexplored. We present the first systematic investigation of model watermarking in HDC by adapting backdoor-based embedding to three classifiers with different prototype-learning mechanisms. Across four benchmarks, with conventional neural networks as references, we examine whether watermarks can be learned without compromising task accuracy and whether they survive subsequent model modification and replication. The error-driven classifiers, AdaptHD and OnlineHD, support verification at a nominal trigger exposure of 1% with at most 0.58 percentage points of mean accuracy loss, whereas single-pass Vanilla HDC requires greater exposure and can incur substantial accuracy costs. The error-driven models retain verification under native continued training, but extraction produces surrogates that reproduce much of the task behavior without verifiable watermarks. Watermark overwriting readily embeds attacker watermarks into HDC prototypes and erodes the owner’s watermark more strongly than in the neural network references, although the strongest erosion coincides with large accuracy losses. We further deploy watermarked AdaptHD on an FPGA and stress its execution through undervolting and overclocking. Verification survives substantial computation errors, whereas more severe stress either causes verification failure together with accuracy collapse or results in inference timeouts. These findings establish the feasibility of watermarking in HDC while identifying limitations in retention under extraction and sensitivity to overwriting.
[LG-58] A Certified Robustness Framework for Backdoor Poisoning Attacks using f-Divergences
链接: https://arxiv.org/abs/2610.34934
作者: Evan Rose,Alina Oprea
类目: Machine Learning (cs.LG); Cryptography and Security (cs.CR)
*备注: 39 pages, 3 figures, 7 tables
Abstract:Machine learning systems trained on untrusted data are vulnerable to data and backdoor poisoning attacks, in which an adversary corrupts training and test samples to manipulate predictions at inference time. Randomized smoothing yields provable defenses against such attacks, but existing applications to train-time robustness are limited to specific threat models and smoothing mechanisms. We propose a general framework that allows expressing complex adversary constraints on the number of corrupted training examples, feature and label perturbation magnitudes, and test-time corruptions. Our framework randomizes inputs by combining data subsampling with feature, label and test-sample noising. We develop a composable certification procedure based on f -divergences that handles arbitrary combinations of threat models and noising strategies, and recovers several prior certification regimes exactly as special cases. We demonstrate our framework experimentally by certifying, for the first time, robustness under general \ell_p -constrained poisoning adversaries under bagging with additive Laplace, Gaussian, and Uniform noises.
[LG-59] A Hierarchical Method for Environmental Sounds Recognition in Complex Acoustic Scenes
链接: https://arxiv.org/abs/2610.34931
作者: Zhouwen Li,Haijun Lin,Bowen Lv,Yupeng Chen
类目: ound (cs.SD); Machine Learning (cs.LG)
*备注:
Abstract:Environmental noise identification is essential for noise control, yet conventional single-stage classifiers suffer from class confusion under imbalanced multiclass conditions. This paper proposes a lightweight hierarchical model integrating a Frequency-Time Adaptive Network (FTANet) with a two-stage classification mechanism. FTANet employs parallel asymmetric one-dimensional convolutions to capture temporal dynamics and frequency-band distributions, while adaptive fusion attention (AFA) enhances discriminative acoustic features. In the first stage, FTANet classifies sounds into five coarse-grained categories covering 25 fine-grained classes. In the second stage, samples are routed to the corresponding category-specific classifiers for fine-grained recognition. On a self-constructed dataset, FTANet achieves 95.45% coarse-grained accuracy with only 0.2004M parameters. On UrbanSound8K, it achieves 79.24% mean accuracy and 80.37% mean F1-score in ten-fold evaluation, outperforming compared convolutional models. Field tests on the Tianjin Online Acoustic Environment Monitoring Platform yield 87.94% coarse-grained accuracy and 97.21% conditional fine-grained accuracy for natural sounds correctly routed to C2. The complete hierarchical model achieves 87.18% overall accuracy and 87.40% weighted-average F1-score, demonstrating its feasibility for urban environmental noise monitoring.
[LG-60] Attack logics not outputs: Concept-based Property Attacks for Shortcut Elimination in Neural Networks NEURIPS2026
链接: https://arxiv.org/abs/2610.34923
作者: Raik Dankworth,Felix Mächtle,Thomas Eisenbarth,Gesina Schwalbe
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: Rejected by NeurIPS 2026
Abstract:Deep neural networks (NNs) for computer vision are vulnerable to adversarial attacks, such that minuscule malicious changes to inputs may induce unintuitive outputs. One key approach to verify and mitigate such robustness issues is to falsify expected output behavior, which enables local robustness guarantees and adversarial training. Due to the black-box nature of NNs, current attacks only falsify a class of the final output, such as flipping from \text\textttstop_sign to \neg \text\textttstop_sign . In this work we argue that robustness analysis should focus more generally on logical inconsistencies in concept-level reasoning rather than solely on output misclassification. We formalize this perspective through concept-based properties, i.e., constraints over human-interpretable concepts such as \text\textttred \land \text\textttoctogonal \implies \text\textttstop_sign and study the falsification of such properties as a principled form of adversarial attack. Furthermore, an implementation of concept-based properties on already trained NNs is proposed using techniques from concept-based explainable artificial intelligence (C-XAI). For that we formally prove that attacks on concept-based properties are expected to have a reduced search space compared to simple class falsification, whilst arguably be more aligned with intuitive robustness targets. The practical effectiveness of our approach for logically informed adversarial training is shown in an empirical study across multiple models and datasets, including both vision and a security-relevant code processing task. Our results demonstrate that concept-level falsification generalizes classical adversarial attacks: Retraining on our novel attacks maintains robustness gains as compared to classical adversarial retraining whilst better eliminating semantic biases.
[LG-61] End-to-End MXFP8 Reinforcement Learning on Blackwell: What Matching Buys When It Pays Off and How It Fails
链接: https://arxiv.org/abs/2610.34915
作者: Wengee Zhang
类目: Machine Learning (cs.LG)
*备注: 36 pages, 8 figures, 13 tables
Abstract:Low-precision arithmetic can make reinforcement learning (RL) for large language models cheaper, but RL requires the sampler and the learner to agree. Several frameworks now run FP8 on both sides. We implement end-to-end MXFP8 RL in verl (Megatron/Transformer Engine training; vLLM and SGLang rollout; dense and MoE models) and measure on B200 GPUs, with Qwen3-8B and Qwen3-30B-A3B, what this design buys, when it is faster, and how it fails. (1) Matching. For an MXFP8 learner, a sampler on the learner’s grid is more consistent than a higher-precision BF16 sampler: in one engine at fixed weights, train-rollout KL is 11% lower with half of the sampler’s linear layers on the grid and 16% lower with all of them, and the synchronized weights we checked equal the learner’s FP8 forward weights byte for byte. On the MoE model, an FP8 pipeline whose sampler uses another grid has twice the shared-grid pipeline’s KL from the first step. Without a quantized KV cache, BF16 end to end remains 2.5-10x more consistent. (2) Speed. MXFP8 training adds a fixed cost per micro-batch and saves a per-token cost, so in our dense setting it pays off only above about 4.5k tokens per micro-batch; a micro-batch sweep on identical data confirms the per-micro-batch cost, and a trace shows the GPU waiting for the host for most of the added time. End to end, MXFP8 cuts MoE step time by 10.7% and generation time by 24%, and slows a short-sequence dense workload by 5%. (3) Correctness. We state seven conditions every weight synchronization must preserve and checks that test them against the live engine. Each of the seven failures we met on hardware broke one condition, five silently; the checks caught two at their first opportunity, each missed by the other checks, and CPU fault-injection tests reproduce five. All runs are single-seed and at most 60 steps long; we make no claim about final model quality.
[LG-62] Checkpoint Reload Explains Most of the Measured State Gap in Two State Discovery Reproductions NEURIPS2026
链接: https://arxiv.org/abs/2610.34891
作者: Weijia Han,Lisha Qu
类目: Machine Learning (cs.LG)
*备注: Equal contribution, 11 pages, 3 figures, 7 tables. Accepted at the NeurIPS 2026 Workshop on Foundation Models for Temporal Systems (FMTS 2026). Equal contribution
Abstract:Temporal models that maintain per-sequence latent states during training must reconstruct those states when the stored table is unavailable at inference. This creates a basic consistency question: do the reconstructed states agree with those formed during training? We study this question for State Discovery in World Machine, which updates stored states in parallel each epoch but performs inference by sequential rollout from an all-zero start. On Toy1D, we analyze two paired-checkpoint runs whose decoded-state MSE is 27 to 36 percent above the published value. We compare the stored table with rollouts under the best-validation checkpoint reloaded after training and the last-epoch weights that wrote the table. Matching the rollout to the table-writing weights removes 90 to 91 percent of the mean per-dimension Wasserstein-1 distance. Causal distillation and frozen SD updates do not satisfy their pre-specified criteria under either checkpoint. Overall, checkpoint reload accounts for most of the measured gap in these runs.
[LG-63] AgriDefend: A 205 Offline Edge Device that Detects Crop Stress Before Visible Symptoms via Multimodal AI Fusion
链接: https://arxiv.org/abs/2610.34861
作者: George Xu,Nikhil Vincent,Elizabeth S. Forbes
类目: Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: 3 pages, 3 figures
Abstract:Most in-field crop-monitoring devices sense a single modality – either a camera looking at the canopy or a probe reading the soil – and so cannot detect stress that begins below ground before it becomes visible above it. We present AgriDefend, a self-contained consumer-grade edge device that fuses both. A MobileNetV2 convolutional neural network (CNN) analyses canopy imagery, and a Random Forest classifier integrates trend features from a single soil probe: moisture, the rate of change in pH and in N, P, K, and belowground CO2 relative to ambient (a proxy for soil respiration). A logistic-regression meta-learner fuses the two into a five-class crop-health prediction, all on a Raspberry Pi 5 with no cloud or cellular link. On a controlled greenhouse drought trial with wild radish, the fused model reached 87% accuracy (R^2=0.88, MAE=0.05) at 1.4 s INT8 TensorFlow Lite inference, and its soil signals preceded visible wilting by about four days. The complete bill of materials is 205, roughly a fifth of comparable commercial nodes, positioning multimodal, sustainable crop sensing as a practical consumer-electronics product for connectivity-limited settings.
[LG-64] When Respondents Are Scarce: LLM -Derived Semantic Priors for Bayesian Confirmatory Factor Analysis
链接: https://arxiv.org/abs/2610.34843
作者: Shaohuai Feng,Mohd Wira Mohd Shafiei,Mohd Sukri Shafie,Siyang Tian
类目: Machine Learning (cs.LG)
*备注: 8figures
Abstract:Across social and behavioral research, measuring latent constructs can be challenging when target populations are small or difficult to reach. Recruiting additional respondents may be costly or infeasible, leaving researchers with limited data for estimating questionnaire factor structures. Yet item wording contains structural expectations before data collection, and whether these can improve small-sample estimation without replacing empirical validation remains unclear. We introduce the Semantic-Empirical Updating Framework (SEUF), which converts large language model paraphrases and embeddings, aligned with hypothesized item-factor assignments, into prespecified priors for Bayesian confirmatory factor analysis. We evaluated SEUF using a 38-item survey of 617 industrial-park stakeholders and two public surveys, FEVS and RIASEC. Semantic structure persisted after wording controls across all three instruments. With 50 respondents, SEUF achieved loading precision comparable to conventional CFA with 70-82 respondents, equivalent to 29-39% fewer respondents at the specified precision benchmark. Results showed lower loading errors and improved credible-interval coverage relative to conventional CFA in both public surveys. Gains were greatest when respondent data were limited, highlighting the value of semantic priors under information scarcity. By harnessing structural information already embedded in questionnaires, SEUF advances a more resource-efficient approach to measurement and opens new directions for study design and empirical inference in hard-to-reach populations.
[LG-65] ransporting Quantiles to Codebook: Scalable Vector Quantization without Codebook Collapse
链接: https://arxiv.org/abs/2610.34828
作者: Yuwei Zeng,Zekun Shi
类目: Machine Learning (cs.LG)
*备注:
Abstract:Effective discrete representation learning with VQ-VAE serves as a fundamental component of modern autoregressive generative models. However, it is widely known to suffer from codebook collapse and inefficient code utilization. To address this issue, we propose FlowVQ, a simple yet effective alternative to standard vector quantization by formulating neural quantization as a joint modeling and clustering process. Specifically, we learn the latent space using a lightweight flow matching model alongside training, corresponding to a probability flow ODE that transports an often fixed well-behaved source distribution while preserving probability mass. Based on this, we define the codebook C_1 by transporting even-quantile partitioned prior codes C_0 from the source space through the learned flow, and obtaining the Voronoi centroids under the induced distribution, leading to a structured discretization. We evaluated on image tokenization and generation, as well as on two tasks requiring high-precision tokenization, demonstrating scalability and improved expressivity of the codebook with full utilization.
[LG-66] Can System-1-Style Decision Models Replace Dedicated Rerankers in RAG ? A Systems Evaluation
链接: https://arxiv.org/abs/2610.34816
作者: Qiaomu Li,Ying Xie
类目: Machine Learning (cs.LG)
*备注:
Abstract:Retrieval-augmented generation (RAG) systems often use a dedicated reranker to choose which retrieved candidates reach the language model. Reranking improves relevance but adds latency, so teams look for faster alternatives. System-1style decision models make direct, structured predictions instead of generating long responses. Could one provide comparable reranking quality with less inference overhead? We first screen 21 open-weight models on 11 benchmarks, then compare a promising decision model with a dedicated reranker on identical candidate lists. The default decision-model setup fails on many long candidate sets, so scoring only the requests it completes gives a misleading ranking. Limiting candidate text length largely restores reliability and makes some configurations competitive. A frozen evaluation on new retrieval data supports the reliability and ranking results, but confirms the latency requirement for only one of the two selected configurations. Across workloads, which model ranks better changes, and a tuned reranker remains faster and uses far less GPU memory. A decision model can perform competitive reranking, but replacing a reranker requires weighing quality, latency and memory together, and teams should tune the reranker they already run first.
[LG-67] When Does Data Value Reduce to Class Balance? A Coverag e View of Per-Point Data Valuation
链接: https://arxiv.org/abs/2610.34801
作者: Yingjie Bai,Weiming Zhi
类目: Machine Learning (cs.LG)
*备注: Published in Transactions on Machine Learning Research (TMLR), 2026. 34 pages. Code: this https URL . Video: this https URL
Abstract:Per-point data-valuation scores – Data Shapley, influence, leverage – are routinely used to select training data: keep the points with the top k scores. We argue that top- k by a fixed, mode-blind per-point score is the wrong selection rule precisely in the regime where data selection matters. The reason is that a point’s value is not a fixed number but depends on what else has been selected – a near-duplicate of an already-chosen point adds almost nothing, however high its score – and a per-point scalar cannot express this interaction. The theory quantifies this failure exactly: in an equal latent-mode model, fixed top- k selection incurs a closed-form excess risk over adaptive coverage because it repeatedly spends budget on already-covered modes. The gap is largest when the budget is tight, the learner representation is redundant, and missing a mode is costly. The practical rule is simple: use balance when the labels available at selection identify the learner’s modes, and use explicit coverage when they do not. On standard classification tasks with strong pretrained features, class-balancing the same per-point scores recovers the missing coverage and matches or outperforms generic set coverage across our datasets. Without labels, label-free coverage improves accuracy by up to +0.20 while raw top- k can fall below random; when labels are coarser than the modes, coverage within coarse strata performs best. We turn this distinction into a pre-specified diagnostic and validate it across both favorable and stress-test settings. The limits are part of the claim: the advantage is representation-relative. Coverage fails when the relevant modes are not expressed in the learner geometry, as with spurious subgroups, or when data quality varies independently of coverage – exactly as the theory predicts. Comments: Published in Transactions on Machine Learning Research (TMLR), 2026. 34 pages. Code: this https URL . Video: this https URL Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.34801 [cs.LG] (or arXiv:2610.34801v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.34801 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Journalreference: Transactions on Machine Learning Research, 2026 Submission history From: Yingjie Bai [view email] [v1] Mon, 26 Oct 2026 03:42:03 UTC (598 KB)
[LG-68] meWorld: Towards Action-Conditioned Multimodal World Models for Time Series Forecasting
链接: https://arxiv.org/abs/2610.34797
作者: Hang Ni,Weijia Zhang,Hao Liu
类目: Machine Learning (cs.LG)
*备注: under review
Abstract:Time series forecasting supports decision-making in web applications such as online retail and network management, where planned actions can reshape future dynamics. World models provide a natural paradigm for predicting such action-dependent changes beyond extrapolating historical patterns. However, existing time series world models (TSWMs) typically rely on fixed, domain-specific covariate schemas that limit flexible intervention specification, while multimodal forecasters treat text primarily as passive predictive context. To investigate these gaps, we introduce TSWM-Bench, a cross-domain benchmark for multimodal TSWMs spanning seven real-world datasets, with textual interventions grounded in recorded action–outcome data. Our evaluation reveals that conventional multimodal forecasters underperform history-only time series foundation models (TSFMs) despite dataset-specific training, whereas simple text-prefix adaptations improve forecasting across all evaluated TSFM backbones. These findings suggest that pretrained TSFMs encode implicit world knowledge of temporal dynamics, offering a strong foundation for action-conditioned modeling beyond conventional multimodal fusion. Building on this insight, we propose TimeWorld, a plug-in framework that adapts pretrained TSFMs through explicit action-response modeling. Specifically, the Prototype-Guided Latent Modulation (PLM) module grounds textual actions in temporal dynamics by composing representative response prototypes to modulate future representations based on the action and historical patterns. The Action Contrastive Learning (ACL) component strengthens action-specific response modeling by contrasting these updates under alternative interventions sharing the same history, distinguishing action types while preserving relative intensity relationships.
[LG-69] Sparse Reconstruction by Direct Measurement-to-Field Generation in Function Space
链接: https://arxiv.org/abs/2610.34709
作者: Linzheng Wang,Jason Chen,Nicolas Tricard,Zituo Chen,Xingsen Guo,Sili Deng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Understanding physical systems often requires recovering their full spatial state from measurements available only at sparse locations. This reconstruction becomes challenging when physical boundaries are unknown, high-resolution training fields are unavailable, or important quantities cannot be measured directly. Sparse observations may also be consistent with multiple plausible states, leading to significant reconstruction uncertainty. Here we introduce Direct Measurement-to-Field Generation (\DMF), a functional rectified flow framework that learns how sparse measurements drive physical-field generation end to end. Instead of relying on grid-based rasterization or runtime guidance, \DMF jointly learns a sensor–global–sensor representation with the generative transport, dynamically fusing local grounding with global context to evaluate physical fields at arbitrary query coordinates. Across benchmarks involving unobserved boundaries, missing high-resolution training fields, and unmeasured physical quantities, \DMF outperforms alternative baselines by significantly reducing reconstruction errors and preserving the fidelity of physical properties. Furthermore, its stochastic ensemble generation reliably quantifies uncertainty while improving continuous ranked probability scores. These results extend sparse reconstruction from fixed-setting point estimation toward measurement-conditioned generation of physical fields under incomplete and changing information.
[LG-70] Stable Continuous-Time Consistency Distillation: An Empirical Study with a Multistep Extension
链接: https://arxiv.org/abs/2610.34704
作者: Vinay Saji Mathew,Soundar R. Kumara,Gretta D Kellogg,William KM Lai
类目: Machine Learning (cs.LG)
*备注: Published in TMLR (2026). Reproducibility Certification
Abstract:Continuous-time consistency models are an attractive route to few-step generation. They remove the discretization error inherent to the timestep grid that discrete-time consistency models depend on. Until recently, however, training them was unstable. Lu Song (2025) address this with the TrigFlow formulation and a coordinated set of architectural and training-objective changes. We conduct an independent empirical study of this framework on CIFAR-10 and ImageNet-64. No open-weight TrigFlow teachers exist at these scales, so we train them from scratch and then verify the consistency distillation (sCD) and consistency training (sCT) behavior of students initialized from these teachers. On CIFAR-10, our students reproduce the reported FID at one and two steps. On ImageNet-64, they fall short, a gap that appears related to our teacher and to architectural choices the paper leaves unspecified. We also report a parameter-count gap of about 5% between the model size stated in the paper and a faithful re-implementation, which we attribute to the scale-and-shift conditioning projections in Adaptive Double Normalization. We then propose Multistep sCD (MS-sCD), a continuous-time, segment-conditioned multistep generalization of sCD. The segment-conditioning idea is borrowed from the discrete-time multistep consistency distillation of Heek et al. (2024); we build on their preliminary continuous-time experiments within the TrigFlow/sCM framework. The rest of the algorithm follows sCD. MS-sCD replaces sCM’s two-step intermediate time with a fixed segment-boundary schedule and recovers sCD at M = 1. We report FID at M = 2, 4, 8 and compare against the discrete-time MSCD baseline. Code, model weights, and training scripts are available on GitHub (Mathew et al., 2026).
[LG-71] IsoAct: Geometry-Aware Post-hoc Debiasing via Isometric Actions
链接: https://arxiv.org/abs/2610.34690
作者: Sumin Park,Taero Kim,Minhyeong Cho,Subeen Park,Jong Hwan Lee,Kyungwoo Song
类目: Machine Learning (cs.LG)
*备注:
Abstract:Learning representations that are robust to nuisance factors is essential for reliable deployment, but remains challenging when the representations lie on non-Euclidean manifolds. Existing debiasing methods typically assume Euclidean representations or apply linear interventions, which can distort the geometry of structured latent spaces. We propose IsoAct, a post-hoc framework that accounts for geometry in both nuisance identification and representation editing. It identifies nuisance-predictive directions using Fisher discriminant analysis (FDA) in tangent coordinates and applies corrections through manifold-native actions that preserve manifold membership. Controlled and real-world experiments on hyperspherical, hyperbolic, and \mathrmSE(3) representations evaluate nuisance reduction, utility preservation, and manifold validity jointly. IsoAct reduces nuisance recoverability while retaining competitive utility. Its clearest joint result is on KLIFS, combining reduced kinase-group recoverability, strong structural prediction, and valid rigid poses.
[LG-72] Beyond Deduplication: Sharp Retrieval Limits for Attention with Correlated Distractors
链接: https://arxiv.org/abs/2610.34682
作者: Heehyun Park
类目: Machine Learning (cs.LG); Probability (math.PR)
*备注: 29 pages, 2 figures, 3 tables
Abstract:Redundant context affects attention through two distinct mechanisms: repeated items add normalization mass, while score variation within correlated groups can produce extreme individual scores. We separate these effects in a planted retrieval model with exponentially many Gaussian distractors arranged in known, exponentially large clusters. The model is a two-level generalized random energy model, whose classical free energy yields exact retrieval thresholds. Gaussian interpolation further shows that increasing positive correlation at fixed token count lowers the distractor free energy, unlike adding duplicates. Our main result identifies a sharp limitation of subtracting log cluster size from pre-softmax scores: this correction removes multiplicity but retains within-cluster extremes. Even a temperature chosen after observing all scores cannot lower the required signal below \sqrt2\alpha_1 , where \alpha_1 is the exponential rate of the number of clusters. Averaging scores before exponentiation instead suppresses these fluctuations and achieves \sqrt2\rho\alpha_1 , where \rho is the within-cluster correlation. The gap persists after adding exponentially many singleton distractors, so group size alone does not identify the target, and extends to fixed finite hierarchies of light-tailed independent-increment scores. Finite-sample, stability, and participation results complement the asymptotic analysis. A controlled retrieval experiment with frozen sentence embeddings supports this distinction while delineating its scope: averaging helps when the relevant item is a singleton, but can discard useful evidence when a relevant group contains multiple matching items.
[LG-73] AT-AT: Automatic Tool Optimization from Agent Traces
链接: https://arxiv.org/abs/2610.34679
作者: Maya Ravichandran,Pragya Tripathi,Dharamendra Kumar,Adith Swaminathan,Nathan Kallus
类目: Machine Learning (cs.LG)
*备注:
Abstract:Efforts to improve LLM-based agents typically focus on modifying the agent itself, yet success also depends on the environment in which an agent acts. Tools mediate this interaction: a tool’s description shapes how an agent calls it, and its implementation determines the results and errors the agent receives afterwards. Many tools, such as APIs and MCP servers, are shared by many agents and maintained by providers who cannot change those agents. We introduce AT-AT (Automatic Tool Optimization from Agent Traces), an agentic tool optimizer that improves a shared toolset offline from historical agent executions while holding the agents fixed. A lightweight coding agent reads agent trajectories and their evaluation results, and selectively edits tool descriptions and, when permitted, implementations. On \tau^2 -bench-verified, AT-AT achieves the highest task completion rates in the Airline and Telecom domains at lower cost than strong prompt- and tool-optimizer baselines, and its Telecom gain is statistically significant under paired permutation tests. Optimizing the tools within a coding agent harness also improves held-out performance on TerminalBench-2 and TBLite. When applied to three production MCP servers at a large enterprise, AT-AT lowered cost or tool error rates on every server and increased held-out task passes on one. By turning accumulated execution traces into updates to shared tool environments, AT-AT provides a path toward continual improvement of agent systems through environment-side learning.
[LG-74] Valid Early Detection of Memorization in Diffusion Models NEURIPS2026
链接: https://arxiv.org/abs/2610.34617
作者: Yixiao Jing,Chia-Yu Hsu,Milad Hoseinpour,Zekai Zhang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: Extended version of a 4-page paper accepted as a poster at the NeurIPS 2026 Workshop on E-Values: From Statistics to ML. 13 pages, 3 figures, 2 tables
Abstract:Diffusion models can sometimes reproduce examples from their training data, a phenomenon known as memorization. Existing memorization detection methods for unconditional diffusion models typically operate at a pre-selected timestep and therefore cannot adapt the detection time to the evolving denoising trajectory. Sequential monitoring enables adaptive detection but repeated testing across timesteps complicates trajectory-level false-alarm control. We study two ways of combining split conformal calibration with sequential monitoring for prompt-free memorization detection: per-step Bonferroni calibration and direct trajectory-level calibration of the maximum memorization score. The former controls repeated testing by splitting the false-alarm budget across timesteps, whereas the latter calibrates the event of ever crossing a threshold directly through the trajectory-level maximum. We evaluate this formulation on an unconditional CIFAR-10 EDM using two complementary per-step statistics based on geometry and representation spikiness. Experiments show that sequential monitoring improves the power-timing tradeoff over fixed-time detection. Compared with Bonferroni calibration, trajectory-level calibration achieves comparable detection power while using the false-alarm budget less conservatively and enabling earlier detection at higher noise levels.
[LG-75] Sparse Sum-of-Squares Layers for Certified Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.34590
作者: Ege Gursoy,Nicolas Mansard
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: Accepted to NeurIPS 2026 - Main track
Abstract:A central limitation of modern machine learning systems is the lack of formal certificates guaranteeing their behavior beyond empirical, sample-based evaluation. Sum-of-squares (SOS) methods provide such guarantees by certifying polynomial inequalities over continuous domains via semidefinite programming (SDP). While prior work has introduced SDP layers as differentiable components in neural networks, repeated learning calls can lose the computational benefit of the forward program’s sparse representation. We propose Sparse SOS Layers, differentiable value layers whose backward passes reuse the local coordinates of the forward solve. Under a fixed template and learned objective coefficients, each layer avoids building dense certificates and reuses its sparse structure across training steps. Compared with existing methods, our approach achieves up to 2.7 times speedup in total warm-call time, with nearly 100 times lower memory usage (1.7 vs. 160 GiB) for relatively large problem sizes for the tested dense parameterization. Experiments cover sparse polynomial certification, pendulum control, combinatorial relaxations, robotics projection, oscillator chains, and pose graph synchronization. A design study achieves zero certified violations across unseen problem instances and sizes, while requiring roughly 7 times fewer optimization solves than estimating gradients by perturbing the objective coefficients and re-solving the problem (finite differences). These results show that sparse certificate values can be effectively integrated into learning pipelines when the backward pass respects the verifier’s structure.
[LG-76] Cumulative Marginal Matching for One-Step Diffusion
链接: https://arxiv.org/abs/2610.34575
作者: Paribesh Regmi,Rui Li
类目: Machine Learning (cs.LG)
*备注:
Abstract:The sampling procedures of diffusion and flow-matching models involve hundreds of denoising steps incurring substantial inference time, which has motivated a growing body of work on training schemes for few-step and one-step generation. Existing approaches primarily seek to align the generation trajectory of few-step and base models, yet their training objectives only involve discrepancies at a single point in the trajectory, encouraging only local \emphpoint-level alignment. While recent methods such as \emphShortcut-CSL introduce cumulative objective that enforce alignment over the entire trajectory, this is suboptimal as they do not directly enforce alignment between the distributions induced by the two models throughout the generation process. We propose \emphCumulative Marginal Matching (CMM), a principled objective for few-step generation: by viewing flow-matching as a sequence of intermediate marginal distributions connecting the prior and data distributions, CMM directly optimizes the cumulative discrepancy between the base and few-step marginal distributions over the entire sequence. Specifically, we formulate this discrepancy using the Wasserstein-2 distance. By exploiting the pushforward representation of flow-matching, we transform an intractable cumulative Wasserstein objective into a tractable surrogate that can be optimized using samples from the intermediate marginal, which are readily generated through the conditional interpolation process of flow-matching. Furthermore, we show that the existing \emphShortcut-CSL objective arises as a first-order approximation of that of the CMM, establishing a theoretical connection between the two objectives. Experiments on standard benchmarks demonstrate improvement in generation quality over \emphShortcut-CSL and other state-of-the-art one- and few-step baselines under identical training budgets.
[LG-77] Conditional Diffusion Generation of DVL and Inertial Signals for Underwater Navigation
链接: https://arxiv.org/abs/2610.34574
作者: Michal Levin,Itzik Klein
类目: Machine Learning (cs.LG)
*备注:
Abstract:Autonomous underwater vehicles (AUVs) navigate without satellite positioning and rely on Doppler velocity log (DVL) and inertial sensors for motion estimation. Deep learning methods for processing these sensors require training data covering the motion the AUV encounters, commonly obtained through costly sea trials. To overcome this limitation, we present a conditional diffusion framework that generates synthetic DVL and inertial signals to replace real recordings in training. The generator employs an Elucidated diffusion model conditioned on three per-channel motion descriptors. Three configurations are evaluated: DVL beams alone, inertial channels alone, and all channels jointly from one model. Across thirteen trajectories of the real dataset, the generated signals followed the requested conditions with Pearson correlations between 0.72 and 0.997, transferred to trajectories withheld from generator training, and did not replicate any training window. A BeamsNet velocity estimator trained on generated signals alone, with no real recording retained, matched the velocity error obtained with the full real training set. Replacing up to all real windows with generated ones caused no measurable loss. Thus, this approach reduces the dependence of learned AUV navigation on sea-trial data, enables training when recordings are scarce, and facilitates developing and evaluating deep learning methods for underwater navigation.
[LG-78] Error Diffusion Replaces Latent Weights: Training Binary-Weight Language Models on a Few Bits of Dithered State NEURIPS2026
链接: https://arxiv.org/abs/2610.34566
作者: Pierre-Olivier Bonin
类目: Machine Learning (cs.LG)
*备注: Accepted at SLM-Agents: 1st Workshop on SLMs for Agentic Systems, NeurIPS 2026, Paris, France. 16 pages, 4 figures. Code: this https URL
Abstract:Binary-weight language models store one bit per weight, but training them keeps a full-precision latent copy of each weight plus Adam state, about 96 extra bits per weight, so a deployed binary model cannot be adapted on the device that runs it. We replace the latent weight with a 2-8-bit residual buffer quantized by stochastic rounding, which cuts the training state per weight from 97 bits to 3-9, up to 19x less. The resulting error-diffusion (ED) optimizer moves the residual-carrying loop of image halftoning onto the weight-update path. Each buffer accumulates the unrealized part of every update and flips the weight when the accumulated value crosses a threshold, and it works only when its quantization grid has a level exactly at that threshold. Under this threshold-alignment rule, 4-bit buffers match 8-bit buffers on a binary multilayer perceptron benchmark, and the same rule lifts a different latent-free optimizer, Bop, from a 10-point deficit at 4 bits to full-precision accuracy. On character-level transformers with 97% of parameters binary, ED trains within 5-11% of the latent-weight reference at a 2,000-step budget and within 1.7% at convergence. On scaling ladders over enwik8 and wikitext-103, up to 19M parameters on enwik8 (70M with the wikitext-103 embeddings), the latent-free loss stays within -2.3% to +2.9% of the reference at all six sizes once a per-step cap on the buffer update removes a launch instability; without the cap it trails by 10% at the largest tokenized size. A deployed binary model continues training under ED at one tenth of the latent training state, matching latent-weight continuation, and a fused CPU kernel completes the adaptation within a 275 MB training-memory budget, which latent-weight training cannot. A 250-step probe sets the flip learning rate at any new scale.
[LG-79] MOFjson: An Explicit Topological Representation of Metal-Organic Frameworks for Language Models
链接: https://arxiv.org/abs/2610.34549
作者: Anabel Yong,Matthew Hart,Philippe Schwaller
类目: Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注:
Abstract:Language models for metal-organic frameworks (MOFs) rely on formats that separate atomic geometry from structural topology, forcing models to choose between unstructured coordinate lists and non-spatial string identifiers[1]. We present MOFjson, the first representation separating structural data into a raw atomic layer with explicit bond shift vectors and a derived layer exposing secondary building units (metal nodes and organic linkers) and network topology. Additionally, we introduce a benchmark with crystal-derived ground truth across four question categories. Across four models, MOFjson beats all three baselines in every category: 100% vs. 18-47% on reaching a block’s nearest periodic copy, 96% vs. 71% on crystal system characterisation, and 60% vs. 48% on metal node and organic linker coordination. Ablating shift labels drops accuracy to 12.7%, proving models read the periodic graph rather than the surface format. MOFjson gives language models structure to read rather than reconstruct, establishing a foundation for agentic tool workflows and constrained MOF generation.
[LG-80] Multi-view Semantic-aware Resource Allocation for Edge AI
链接: https://arxiv.org/abs/2610.34532
作者: Jiawei Luo,Hussein Saleh,Mohamad Assaad
类目: Information Theory (cs.IT); Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: Submitted to IEEE Wireless Communications Letters. Jiawei Luo and Hussein Saleh contributed equally to this work
Abstract:Semantic communication is emerging as a pivotal paradigm to overcome wireless resource scarcity in 6G networks. This paper proposes a novel end-to-end framework for semantics-aware resource allocation in multi-user Edge AI. We extend the Opportunistic Information Bottleneck (OIB) to a multi-view setting, termed the Multi-view Gaussian Information Bottleneck (MGIB), by fusing complementary source views without variational approximation or network retraining. A resource allocation framework is then formulated for joint compression-level selection, transmission mode selection (i.e., the number of active sources per group), and PRB assignment. The problem is formulated as a Mixed-Integer Linear Program (MILP), solved efficiently via a proposed Lagrangian Relaxation (LR) algorithm, and benchmarked against the NP-hard optimal solution. Simulations demonstrate that the proposed LR scheme achieves near-optimal utility, and doubles PRB efficiency versus average allocation.
[LG-81] Ambiguous Fits: The Rashomon Effect and Challenges in Scaling Law Estimation
链接: https://arxiv.org/abs/2610.34514
作者: Bruno Mlodozeniec,Spencer Frei
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 31 pages
Abstract:Estimating neural scaling laws involves an end-to-end pipeline: designing a scaling recipe, training models across scales, evaluating models on a target metric, and deploying an algorithm to fit a scaling law to the resulting data. This final fitting stage is often treated as routine: standard procedures minimize a non-convex fitting loss using local optimization from a grid of initializations, and return the best solution found. Yet such heuristic searches can fail in two distinct ways: they offer no guarantee of returning a nearly optimal solution, and they obscure whether fundamentally different parameter choices fit the data nearly as well. Indeed, we show that standard scaling law fitting is subject to a Rashomon effect (Breiman, 2001): the sublevel sets of near-minimizers correspond to broad parameter regions whose extrapolations diverge substantially at scale. We resolve these issues with a deterministic branch-and-bound algorithm that guarantees nearly optimal fitting loss and exhaustively maps the near-minimizers to the implied range of scaling curves.
[LG-82] Best Policy Tracking in Reinforcement Learning
链接: https://arxiv.org/abs/2610.34501
作者: Judith Echevarrieta,Etor Arza,Aritz Pérez,Josu Ceberio
类目: Machine Learning (cs.LG)
*备注: Preprint submitted to Transactions on Machine Learning Research (TMLR), this https URL
Abstract:Reinforcement learning commonly exhibits non-monotonic performance dynamics in single-task settings. During learning, a sequence of policies is generated whose true performance is unknown and may oscillate significantly across iterations. Despite this well-known phenomenon, standard experimental practice typically reports either the final visited policy or the best among periodically evaluated ones, implicitly assuming that these selection criteria reliably approximate the best-performing policy visited throughout the entire sequence. In this work, we present an empirical study on the implications of non-monotonicity for policy selection in single-task policy-based reinforcement learning. We introduce post-hoc diagnostic metrics to characterize performance degradation and assess the accuracy and efficiency of different selection criteria. These metrics allow us to evaluate commonly used selection strategies across standard continuous-control benchmarks using a well-established on-policy actor-critic method. Our results show that, in this representative experimental setting, widely adopted selection practices can deviate considerably from the best policy encountered during learning, leading to underperforming estimates and suboptimal use of computational resources. As an alternative, we recommend a validation-based policy selection strategy constrained by computational cost, which provides a more consistent and accurate approximation of the best visited policy while improving selection efficiency. Overall, our findings highlight policy selection as an underexplored factor in the evaluation protocols of reinforcement learning and motivate further study on explicit and standardized validation-based selection procedures. Comments: Preprint submitted to Transactions on Machine Learning Research (TMLR), this https URL Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.34501 [cs.LG] (or arXiv:2610.34501v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.34501 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Journalreference: Transactions on Machine Learning Research, 10/2026
[LG-83] Approximate Unlearning in Prototype-based-classifiers
链接: https://arxiv.org/abs/2610.34495
作者: Sreejita Ghosh,Mohammad Mohammadi
类目: Machine Learning (cs.LG)
*备注: Submitted at the moment to Knowledge-Based Systems
Abstract:The recent data and AI regulations and governance needs have highlighted the need for removing specific data even from models trained on or with them. While retraining the entire model is a theoretically trivial solution, it is computationally expensive, thereby motivating research in Machine Unlearning. This paper introduces a robust approximate unlearning technique for the family of prototype-based classifiers called Generalized Learning Vector Quantization, aimed at removal of effects of specified samples, either due to deletion request or due to them being identified as noisy or corrupt. We first show the impact of the different sizes of the forget set, as well as the characteristics of the data in this set, on the unlearned model. To address the sources of instabilities detected in this unlearning method we propose an adapted version of this technique, such that the unlearning is robust to the characteristics of the forget set. Thereafter using the intrinsic interpretability property of GLVQ and appropriate metrics defined in unlearning literature, we show the efficacy (approximately 0 deviation of parameters and performance of unlearned model relative to retrained model) and efficiency (6X for smallest dataset to 10^(5)X in the largest dataset) of the proposed unlearning techniques compared to naive retrained, on six benchmark datasets for tabular data recommended by literature. Although in this work we investigated only binary class problems, the proposed robust unlearning strategy can be readily applied for removing sample effects from any trained GLVQ model, i.e., it does not require any modification in the training scheme of GLVQ.
[LG-84] Models Verify Sufficiency Not Validity: Usable Falsehoods Evade the Checks that Gaps Trigger NEURIPS2026
链接: https://arxiv.org/abs/2610.34488
作者: Naveen Madiraju
类目: Machine Learning (cs.LG)
*备注: Accepted at the NeurIPS 2026 Workshop on Self-Evolving Diversity-Driven Search for Robust AI Systems (EvoRobust). 5 pages plus references, 3 tables
Abstract:Ask a model “how many settlements did the Normans control?” and it answers that the passage does not say. Tell it “given that the Normans controlled 12 settlements of 400 inhabitants, what is the total?” and it computes 4,800. Nothing was added to the passage; the unsupported fact merely became usable. Benchmarks for unanswerable questions are built almost entirely by omission, deleting the fact a question needs and scoring refusal, so they never probe this. Holding source support fixed and varying only whether an unsupported premise is usable, flagging drops 8-30 points and commitment to an answer rises 1.5-2.3x on GSM8K across four models spanning two families and 2B-26B; all 16 paired contrasts remain significant after Holm correction. Two further manipulations reproduce it, one also on SQuAD: a fact asserted rather than asked about, and a corrupted intermediate value rather than a removed one. Models articulate the exact gap they then ignore, stating that “there is no financial information provided” for an inert premise and silently adopting a fabricated price for the same item. We read these as one behaviour: whether a model checks depends substantially on whether usable material is present, even when source support is held fixed. Because the main design never instructs models to reject premises introduced in the question, adopting such a premise can also be read as conditional reasoning, and we delimit what the design does and does not show. In the judging seat the pattern differs: asked whether the answer follows, judges pass solutions with a required step deleted 78-99% of the time at every scale tested. A natural prompt-level mitigation lowers accuracy by 22 points without improving detection.
[LG-85] Finite-Sample Sensitivity of Kernel Metrics for Generative Evaluation
链接: https://arxiv.org/abs/2610.34477
作者: Tomasz Radzikowski,Mateusz Modrzejewski
类目: Machine Learning (cs.LG); Sound (cs.SD)
*备注: 21 pages, 2 figures, 20 tables
Abstract:Kernel distances such as the Kernel Inception Distance (KID) and the Kernel Audio Distance (KAD) compare generated and reference samples in an embedding space, and model comparisons rest on their values. Each ships with one fixed kernel, and a reported value does not say which changes that kernel detects at the sample size used. We run each default as a permutation two-sample test on real embeddings of symbolic music, audio and images, comparing a sample with an independent sample changed by a mean shift, an isotropic contraction, or a rotation that keeps the mean and the covariance trace and is matched to the shift in transport distance. At 780 samples per side, both defaults rarely detect the rotations in five evaluation spaces, while they detect the shifts in every run. The median-bandwidth Gaussian kernel of KAD detects the contraction and regains power against rotations on image embeddings with an order of magnitude more samples. The cubic kernel of KID contains second- and third-moment information but misses the contraction in most runs, and our experiments do not establish a high-power rotation regime for it over the tested sample sizes. Narrower Gaussian bandwidths restore sensitivity in music and audio, an \ell_1 Laplace kernel does so for images, and aggregated tests combine both at nominal level. We recommend that a reported kernel distance state its sample budget and the changes its configuration detects at that budget.
[LG-86] IntNet: A Communication-Driven Multi-Agent Reinforcement Learning Framework for Cooperative Autonomous Driving
链接: https://arxiv.org/abs/2610.34466
作者: Leandro Parada,Kevin Yu,Panagiotis Angeloudis
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注:
Abstract:Achieving safety in autonomous driving through Multi-Agent Reinforcement Learning (MARL) is a critical yet challenging task due to non-stationarity, partial observability, and the need for effective coordination among agents. Although earlier cooperative MARL methods have aimed to improve coordination by sharing encoded state observations, we find these strategies are insufficient in safety-critical scenarios. To address this gap, we present IntNet, a novel MARL framework that enhances safety by incorporating the transmission of vehicle intent and adaptive communication scheduling into a unified end-to-end learning paradigm, jointly optimising all components for safe coordination. Central to our approach is an observation prediction module that enables agents to forecast subsequent policy outputs, which they can then share across a vehicle-to-vehicle network. Our model-free architecture employs a Graph Attention Network to process incoming messages and a scheduler component to dynamically optimise the communication graph for bandwidth efficiency. We compare IntNet against state-of-the-art communicative MARL methods in complex urban environments with both autonomous and human-operated vehicles, achieving improved coordination, lower collision rates, and reducing communication efforts by up to 60%. Through extensive experiments, we assess the framework’s learning efficiency, scalability, and the balance between information sharing and bandwidth usage for collision-free trajectories.
[LG-87] From Explaining to Enabling: Actionable Transparency in Algorithmic Insurance Pricing
链接: https://arxiv.org/abs/2610.34459
作者: Agathe Fernandes Machado,Fei Huang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine learning (ML) systems increasingly make consequential decisions about individuals, including credit, hiring, and insurance. Research on ML-driven insurance pricing has asked how the pricing mechanism itself can be made accountable. Yet it remains unclear how an individual policyholder’s algorithmically determined premium should be explained to them. We argue that a useful explanation must be valid, actionable, and reliable. A valid recommendation lowers the premium according to the model. An actionable recommendation changes only characteristics the policyholder can control. A reliable recommendation has an actual outcome that is likely to meet the target, even in sparsely observed regions of the feature space where the model is uncertain. We formalize these three requirements as axioms and develop a method that turns an algorithmic premium into recommendations that meet them. The method meets the third requirement with a Conformal Prediction constraint, which screens out recommendations that the model is too uncertain about and thereby bounds the rate of unreliable recommendations. We apply four Counterfactual Explanation techniques to 57,832 real Florida flood-insurance claims and 98,058 synthetic driver-telematics policies. We adapt these techniques for insurance applications, so that their recommendations change only controllable characteristics and resemble policyholders observed in the data. In our illustrations, the constrained recommendations for high-risk policyholders are both more reliable and associated with larger predicted risk reductions than the unconstrained ones. These results suggest that insurers can give policyholders explanations that are both actionable and trustworthy. The work contributes to research on algorithmic accountability and offers implications for insurers and regulators.
[LG-88] Speculative Decoding for Mixture-of-Experts Models: Expert Bytes Kernel Bandwidth and the Best Draft Length
链接: https://arxiv.org/abs/2610.34440
作者: Yuzheng Lan
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF)
*备注: 35 pages, 7 figures, 8 tables
Abstract:Speculative decoding checks several drafted tokens in one verification pass. On a mixture-of-experts (MoE) model, consecutive tokens often select the same experts. A deduplicating engine reads each selected expert once per pass, so it requests fewer bytes than a per-slot engine, which reads it once per token. We write a step’s time as the bytes each kernel requests over its effective bandwidth, plus fixed time. Under geometric acceptance, we derive from this step time when speculation is faster than plain decoding. The best draft length is a monotone staircase in the acceptance rate. We bound how a bandwidth error moves each result, and we check the main steps in Lean. Deduplication is faster exactly where its share \bar u(n) of the per-slot bytes is below its ratio r(n) of the per-slot bandwidth. On this http URL Metal, MLX-LM and this http URL CUDA, which we modified to deduplicate, the per-slot kernel has the higher effective bandwidth (r = 0.32 to 0.75). Thus deduplication is faster only from 5 to 8 tokens per pass, or never. We fit the bandwidths on one layer’s expert operation, run alone, in under an hour per model. Bytes from routing traces over these bandwidths predict the time that deduplication saves. The measured saving lies inside its error bound at every point on CUDA and MLX-LM and at 36 of 48 points on Metal. The crossover is the first pass size at which deduplication is faster. The prediction places it within one pass size in 10 of 12 configurations of engine, model and depth. We computed these predictions after the measurements. On DeepSeek-V2-Lite, operation times measured before its speculative runs give the best draft length on CUDA within one at all 22 acceptance values.
[LG-89] ScaleRLT: Exact Log-Probabilities and Fast Deferred Rollouts for Reinforcement Learning on Looped Transformers
链接: https://arxiv.org/abs/2610.34427
作者: Dezhen Lu,Zixiang Ni,Runguo Li
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注:
Abstract:Looped language models such as Ouro reuse one block of layers several times per token. A rollout engine can batch the deeper loops of several response positions and run them only when a deep readout is due. This deferred schedule is fast, but the learner of an on-policy RL algorithm recomputes the sampled tokens’ log-probabilities with a dense pass, and the two computations differ in floating point: in FP32 at 4096 tokens about 18% of the sampled tokens miss a 10^-5 same-weight gate, and with bf16 weights nearly every token does. We use a hybrid learner log-probability, \ell=\ell_\mathrmdense+\mathrmsg(\ell_\mathrmsched-\ell_\mathrmdense) , where \mathrmsg denotes stop-gradient. Its value is a no-grad replay through the rollout’s own schedule, and its gradient is the dense learner’s. The replay reproduces the rollout bit for bit in all 167 FP32 and 28 bf16 real-weight cells we ran on three platforms, including rollouts whose batch shrinks at stop tokens, and the gradient has the same FP32 error as other dense implementations in all 102 pre-registered comparisons from 96 to 4096 tokens; the full GRPO loss of a training step matches a float64 reference to 3.1\times10^-4 . On one GPU, deferral raises rollout throughput by 1.47 to 1.56\times , and by 1.18 to 1.54\times once every step runs from CUDA graphs, bit for bit. In a two-node GRPO loop, one-wave KV admission and CUDA graphs shorten an update by 2.71\times with the gate bitwise at every update, and four nodes keep the update time within 2% at the same per-GPU load. In 100-update runs held-out GSM8K accuracy rose from 23% to 82% with the mixed-depth readout that deferral relies on and from 34% to 90% with the native readout, which reached a higher accuracy in less time. For looped models beyond one node we analyse a ring mapping of layers to GPUs and measure its communication paths.
[LG-90] From GPU to TPU with Coding Agents : What Works What Fails and What It Costs? ICSE2027
链接: https://arxiv.org/abs/2610.34422
作者: Hui Sun,Anderson Uchôa,Wesley K. G. Assunção
类目: oftware Engineering (cs.SE); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注: 12 pages, 2 figures, 2 tables. Submitted to the ICSE 2027 Software Engineering in Practice (SEIP) track
Abstract:A coding agent can migrate a machine learning workload, report success, and still change how the model trains. Rising compute costs, limited accelerator capacity, and vendor dependence are pushing organizations toward alternatives to graphics processing units (GPUs); a key question is whether agent-generated migrations to tensor processing units (TPUs) can be trusted and at what cost. We build a dataset of 148 PyTorch ML research projects and use Claude Code to migrate ten of them from NVIDIA GPUs to Google Cloud TPUs. We independently assess whether each migration runs, preserves numerical behavior, executes on the intended hardware, maintains the project’s goal, and delivers acceptable performance. We also examine failures, human effort, and costs. Eight projects meet our project-goal criterion, and migrations require 2.4 human interventions on average. Of 70 execution attempts, 53 fail (76%), and 26 of these failures (49%) raise no error. The most frequent cause is repeated compilation and other execution inefficiencies, while interactions with third-party libraries can silently change the computation: in one four-chip migration, the agent reports success despite a change in gradient clipping that reduces accuracy from 86.53% to 84.89%. Execution costs vary widely: in one workload, TorchTPU, a PyTorch execution backend for TPUs, trains 1.86 times faster and 25% cheaper per run than the GPU. For practitioners, these findings motivate independent reference runs, verification under the intended hardware configuration, and workload-specific acceptance and stopping criteria that account for both computational and human costs.
[LG-91] From Human Intent to In-Orbit Action: A Self-Learning Large Language Model Agent for 6G Satellite Networks
链接: https://arxiv.org/abs/2610.34387
作者: Zhengqian Xu,Yasheng Zhang,Chenhua Sun,Shan Lu,Chengcheng Li
类目: Machine Learning (cs.LG)
*备注: 18 pages, 9 figures
Abstract:Autonomous control of sixth-generation (6G) low Earth orbit satellite networks is hindered by the gap between high-level human intent and low-level control logic: translating a request such as “uninterrupted Arctic video” into a valid reinforcement learning reward still takes days of manual trial and error, and existing large language model translators cannot learn from the failures of the policies they generate. This article presents ALMOND (Agentic Large-Model-Orchestrated autonomous Network Design), a cognitive framework that closes this intent-to-action loop under on-orbit compute constraints. The contribution in artificial intelligence is a bidirectional semantic engine that translates natural-language demands into multi-objective reward functions and, conversely, mines implicit operational constraints from failed reinforcement-learning trajectories through counterfactually validated forensics, together with a large language model-guided graph neuroevolution that seeds and critiques policy architectures. The contribution in engineering is a ground-centric, compute-frugal architecture for autonomous 6G non-terrestrial network operation, in which satellites execute only distilled policies below 50,000 parameters with sub-15 ms inference and no on-board large language model. On a 350-satellite digital twin driven by real ephemeris data, ALMOND raises polar handover success from 55% to over 95%, converges about three times faster than baseline reinforcement learning, and reduces reward engineering from 48 hours to 4.5 hours, demonstrating a practical path toward trustworthy in-orbit autonomy.
[LG-92] Does Notation Matter? MusicXML Tokenisation for Symbolic Music Generation and Understanding
链接: https://arxiv.org/abs/2610.34247
作者: Andrea Poltronieri,Matteo Spanio,Simone Chieppa,Xavier Serra,Martín Rocamora
类目: ound (cs.SD); Machine Learning (cs.LG)
*备注: 9 pages, 1 figure, 4 tables
Abstract:Symbolic music representations encode musical content at varying levels of completeness, from performance-oriented formats such as MIDI to fully notated scores such as MusicXML. Yet the impact of this representational richness on deep learning models remains poorly understood. While tokenisation has been extensively studied for MIDI, score-based formats have received considerably less attention, and MusicXML in particular remains largely unexplored despite being the de-facto standard for score interchange. In this paper, we introduce the first general-purpose tokenisation framework for MusicXML, and conduct a systematic empirical study to determine whether notational richness confers a measurable advantage for deep learning on symbolic music. We train 26 autoregressive and masked language models, comparing MusicXML tokenisation against MIDI and ABC-based representations, probing the effect of different tokenisation strategies within MusicXML, and testing whether music-grammar-aware tokenisation outperforms treating scores as plain text. We evaluate music understanding tasks using linear probing across composer classification, difficulty estimation, and accompaniment suggestion, and generative tasks using objective statistical metrics and a human pairwise preference study. Our results show that MusicXML-tokenised models outperform MIDI and text-based counterparts on understanding and generative tasks.
[LG-93] Alignment via Generalization
链接: https://arxiv.org/abs/2610.34242
作者: Shai Shalev-Shwartz,Amnon Shashua
类目: Machine Learning (cs.LG)
*备注:
Abstract:AI systems can become better at achieving the goals we give them without becoming better at doing what we actually want. Studying this alignment problem rigorously requires a framework connecting measurable performance to intended behavior. We propose Aligned-PAC learning, whose central idea is to reduce alignment to generalization: does success on the requirements used for training extend to requirements the system has not encountered? Rather than express our entire intention through a single training objective, we represent its different aspects through a collection of specific requirements. This perspective already yields strong theoretical results. We show how independent evaluation can provide rigorous evidence of alignment across the represented requirements, and establish how much evaluation is needed. We also develop an optimization method that, under explicit assumptions, combines learned models to satisfy the training requirements. The framework does not solve alignment in full: choosing requirements that faithfully capture our intentions, and understanding when learning will generalize to unseen requirements, remain open challenges. It provides a foundation for making these questions precise and studying them with the tools of learning theory.
[LG-94] Interpretable and Steerable Molecular Graph Generation using Concept Bottlenecks NEURIPS2026
链接: https://arxiv.org/abs/2610.34227
作者: Sarabeshwar Balaji,Ritesh Kumar,Lav R. Varshney
类目: Machine Learning (cs.LG)
*备注: Accepted at AI for Drug Discovery Workshop @ NeurIPS 2026. Additional experiments added
Abstract:Deep graph generators (DGGs) largely lack interpretability, which is a significant limitation for one of their most crucial applications, drug discovery, where interpretable and transparent generation could provide greater control over the desirability of generated molecules, thereby reducing trial-and-error in candidate selection. Common approaches to interpretability, such as disentanglement, cannot guarantee that the learned representations are human-interpretable. To address this, we propose incorporating concept bottlenecks into DGGs, allowing users to interpret, debug, and steer the generation process. We show that this consistently improves distributional fidelity (lower FCD), with only modest, non-systematic effects on validity and uniqueness. We further explore Concept Controller, an optimization-based steering approach that has proven to be effective for steerable image generation. We evaluate our approach on a one-shot molecular graph generation framework, though the underlying ideas can be transferred to sequential graph generation as well.
[LG-95] AudioFM: Towards General Purpose Audio Representation Learning
链接: https://arxiv.org/abs/2610.34183
作者: Yamin Li,Ruidong Zhang,Anran Wang,Eduardo Fonseca,Dan Ellis,Zechen Li,Simon A. Lee,Tao Pang,Jake Garrison,Maxwell A. Xu,Yicheng Wang,Jitian Zhang,Weizhi Zhang,Haichao Zhang,Catie Chang,Xin Liu,Anupam Pathak,Daniel McDuff,Shwetak Patel,Jake Sunshine,Yuzhe Yang,Xiaoran Fan
类目: ound (cs.SD); Machine Learning (cs.LG)
*备注:
Abstract:Learning audio representations that effectively bridge holistic global soundscapes with fine-grained acoustic moments is a long-standing goal in self-supervised learning. Recording-level contrastive objectives encourage agreement between pooled audio representations, but do not explicitly supervise local time-frequency representations. We introduce AudioFM, a self-supervised framework that combines recording-level contrast with patch-wise masked latent discrimination. Using visible acoustic context, a student predicts representations of masked patches and identifies their corresponding full-view teacher targets among candidates from other recordings. We further incorporate weaker supervision from eligible temporal neighbors to encourage local temporal consistency. Extensive evaluations show that AudioFM achieves state-of-the-art performance across diverse audio domains in linear probing, fine-tuning, and zero-shot transfer, demonstrating superior event detection, robust scene-level understanding, and strong domain generalizability.
[LG-96] Last-Iterate Stationarity of Muon with Finitely Many Newton-Schulz Iterations
链接: https://arxiv.org/abs/2610.34158
作者: Ping Li,Vincent Y. F. Tan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Muon has achieved strong empirical performance in large-scale language-model training, motivating a better theoretical understanding of its practical implementation. Unlike the idealized polar-factor update, practical Muon uses a fixed number of quintic Newton–Schulz iterations. Existing analyses mainly establish average- or best-iterate stationarity guarantees, leaving last-iterate convergence less understood. We study last-iterate gradient convergence of practical finite-step Muon on smooth, lower-bounded nonconvex objectives. For deterministic gradients, we show that without input stabilization, Muon can cycle even on a smooth convex quadratic. With positive stabilization, we establish last-iterate gradient convergence for any fixed momentum coefficient below one and a sufficiently small constant learning rate, together with square-summability of gradients. For stochastic gradients, we first show that persistent bounded, unbiased noise can prevent convergence with fixed learning rates and momentum, even with positive stabilization. We then establish almost-sure and mean-square last-iterate gradient convergence by combining decreasing learning rates with increasing momentum under bounded-variance stochastic gradients. Finally, we construct a smooth counterexample showing that neither convergence theorem guarantees a pointwise polynomial rate. We further establish density-one convergence rates, guaranteeing polynomial gradient decay for all but a vanishing fraction of iterations. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.34158 [cs.LG] (or arXiv:2610.34158v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.34158 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-97] Edge-Centric Adaptive Neural Activation via Parametric Tanh Functions
链接: https://arxiv.org/abs/2610.34154
作者: Andreas Falkenberg
类目: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
*备注: 11 pages, 4 figures
Abstract:Within the broader landscape of artificial intelligence architectures, this paper introduces a compact, edge-centric neural framework that resolves the computational bottlenecks of modern edge-activated paradigms like Kolmogorov-Arnold Networks (KANs) while retaining their expressiveness over traditional Multi-Layer Perceptrons (MLPs). While KANs demonstrate remarkable representational capacity by placing learnable univariate functions on connection edges rather than hidden nodes, their reliance on non-local B-splines introduces severe optimization hurdles, high memory overhead, and poor hardware scaling. To bridge this gap, we propose the Parametric Tanh Connection Function, a smooth, bounded architectural primitive defined as a_\tanh(x;\delta,\alpha,\beta,\gamma) = \delta + \alpha \tanh(\beta x + \gamma) . This formulation unifies offset, amplitude, local slope, and horizontal displacement into a lightweight, 4-parameter differentiable mapping optimized edge-by-edge via backpropagation. Evaluated on dense architectures using MNIST and CIFAR-10, our parametric tanh connection model achieves significant absolute accuracy gains over traditional affine baselines-reaching 95.63% \pm 0.08% on MNIST and 44.74% \pm 0.18% on CIFAR-10. Crucially, it matches or exceeds the expressiveness of unconstrained 7-parameter exponential connection baselines while requiring up to 85% less training runtime. By replacing complex splines or hyper-flexible exponents with a bounded, analytically smooth hyperbolic tangent kernel, this work delivers a highly scalable, hardware-friendly compromise for heterogeneous connection-level deep learning.
[LG-98] Execution Logs Sway VLM Judges of Robot Progress
链接: https://arxiv.org/abs/2610.34151
作者: Weiyan Zhao,Sunny Yang
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: SPAIS Workshop @ CoRL 2026
Abstract:Vision-language models now grade robot videos, and text placed beside the frames can change their verdicts. On clips where the robot is losing progress, we add one log line saying the robot is making progress. Across Gemini, Qwen and GPT judges, this flips correct verdicts to progress 10 to 35 percentage points more often than a neutral line does. A one-line warning that the log may be wrong removes most of this effect. Selected fake “task complete” claims flip half of Qwen’s correct verdicts, against a tenth for unrelated log lines. They have no detectable effect on Gemini.
[LG-99] Bug Severity Classification with Fine-Tuned Open LLM s: A Paired Comparison with TF-IDF and CodeBERT Baselines on Eclipse Reports
链接: https://arxiv.org/abs/2610.34127
作者: Asset Kanatov,Talgar Bayan,Askar Boranbayev
类目: oftware Engineering (cs.SE); Machine Learning (cs.LG)
*备注: 17 pages, 5 figures, 8 tables. Code and data: this https URL
Abstract:Open large language models (LLMs) small enough to run on a laptop are an obvious candidate for automatic bug severity prediction. We measure what they add over cheaper classifiers on one task: assigning one of four Bugzilla severity levels (Critical, Major, Minor, Trivial) to Eclipse bug reports. From a public dump of 361,006 unique Eclipse reports we draw a class-balanced sample of 20,000 rows and compare eleven configurations of nine model types: six TF-IDF classifiers, a fine-tuned CodeBERT encoder, and Llama 3 8B and Qwen 2.5 7B, each run zero-shot and after QLoRA fine-tuning. All eleven are scored on the same 2,000 held-out reports, with bootstrap intervals and McNemar tests. The zero-shot runs reach macro-F1 0.305 and 0.321, below every TF-IDF model (0.430-0.489). The fine-tuned runs reach 0.526 and 0.485, but the zero-shot and fine-tuned setups also differ in checkpoint, serving runtime and decoding, so this gap cannot be credited to fine-tuning alone. Fine-tuned Llama 3 shows no detectable difference from CodeBERT (0.518; difference +0.007, 95% CI -0.017 to +0.032), and fine-tuned Qwen 2.5 is below it on macro-F1. The models differ more by class than in aggregate: fine-tuned Llama 3 finds 36% of Critical reports, Qwen 2.5 finds 69% but only 21% of Minor ones. The dump repeats 60,000 rows, and 8% of the test reports duplicate a training report; removing them lowers TF-IDF scores by up to 0.031 and leaves the LLMs almost unchanged. We release split identifiers, per-report predictions and the analysis script at this https URL, together with CyberTriage, a tool that serves the fine-tuned classifier locally.
[LG-100] Enhancing Type 2 Diabetes Diagnosis with Evolutionary Algorithms and Machine Learning
链接: https://arxiv.org/abs/2610.34123
作者: Parisa Rezaei,Mohsen Saffar,Hamid Reza Naji,Mohammad Mehdi Faghih,Rasoul Nouriazar
类目: Machine Learning (cs.LG)
*备注:
Abstract:Type 2 diabetes is a widespread chronic disease that requires early and accurate diagnosis to prevent serious complications. Traditional diagnostic methods often lack sufficient accuracy, which highlights the need for more reliable computational solutions. In this study, we propose a model that integrates the Random Forest classifier with the Bat Optimization Algorithm for simultaneous feature selection and hyperparameter tuning. The SMOTE-ENN method was first applied to the Pima Indians Diabetes Database to correct class imbalance and remove noisy or ambiguous samples, producing a more balanced and cleaner dataset. The optimized model trained on this refined dataset achieved 89% accuracy, 88% precision, 90% recall, and an F1-score of 89%, clearly outperforming the baseline Random Forest and other existing approaches. These results demonstrate the potential of combining evolutionary algorithms with ensemble learning to provide a practical and cost-effective tool for early detection of type 2 diabetes in clinical practice.
[LG-101] INTRA: Interleaved Non-contiguous Token spaRse Attention
链接: https://arxiv.org/abs/2610.34056
作者: Yuxuan Xia,Wenbo Lu,Shaoyi Zheng,Shengjie Wang
类目: Machine Learning (cs.LG)
*备注: 26 pages, 8 figures, including appendices. Code: this https URL
Abstract:Transformers achieve strong performance across modalities but are bottlenecked by the quadratic cost of attention. Sparse attention can reduce this cost, but effective sparsity must preserve useful long-range communication and map well to GPU hardware. We introduce INTRA–Interleaved Non-contiguous Token spaRse Attention–a hardware-aware static sparse attention design based on a relay-communication view under Computational Query Set (CQS) constraints. Rather than directly covering all token-to-token interactions within each sparse layer, INTRA uses complementary sparse groups to provide short mediated communication paths across layers. Scatter forms relay groups that sample broadly across local neighborhoods, while Gather reconnects these groups in subsequent layers. Under a partition model, this structure guides CQS-compatible sparse patterns with global reachability. Our CQS-aware FlashAttention-style kernel computes structured indices at runtime, loads selected non-contiguous tokens, and packs them densely for computation. On an A100, INTRA achieves a 1.53\times speedup in 2K FLUX.1-dev generation, with better measured image quality than evaluated CLEAR configurations after LoRA self-distillation. On LLaMA-3.1-8B, it improves long-context prefill efficiency with strong teacher-forced retrieval performance under matched LoRA adaptation. Attention-only benchmarks further show up to 8.4\times and 10.4\times speedups over MoBA and density-constrained NSA, respectively.
[LG-102] Leverag ing Dales Principle as an Inductive Bias in Recurrent Neural Networks
链接: https://arxiv.org/abs/2610.34052
作者: Jiyi Wang,Chenxiao Yang,Yujia Zhao,Jingzhao Zhang,Lu Mi
类目: Machine Learning (cs.LG)
*备注: 48 pages, 38 figures
Abstract:One of the fundamental differences between biological neural networks and artificial neural networks (ANNs) is that biological neurons obey Dale’s principle: each neuron projects either exclusively excitatory or exclusively inhibitory outputs. While this constraint is central to biological circuits, previous attempts to impose Dale’s principle in ANNs have often led to training instability or degraded performance. In this work, we introduce EISep RNN, a recurrent neural network architecture that implements Dale’s principle through a learnable excitatory-inhibitory separation. Across a series of learning settings, including multi-task learning, continual learning, and reinforcement learning, EISep achieves more stable training dynamics, competitive or improved performance, and sparser, more modular connectivity than vanilla RNNs and previous Dale-constrained models. EISep also exhibits stable and informative latent dynamics. These results establish Dale’s principle as a biologically grounded and computationally effective inductive bias for recurrent neural networks.
[LG-103] Minimax Regret for Sobolev Kernel-Based Reinforcement Learning
链接: https://arxiv.org/abs/2610.34045
作者: Haichen Hu,David Simchi-Levi
类目: Machine Learning (cs.LG); Statistics Theory (math.ST); Machine Learning (stat.ML)
*备注:
Abstract:We study the minimax regret rate for episodic kernel-based reinforcement learning, where the expected reward and transition density belong to a Sobolev reproducing kernel Hilbert space, with continuous state and action spaces, and adversarial initial states. For Sobolev smoothness parameter \nu0 , state dimension d_S , and action dimension d_A , we prove that the minimax expected regret over T episodes is \widetilde\Theta(T^\max\e_A,e_B) , where e_A=\frac\nu+d_A2\nu+d_A , e_B=\frac\nu+d_A+3d_S/22\nu+d_A+2d_S with the other model parameters fixed. The two exponents reflect the distinct difficulties of action selection and learning across states. We introduce Sobolev Optimal Value Iteration (SOVI), which attains this rate from observed rewards and transitions. For fixed smoothness and reference-kernel lengthscale, its regret bound has a prefactor polynomial in the state-action dimension d=d_S+d_A , horizon H , transition and reward RKHS norm bounds u,B_r , and kernel norm-comparison factor \Lambda , up to logarithmic factors. The algorithm separates initial action selection from continuation-value estimation and combines local polynomial approximation with the shared RKHS structure of the approximation errors. Our matching upper and lower bounds determine the optimal dependence on the number of episodes for Sobolev kernel-based reinforcement learning. Thus, we partially address the open problem in Vakili(2024) under the condition of the Sobolev kernel setup.
[LG-104] he Symmetry of Differential Neural Tangent Kernel under the Orthogonal Group
链接: https://arxiv.org/abs/2610.34023
作者: Bangti Jin,Longjun Wu
类目: Machine Learning (cs.LG); Group Theory (math.GR)
*备注: 25 pages, 2 figures
Abstract:The differential neural tangent kernel (DNTK) is a tool for analyzing the training dynamics of overparameterized physics informed neural networks. In this work, we investigate the symmetry of the infinite DNTK under an orthogonal group, for two layer and deep neural networks separately, based on the explicit expression of the DNTK and recursive formula, respectively. The results are for all activations with or without a bias term, and shed insight into the structure of the infinite width DNTK. Then we prove that the integral operators over the unit sphere induced by operators from the same group orbit share the spectrum, and their reproducing kernel Hilbert spaces differ only by an orthogonal transformation, and that the operator fixed DNTK exhibits zonal symmetry on the unit sphere for symmetric operators. Thus the eigenvalue decay rate of the associated integral operator can be derived for shallow networks with RePU activation and zero bias initialization.
[LG-105] he Information Geometry of Representation Formation
链接: https://arxiv.org/abs/2610.34015
作者: Sanjukta Krishnagopal
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 33 pages, 9 figures. Code available at this https URL
Abstract:Most analyses of neural representations focus on trained endpoints, leaving the process by which internal structure forms comparatively unmeasured. We instead treat representation formation itself as an object of study, asking whether the path a representation takes through training reveals information that its endpoint does not. We formalize this idea by viewing representation learning as a geometric trajectory, using Bures geometry to measure movement through representation space and quantum Fisher information to characterize local sensitivity to input factors. We find that learning concentrates local sensitivity on task-relevant factors, and that changing only the prediction target reorganizes this geometry around the corresponding visual factor. More strikingly, representation trajectories become systematically less direct as the learned computation requires higher-order interactions among input factors. This effect persists under matched-performance and training-duration controls and transfers from synthetic vector inputs to residual CNNs on rendered and natural images. Importantly, formation history contains predictive information about the stability of the learned representation: controlling for interaction order, path tortuosity predicts its subsequent drift under continued training and sensitivity to parameter perturbations, whereas endpoint displacement does not. Finally, representation-aware optimization lets us actively shape these trajectories, showing that formation geometry is not only measurable but actionable. Together, these results establish representation formation as a new source of information about learning: its geometry reveals properties of the computation being constructed, predicts subsequent representational stability, and provides a target for intervention during optimization.
[LG-106] Cross-User Adaptation for sEMG Gesture Recognition via Continual Learning
链接: https://arxiv.org/abs/2610.34011
作者: Aran Ramakrishnan,Shubhi Mishra,Tyler Ham,Assel Kembay,Jason Eshraghian
类目: Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: 4 pages, 2 figures, 3 tables. Accepted at IEEE BioCAS 2026
Abstract:Prosthetic and assistive interfaces increasingly depend on decoding hand gestures from surface electromyography (sEMG). Because the signal is shaped by person-specific factors, the recurrent neural networks (RNNs) commonly used to model these temporal streams must adapt to new wearers and new gesture classes after deployment. Continual learning (CL) enables this adaptation without retraining from scratch, yet applying it to recurrent sEMG models is not straightforward. Standard recurrent training practices can destabilize learning, and which CL strategy holds up under realistic cross-user drift is unestablished. We evaluate three representative CL strategies under domain- and class-incremental drift on NinaPro DB1, and find experience replay to outperform elastic weight consolidation (EWC) and naive fine-tuning. It reaches 83.3% final accuracy for class-incremental drift while limiting forgetting to 1.8-5.8% accuracy drop, as against 16.2-28.5% for EWC and naive training. We further show these models must be trained on shuffled, independently-windowed segments. Streaming windows in temporal order collapses the network into predicting a single dominant gesture, while a shuffled, stateless regime trains stably. Finally, hand-crafted time-domain features raise per-task accuracy by up to about 40% over raw signals, but only replay preserves that gain under class-incremental drift.
[LG-107] Resolving the Mobility-Equity-Emissions Trilemma: A Causal Machine Learning Framework for Dynamic Tolling Policy
链接: https://arxiv.org/abs/2610.33994
作者: Christian Mulomba Mukendi
类目: Machine Learning (cs.LG); Applications (stat.AP); Other Statistics (stat.OT)
*备注:
Abstract:We propose a causal machine learning framework to integrate dynamic tolling for mobility, equity, and climate objectives, resolving a long-standing policy trilemma. Using Double Machine Learning, we estimate zone-type-specific treatment effects, showing that commercial zones yield nearly twice the delay reduction of residential zones. Our adaptive quota mechanism expands residential coverage 34-fold while preserving 93% of total delay savings and sustaining high return on investment. The framework delivers climate co-benefits, reducing emissions by 743.5 metric tons of CO2 (valued at 141 USD) via congestion-sensitive pricing under moderate traffic conditions with sensitivity bounds from 557.6 to 929.3 tons. Crucially, we uncover a policy-invariant toll range where performance remains robust across demand scenarios, simplifying implementation. This study reveals critical, previously unidentified insights, such as a policy-invariant toll range, that can significantly simplify real-world implementation. Validated on synthetic data mirroring real-world networks, our methodological advances lay the foundation for future public deployment.
[LG-108] Constrained Gradient Equilibrium: Approachability under Locally Constrained Adversaries
链接: https://arxiv.org/abs/2610.33985
作者: Kaicheng Zhang,Lujing Zhang,Yifei Pang,Yuxiao Wen,Zhun Deng,Yanjun Han
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 78 pages, 7 figures
Abstract:Gradient equilibrium (GEQ) is a recent framework in online learning that requires the time-averaged gradients along the learner’s trajectory to vanish. We study GEQ when the learner is constrained and ask when such an equilibrium remains achievable in an adversarial environment. We first show that constrained GEQ (CGEQ) can be impossible even under simple and slowly varying constraints, motivating imposing local temporal structure on the adversary. We then develop an approachability framework for CGEQ under locally constrained adversaries, and obtain necessary and sufficient characterizations in the spirit of classical Blackwell approachability. Beyond these theoretical characterizations, we provide an algorithmic principle that converts window-wise no-regret learners into CGEQ algorithms under a local feasibility condition with margin. Experiments on synthetic and real data demonstrate that the resulting algorithms perform strongly in both achieving GEQ and respecting the constraints.
[LG-109] EmoNet: A Lightweight Attention-Based CNN for Speech Emotion Recognition
链接: https://arxiv.org/abs/2610.33977
作者: Muhammad Moiz Khalid,Abdur Rehman Tariq,Muhammad Okasha Yamin
类目: ound (cs.SD); Machine Learning (cs.LG)
*备注: 6 pages, 3 figures, 8 Tables
Abstract:Speech Emotion Recognition (SER) is a fundamental task in affective computing that enables applications such as mental health assessment, human-computer interaction, and automated customer support systems. Accurate emotion classification from speech remains challenging due to variations in speaker style, acoustic conditions, and the high computational cost of modern deep learning approaches. Many existing systems depend on large transformer-based architectures that require substantial computational resources, making them less suitable for real-time or resource-constrained environments. This research proposes EmoNet, a lightweight deep learning framework for speech emotion classification across seven emotion categories. The model is built on a four-block depthwise separable CNN encoder with squeeze-and-excitation attention and residual connections, combined with a Bidirectional LSTM for temporal feature learning and a multi-head attention pooling mechanism for improved representation. A comprehensive 134-channel acoustic feature set alongside data augmentation techniques including audio-level perturbations and SpecAugment is employed to improve robustness under diverse acoustic conditions. Evaluated on a combined dataset of TESS, RAVDESS, and SAVEE, EmoNet achieves a test accuracy of 96.75% with only 4.11 million parameters. The proposed model outperforms its reproduced baseline and remains competitive with significantly larger state-of-the-art architectures.
[LG-110] Performance of Neural Operator Baselines on The Well Dataset
链接: https://arxiv.org/abs/2610.33967
作者: Yash Pincha,Andy Dimnaku,Andrew Chiang,Adarsh Ganeshram,Valentin Duruisseaux,Anima Anandkumar
类目: Machine Learning (cs.LG)
*备注:
Abstract:The Well, a 15-terabyte dataset spanning 16 physical simulation domains, provides an extensive benchmark for developing and evaluating machine learning surrogate models for complex physical systems. The baselines released with the benchmark were presented by their authors as illustrative examples obtained at small computational cost, trained with zero to very limited hyperparameter tuning and under a highly constraining 12-hour per-run wall-clock limit. These results have subsequently been treated as fully tuned benchmarks, and in some cases effectively interpreted as unconstrained performance baselines. We perform a short non-exhaustive hyperparameter search for the Tucker-factorized Fourier Neural Operator (TFNO) under the same 12-hour per-run wall-clock limit, while using only a single input timestep rather than the history of up to four used by the published baselines. Despite this more restrictive setup, tuning alone reduces VRMSE below both published FNO and TFNO baselines on 15 of 17 tasks, sometimes by several orders of magnitude. This sensitivity to a modest tuning effort underscores that the released neural operator baselines should not be treated as architectural benchmarks.
[LG-111] Stable and Calibrated Latent Force ODE Filters
链接: https://arxiv.org/abs/2610.33939
作者: Paul Fischer,Jonathan Schmidt,Philipp Hennig
类目: Machine Learning (cs.LG)
*备注: 21 pages, 10 figures. Code available at this https URL
Abstract:Estimating unobserved forces in mechanistic systems is a central task of scientific machine learning. Approximate Gaussian filtering methods that synthesize mechanistic and empirical information have shown remarkable capacity in jointly identifying latent processes together with the system state in a single forward-backward pass. The resulting latent force ODE filter framework channels information from observational data specifically into the unknown parts of the system. This work addresses open questions about calibration and stability of the method. We show that the joint posterior over state and force is the reweighted conventional latent force posterior. This uncovers a crucial dependence of the stability of the method on the prior variance of the system state, which effectively controls the flow of information from observations to the latent force through the mechanistic system. From this insight, we construct an improved algorithm that provides stable and calibrated plug-and-play inference. We demonstrate the improved framework on several simulated experimental settings.
[LG-112] EnerGNN: Learning Optimization-Compatible Energy Functions for Exact Constrained Combinatorial Inference NEURIPS2026
链接: https://arxiv.org/abs/2610.33930
作者: Niki Triantafyllou,Maria M. Papathanasiou
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: Accepted at NeurIPS 2026
Abstract:Learning-based approaches to combinatorial optimization often predict solutions directly or use predictions to restrict search, while expressive energy-based models typically rely on approximate inference. We introduce EnerGNN, a framework that learns energy functions over complete binary configurations and minimizes them under explicit hard constraints through mixed integer programming (MIP). EnerGNN combines a graph neural network encoder with an optimization-compatible readout and a nonlinear energy head, enabling exact constrained inference over the learned energy. An optimality-aware contrastive loss trains the energy to reflect solution quality and constraint violations. Across Quadratic Knapsack, Combinatorial Auctions, and a personalized cell therapy supply chain problem, EnerGNN achieves near-optimal solutions with 100% feasibility. On the supply chain problem, it achieves a 0.05% mean primal gap and a 17.4 \times speedup over the monolithic MIP. The learned energy also generalizes to larger instances and supports zero-shot constraint injection, adapting to changing constraints without retraining while preserving feasibility.
[LG-113] Learning Population Dynamics with Variational Doob-Wasserstein Flow
链接: https://arxiv.org/abs/2610.33922
作者: Weizhu Qian
类目: Machine Learning (cs.LG)
*备注: 17 pages
Abstract:This paper introduces a novel simulation-free framework for learning unbalanced population dynamics on the Riemannian Wasserstein-Fisher-Rao (WFR) manifold, termed Variational Doob-Wasserstein Flow (VDWF), built upon the theory of the unbalanced mean-field Schrödinger bridge (MFSB). First, by independently learning a geometry model equipped with a data-dependent metric, our model captures the intrinsic geometric structure of population dynamics. Furthermore, leveraging Doob’s h-transform, the proposed method incorporates boundary information to improve flow-based trajectory inference. For unbalanced transport, a growth model is employed to learn the dynamics while constructing intermediate mass samples in the WFR space without simulation. Additionally, we analyze the convergence rate of the proposed framework using a uniform Log-Sobolev inequality (LSI). Finally, we validate the proposed approach on multiple real-world datasets, demonstrating competitive performance in learning both balanced and unbalanced population dynamics compared with state-of-the-art methods.
[LG-114] Real-Time Human Trajectory Prediction from RGB-D for Mobile Robots Using Bi-Mamba
链接: https://arxiv.org/abs/2610.33916
作者: Basheer Al-Tawil,Magnus Jung,Ayoub Al-Hamadi
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Accepted at the 14th International Conference on Control, Mechatronics and Automation (ICCMA 2026), Bari, Italy, November 3-5, 2026. 7 pages, 6 figures
Abstract:Predicting the short-term future trajectory of a person from RGB-D cameras is a key capability for mobile robots navigating in dynamic and human populated environment. The key challenge to deploying learned predictors on real robots is the train-deploy pipeline gap. Training data are often prepared with an idealized, smoothed ground truth, whereas at inference, the model is fed noisy estimates from pose detectors and depth sensors. The model is therefore evaluated on an input distribution it never encountered during training. Most existing models rely on recurrent neural networks (RNNs) or transformers, which either struggle to capture short-term temporal dependencies or suffer from computational cost. We present TrajMamba, a lightweight, bidirectional state space model for short-term human trajectory prediction using an RGB-D camera. The model uses a single shared processing path for both training and deployment, aligning human pose extraction, depth back projection, and smoothing processes. It encodes a short window of persons’s hip center velocities with stacked bidirectional Mamba blocks and decodes future velocities with a GRU head. On the ETH/UCY benchmark, TrajMamba reaches an average ADE of 0.29 m FDE of 0.51 m. With 262K parameters, it runs in real time on modest hardware and is deployed on a TIAGo robot with an onboard NVIDIA Jetson Orin.
[LG-115] Min-Marginal Power Indices for Multi-Objective Feature Selection
链接: https://arxiv.org/abs/2610.33885
作者: Wenjie Shi,Chi Zhao
类目: Machine Learning (cs.LG)
*备注: 12 pages, 3 figures
Abstract:Game-theoretic feature selection scores each feature by its average marginal contribution across coalitions. An average shows whether a feature helps overall, but not whether, in some coalition, adding it turns an acceptable model into an unacceptable one. Building on minimum marginal aggregation, we define the Min-Marginal Power Index (MinPI) on a binary utility-threshold game. For a fixed quota, MinPI takes the value -1, 0, or +1 and labels each feature harmful, neutral, or decisive. A harmful label carries its own evidence, an evaluated coalition in which adding the feature drops utility below the quota, so no score cutoff is needed; finite-budget estimators keep this property but may miss harmful features. We use MinPI to propose feature removals, which are accepted only if they pass a validation gate, and to prefilter multi-objective search. On six benchmark datasets over five seeds, MinPI removed 12.5% of analyzed features with almost no loss of held-out utility. On Madelon and Abalone, where it removed the most, its removals cost far less held-out utility than random removals of the same size. Prefiltering increased the dataset-averaged hypervolume of multi-objective search in all five seeds. MinPI is therefore a conservative feature-deletion step whose gains are largest where the data contain many removable features.
[LG-116] DIGEST: Discovery-Guided LLM Tuning for Symbolic Regression
链接: https://arxiv.org/abs/2610.33880
作者: Kaifeng Jin,Ruizhong Qiu,Zihao Li,Xiao Lin,Dongqi Fu,Yinglong Xia,Hong Li,Hanghang Tong
类目: Machine Learning (cs.LG)
*备注: 29 pages, 6 figures, 12 tables
Abstract:Symbolic regression (SR) seeks to discover mathematical expressions directly from data, without requiring task-specific scientific descriptions. Despite tremendous progress in both conventional search-based methods and recent LLM-based methods, two critical limitations largely remain. First (Limited Generation Flexibility), conventional search-based methods rely on either fixed search rules or specialized learned generators, severely limiting adaptive candidate generation across diverse expression structures. Second (Underutilized Search Discoveries), most LLM-based methods cannot reliably recover ground-truth expressions because they fall short in effectively leveraging the LLM itself to thoroughly digest past expressions or adaptively guide future search. To address these limitations, we propose DIGEST. To render greater flexibility in generation, DIGEST repurposes a pretrained decoder block from a lightweight LLM into a symbolic policy for Genetic Programming (GP) population seeding, thereby leveraging the broader modeling capacity of the LLM. To more thoroughly utilize search discoveries, high-quality expressions discovered by GP are fed back to update the LLM-derived policy online, thereby enabling the LLM to progressively internalize past search discoveries and translate them into improved subsequent population generation. Experiments show that on two ground-truth benchmarks, DIGEST outperforms the strongest competitor by up to 2.59% in recovery. Among LLM-based baselines, it improves recovery by at least 54.91% while using at most 17.4% of the strongest competitor’s mean search time. On 122 BlackBox datasets, DIGEST achieves the highest accuracy while maintaining a moderate model complexity.
[LG-117] Software Engineering LLM s: Why We Need Them and How Software Engineering Helps Build Them
链接: https://arxiv.org/abs/2610.33851
作者: Amir M. Ebrahimi,Mohammed Mehedi Hasan,Seyed Shayan Daneshvar,Jiahuei Lin,Gopi Krishnan Rajbahadur,Ahmed E. Hassan
类目: oftware Engineering (cs.SE); Machine Learning (cs.LG)
*备注:
Abstract:Software engineering (SE) expertise can shape how coding large language models (LLMs) are trained and which engineering capabilities they acquire. We examine this role through an industrial case study of capability-focused training for a 92B-parameter coding LLM serving more than 100,000 developers. We demonstrate how established SE knowledge can guide the selection of engineering capabilities, the construction of verifiable training data, and the evaluation of model improvements. To target repository-level refactoring, we developed RefactorForge, a pipeline that combines refactoring knowledge, established SE tools, and verification techniques to construct training data. Across 306 Python repositories, RefactorForge generated 5,704 executable tasks and retained 4,572 verified teacher trajectories totaling 72.51 million tokens. Incremental supervised fine-tuning on these trajectories, accounting for 0.46% of the preceding coding mixture’s tokens, improved RefactorBench performance by 18.00 percentage points and SWE-Bench ProMax performance by 7.06 points. Under OpenCode, these improvements extended to SWE issue resolution, with gains of 4.97 points on SWE-Bench Verified and 1.23 points on SWE-Bench Pro. Improvements also persisted across other coding-agent harnesses, while greenfield repository generation showed mixed results. We report lessons from constructing capability-focused training data and evaluating its effects across engineering workloads. Building on this industrial experience, we propose a research agenda for expanding SE’s role in transforming coding LLMs into SE LLMs: incorporating SE knowledge into model training, applying engineering methods to model development and maintenance, and systematically defining the engineering capabilities these models should acquire.
[LG-118] When is early stopping beneficial under label noise? A phase diagram for kernel gradient flow
链接: https://arxiv.org/abs/2610.33846
作者: Fabrizio Boncoraglio,Lenka Zdeborová
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Machine Learning (stat.ML)
*备注:
Abstract:Early stopping is commonly used to prevent fitting label noise from degrading generalization. Does this mean that, when a positive fraction of the training labels is intrinsically unpredictable, the best test error must occur before interpolation? Experiments on a CIFAR10 classification task with randomly labeled examples motivate the opposite possibility: although fixed-feature and shallow models exhibit late-time deterioration of the test error, the need for early stopping essentially disappears in larger, deeper convolutional networks even after the random labels have all been classified correctly. We determine when late-time deterioration disappears entirely in a solvable setting. For kernel gradient flow on a synthetic noisy classification task, we derive sharp high-dimensional learning curves and obtain a phase diagram controlled by the balance between the linear kernel component that learns the shared relation between inputs and labels and the higher-order components that enable fitting the label noise. In one regime, generalization is optimized at finite time and early stopping is necessary. In another, generalization improves until interpolation, but the endpoint is suboptimal across kernels. At the boundary, the interpolating endpoint attains the lowest generalization error across the considered kernel family and all training times. Thus, even with extensive irreducible label noise, interpolation can outperform early stopping. The neural-network experiments leave open how learned representations acquire an analogous favorable geometry.
[LG-119] Approximate Universal Prediction under Interaction
链接: https://arxiv.org/abs/2610.33842
作者: Richard Cornelius Suwandi
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Bayesian mixtures, and Solomonoff induction in particular, provide an idealized model of prediction learned by minimizing population log loss, and they underlie universal agents such as AIXI. An agent that plans with an approximation of such a predictor decides where that predictor is evaluated, and a planner can be drawn toward branches on which its model is optimistic. We study which approximation guarantees survive this feedback. For normalized finite-horizon causal mixtures, the transfer coefficient factors into a policy likelihood ratio and task evidence available before each prediction, so task rarity and action-branch rarity can compound. Within one fixed universal mixture, infinitely many computable environments admit excess prediction error of order \kappa\eta,2^K(\mu) , where \eta is the training KL budget and \kappa measures policy coverage. This order is sharp, and for each target a computable predictor attains it at a suitable finite horizon, so computability alone does not remove the penalty. For bounded returns, value error is controlled by the probability of entering a poorly covered branch plus a square-root term in \eta . Both terms are necessary, and finite maximum coverage ratios are not required. Motivated by the mentor override of Golden Handcuffs, we derive margins for preserving deferral decisions and conditional KL budgets with an unavoidable inverse-history-probability penalty, and a counterexample separates agreement of deferral decisions from task-wise event avoidance. The guarantees apply to any predictor architecture that meets explicit population approximation and coverage conditions, which gives concrete targets for implementing idealized universal agents.
[LG-120] Sliding Networks: A Family of Residual Structures Preserving Feature Learning and Feature Diversity
链接: https://arxiv.org/abs/2610.33840
作者: Datong Zhou,Blake Bordelon
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn)
*备注:
Abstract:Deep learning models tend to perform best at large widths and depths. The growing interest in scaling laws, especially in the language modeling domain, has motivated procedures for scaling up network widths and depths in order to preserve stability and computational efficiency of training dynamics. Existing depth-scaling approaches force a choice between retaining locally maximal \emphfeature learning and preserving a form of \emphfeature diversity in the residual stream arising from random differences between channels. We examine a \textitjoint scaling recipe that preserves both. With residual width N_\flat , hidden width N_\sharp , and depth L , the key quantity is, by default, the limiting ratio N_\flat/(N_\sharp L)\to\alpha , while correlated initialization (Half Copy) can partially relax the resulting shape constraints. Theory and residual-MLP experiments support convergence at fixed positive \alpha to an SDE limit with locally maximal feature learning. Hidden computations can then be redistributed between width and depth while preserving this limit, like turning melody into harmony and back. Language-model experiments nevertheless show performance differences under regrouping at fixed parameter count. These results distinguish approximating a limit (ODE or SDE) from scaling for better performance.
[LG-121] Hessian-Similar Optimization with Capped Participation: Lower Bounds and Virtual Masters
链接: https://arxiv.org/abs/2610.33835
作者: Rui Wang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 62 pages; JMLR preprint format
Abstract:What is the round complexity of Hessian-similar optimization when each round can contact at most s of m clients? Prior rates permit a free master or full participation and do not characterize strict caps. We count every one-vector query and reply. A hidden-client chain gives a lower bound for zero-respecting local algorithms with adaptive cohorts and random stopping: on strongly convex quadratics, the required rounds are \Omega((m/s)\sqrt\min\delta_a,\bar\delta/\sqrt m/\mu,\log(1/\epsilon)) for s\leq(m-2)/2 in the high-accuracy, high-dissimilarity regime. Here \mu is strong convexity, while \delta_a and \bar\delta bound anchor and average Hessian variation. To remove dependence on a poor anchor, virtual-master SVRS uses independent estimator and proximal cohorts and averages the proximal replies. With Catalyst, it reaches relative gap \epsilon in \tilde O(m/s+(\sqrtm/s+m^3/4/s)\sqrt\bar\delta/\mu) expected rounds on quadratics with strongly convex clients, independent of \delta_a . Combining this method with charged analyses of existing algorithms matches the lower bound’s polynomial dependence for s\leq\sqrt m and s=\Omega(m) ; the worst gap is O(m^1/8) . A held-out 12-seed synthetic study finds fewer rounds than tuned SVRP in nine of ten dip-family incomplete-cohort cells, versus two of 30 other cells. The lower bounds concern zero-respecting algorithms and do not cover unrestricted messages.
[LG-122] Reactive Optimizers
链接: https://arxiv.org/abs/2610.33826
作者: Xingjian Bai,Jia-Hao Liang,Jayden Teoh,Vincent Sitzmann
类目: Machine Learning (cs.LG)
*备注: This document is a static version of our blog post. We recommend the original interactive version at this https URL . We release the code at this https URL
Abstract:We study a reactive optimizer that uses a pretrained language-model agent to inspect training dynamics and rewrite executable optimizer code during a single training run. The agent can change update rules and schedules for individual parameter groups, without optimizer-specific meta-training or memory transferred from other runs. We evaluate this approach, OPAL, on language modeling and image generation when a recipe has not been tuned for the target setting. Across the evaluated settings, it matches or improves on selected fixed recipes. Analysis of its decisions shows changes to learning-rate schedules, optimizer structure, numerical precision, and momentum. Controlled experiments also show adaptation to repeated data corruption and reduced overfitting. We report these behaviors together with the agent’s computational cost and implementation requirements.
[LG-123] Greedy Orthogonal Polynomial Manifolds with Lipschitz Control for Dimensionality Reduction and Reduced Modeling
链接: https://arxiv.org/abs/2610.33784
作者: Tomoki Koike,Benjamin Peherstorfer,Elizabeth Qian
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注:
Abstract:Polynomial manifolds provide nonlinear low-dimensional representations through polynomial decoders. We propose to represent the polynomial terms of the decoder using features that are orthonormal on the training data. This empirical orthogonality enables a greedy algorithm, Greedy-OPM, which jointly selects encoder modes and polynomial features while minimizing the reconstruction error under a prescribed bound on the decoder’s Lipschitz constant. Numerical experiments show that this Lipschitz constraint improves generalization away from the training data while retaining high reconstruction accuracy.
[LG-124] A quantum k-nearest neighbors implementation using magnitude-aware comparisons
链接: https://arxiv.org/abs/2610.33739
作者: Christopher Rector,Jiangwen Sun
类目: Machine Learning (cs.LG)
*备注:
Abstract:Much attention has recently been devoted to the k-nearest neighbors (KNN) algorithm as a case study for quantum computing algorithms. Several algorithms have devised methods to compute a similarity metric to solve Quantum KNN (QKNN). This paper presents a practical quantum encoding technique that enables Euclidean distance calculation between vectors with different L2 norms by appending a single component to each vector, then uses it in the QKNN problem. It is arguably simpler to understand and implement than comparable methods and uses fewer qubits. Further, an implementation of this algorithm is provided and is used to benchmark it against a recent Euclidean-distance QKNN method on standard labeled datasets. In addition to demonstrating proper formulation, the results show marked improvement over similar implementations during simulation with realistic noise due to the use of fewer qubits and thus more coherent circuits.
[LG-125] Polynomial Residual Networks are Generically Identifiable
链接: https://arxiv.org/abs/2610.33738
作者: Isabel Dahlgren,Giovanni Luca Marchetti,Kathlén Kohn
类目: Machine Learning (cs.LG); Algebraic Geometry (math.AG)
*备注:
Abstract:We prove identifiability results for constant-width residual networks with polynomial activations and bias terms. For activations of degree at least five with generic coefficients, the fiber over a generic parameter is a singleton. For activations of degree at least three with no linear term after a suitable translation of the input, the generic fiber is given by translating biases and then scaling weights and translated biases by real roots of unity. These results specialize to networks without bias terms. With the same proof technique, we further derive necessary conditions for critical points of the ResNet parametrization, that are central in singular learning theory.
[LG-126] How Geometry Shapes Generative Model Evaluation on Riemannian Manifolds NEURIPS2026
链接: https://arxiv.org/abs/2610.33731
作者: Misha Oberski,Valentijn Oldenburg,Xanthos Kekkos,Soham Chaudhuri,Emma Oosterhuis,Floris de Kam,Sagi Schwartz,Alison Pouplin,Federica Valeau
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: NeurIPS 2026, Geometric Distributional Deep Learning workshop
Abstract:On Riemannian manifolds, generative-model evaluation is geometry-dependent since curvature and dimension jointly determine distances and volume growth. Our framework systematically compares geodesic, ambient, and tangent-space versions of Wasserstein and MMD across different geometries. Using controlled-noise perturbations and Riemannian Flow Matching samples, we study metric agreement, noise sensitivity, and their relation to geometry-induced volume scaling. We find that no metric is uniformly reliable: evaluation distance can affect both measured discrepancies and noise sensitivity, varying across curvature and dimension. Our analysis leads to practical recommendations for reporting evaluation distances, comparing metric types, and calibrating metrics to the geometry of the samples.
[LG-127] Learning Representative Port Subsets for Efficient Fluid Antenna Systems
链接: https://arxiv.org/abs/2610.33730
作者: Quan Zhou,Ming Zeng
类目: Information Theory (cs.IT); Machine Learning (cs.LG)
*备注:
Abstract:Fluid antenna systems (FASs) provide many candidate ports, but probing all ports for each channel realization incurs substantial online overhead. We study a complementary offline problem: compress the candidate-port space once and reuse the compressed set across future channel realizations. The method collects optimal ports from sampled training realizations, requiring only one scan of the candidate ports per realization. On a DeepMIMO 7-by-7 FAS with 49 candidate ports, 120 training realizations yield 27 representative ports on average while retaining 99.995% of the channel power gain achieved by exhaustive port selection.
[LG-128] Latent dynamics for reconstructing fine-scale PDE trajectories from coarse observations
链接: https://arxiv.org/abs/2610.33722
作者: Chenhao Si,Zishan Qian,Shing-Tung Yau,Rongling Wu,Ming Yan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Coarse PDE observations omit spatial detail and intermediate states of evolving fields. We present Structured Latent Dynamics (SLD), a framework that reconstructs high-resolution trajectories from spatially and temporally downsampled simulations. SLD interpolates the input in time, adds a learned temporal correction and uses the resulting sequence to guide a hidden state on the coarse grid. Each update of the latent state combines a learned gated response with horizontal and vertical central differences and a five-point Laplacian of the state. The three spatial terms have learned channel-wise coefficients. Two Haar wavelet synthesis branches map the hidden state and refined coarse input to the high-resolution field. We evaluate SLD on two-dimensional Gray–Scott, complex Ginzburg–Landau and viscous Burgers systems, reconstructing 100 output frames from 25 input frames with eightfold refinement in each spatial direction. SLD achieves lower mean relative L_2 error and mean absolute error than unstructured latent dynamics and ConvLSTM baselines with comparable parameter counts, both on held-out trajectories and on later windows of trajectories used during training. SLD also reduces errors in regions with large reference gradients and across most or all of the evaluated later time intervals. A representative Gray–Scott comparison shows lower spectral error over most intermediate and high wavenumbers. Removing the fixed spatial response increased reconstruction error, supporting its contribution to the SLD architecture.
[LG-129] RoboJev: An Open Vision-Language Judge for Robot Success Detection
链接: https://arxiv.org/abs/2610.33718
作者: Josef Liyanjun Chen
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 15 pages, 5 figures
Abstract:Many robot manipulation videos do not indicate whether the task was successfully completed. RoboJev-27B is an open-weight vision-language model that takes a task instruction, 10 frames of a manipulation attempt, and (if available) final frames from auxiliary cameras and predicts whether the robot accomplished the task. RoboJev-27B is a variant of JEV-27B-VL, trained with a rank-32 LoRA adapter on the language model only (keeping the vision encoder frozen) using 52,598 episode records. It also uses a source-specific KL penalty that applies only to research data, anchoring predictions closer to the original model on those sources. It uses a calibrated two-token readout that outputs success/failure probabilities in a single forward pass. On 268 held-out episodes of the same robot model (SO-101), RoboJev-27B attains 95.2% balanced accuracy and 0.993 AUROC. It also generalises to episodes collected by unseen labs/robots (AUROC = 0.890 over 459 episodes) as well as to clearly labelled community-collected episodes (accuracy = 93.3% over 135 episodes), where the base model attains an AUROC = 0.900 on the unseen-lab/robot subset with the same inputs. It takes 0.26 seconds per episode (median, serial) to infer on a single H100, which is about \ 0.19 per 1,000 success predictions at \ 2.69 per GPU-hour (not including media processing and serving costs). RoboJev-27B is released together with its inference code, along with a transparent evaluation: 20,628 predictions organized into 29 model/protocol entries are made available to facilitate browsing and to enable reproducible benchmarking of success detection. We also compare against strong API-based frontier VLMs, which take in fewer input frames (4 images from the primary camera only); note that this comparison is not strictly like-for-like because RoboJev-27B is given more frames at inference.
[LG-130] VulDec: Vulnerability Detection in Binaries via Neural Decompilation
链接: https://arxiv.org/abs/2610.33716
作者: Francesco Crescenzo Grasso,Pietro Liguori,Roberto Natella,Domenico Cotroneo
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注: Submitted to Automated Software Engineering
Abstract:Automated vulnerability detection in binary executables is challenging due to the lack of source code and the limited semantic information available at the assembly level. Despite significant progress in machine-learning-based code analysis, the field still lacks methodologies and large, high-quality datasets that provide vulnerable code jointly at assembly- and source-level with precise vulnerability labels. To address this gap, we introduce VulDec, consisting of (i) a new large-scale, carefully curated dataset aligning real-world Assembly-C function pairs with detailed CWE annotations across multiple architectures, and (ii) the first end-to-end framework that leverages neural decompilation for binary vulnerability detection. Tuned LLM-based neural decompilers trained on our dataset achieve strong source-level recovery across architectures (Edit Similarity of approximately 56%, BLEU-4 of approximately 43%), while our tuned LLM-based detectors reach a macro F1-score of 0.61 on an in-distribution realistic test set (3% vulnerability rate). Our findings provide evidence that neural decompilation offers a powerful alternative to conventional low-level binary representations, facilitating robust binary vulnerability detection.
[LG-131] Beyond Affine Reconstruction: Multi-Resolution Transformer DeepONets for the Radiative Transfer Equation
链接: https://arxiv.org/abs/2610.33668
作者: Tanguy Azéma,Marc Bocquet,Fatiha Nmira,Xiaoyi Lu,Jean-Louis Consalvi,Sibo Cheng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Radiative heat transfer plays a dominant role in fire simulations, where solving the radiative transfer equation (RTE) is a computational bottleneck. Neural operators can accelerate these computations by learning solution operators from offline simulation data. However, conventional DeepONet surrogates rely on fixed-grid inputs and affine reconstruction, limiting their flexibility across resolutions and their approximation accuracy at a fixed latent dimension. In this paper, we develop a transformer-based DeepONet for multi-resolution RTE surrogate modelling. The proposed approach combines a Perceiver encoder, which maps temperature and absorption coefficient fields at different spatial resolutions into a fixed-size latent representation, with a cross-attention decoder that enables non-affine reconstruction of the radiative intensity. This architecture allows a single model to process inputs at varying resolutions without resampling them onto a common grid. We formulate a computable empirical version of a previously derived theoretical lower bound for affine reconstruction, enabling direct comparison with model errors on the same dataset. Experiments on a two-dimensional pool-fire dataset demonstrate that the proposed approach outperforms autoencoder-based DeepONets in predicting both radiative intensity and incident radiation, with fewer trainable parameters. Power spectral density analysis shows improved preservation of small-scale structures. Both affine baselines remain above the corresponding lower bound at all evaluated ranks, whereas the proposed model achieves a training error below it at one matched latent dimension, demonstrating the potential of non-affine reconstruction to overcome this representational limitation. Experiments on a dataset combining three spatial resolutions confirm the accuracy advantage of the approach at each input resolution.
[LG-132] Geodesic-SVDD Latent Mixup for One-Shot Distributed Tabular Learning
链接: https://arxiv.org/abs/2610.33619
作者: Andrea Moleri,Christian Internò,Markus Olhofer,Barbara Hammer,Ali Raza
类目: Machine Learning (cs.LG)
*备注:
Abstract:Distributed Learning on structured data is a key and widely adopted approach in privacy-sensitive sectors, ranging from financial fraud prevention to federated healthcare diagnostics. While Gradient Boosted Decision Trees are widely used in centralized settings due to their efficiency, they are incompatible with standard gradient-based federated learning protocols because of their non-differentiability. Conversely, standard deep classifiers that are amenable to federated optimization often suffer from catastrophic forgetting under label distribution skew. In this work, to address these gaps we propose Geodesic-SVDD, a framework for Distributed Spherical Prototype Learning that recasts each client’s task as one-class density estimation on the hypersphere, so that local models can be aggregated as probabilistic manifolds in a single round. Specifically, for each client, we employ a Residual Spherical Encoder and Geodesic Latent Mixup that enforces strict hyperspherical constraints ( |\phi(x)| \equiv 1 ), preventing the manifold collapse observed in linear approximations. Furthermore, leveraging this insight, we propose a novel One-Shot “Product of Experts” protocol that enables privacy-preserving aggregation without iterative synchronization. Finally, we mechanistically analyze the trade-off between interpolation strategies, demonstrating that while linear mixup offers an enticing computational shortcut, it destabilizes optimization gradients by orders of magnitude ( 10^9\times ). Across five tabular benchmarks and four levels of label skew, Geodesic-SVDD attains the highest mean AUROC among seven federated methods, outperforms federated tree ensembles by up to 21 AUROC points on Covertype, and its One-Shot variant matches centralized Deep SVDD under IID and mild skew with a single communication round.
[LG-133] H-GLow: Accelerating Graph Generation via Hazard-Guided Discrete Flow Matching
链接: https://arxiv.org/abs/2610.33599
作者: Sirine Ayadi,Stephan Günnemann
类目: Machine Learning (cs.LG)
*备注:
Abstract:Discrete flow matching (DFM) offers a promising approach to graph generation; however, generating high-quality graphs with few steps remains challenging due to the strong dependencies between nodes and edges. Although DFM decouples training from sampling, enabling flexible sampling strategies without retraining, existing graph DFM samplers often rely on task-specific heuristics, substantially expanding the hyperparameter space. Moreover, graph DFM models still require sampling budgets comparable to those of diffusion-based approaches, leaving their potential for accelerating graph generation only partially realized. In this work, we introduce H-Glow, a hazard-guided framework for accelerating graph generation built on the decomposition of Continuous-Time Markov chain (CTMC) generators into hazards and conditional jump distributions. H-Glow leverages these quantities to plan graph edits and dynamically adjust the sampling time to graph activity. H-Glow can be adapted to pretrained graph DFM models, substantially improving the generation quality with fewer steps. We further introduce a novel graph DFM model based on the conditional generator matching framework, which directly predicts hazards and conditional jumps, and show that its training objective decomposes exactly into hazard matching and hazard-weighted jump prediction. Experiments on molecular and synthetic graph benchmarks demonstrate improved generation quality with substantially fewer sampling steps.
[LG-134] Variational Atlas Autoencoders and Self-Supervised Chart Selection
链接: https://arxiv.org/abs/2610.33577
作者: Arno Coomans,Merlin Nimier-David,Eugene d’Eon,Boris Bonev
类目: Machine Learning (cs.LG)
*备注:
Abstract:Complex, high-dimensional data lie on manifolds that a single global latent coordinate system cannot represent without distortion. We introduce Variational Atlas Autoencoders, a mixture-of-charts framework that learns an atlas of local latent representations, each responsible for one region of the data manifold. The central difficulty in learning an atlas is deciding which chart should be responsible for which data point without labels. We show that this decision need not be a heuristic. By treating the chart index as a latent variable, the evidence lower bound yields the training objective and casts chart assignment as posterior inference, valid for any routing distribution. Within this framework, we propose latent-evidence routing, which assigns a point to a chart in proportion to the evidence of its encoding under that chart’s latent prior. This has a simple interpretation: the closer a point lands to the center of a chart’s latent space, and the more confidently it is placed there, the more likely it is to belong to that chart. Chart assignment thus emerges as posterior inference, depends on the encoder alone, and is fully self-supervised, so the charts discover structure in the data on their own, separating classes and subclasses such as distinct types of shoes or faces without access to labels. We further recover an atlas from the trained charts by identifying their overlaps and transition maps, enabling continuous interpolation within and across charts that stays close to the data manifold rather than traversing unsupported regions of latent space. Compared to capacity-matched baselines, Variational Atlas Autoencoders yields better-structured latent spaces and higher sample quality on image datasets including CelebA, while sampling requires evaluating only a single chart’s decoder.
[LG-135] Joint End-to-End Training of Weighted BP Decoder Ensembles
链接: https://arxiv.org/abs/2610.33559
作者: Felix Krieg,Paul Bezner,Stephan ten Brink
类目: Information Theory (cs.IT); Machine Learning (cs.LG); Signal Processing (eess.SP)
*备注: 6 pages, 7 figures, submitted to an IEEE conference for possible publication
Abstract:Ensemble decoding improves decoding reliability by combining component decoders with complementary failure events. Existing approaches typically obtain this diversity from code-specific structure, such as code automorphisms, multiple parity-check representations, or targeted modifications of the message-passing graph. In this work, we instead learn the diversity of an ensemble of weighted belief propagation decoders by jointly optimizing trainable Tanner-graph edge weights. A central design question is how the training objective should distribute gradient among the ensemble members to promote complementary decoding behavior. To this end, we compare four ensemble objectives under direct gradient-based optimization and identify best-member training as the most effective approach. For a (64,32) progressive edge-growth (PEG) low-density parity-check (LDPC) code, the resulting ensemble outperforms the considered trained and untrained belief-propagation-based ensemble baselines. With 10 members, it provides a gain of approximately 0.66 dB over a single belief propagation decoder at a target frame error rate (FER) of 10^-4 . Similar behavior is observed for a (128,64) CCSDS LDPC code, indicating that the learned complementarity is not specific to PEG. Although the ensemble performance improves substantially, all trained members show a higher standalone FER. The gain therefore results from learned specialization and complementary failure events rather than uniformly improved component decoders.
[LG-136] Nowcasting aerosol particle size distributions from PM2.5 with physics-regularized flow matching
链接: https://arxiv.org/abs/2610.33557
作者: Zhi-Song Liu,Jenni Köykkä,Wenqing Peng,Ammar Kheder,Petri Clusius,Michael Boy
类目: Machine Learning (cs.LG)
*备注: 42 pages, 20 figures
Abstract:The mass concentration of particles smaller than 2.5um (PM2.5) is cheap to measure and widely monitored, whereas the particle number size distribution (PSD) requires costly instruments and is measured at only a few research stations. Yet many aerosol effects, such as vapor uptake, cloud droplet formation and deposition in the human respiratory tract, depend on the PSD rather than on mass. We explore whether recent generative deep learning models can nowcast the PSD: from the PM2.5 record and eight meteorological parameters over a 24h window, a model generates an ensemble of plausible hourly size distributions for the same window, reflecting that one mass can arise from many size distributions. Rather than proposing a single best model, we use two contrasting Finnish stations, the boreal forest site Hyytiälä and the urban site Kumpula, as a controlled testbed. On it, we benchmark six generative models against deterministic and climatological baselines, locally and after transfer between sites, and test physical constraints for neural network training. Flow matching performs best overall; its guided variant has the lowest continuous ranked probability score (CRPS) at both sites. A constraint based on a reduced aerosol general dynamic equation (GDE) lowers the CRPS by 5.5-11.5% at the training site and reduces errors after transfer between sites. Transfer remains harder than local estimation, identifying training on large multi-station data sets as the next step towards PSD estimates at unmonitored locations. This benchmark establishes physics-informed generative models as a promising route from routine air-quality monitoring to size-resolved aerosol information.
[LG-137] Clustering of High-Dimensional Human Gene Score Data for Target Cell Classification
链接: https://arxiv.org/abs/2610.33531
作者: S. Das,V. Madala,P. Majumder,N. S. Sinha
类目: Machine Learning (cs.LG)
*备注: 9 pages, 10 figures
Abstract:The goal is to investigate whether the total human gene-score dataset and the selected human kinase-gene dataset exhibit the same number of clusters when analysed using the same set of 769 features. Two datasets were considered: the complete gene score dataset and a subset containing selected kinase genes. Two missing-data strategies were investigated. In Case I, rows containing missing gene-score values were removed. In Case II, rows with more than 20% missing values were excluded and the other missing values were replaced by the corresponding column mean. For each case, the data were scaled and principal component analysis (PCA) was performed before clustering. Four clustering approaches were considered: K-Means, Hierarchical Clustering, Fuzzy C-Means, and Graph-based Clustering. Across both datasets and both missing-data treatments, the optimal number of clusters was found to be two. In the absence of an optimal number of clusters, the quality of clustering was assessed using the Silhouette score. Relatively high Silhouette scores were obtained in all cases.
[LG-138] Collective Collaboration of Recursive Autonomous Agents
链接: https://arxiv.org/abs/2610.33515
作者: Ziqing Zhou,Linqiang Hu,Yuning Chen,Hongda Zhang,Zhao Huang,Ding Zhou,Haining Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Smart factories must translate order-level objectives into production tasks at multiple scales and combine the capabilities of individual units into an overall delivery capability. To address repeated adaptation of organizational mechanisms across equipment, workshops, and factories, as well as coordination across levels under changing production conditions, this paper develops a framework for the collective collaboration of recursive autonomous agents (RAAs). A common task and capability interface supports the recursive decomposition of production objectives from the top down and the aggregation of conditional capabilities through peer collaboration from the bottom up. Composite units can therefore participate in higher-level organizations as agents of the same kind. We formulate organizational models for leaf and composite units, specify interface closure under composition and conditions for finite task expansion, and develop operational mechanisms linking task decomposition, resource commitments, result aggregation, and coordination within the scope affected by change. A production network integrating machining, inspection, and logistics across multiple workshops illustrates how a common organizational kernel is instantiated at the equipment-service, workshop, and factory scales. The framework connects recursive organization with explicit task responsibilities, capability conditions, and execution interfaces for manufacturing systems.
[LG-139] Overlap-Aware Sub-hypergraph Pattern Frequencies for Hypergraph-Level Classification
链接: https://arxiv.org/abs/2610.33514
作者: Nafiul Hasan Anik,Nazira Jesmin Lina,Md Tanvir Alam,Md. Fahim Arefin
类目: Machine Learning (cs.LG)
*备注: 23 pages, 4 figures, 10 tables. Code: this https URL
Abstract:Classifying whole hypergraphs requires a fixed-length description of inputs whose numbers of vertices and hyperedges vary. Frequent sub-hypergraph patterns provide such a description, but previous work records only whether each pattern is present, which loses two kinds of information: how large the overlap between two hyperedges is, and how much of a hypergraph each pattern makes up. We extend frequent sub-hypergraph mining with overlap labels, the binned intersection size of two pattern hyperedges inherited from the target hypergraph, and represent each hypergraph by the relative frequency of each mined pattern. The vocabulary is mined from training hypergraphs only and then frozen, and validation and test hypergraphs are matched against it. A support rule that depends on pattern size keeps the vocabulary small. On the six hypergraph-classification datasets of DHG-Bench, with a single configuration chosen on validation scores, the method has a higher mean Macro-F1 than the best published mean on five datasets, by 18.67 points on Twitter-Friend and 10.02 points on Steam-Player; on RHG-3 it is 0.21 points lower. Ablations show that on Steam-Player and Twitter-Friend, replacing relative frequencies with presence lowers Macro-F1 by about 18 points, and that removing overlap labels lowers it by 4 to 9 points.
[LG-140] DualPace: Decoupled Scheduling and Cooperative GPU–CPU Decoding for LLM Serving
链接: https://arxiv.org/abs/2610.33501
作者: YuXuan Lin,KeQiang Wu,Kingsum Chow
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Large language model (LLM) serving relies on batching to improve inference throughput, but the key-value (KV) cache grows with sequence length and limits GPU-resident concurrency. CPU offloading can expand KV capacity using host memory, yet existing systems remain largely GPU-gated: CPU-resident requests make progress only when the GPU scheduler exposes suitable overlap opportunities. As a result, host memory and CPU compute cannot act as an independent serving resource, leading to bursty CPU-side progress and inflated tail inter-token latency (ITL) under load. We present DualPace, an LLM serving runtime that turns the host into an independently scheduled decode engine. DualPace separates scheduling authority across two engines: the GPU engine executes prefills and GPU-resident decodes, while an AMX-enabled CPU engine independently batches and advances CPU-resident decodes. The engines are decoupled in scheduling but cooperative in execution: the CPU can execute a complete decode iteration locally or selectively invoke GPU Dense Assist for dense stages, while offline profiles guide hardware-specific batching and routing. Our evaluations show that, in decode-oriented scenarios, DualPace improves serving throughput by 6.9–43.2% over NEO, a state-of-the-art open-source CPU-offloading system, and by 11.5–90.6% over vLLM, a widely used GPU serving system, while maintaining the same level of latency. DualPace also preserves NEO’s P99 time-to-first-token (TTFT) advantage over vLLM while sustaining 7.3–50.5% higher request rates than NEO at the same level of P99 ITL. Subjects: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG) Cite as: arXiv:2610.33501 [cs.DC] (or arXiv:2610.33501v1 [cs.DC] for this version) https://doi.org/10.48550/arXiv.2610.33501 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-141] FedSEAM: Federated Continual Learning of LLM s via Subspace Expert Agreement Merging
链接: https://arxiv.org/abs/2610.33496
作者: Fatema Siddika,Abhijit Chunduru,Zilinghan Li,Ali Jannesari,Ravi Madduri
类目: Machine Learning (cs.LG)
*备注:
Abstract:Federated continual learning trains a model on a sequence of tasks across several clients without sharing their data while retaining knowledge of earlier tasks. Sparse mixture-of-experts models support this setting by adapting selected parameters on a frozen backbone. However, clients learn different expert content as tasks evolve, making it difficult to share knowledge across clients without disrupting task-specific and client-specific specialization. Matching tile coordinates alone does not establish whether their learned content is compatible. We present FedSEAM, a federated continual learning framework that organizes task knowledge into sparse experts and selectively shares their content across tasks and clients. On each client, task-aware tile selection allocates capacity to a new task, while a subspace-based compatibility test determines which existing expert tiles to reuse and where to introduce fresh capacity. On the server, Subspace Expert Agreement Merging (SEAM) compares client updates as subspaces and moves each client towards consensus only along agreed-upon directions, leaving the rest of its update unchanged. We identify two remaining sources of forgetting, routing drift and merge erasure, and address them with three protections, one of which exactly preserves tile components in a selected prior-overlap subspace at every merge. We evaluate FedSEAM on TRACE using Qwen3-4B and Llama-3.2-3B, with component-wise ablations and studies of data size to assess its balance between learning new tasks and retaining prior knowledge.
[LG-142] RBGA: Residual-Budget Guide Arbitration for Offline-to-Online Safe Reinforcement Learning in Dexterous Manipulation
链接: https://arxiv.org/abs/2610.33494
作者: Zongyou Yu,Ao Zhou,Xiongbo Shi,Jiayi Guan,Shihang Du,Xudong Zhang,Shangding Gu,Junqiao Zhao,Hanli Wang,Guang Chen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Dexterous manipulation requires robots to leverage existing datasets for rapid skill acquisition while maintaining safety during real-world deployment to prevent damage to the robot or its surroundings. Offline-to-online safe reinforcement learning (O2O safe RL) provides a promising solution for this problem by leveraging prior experience for online improvement under safety constraints, thereby mitigating the sample inefficiency and exploration risk of learning safe policies from scratch. However, existing O2O safe RL approaches do not explicitly account for the safety budget already consumed when making decisions within an episode. Excessive cost incurred early can therefore leave little room for subsequent exploration and policy improvement. To address this limitation, we propose \textbfResidual-\textbfBudget \textbfGuide \textbfArbitration (RBGA), an algorithm that explicitly tracks the remaining episode safety budget and uses it to adaptively select between offline guidance and the evolving online policy. At each step, RBGA compares the cost-to-go estimates of the guide and online actions against the remaining budget, enabling step-wise action arbitration throughout the episode. We provide theoretical analysis characterizing the safety implications of residual-budget-based selection and its effect on guide-assisted exploration. Extensive experiments demonstrate that RBGA enables effective online policy improvement while maintaining strong safety performance. Our code is available at this \hrefthis https URLlink.
[LG-143] Climate downscaling with likelihood functional flow matching
链接: https://arxiv.org/abs/2610.33482
作者: Zhi-Song Liu,Helmi Toropainen,Michael Boy
类目: Machine Learning (cs.LG)
*备注: 29 pages, 14 figures
Abstract:Decisions about heat, frost and flooding are made locally, yet reanalyses and climate models resolve tens of kilometres. Generative downscalers add kilometre-scale detail with uncertainty, but are tied to a grid, costly to sample and lack a tractable likelihood. Here we introduce CD-FFM, a functional flow-matching model that treats the high-resolution field as a continuous random field that one model can query at any location and resolution. Its velocity acts pointwise, so spatial variability comes from a noise prior fitted to the observed spectrum and every observation has an exact likelihood. Downscaling ERA5 to 1 km daily temperature over Fennoscandia, CD-FFM lowers the continuous ranked probability score by 27-31% against regression and 12-18% against a forty-times larger neural operator in each of three test years, and outperforms a CorrDiff-style diffusion model while sampling about fifty times faster. The same model transfers to precipitation and makes large kilometre-scale ensembles affordable.
[LG-144] FG-ConvNO: A Geometry-Aware Neural Operator for Propeller CFD Prediction AAAI2026
链接: https://arxiv.org/abs/2610.33456
作者: Yichen Di,Valentin Duruisseaux,Di Zhou,Xinyi Li,Daniel Leibovici,Jean Kossaifi,Anima Anandkumar
类目: Machine Learning (cs.LG)
*备注: 18 pages, 5 figures. Presented at the AAAI 2026 AI2ASE Workshop (non-archival)
Abstract:Computational Fluid Dynamics (CFD) is essential in contemporary engineering design, yet it remains expensive to compute and can take thousands of CPU/GPU hours or more for a single high-fidelity simulation. Recent developments in machine learning for scientific modeling have led to powerful surrogate frameworks that can approximate flow fields and physical quantities by using a fraction of the computational cost. However, most of these methods require largescale homogeneous datasets or simple geometries, exhibiting severe generalization gaps when applied to complex and data-scarce configurations. To address this, we propose the Factorized Geometry Convolutional Neural Operator (FG-ConvNO) for robust geometry-aware propeller pressure prediction under limited training data. We introduce geometryaware encoding, efficient interpolation-based decoding, and auxiliary global quantity supervision. In addition, we incorporate DISCO blocks that use discrete continuous convolutions to make the model discretization agnostic. In our experiments, FG-ConvNO achieves the highest accuracy for both surface pressure and global thrust prediction on a procedurally generated propeller CFD dataset. Geometry-aware encoding and architectural simplification improve stability, while random vertex subsampling enhances robustness. Overall, adding geometric priors to grid-based convolutional operators reduces overfitting on complex propeller geometries and improves generalization in low-data regimes.
[LG-145] From Intelligence to Autonomy: Foundations and Construction of Autonomous Agents for Future Intelligent Systems
链接: https://arxiv.org/abs/2610.33445
作者: Hongda Zhang,Ziqing Zhou,Yuning Chen,Linqiang Hu,Yi Liu,Ding Zhou,Zhao Huang,Haining Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Artificial intelligence is moving from one-off perception and generation tasks toward long-term planning, tool use, and interaction with real environments. Greater task competence, however, does not automatically yield autonomy: an external actor may still have to set goals, sustain operating conditions, detect failures, and approve updates. We define an autonomous agent as a system that, within explicit identity and authority boundaries, maintains both task goals and viability conditions, acts continuously using models of the world and itself, and adjusts its control mechanisms and long-term capabilities through feedback. We trace the conceptual path from homeostasis and feedback control to autopoiesis and biological autonomy. We then distinguish six lines of operationalization: computational autopoiesis, behavior-based robotics, agent formalization, adaptive software, embodied adaptation, and robotic self-modeling. On this basis, we propose a one-boundary, four-loop framework in which Operational, Viability, Reflexive, and Learning loops organize task execution, self-maintenance, metacontrol, and governed learning. Five paired simulation experiments test permission constraints, road-closure response, health maintenance, capability re-estimation, and cross-task learning. Their results provide preliminary evidence for the feasibility of these mechanisms, while broader environments and physical systems remain to be tested.
[LG-146] Combinatorial Lock for Trojan-Resilient AI Engines via Hash-Assisted Spectral Filtering
链接: https://arxiv.org/abs/2610.33437
作者: Soumita Chatterjee,Sudip Ghosh,Tamal Ghosh,Bhargab B. Bhattacharya,Hafizur Rahaman
类目: Cryptography and Security (cs.CR); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注: 7 pages, 5 figures, 4 tables
Abstract:Multi-trigger backdoor attacks, in which heterogeneous trigger mechanisms (e.g., patch, blend, warping, and the latent-domain Spy-Watermark) coexist within a single poisoned model, cannot be addressed by existing shortcut-based defenses that assume a single, static attack mechanism. Current multi-trigger remedies (BU+FT, BU+ANP, and BU+ABL) rely on backdoor unlearning, which incurs high training costs and loses effectiveness at low poisoning ratios. This paper proposes Combinatorial Lock, a training-free, reference-calibrated defense that reduces multi-trigger identification to a combinatorial problem over binary activation signatures. This formulation is the key novelty: it yields a multi-trigger defense that is training-free, trigger-oblivious, and directly implementable in hardware, since its decision rules can be evaluated with integer and logic operations. The framework combines anomaly-gated spectral scrubbing, which removes the malicious subspace only from inputs identified as anomalous, with a hash-compressed combinatorial signature verified by an ensemble of detection experts, including a compound-attack-aware expert for multi-trigger inputs. A joint threshold-shrinkage search identifies an operating point ( \tau at the 97.1st percentile, \gamma = 0.78 ) that reduces the attack success rate (ASR) to as low as 0.1% while preserving 97.8% clean accuracy. Evaluation on six models (ResNet18, VGG16, MobileNet, AlexNet, ViT-Small, ViT-Base) shows strong generalization: near-100% ASR drops to single digits across poisoning rates of 1-10% and 1-9 simultaneous triggers, with only gradual degradation in clean accuracy. ASIC and FPGA implementations confirm practicality, adding a consistent 19-20% latency overhead across all evaluated architectures and establishing Combinatorial Lock as a hardware-native approach to multi-trigger backdoor mitigation.
[LG-147] An Empirical Study of Reinforcement Learning Scaling Laws
链接: https://arxiv.org/abs/2610.33420
作者: Zeyu Zheng,Clare Lyle,Khimya Khetarpal,Will Dabney
类目: Machine Learning (cs.LG)
*备注:
Abstract:The discovery of scaling laws enables prediction of the performance, the optimal resource allocation strategy, and the best hyperparameter configuration of a large-scale training job based on inexpensive experiments at smaller scales. Guided by these empirical scaling laws, the deep learning community has achieved great success in training large (self-)supervised models on Internet-scale datasets. In contrast, despite its potential and wide application, reinforcement learning (RL) has yet to find a guiding principle for predictable scaling. In this work, we conduct an empirical study of RL scaling laws of an actor-critic algorithm in two procedurally generated environments, Starpilot and Craftax. We discover empirical scaling laws with strong predictive performance in both environments. The discovered scaling laws accurately predict the aforementioned three quantities for larger-scale training jobs up to two orders of magnitude in total training compute. In particular, we employ a variant of Elo scores for the single-agent RL paradigm as an alternative performance metric than the commonly used mean episode return. Our results show that the Elo score is a much more predictable metric than the mean episode return, making it a better choice for fitting scaling laws. By applying our discovered scaling laws to increase training compute, we establish a new state of the art in the challenging Craftax environment. Beneath the surface of the global scaling trend, we also observe a non-monotonic scaling pattern of the compute-optimal model size in Craftax. We connect this observation to a pathology known as Ray interference.
[LG-148] Performance Gaps Between Open-Loop Certainty-Equivalent and Optimal Control Under Action-Dependent Observations
链接: https://arxiv.org/abs/2610.33412
作者: Leo Maynard-Zhang,Sunmook Choi,Yahya Sattar,Yassir Jedra,Sarah Dean,Maryam Fazel
类目: Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注:
Abstract:In dynamical systems, feedback can improve performance by allowing decisions to depend on past observations, rather than being designed a priori. Under full observation, the current state is sufficient for decision-making; under partial observation, the controller must instead estimate the state from measurements. When actions also affect observations, control can have a dual role: actions influence both the state and the information available for future decisions. This interaction can invalidate certainty-equivalent (CE) control, a widely used strategy that applies state feedback policies as if the estimated state were the true state. We ask when this suboptimality leads to a quantitative performance gap that grows with the horizon. We study this question in a linear-quadratic system with bilinear observations by comparing the optimal policy, certainty-equivalent control, and the optimal open-loop (OL) policy. We identify an instance-dependent information-strength parameter q and establish distinct performance regimes as q varies. When q is sufficiently small, we show that no policy can improve over OL control by more than O(1) cumulative cost. For large values of q , we find that CE achieves an \Omega(T) advantage over OL control. Our main finding is that for intermediate values of q , we can construct a policy that occasionally probes for information and achieves an \Omega(T) advantage over OL control while CE only achieves an O(1) advantage, establishing an \Omega(T) gap between CE and optimal control. Our analysis relies on a regret decomposition that captures the tradeoff between the immediate cost of gathering information and its future benefit.
[LG-149] COPE: Composable Performance Emulation for Mixture-of-Experts Offloading Policies
链接: https://arxiv.org/abs/2610.33404
作者: Hanming Zhang,Erhu Feng
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF)
*备注:
Abstract:As LLMs scale, mixture-of-experts (MoE) architectures increase model capacity while keeping per-token computation manageable through sparse expert activation. This sparsity also enables memory-efficient deployment by retaining most expert weights in CPU memory and transferring or executing only the routed experts on demand. Existing MoE offloading systems typically target specific model-platform configurations, yet differences in model sparsity, CPU and GPU throughput, VRAM capacity, and host-device bandwidth can change the policy ranking. Because implementing and exhaustively evaluating all candidate policies incurs substantial engineering cost, efficiently selecting the best-performing policy for each model-platform configuration remains a key challenge for practical MoE offloading. We present COPE (Composable Offloading Policy Emulator), which selects the best-performing MoE offloading policy for a target model-platform configuration before runtime integration. COPE comprises two components. Its Policy Matrix factors the design space into two independent axes, expert residency and MoE execution strategy, enabling mechanism reuse across candidate policies and exposing four new candidates. Its Virtual Heterogeneous Machine profiles combined-operator schedules on the target platform and combines these measurements with offline routing traces to estimate candidate performance. Compared with the evaluated baselines, COPE improves fastest-policy selection accuracy by 24%-100% and reduces performance-prediction error by 1.7\times to 6.8\times . By identifying the preferred policy before integration, COPE further reduces policy-specific implementation effort by up to 74%. Subjects: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF) Cite as: arXiv:2610.33404 [cs.LG] (or arXiv:2610.33404v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.33404 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-150] Faithful Federated Tabular Data Synthesis
链接: https://arxiv.org/abs/2610.33403
作者: Zilong Cao,Hai Zhang,Xuan Bi
类目: Machine Learning (cs.LG); Statistics Theory (math.ST); Methodology (stat.ME); Machine Learning (stat.ML)
*备注:
Abstract:Synthetic data is an important data source in the AI era. This paper studies the generation of mixed-type tabular synthetic data in federated settings, where two key challenges arise: ensuring distributional fidelity and effectively handling mixed-type data. We propose Federated Integrative Two-Step Synthesis (FITS), a framework for mixed-type tabular data synthesis. At its core is a two-step federated algorithm for estimating conditional cumulative distribution functions (CDFs) using local linear regression. Building on this algorithm, we develop a federated synthesis procedure based on the probability chain rule to ensure distributional fidelity. To handle mixed-type data, FITS introduces a continualization method that lifts discrete variables into a continuous space and recovers their discrete distributions without loss at both the density-function and CDF levels. This transformation provides a unified continuous representation of mixed-type variables, facilitating the application of local linear regression and synthetic data generation. Our theoretical analysis characterizes the conditional density and cumulative distribution functions of the continualized mixed-type variables and establishes the uniform convergence of the resulting federated conditional CDF estimators. Experiments demonstrate that FITS-generated synthetic data improves downstream task performance for all clients, with the benefits depending on how the synthetic data is used.
[LG-151] he Cost of an Unknown Horizon for the Last Iterate of Clipped SGD under Heavy-Tailed Noise
链接: https://arxiv.org/abs/2610.33381
作者: Yuxing Peng,Zhiqing Tang,Weijia Jia
类目: Machine Learning (cs.LG)
*备注: 48 pages, including appendices and references. Lean 4 proofs of Theorems 5 and 11(i) and of the known-horizon bound of Theorem 1, and the script of the numerical illustration, are included as ancillary files; the Lean code is also at this https URL
Abstract:Training is often stopped at a data-dependent time, as in early stopping, and the last iterate is kept. Guarantees valid at such times must hold at all times simultaneously. We determine the optimal time-uniform rate for the last iterate of clipped SGD on smooth convex objectives under heavy-tailed noise with a bounded central p th moment, 1p\le2 . A schedule without horizon or confidence inputs attains \tilde O(t^-\beta_p) at all times, with \beta_p=2(p-1)/(3p-1) , also almost surely, and no rule that steps along the stochastic gradient at its current iterate attains \tilde O(t^-r) for any r\beta_p , even with adaptive step sizes and thresholds. This rate lies between the optimal rate O(T^-\alpha_p) for a known horizon T , \alpha_p=(p-1)/p , and the previous time-uniform rate \tilde O(t^-2(p-1)/(3p)) (Chezhegov et al., ICLR 2026); under finite variance, the exponents are 1/2 , 2/5 , and 1/3 . Their conjecture that a better schedule reaches the optimal rate thus holds with a known horizon, as shown by Liu and by Jiang et al., but fails without one. Querying gradients at the running average of the iterates attains \tilde O(t^-\alpha_p) at all times, whereas heavy-ball momentum with geometric forgetting and polynomially bounded steps cannot improve on \beta_p .
[LG-152] One Call Ahead: Packed Self-Speculative Decoding with Diffusion Drafting
链接: https://arxiv.org/abs/2610.33379
作者: Yuxiao Chen,Changxiao Cai
类目: Machine Learning (cs.LG)
*备注:
Abstract:Autoregressive (AR) language models generate tokens sequentially, so generating long outputs is costly. Speculative decoding mitigates this by using a lightweight drafter to propose multiple tokens for the target AR model to verify in parallel, while a diffusion-based drafter can produce an entire draft block in one forward pass. In models with AR and diffusion modes, one network both drafts and verifies, but drafting still waits for verification, so the standard schedule needs two sequential model calls per iteration. Existing one-call schedules avoid the wait but incur input size quadratic in the speculative width. Motivated by this efficiency bottleneck, we introduce Packed-SS, a self-speculative decoding schedule that verifies and drafts in a single model call with input size linear in the speculative width, while preserving the target model’s distribution. The key idea is to let the diffusion drafter run one call ahead: while verifying the current draft, the model simultaneously constructs the next draft from an older verified prefix and keeps its unresolved suffix for the next iteration. Under a mild monotonicity assumption on the drafter, we further show that Packed-SS requires no more model calls than the standard two-call decoding in the greedy setting. Experiments on Nemotron 3B/8B and SetDLM 1.7B show higher throughput and lower mean wall time than both baselines under greedy decoding.
[LG-153] Multiple Importance Sampling for Guided Speculative Inference
链接: https://arxiv.org/abs/2610.33368
作者: Joseph Rowan,Ashish J. Khisti
类目: Machine Learning (cs.LG); Information Theory (cs.IT)
*备注: 9 pages, 1 figure. Presented at the IEEE ISIT 2026 Workshop on Information Theory for Large Language Models (IT4LLM)
Abstract:We study an extension of the Guided Speculative Inference (GSI) algorithm for test-time-scaling of large language models (LLMs) to a novel setting where an ensemble of smaller LLMs proposes draft reasoning steps which are later verified by a larger base model and a process reward model. Using the established framework of multiple importance sampling (MIS), we introduce weighting schemes to aggregate samples from a diverse collection of proposal distributions. We provide new bounds on the KL divergence from the optimal reward-tilted target distribution and the expected reward degradation, then demonstrate that the well-known balance heuristic is optimal. Compared to GSI with a single well-tuned drafter, combining several diverse LLMs while retaining the same total number of samples maintains competitive pass-at- N downstream performance while being mathematically robust to situations where the support of a single draft model does not cover that of the target, where GSI may fail.
[LG-154] LSTM-Style Looped Transformers with Shared KV Caches
链接: https://arxiv.org/abs/2610.33360
作者: Andreas Kirsch
类目: Machine Learning (cs.LG)
*备注: 12 pages + 14 pages appendix
Abstract:Recurrent transformers reuse weights to perform deep computation with fewer parameters. We study an LSTM-style transformer that also shares one key–value cache across its recurrent core. At width 1024, it uses 40.4% fewer parameters than an untied control at nearly matched forward FLOPs, and reduces persistent cache sets from fifteen to four. Parallel training supplies cache entries from a standard-attention pass, whereas continuation reads entries produced recursively using shared attention. We address this mismatch by aligning the two training passes’ cache entries with a consistency loss during the final 10% of a fixed training-token budget. Across three model widths on FineWeb, this removes 81–89% of the continuation-loss penalty relative to full-cache evaluation. The improvement replicates on disjoint documents for two training seeds. Across four widths and 1.05–4.19B training tokens, the loss cost of weight tying is smallest at lower token budgets, suggesting an opportunity to use recurrent computation when data and model storage are limited.
[LG-155] he Clustering Occupation Kernel (CLOCK) Method for Time Series Arising from Dynamical Systems
链接: https://arxiv.org/abs/2610.33354
作者: Michael L. Wells,Kamel Lahouel,Victor Rielly,Bruno Jedynak
类目: Machine Learning (cs.LG)
*备注: 19 pages, 3 figures
Abstract:We introduce the Clustering Occupation Kernel (CLOCK) method, a novel representation-based framework for clustering time series arising from dynamical systems. CLOCK embeds observed sequences into a reproducing kernel Hilbert space (RKHS) via a time-averaged occupation kernel, enabling trajectory comparison through their RKHS norm. For systems governed by differential equations, this formulation rigorously approximates the kernel mean embedding of the invariant Sinai-Ruelle-Bowen (SRB) measure, meaning our dissimilarity metric approximates the maximum mean discrepancy (MMD) between attractors. To accommodate general time series, we introduce time lags via Takens’ Embedding Theorem to reconstruct the phase space. Crucially, CLOCK natively handles trajectories of unequal lengths and irregularly-spaced observations. We validate the proposed metric using spectral clustering, demonstrating superior performance against benchmark methods on both synthetic datasets and a real-world intracranial large-scale electroencephalogram (EEG) dataset. Finally, ablation studies on the Lorenz-96 system are consistent with the theoretical convergence rates of the occupation kernels and highlight the method’s strong robustness to zero-mean Gaussian noise.
[LG-156] Information-Theoretic Scaling Laws for Generative Diffusion Training
链接: https://arxiv.org/abs/2610.33337
作者: Druv Pai,Dejan Stancevic,Yi Ma,Luca Ambrogioni
类目: Machine Learning (cs.LG)
*备注: 42 pages, 15 figures
Abstract:Diffusion models trained on increasingly large datasets sampled from the same data distribution will converge to the same learned distribution. However, in practice, a model trained on finite data depends on the particular training dataset; this dependence is reflected in properties of the model’s outputs, such as memorization and sample quality. To understand this dependence through an information-theoretic lens, we study \textitmodel information, a computationally tractable measure of how much samples from the trained model depend on the particular training dataset. We explore its scaling behavior as model size and training dataset size vary. Specifically, we uncover two qualitatively distinct scaling regimes: (1) a nonlinear model information regime marking the transition from pure memorization to imperfect generalization, and (2) a generalization regime characterized by linear scaling of the model information and progressively improved generalization and sample quality. We then posit an information-theoretically principled scaling law to quantify these two regimes, and show that this law yields induced scaling laws for memorization and sample quality. We finally demonstrate that these induced scaling laws extrapolate well outside the fit’s domain. Our results demonstrate information-theoretic principles for memorization and sample quality scaling laws in diffusion models. Our code is publicly available on \hrefthis https URLGitHub.
[LG-157] Gaussian Approximation for Composition and Subsampling of Differentially Private Algorithms
链接: https://arxiv.org/abs/2610.33333
作者: Vitaly Feldman,Moshe Shenfeld
类目: Machine Learning (cs.LG)
*备注:
Abstract:The privacy analysis of differentially private stochastic gradient descent (DP-SGD) relies on bounds for the composition of subsampled mechanisms. The privacy of such a composition is bounded by that of a pair of product distributions, whose privacy loss is a sum of independent terms and has been analyzed via a variety of notions and concentration results. Yet, existing closed-form bounds either lose a logarithmic factor or require restrictive conditions on the noise scale and delicate analysis of subsampling. Further, while it is well known that sums of independent random variables are approximately Gaussian, existing CLT-based techniques give an approximation with a large error in \delta . We show that the privacy loss of this pair is approximately Gaussian as long as each step has a small privacy loss except with small probability. Specifically, its privacy loss distribution is close to a Gaussian with the same mean and variance, after truncating the rare large privacy losses. The closeness in both directions is up to a shift of \varepsilon that grows only as \ln(1/\beta) for an additive error \beta in \delta and is a lower-order term in typical applications. In particular, the composition satisfies Gaussian differential privacy up to a lower-order shift of \varepsilon and the additive error \beta . We also give a simple way to analyze the resulting variance-based bounds for subsampled mechanisms which relies on the \chi^2 divergence of a single step. For the Gaussian mechanism at sampling rate 1/t and t\ln^3(1/\delta) , this gives the predicted rate for \sigma\gtrsim\max\1,\sqrt\ln(1/\delta)/\ln(t)\ , whereas the standard Rényi-divergence analyses need \sigma\gtrsim\sqrtt\ln(1/\delta)/\ln(t) . We extend these results to random allocation (balls-and-bins sampling), and show that it explains the high accuracy of the connect-the-dots method. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.33333 [cs.LG] (or arXiv:2610.33333v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.33333 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-158] Augmentation-Guided Adversarial Training via Local Stability-Based Sample Selection
链接: https://arxiv.org/abs/2610.33331
作者: Daniel Agyei Asante,Mark Tetteh,Deborah Asamoah,Bernard Asante,Regina Esi Turkson
类目: Machine Learning (cs.LG)
*备注:
Abstract:Adversarial training remains one of the most effective defenses against adversarial attacks, but its standard PGD-based training is computationally expensive because adversarial examples are generated for every training sample. Selective adversarial training reduces this cost by perturbing only a subset of samples. However, existing selective techniques typically rely on point-wise difficulty signals such as loss, entropy, margin, or gradients. We show that these point-wise criteria do not capture local instability in a sample’s neighborhood. Motivated by this, we propose Augmentation-Guided Adversarial Training (AGAT), a selective adversarial training framework that uses augmentation-induced local variation to allocate samples with unstable neighborhoods for adversarial training. AGAT ranks samples using Augmentation-based Local Stability (ALS), a lightweight score that measures the normalized change in model logits between an input and its label-preserving augmented variant. Experiments across different architectures on MNIST, CIFAR-10, CIFAR-100, and Tiny-ImageNet-200 demonstrate that AGAT achieves comparable or higher robust accuracy while applying adversarial supervision to only 20-30% of the training samples in each minibatch. AGAT also requires less training time than all other adversarial training baselines in the computational cost evaluation.
[LG-159] MLIPDiff: Guiding Denoising Diffusion Models for Structure-Based Drug Design via Machine-Learned Interatomic Potentials
链接: https://arxiv.org/abs/2610.33315
作者: Yiyan Liao,Amin Tavakoli,Yu Zhang,Anima Anandkumar
类目: Machine Learning (cs.LG)
*备注:
Abstract:Diffusion models have become a dominant approach to structure-based drug design. However, because they are trained primarily to fit reference structures, they often place ligands in physically implausible poses that clash with the protein. We introduce MLIPDiff, an inference-time guidance method that can be applied to pretrained diffusion models for ligand generation. The method uses physics-informed guidance from UMA (Universal Models for Atoms), a machine-learned interatomic potential (MLIP). The guidance combines four weighted energy terms (protein-ligand interaction, ligand strain, steric clash, and pocket anchoring) and uses the original model checkpoint without retraining. On CrossDocked2020, MLIPDiff completely eliminates protein-ligand collisions while preserving the ligand’s local geometry, improving bond-length distributions and physical plausibility relative to the base model. Compared to models such as NucleusDiff that are specifically designed to reduce atomic clashes, MLIPDiff achieves clash-free structures without compromising local geometry. It maintains the base model’s PoseBusters pass rate and attains the best median Vina Score, Vina Min, and Vina Dock, surpassing both the base model and NucleusDiff, with a median of 59.6% of ligands per pocket binding more strongly than the reference. MLIPDiff thus addresses a limitation of prior methods by generating physically plausible structures with favorable predicted binding affinity using a single frozen model and no additional training.
[LG-160] ree Depth and Sample-Quality-Associated Features: Two Hidden Factors in Boruta Selection for RNA-Seq Data Classification
链接: https://arxiv.org/abs/2610.33307
作者: Tareque Mohmud Chowdhury,Abu Raihan Mostofa Kamal
类目: Machine Learning (cs.LG)
*备注: 19 Pages, 6 Figures, 4 Tables
Abstract:Boruta is a widely used method for selecting gene panels from RNA sequencing data, and its most used implementation recommends Random Forest trees of depth 3 to 7, an untested setting. We tested whether this setting changes which genes are selected, using pan-cancer data from The Cancer Genome Atlas (10,469 tumors, 33 cancer types) and breast cancer PAM50 subtyping (981 tumors, 5 subtypes). A leakage-free protocol fitted every label-dependent step inside training folds, measured every effect against seed-to-seed variation, and held out a locked test set per task. In pan-cancer data, shallow trees changed the selection far more than a new seed did: the top-500 overlap with unlimited-depth selections was 0.37 at depth 5, against 0.84 to 0.93 between seeds at depth 5 or unlimited depth. Shallow trees confirmed 3.5 times fewer genes, missed markers of large classes and produced less accurate 50-gene panels (balanced accuracy 0.809 vs 0.889). The effect vanished once the cap reached about 60 to 75% of the depth that unrestricted trees grow to, set mainly by the number of classes, not sample size, and leaving depth uncapped cost no accuracy in either task. The effect replicated in a second cross-validation repeat and under permutation importance. About a third of pre-filtered candidates also tracked sample quality within class; a within-class filter removed them without loss of test accuracy (balanced accuracy 0.943 vs 0.942 with 1,000 genes). We recommend uncapped trees, or a cap chosen from the realized tree depth, and a within-class quality filter before Boruta.
[LG-161] All-in-One: A Composable Attack on LLM Transformations
链接: https://arxiv.org/abs/2610.33298
作者: Xiaohua Zhan,Kazuki Egashira,Robin Staab,Mark Vero,Martin Vechev
类目: Machine Learning (cs.LG)
*备注:
Abstract:Quantization, pruning, and model merging are popular transformations applied to deployment-ready LLMs to improve efficiency and downstream performance. Recent work has shown that these transformations can introduce a security threat: an adversary can train a model that passes safety checks in its pre-transformed state, only turning malicious once it is transformed for deployment. However, prior attacks target one specific transformation (e.g., quantization) and generally rely on transformation-specific techniques, making them difficult to combine in practice. In this work, we introduce a simple, composable attack framework that targets multiple transformations simultaneously and triggers when any one of them is applied by a user. Our key insight is that the effect of each transformation can be approximated as a perturbation to the model weights, allowing us to jointly train the model with a benign loss on the original model and a malicious loss on each simulated transformed model. These perturbations are simple enough to simulate at every training iteration across diverse transformations, making them practically combinable within a single training objective, enabling a unified attack. Through extensive experiments across attack scenarios, model families, and a wide range of deployment-time model transformations, we show that our attack achieves high success rates. This demonstrates that model transformations may be utilized to trigger significantly different behavior from the original model and reinforces the associated security concerns by demonstrating that a single attack can jointly target diverse transformations.
[LG-162] Individual Fairness via Local Predictive Consistency
链接: https://arxiv.org/abs/2610.33285
作者: Zhouting Zhao,Tin Lok James Ng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Individual fairness in machine learning requires that similar individuals receive similar predictions. Most existing definitions are pairwise in nature, imposing a Lipschitz constraint that must hold either globally across all pairs of data instances or locally within specified neighborhoods. In practice, however, such pairwise constraints may be overly restrictive and may not directly identify which individuals receive locally inconsistent predictions. We study local predictive consistency as a node-level relaxation of pairwise individual fairness, where each prediction is compared with a similarity-weighted average of predictions in its local neighborhood. This yields an interpretable node-level fairness score and a corresponding regularized learning approach based on local residuals. Since graph-based individual fairness depends critically on the underlying similarity structure, we make the construction of the neighborhood graph explicit and discuss threshold selection through graph-coverage diagnostics. Our approach applies to binary classification, regression, and time-to-event prediction. We further compare this formulation with pairwise constraint-based approaches and analyze the resulting fairness–performance trade-offs.
[LG-163] RadonBench: A Benchmark of Classical and Learning-Guided Search Techniques for Problems in Combinatorial Geometry
链接: https://arxiv.org/abs/2610.33259
作者: Xiaoan Xu,Ethan X. Fang,Jesus A. De Loera,Junwei Lu,Luze Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Classical optimization and learning-guided search offer complementary approaches to open problems in discrete mathematics, but their outputs are difficult to compare when search spaces, validity checks, and stopping rules differ. We present \emphRadonBench, a benchmark for constructing maximum-cardinality Radon-free lattice configurations: finite subsets of \mathbbZ^d admitting no integral Radon partition. It remains open whether Radon-free configurations with 11 points in \mathbbZ^3 or 21 points in \mathbbZ^4 are possible. If they exist, they would improve the corresponding lower bounds on the integer Radon numbers. We compare exhaustive extension, greedy local search, PatternBoost, AlphaZero, and AlphaEvolve. All outputs use one success predicate: a convex target-sized configuration succeeds exactly when its number of integral Radon points is zero. PatternBoost and AlphaZero construct verified Radon-free configurations of sizes 10 and 20 , one point short of the respective targets; AlphaZero’s best convex target terminals have two integral Radon points in both dimensions. Exhaustive extension over fixed seeds and declared finite search boxes likewise reaches a minimum of two integral Radon points. Greedy local search and AlphaEvolve each produce target-sized near misses with one integral Radon point. No audited run reaches zero. These results establish reproducible near-miss baselines over explicitly scoped search domains.
[LG-164] Neyman Orthogonal Activation Steering
链接: https://arxiv.org/abs/2610.33217
作者: David Arbour,Erfan Jahanparast,Serina Chang,Avi Feller
类目: Machine Learning (cs.LG)
*备注:
Abstract:Activation steering is a popular approach for shifting LLM output toward a desired behavior, and increasingly for generating realistic synthetic data. The standard recipe estimates a steering direction by contrasting LLM activations on prompts with and without a target concept, such as truthfulness. In this paper, we separate the estimand (the steering direction these methods target) from the estimator used to recover it, two notions the literature often conflates. We frame nearly all existing methods as plug-in estimators of the average gradient of an outcome regression, with the binary contrastive label fixed as the outcome. Using this framing, we can immediately incorporate flexible modeling approaches and extend existing methods to allow for continuous outcomes such as log-probabilities or oracle ratings. Flexible ML models that fit the outcome well, however, might not estimate the gradient well, which motivates estimators that target the gradient directly. Drawing on the debiased machine learning literature, we develop a complementary estimator built on the Stein score, the Riesz representer of the gradient functional. Combining the two routes gives an orthogonalized estimator with classical double-robustness properties. We illustrate the framework on activation steering applied to a number of benchmark datasets.
[LG-165] Model-agnostic predictability quantification of multivariable time series
链接: https://arxiv.org/abs/2610.33209
作者: Yuxuan Shu,Vasileios Lampos
类目: Machine Learning (cs.LG)
*备注: The code is available at this https URL
Abstract:Multivariable time series forecasting plays an important role in various domains, where predictions depend on complex interactions among covariates. Forecasting models are commonly evaluated across numerous benchmark datasets. However, forecasting accuracy reflects both model capability and task predictability, making it difficult to characterise the difficulty of a task independently of the model used. Existing predictability measures often fail to capture nonlinear dependencies or are limited to univariate series. In this work, we propose Multivariable Predictability Score (MAPS), a model-agnostic metric for measuring the intrinsic predictability of multivariable time series forecasting tasks. MAPS uses covariate information to assess predictability across different lookback windows and forecasting horizons, and provides a quantitative assessment of the predictive information available for forecasting. We evaluate MAPS across synthetic and real-world forecasting datasets against 6 baseline predictability measures. The former confirms that our method closely aligns with the expected reference predictability (r=0.963). Empirical results further corroborate that MAPS yields more reliable projections in the majority of experimental configurations.
[LG-166] Lower expected Fisher rank implies better PAC generalization
链接: https://arxiv.org/abs/2610.33208
作者: Marco Miani,Albert Kjøller Jacobsen,Albert Lopez i Serrano,Søren Hauberg,Georgios Arvanitidis
类目: Machine Learning (cs.LG)
*备注:
Abstract:Deep learning architectures tend to generalize well to unseen data, but explaining this capability is still an open area of research. Under the assumption of i.i.d. data, PAC-Bayes provides a bound on generalization error, and its tightness relies on a parameter distribution that covers a large volume of low-loss solutions. The Fisher information metric endows the parameter space with a geometry for which distances reflect functional changes, and whose zero-component captures functional invariances. We define and study a Fisher-aligned parameter distribution for which we can simultaneously guarantee (1) a low-loss, thanks to the exploration constrained on the functional invariances and (2) a large log-volume of parameters, proportional to the number of zero eigenvalues. This distribution results in a high-probability generalization error bound, which scales with the expected rank of the Fisher information matrix, shedding light on the relevance of this quantity.
[LG-167] A Gaussian Process State Space Formulation for Materials Constitutive Modeling: Application to Quasi-brittle Materials
链接: https://arxiv.org/abs/2610.33175
作者: Houlin Xu,Chenyang Li,K.T. Ramesh,Michael D. Shields
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG); Numerical Analysis (math.NA); Probability (math.PR)
*备注:
Abstract:Constitutive models for history-dependent, path-dependent materials typically encode this dependence in prescribed functional forms through internal variables such as damage or plastic strain that are phenomenological, model-dependent, and not directly observable. These rigid forms lack flexibility to match observations, make calibration potentially non-unique, and complicate uncertainty quantification. This paper develops a Gaussian-process state-space model (GP-SSM) that replaces the fixed form of the constitutive model with a flexible GP regression, and the prescribed internal variables with latent state variables learned directly from observable stress-strain histories. As a GP, the model is naturally endowed with a measure of uncertainty and can be constrained to satisfy physical principles. We demonstrate this framework for quasi-brittle materials under triaxial compression, where the latent variable evolution is driven by the current-state stress invariants and the increment of equivalent strain to reflect irreversible, pressure-sensitive history. A sparse multi-output GP then maps the latent state and current strain tensor to the six-component stress response, trained by variational sparse inference. The trained model is exported into Abaqus/Explicit through a user material subroutine. Four verification cases are reported: (1) offline rollout on validation and test histories; (2) a single element under changing confinement; (3) a frozen-memory unloading-reloading extension; and (4) a multi-element structural model compared against the Karagozian and Case Concrete (KCC) model. These results show that the learned state-space update reproduces held-out RVE-scale histories in closed-loop rollout, tracks evolving confinement, supports unloading without violating the irreversibility of the learned history state, and reproduces structural-level behavior consistent with KCC.
[LG-168] Learning the Past from the Future: Retrospective Reconstruction in Temporal Interaction Networks
链接: https://arxiv.org/abs/2610.33171
作者: Amarnath R
类目: ocial and Information Networks (cs.SI); Machine Learning (cs.LG)
*备注: 12 pages, 5 figures, 3 tables
Abstract:Temporal link prediction uses past interactions to predict future links. In practice, interaction records from web and mobile applications can contain gaps caused by logging failures, API limits or lost records, even when later events remain available. To our knowledge, the extent to which later events help recover interactions within such gaps has not been measured directly for this ranking task. We study retrospective reconstruction using the tgbl-wiki and tgbl-comment datasets. We hide a two-hour window, use its interactions as ground truth, and rank the directed pairs active within it. We compare count-based heuristics and logistic regression models using past information alone or together with 1, 6 or 24 hours of later events. Target pairs and candidate lists remain fixed across conditions, so differences in ranking performance cannot be attributed to changes in candidate selection. With 24 hours of later events, the 26-feature logistic regression improves mean reciprocal rank (MRR) by 0.039 on tgbl-wiki under exhaustive ranking, 0.046 under historical-random ranking, and 0.108 on tgbl-comment under historical-random ranking. The 26-feature logistic regression benefits from adding later features to past features, whereas replacing the past pair count with the later pair count reduces MRR by 0.157 on tgbl-wiki under exhaustive ranking. A control removing later interactions from query sources to historically observed destinations reduces the MRR gain by 59% on tgbl-wiki and 10% on tgbl-comment. Because this control removes more than repeated target pairs, these reductions cannot be attributed to pair recurrence alone. Later events improve the ranking of hidden interactions on both datasets, including tgbl-comment, where only 6.4% of target pairs reappeared within 24 hours.
[LG-169] Modeling Liquid-Cooled Data Centers in EnergyPlus
链接: https://arxiv.org/abs/2610.33148
作者: Yun-Dam Ko,Dipashreya A. Sur,Cade Napier,Carlos D. Díaz-Marín,Rishee K. Jain
类目: ystems and Control (eess.SY); Computational Engineering, Finance, and Science (cs.CE); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注: 14 pages, 7 figures, 6 tables. Code: this https URL
Abstract:As AI workloads accelerate data center transition toward high-density racks with direct-to-chip (DTC) liquid cooling, existing simulation tools remain limited in evaluating whole-building and multi-year efficiency. To bridge this gap, we develop a native EnergyPlus object, ElectricEquipment:ITE:LiquidCooled, that models DTC liquid cooling for chilled-water and warm-water architectures as plant-loop components. It couples facility-water loops, rack-side secondary loops, coolant distribution units, and chip-junction temperatures within a single solver. Validation against an independent Modelica implementation across 21 operating conditions, including two full-year runs compared at one-minute resolution, yields loop-temperature RMSE at or below 0.021 K, well within the 0.1 K acceptance threshold. Our case study demonstrates that warm-water liquid cooling reduces annual PUE from 1.43 to 1.25, eliminates on-site water use, and cuts total facility energy by 37.9% relative to an air-cooled baseline. By embedding liquid-cooling logic into EnergyPlus, our work provides whole-building, long-term energy modeling for high-density AI infrastructure within an open-source framework.
[LG-170] Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity NEURIPS2026
链接: https://arxiv.org/abs/2610.33145
作者: Nico Policzer,James Ho,Joseph Soo,Xin Tang
类目: Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
*备注: NeurIPS 2026 poster
Abstract:Understanding how neural activity encodes behavioral and sensory variables is a key challenge in systems neuroscience. This requires models that go beyond predicting activity from labels and instead support source-conditioned counterfactual editing: given a recorded source response, we should be able to edit one measured behavioral or stimulus label and predict how neural activity would change while preserving other observed labels and source-specific variability not explained by the labels. Here, we introduce \textbfNeural-DISCO, a disentangled representation learning framework for counterfactual modeling of neural activity. Neural-DISCO partitions the latent space into label-specific components, each associated with an observed behavioral or stimulus label, together with a residual latent that captures variability not supplied through observed labels. Across simulated and real neural recording datasets, we show that this structure enables selective editing of measured labels to generate counterfactual neural responses while preserving the remaining content of the original activity. Further, to gain insight into how behavioral and stimulus labels are encoded at single-neuron resolution, we pair this disentangled framework with feature attribution methods that identify which neurons are most important for each label. We find that disentanglement enhances the recovery of known functional cell roles in both synthetic and real data. Together, these results support Neural-DISCO as a framework for counterfactual modeling of neural activity and provide a practical interface for probing how neural populations encode behavior and sensory information. Code is available at this https URL.
[LG-171] Physics Informed Neural Networks for Solving the Drift-Diffusion Equations
链接: https://arxiv.org/abs/2610.33133
作者: Luke Ragan,Matteo G. C. Alasio,Enrico Bellotti,Luca Dal Negro
类目: Machine Learning (cs.LG)
*备注:
Abstract:The drift-diffusion equations are the fundamental building block for the modeling, design, and optimization of modern semiconductor devices. Solutions of the drift-diffusion equations rarely give way to analytical forms, and exhibit strong nonlinearity, with the solutions ranging over several orders of magnitude. These attributes have led to research into numerical solvers, and while many commercial Technology Computer Aided Design (TCAD) packages now exist, they may require long simulation times, exhibit sensitivity to initial guesses, and rely on mesh generation. Further, these solvers are not well suited for inverse problems where the design parameters of a device are retrieved from its desired functional characteristics. To address all these challenges, this work demonstrates the usage of Physics Informed Neural Networks (PINNs) for the solution of the drift-diffusion equations with errors comparable to traditional simulation software using advanced training techniques to solve problems beyond the well-behaved equilibrium regime. For this study, we specifically choose an abrupt PN-junction in both one and two dimensions that features extreme gradients of the carrier concentrations at the Ohmic contacts and an abrupt discontinuity in the doping profile. Using the proposed PINN architecture, we are able to retrieve accurate results for all observables under all bias conditions. Additionally, we apply the PINN solver towards the inverse problem of retrieving doping profiles directly from desired electric field distributions or electrostatic potentials under non-zero applied biases. Our work establishes a powerful PINNs-based approach for semiconductor device modeling with the ability to solve inverse problems under non-equilibrium conditions for microelectronics research and discovery.
[LG-172] Inference-Time Prior Adaptation for Imbalanced Data Streams under Concept Drift ECML KDD2026
链接: https://arxiv.org/abs/2610.33126
作者: Tobias Callies(1),Eirini Ntoutsi(1),Arthur Zimek(2) ((1) University of the Bundeswehr Munich, (2) University of Southern Denmark)
类目: Machine Learning (cs.LG)
*备注: 12 pages, 7 figures, accepted to Streaming Continual Learning Workshop at ECMLPKDD 2026
Abstract:Data stream classification under concept drift and class imbalance requires models to adapt continuously while maintaining strong minority-class performance. Many existing approaches rely on active memory management, in- cluding instance removal, memory restructuring, resampling, and synthetic data generation. We present PAI-KNN, a k-nearest neighbor classifier that adapts via inference-time prior adjustment rather than memory modification. PAI-KNN main- tains a simple sliding-window buffer and dynamically adjusts its decision rule using class prior estimates derived from recent observations. Evaluated on a mul- titude of synthetic data streams covering various imbalance and drift scenarios, PAI-KNN achieves competitive performance with state-of-the-art methods, per- forming particularly well under scenarios involving dynamic class imbalance and concept drift. These findings highlight the potential of inference-time adaptation in challenging non-stationary environments and suggest that adaptation in memory- based stream classifiers need not rely exclusively on memory modification.
[LG-173] Evaluating Spiking MoveNet Deployability for Near-Future Onboard Spiking Satellite Pose Estimation
链接: https://arxiv.org/abs/2610.33061
作者: Jonathan Courtois,Christophe Marabotto,Benoit Miramond,Alain Pegatoquet
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
*备注: 9 Pages, 3 figures. This paper presents work originally developed in Jonathan Courtois’s doctoral thesis
Abstract:SNNs are attracting interest due to their potential for energy efficiency and event-driven processing. However, neuromorphic hardware is not yet mature enough to enable embedded deployment of complex spiking network architectures. In this work, we investigate a spiking version of the MoveNet architecture tailored for autonomous spacecraft operation, aiming to evaluate the deployability of this complex architecture compared to the non-spiking version for embedded, fully event-based solutions in space applications. Using the RAPTOR dataset, we compare the performance and operational cost of the proposed S-MoveNet-Lightning architecture with its standard version, MoveNet-Lightning. Our results show balanced advantages in computational resources and architectural settings for spiking U-Net-like architectures. An evaluation of the performance and hardware footprint of the proposed Spiking Neural Network (SNN) approach for pose estimation shows that, on average, 2.2 times fewer multiplication and accumulation operations are required compared to the non-spiking solution deployed on FPGA platforms with the AMD Deep Learning Processor Unit (DPU). These results provide valuable insights into the feasibility and potential of SNNs in complex aerospace applications and highlight their role in balancing efficiency and performance, paving the way for event-driven solutions in embedded space applications.
[LG-174] Organon : From Evolutionary Search to Recursive Self-Improving Agents for Scientific Discovery
链接: https://arxiv.org/abs/2610.33046
作者: Cheikh Ahmed,Zhiyu Zhang,Kun Mao,Yong Zhang,Zirui Zhou
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
*备注:
Abstract:Scientific discovery does not follow a single fixed methodology: research methods vary across domains and evolve through experimentation as investigations unfold. Although LLM-guided evolutionary search has advanced automated discovery, existing frameworks typically rely on fixed search procedures or restrict adaptation to predefined components, limiting their ability to address diverse research problems and adapt to challenges that emerge at runtime. We introduce Organon, a recursively self-improving agent for scientific discovery. During execution, a single persistent agent analyzes its accumulated experience and autonomously revises its own search strategies, population management, diagnostic tools, memory, and worker coordination. These revisions take effect at runtime, allowing the agent to continuously adapt its approach as research progresses. Importantly, the agent can also diagnose and correct flaws in its own adaptation procedures, enabling recursive improvement of the mechanisms that guide its future self-improvement. We evaluate Organon across 21 computational discovery tasks spanning algorithm design, GPU kernel optimization, mathematical construction, and systems optimization. On ALE-Bench Lite, Organon achieves the best performance among competing search frameworks on all 10 tasks, improving mean held-out performance by 20.5% over EvoX. On KernelBench, it achieves the best results on three of four workloads, with speedups of up to 1.67x over the strongest competing method. Analysis of the search trajectories further reveals how the agent develops task-specific search strategies, builds new diagnostic capabilities, and corrects or reverses its own earlier revisions. We implement Organon using DeepSeek Harness and Cordis’s spatiotemporal composability. Code is available at this https URL.
[LG-175] Bayesian Fixed-Confidence Best-Arm Identification with Abstention
链接: https://arxiv.org/abs/2610.30384
作者: Yuqi Huang,Yunlong Hou,Vincent Y. F. Tan
类目: Machine Learning (cs.LG); Information Theory (cs.IT)
*备注:
Abstract:We study the Bayesian fixed-confidence best-arm identification problem in which a learner can abstain from making a terminal recommendation. Subject to a target error probability \delta and an abstention budget \alpha , we propose Screen-and-Confirm (SC) and analyze its expected stopping time. We further derive the minimum expected stopping time for any algorithm that satisfies these constraints. For Gaussian priors and rewards, as (\delta,\alpha)\downarrow(0,0) , we establish matching upper and lower bounds up to multiplicative constants, with rate \kappa_\nu^2[1+\log((\alpha+\delta)/\delta)]/(\alpha+\delta) , where \kappa_\nu is the prior density of the top-two gap at zero, highlighting that nearly tied instances govern the sampling cost. When abstention is disallowed ( \alpha=0 ), SC yields the optimal \Theta(1/\delta) dependence for fixed model parameters, removing the additional \log(1/\delta) factor in the upper bound of Jang et al. (2024) under our Gaussian assumptions. We further characterize the exact multiplicative constant of the minimum expected stopping time for practical regimes including (i) fixed \alpha \in (0,1) and (ii) \alpha/\delta\to\infty , which are also achieved by SC.
[LG-176] On the stability of learning
链接: https://arxiv.org/abs/2610.35856
作者: John J. Vastola
类目: Neurons and Cognition (q-bio.NC); Machine Learning (cs.LG)
*备注: 20 pages, 6 figures
Abstract:Learning algorithms are usually analyzed under the assumption that informative gradients are continuously available. But this is not always true—learning signals in biological neural networks, for example, may be sparse, local, and only intermittently available at any given synapse. In this work, we show that if gradient information is noisy and only intermittently available, learning may be unstable. In particular, even if average learning dynamics converge to the correct optimum, variance in learning may be asymptotically infinite. We explore this point in a number of analytically tractable settings, including single-layer and multi-layer linear networks, and derive explicit conditions under which learning is stable. These conditions complement, but are conceptually distinct from, existing conditions related to the stability of stochastic gradient descent. Our work demonstrates that unbiased credit assignment is not enough for reliable learning; sufficiently frequent and informative feedback is also required, which yields an additional constraint on models of biologically plausible learning.
[LG-177] Model Selection for Off-Policy Evaluation via Pairwise Bellm an Validation
链接: https://arxiv.org/abs/2610.35748
作者: Lars van der Laan,Vasilis Syrgkanis,Nathan Kallus
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Econometrics (econ.EM)
*备注: 48 pages, including appendices
Abstract:Model selection and hyperparameter tuning pose a central challenge in off-policy evaluation (OPE) because errors of fitted Q-functions and occupancy-ratio estimates cannot be directly evaluated from held-out transitions. Bellman-residual and minimax validation methods often require auxiliary regressions or optimization over separate critic classes, adding computational and tuning costs. We introduce pairwise Bellman validation (PBV), a tuning-free minimax procedure for selecting among finite classes of fitted Q-functions and occupancy-ratio estimators using held-out transitions. PBV constructs critics from pairwise candidate differences, requiring only finitely many empirical moment calculations without auxiliary validation models or separate critic optimization. We establish finite-sample oracle inequalities relative to the best available candidate, even under candidate-class misspecification. Our key insight is that Q-function errors and log ratios between candidate and true occupancy ratios provide oracle critics whose Bellman and adjoint Bellman moments control estimation error. Pairwise candidate differences approximate these unknown critics, while penalties account for comparator approximation error. For Q-function selection, we first develop unweighted PBV, which evaluates pairwise Bellman moments under the offline distribution and admits oracle guarantees under a pairwise coercivity condition relating the offline distribution to the target-policy dynamics. We then develop target-aligned PBV (TA-PBV), which uses marginalized importance weighting to provide error control without pairwise coercivity. For occupancy-ratio selection, we develop adjoint PBV (A-PBV), which evaluates adjoint Bellman moments against log-ratio critics. On Hopper and Deep OPE, unweighted PBV performs competitively, while target-occupancy weighting can improve performance in settings with controlled overlap.
[LG-178] Explainable Prediction of High-Speed Solar Wind Streams driven by Coronal Holes via Topological Data Analysis
链接: https://arxiv.org/abs/2610.35549
作者: Francesco Marchetti,Sabrina Guastavino,Daniel Collin
类目: olar and Stellar Astrophysics (astro-ph.SR); Machine Learning (cs.LG)
*备注:
Abstract:High-speed solar wind streams (HSSs) originating from coronal holes (CHs) are an important driver of recurrent geomagnetic activity. Since HSSs are associated to coronal structures visible in extreme-ultraviolet observations, this provides a natural basis for forecasting several days in advance. In this work, we investigate whether the topology of coronal structures observed in extreme ultraviolet (EUV) images contains predictive information on the subsequent occurrence of CH-driven HSSs at Earth. To this end, we consider Topological Data Analysis (TDA), and in particular persistent homology, to classify EUV solar images according to the occurrence of HSSs at Earth four days later. The target is a binary classification of HSS occurrence at a four-day lead time, defined using an HSS interval catalog and solar wind speed thresholds. Several combinations of topological representations and classifiers are evaluated under a common cross-validation framework. The results indicate that the topology of EUV coronal structures alone contains predictive information on subsequent CH-driven HSS occurrence, despite the absence of explicit CH geometric descriptors in the model input. To investigate the origin of this predictive information, we develop a model-level explainability framework based on cross-validation-aggregated feature weights and spatial evidence maps. A fully traced example connects persistence diagrams and persistence-image features to the classifier output and to their spatial localization in the original EUV observation. The proposed framework provides a transparent and computationally lightweight approach for investigating the relationship between coronal topology and subsequent high-speed solar wind streams.
[LG-179] Depth Improves Training-Time Scaling for Difficult Tasks in Deep Linear Networks
链接: https://arxiv.org/abs/2610.35514
作者: Taishi Watanabe,Ryo Karakida,Jun-nosuke Teramae
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 43 pages, 9 figures
Abstract:Neural scaling laws describe the predictable power-law dependence of model performance on training time, dataset size, and model scale, with the exponents quantifying how performance changes with resources. Recent theory has shown how scaling exponents are shaped by the data distribution and the structure of the learning task, but whether network depth can improve training-time scaling has remained unsettled. Through a refined analysis of deep linear networks, we clarify that the task spectrum determines when increasing network depth improves training-time scaling. For spectrally difficult tasks, increasing depth improves the asymptotic training-time scaling exponent, whereas for easy tasks, the exponent remains independent of depth, with depth affecting only a logarithmic correction at the boundary between these regimes. Remarkably, we find that a reparameterization of time governed by the predictor norm provides a unified description of the dynamics across task difficulties. Under this reparameterization, the loss follows a common scaling law whose exponent is independent of depth. Numerical experiments on a task derived from benchmark image data and with nonlinear networks exhibit qualitatively consistent task-dependent effects of depth on training-time scaling. Our findings advance the theoretical understanding of how task structure shapes the effects of network architecture on learning and offer a basis for studying scaling behavior across broader classes of neural networks.
[LG-180] Constraint Learning by Optimizing over Trained Neural Networks: A Prag matic Tutorial
链接: https://arxiv.org/abs/2610.35505
作者: Jiaxiao Fang,Thiago Serra
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Constraint learning is an emerging paradigm of modeling optimization problems with the help of machine learning. In this paradigm, trained machine learning models are embedded into the optimization model. For example, we can learn constraints empirically from available data, or we can avoid using a nonlinear objective function by learning a simpler approximation for it. Neural networks become a compelling option if the machine learning models must capture more elaborate and nuanced relationships. However, optimization models embedding neural networks may become too large or too slow to solve. In this tutorial, we present practical considerations for an effective and efficient use of constraint learning with neural networks and validate it with a case study highlighting the benefit of smaller and sparser networks.
[LG-181] Minimax Rates for Conditional Generative Learning under Wasserstein Loss
链接: https://arxiv.org/abs/2610.35490
作者: Sungwon Park,Kyoung-Sook Moon,Hongjoong Kim
类目: atistics Theory (math.ST); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We study minimax rates for conditional generative learning under integrated Wasserstein loss, with Hölder continuous conditional laws and an unrestricted covariate distribution. We evaluate unrestricted generators through their conditional output distributions. For expected Wasserstein loss, the minimax rate agrees with the rate under uniform covariates and follows from balancing localization bias against unconditional distribution estimation error. Higher moments of the loss can instead be dominated by unobserved covariate regions, which force an additional risk contribution of order n^-1 , where n is the sample size. An adaptive localization argument gives matching upper bounds. The results cover unrestricted response laws and possibly vanishing Besov densities for W_1 , and extend to higher Wasserstein orders under uniform response density bounds and integrability conditions. A multiscale lower bound closes the logarithmic gap in unconditional W_1 minimax rates in dimension two for both unrestricted distributions and Besov density classes. We also establish minimax rates for conditional Brenier maps under density, integrability, and uniform curvature conditions.
[LG-182] XL MIMO Extreme Learning Machines: Wave-Domain Inference with Stacked Metasurfaces
链接: https://arxiv.org/abs/2610.35469
作者: Kyriakos Stylianopoulos,Mattia Fabiani,Davide Dardari,George C. Alexandropoulos
类目: ignal Processing (eess.SP); Emerging Technologies (cs.ET); Machine Learning (cs.LG)
*备注: Submitted for journal publication
Abstract:The recently envisioned goal-oriented communications paradigm requires machine learning inference to be performed directly on wirelessly transferred data, by splitting neural network computations between the transmitter, the Receiver (RX), and the wireless channel. Such approaches aim to offload large portions of the computation onto the channel itself, which is often controlled by reconfigurable MetaSurfaces (MSs). Typical MS-parametrized systems, however, enable only limited computation offloading in the wave domain, usually corresponding to a single neural network layer, due to adverse channel fading effects and the almost linear operation of the channel and conventional MS models. In this paper, we demonstrate that an eXtremely Large (XL) Multiple-Input Multiple-Output (MIMO) system equipped with Stacked Intelligent MSs (SIM) at the RX side, together with a single antenna and its accompanying radio-frequency chain, may act as a two-layer neural network operating exclusively over the air. The first (uncontrollable) network layer corresponds to the transformations offered by wireless fading, while the second (controllable) layer is implemented via SIM operations. In particular, only the first, front-end MS layer needs to exhibit nonlinear response to serve as an activation function. Leveraging the Extreme Learning Machine (ELM) framework, the proposed XL MIMO system architecture is shown to be a universal approximator, and the optimal linear-SIM-based combining is derived in closed form, allowing for fast training even and partial signal observability. Numerical evaluations across diverse datasets showcase that the proposed XL-MIMO-ELM framework achieves classification accuracy comparable to idealized digital models, thereby, proving the viability of low complexity, wave-domain neural computations.
[LG-183] A knowledge-driven nonconvex square-root regularization framework for joint group sparsity and structured smoothness
链接: https://arxiv.org/abs/2610.35428
作者: Wanling Xie,Hu Yang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
*备注:
Abstract:Structured sparse learning is a leading paradigm for high-dimensional data modeling, as it enables rigorous integration of domain-validated structural priors into the learning process. Among these priors, pre-defined grouped structure and piecewise smoothness are ubiquitous in real-world applications. However, most existing structured regularization methods are built on squared loss and convex penalties, rarely jointly model these two complementary structural priors, and suffer from two key limitations. Their optimal tuning parameters are sensitive to the unknown noise level \sigma due to the squared loss. Meanwhile, convex penalties induce inherent estimation bias and severe over-shrinkage of large magnitudes. To address these issues, we propose a knowledge-driven Nonconvex Structured Smooth Adjustment for Square-Root Regularization (NS ^3 R), which unifies the \sigma -free square-root loss with a joint nonconvex \ell_2,q penalty ( 0 q 1 ) to simultaneously encode group sparsity and structured smoothness priors. Theoretically, we establish rigorous non-asymptotic error bounds for the NS ^3 R estimator under a mild (2,q) -group restricted compatibility condition. Computationally, we develop an efficient double-screened IRLS with thresholding (DS-IRLS-th) algorithm, which adopts a nested double-screening strategy to reduce computational cost and improve numerical stability, with provable global convergence for the resulting nonconvex optimization problem. Extensive experiments on synthetic and real-world data demonstrate that NS ^3 R consistently outperforms state-of-the-art methods in finite-sample settings.
[LG-184] Beyond Wasserstein Proximity: Finding Better Transfer Sources
链接: https://arxiv.org/abs/2610.35314
作者: Gordei Verbii,Georgii Killinkarov,Vladimir Yafarov
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Source selection often favors similar data, yet marginal proximity need not identify the most useful pretraining source. We motivate a conditional alternative through domain-adaptation theory and a regression example in which broader source variation improves estimation despite increasing Wasserstein distance. We examine five-day cryptocurrency volatility using ten Binance spot assets, six forecaster families, three seeds, and shared temporal cutoffs. Selecting the three farthest rather than nearest Gaussian-Wasserstein sources improves mean persistence-relative skill by 12.52 percentage points for MLP and 20.90 for Transformer, with positive post-hoc paired date-block intervals. Forward pilot selection improves mean skill over random transfer in all six families. Compact recurrent models achieve the lowest absolute mean errors, separating selection utility from forecasting quality. Retrospective five-candidate screening shows additional gains from farthest-first geometry before pilot ranking for MLP and Transformer. We distinguish across-target associations from within-target decisions and evaluate target heterogeneity, training seeds, and temporal-block sensitivity. Findings are exploratory for a fixed historical panel; the screening analysis measures no runtime saving.
[LG-185] mporal-Difference Calibration of Black-Box Simulators
链接: https://arxiv.org/abs/2610.35311
作者: Min Zeng,Xin Wen,Xi Chen,Yichen Zhang,Wen-Xin Zhou
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We develop a derivative-free temporal-difference (TD) framework for calibrating black-box simulators using one-step transitions from a target system and noisy simulator-generated continuation values. The calibration parameter is defined through Bellman orthogonality, recovering the physical parameter under correct simulator specification and yielding a near-optimal L_2 approximation of the policy value under misspecification. We propose three TD algorithms based on random-direction finite differences, distinguished by their query costs and reuse of noisy simulator evaluations. We establish almost-sure convergence, last-iterate mean-square error bounds, and central limit theorems for Polyak-Ruppert averages across four query-noise regimes. Our analysis reveals a specification-dependent bias phenomenon: for two algorithms, the finite-difference bias of the TD estimating equation vanishes at the calibration target under correct specification, even when the gradient surrogate remains biased. This permits a fixed perturbation radius and rate exponents approaching 1/2 in the simulator rollout budget, whereas misspecification generally recovers the classical Kiefer-Wolfowitz rates. We also develop a few-replicate online bootstrap that yields Student- t calibrated confidence intervals under negligible query noise, and identify a covariance mismatch that can invalidate conventional multiplier inference along Markov trajectories. Finally, a second-stage TD recursion corrects the remaining value discrepancy without additional simulator queries.
[LG-186] Identities Not Pairs: Distribution-Free False-Accept Control for Embedding-Based Verification
链接: https://arxiv.org/abs/2610.35277
作者: Salim Khazem
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Embedding-based verification often calibrates a false-accept threshold from all impostor pairs among m identities, although these pairs share identities. We characterize a worst-case limit: when hub identities score highly against a fraction \rho of the population, a calibration-only threshold with marginal false-accept rate (FAR) at most \alpha under every identity-exchangeable law reaches the hub score on hub-free calibration sets with probability at most (m+2)\alpha/(2\rho) . Uniformly accepting near that score at levels well below \rho/m therefore requires trial information. We propose COVE, which ranks the trial pair among all N=\binomm+22 pairs formed by augmenting the calibration units with the probe and reference. Under identity exchangeability, its marginal FAR is at most \lfloor\alpha N\rfloor/N , with first possible acceptance at 1/N=O(1/m^2) . The rule also retains this guarantee with a fixed symmetric cohort-normalized score using an independent cohort. On 6 benchmarks and 8 frozen encoders, realised FAR was consistent with the exact bound. Where both valid calibration-only baselines accepted nothing, COVE retained a median 100% of the sampled oracle’s true-accept rate over reliable comparisons. In contrast, an empirical quantile over dependent pairs reached 2.87 ,\alpha FAR, and a binomial upper confidence bound over pairs missed its nominal 95% conditional-FAR target in up to 55% of calibration draws. Symmetric CSLS raised COVE’s true-accept rate by up to 4.3-fold in the tested settings. The code is available in this https URL
[LG-187] mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles
链接: https://arxiv.org/abs/2610.35204
作者: Anton E. Dorn,Emile de Bruyn,Fabrice von der Lehr,Philipp Knechtges,Stefan Kesselheim,Alexander Schug
类目: Biomolecules (q-bio.BM); Machine Learning (cs.LG); Biological Physics (physics.bio-ph)
*备注: 24 pages, 9 figures. Dataset and accompanying resources available on Hugging Face under the HelmholtzAI-FZJ organization
Abstract:Learning molecular models of RNA requires simulation data that capture conformational variability and provide consistent physical training signals. Here, we introduce mdRNAcat, a curated dataset of all-atom molecular dynamics (MD) simulations of 48 RNA systems selected to promote sequence diversity. Each system is simulated using replica-exchange with solute scaling (REST2), with 32 Hamiltonian states and production lengths of 200-300 ns per state. The dataset provides atomic coordinates, instantaneous forces, and forces reevaluated under the unscaled reference Hamiltonian, together with structural annotations and replica-exchange metadata. Coordinates and forces are available at 1 ps intervals for the reference state and 100 ps intervals across all states in a standardized HDF5 format. We characterize conformational sampling, exchange behavior, base pairing, and sugar-pucker distributions to document the properties of the simulated ensembles. mdRNAcat provides a reusable resource for developing and evaluating data-driven RNA models, including coarse-grained force fields.
[LG-188] Global mathcalOleft(k-2right) Convergence of Multilevel Gradient-Regularized Newton Methods
链接: https://arxiv.org/abs/2610.35195
作者: Nick Tsipinakis,Matthias Voigt
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注: 28 pages, 3 figures
Abstract:We develop and analyze multilevel gradient-regularized Newton methods for unconstrained convex optimization with Lipschitz-continuous Hessians. By computing Newton corrections in lower-dimensional subspaces, the proposed methods aim to reduce computational costs while preserving the global \mathcalO\left(k^-2\right) convergence rate in function values of the full-space method. To establish this rate for multilevel methods, we introduce a class of row-nested restriction operators that ensures monotonicity of the reduced-gradient norms across levels. Numerical experiments with row-nested operators demonstrate convergence behavior comparable to full-space gradient-regularized Newton, while offering substantial computational savings on large-scale instances.
[LG-189] Sequential versus fully conditional specification for tree-based synthesis of health data
链接: https://arxiv.org/abs/2610.35163
作者: R{é}my Chapelle(CB, CESP, EVDG),Mohammed Sedki(CESP)
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Disseminating synthetic data in place of original ones is a promising way to enforce privacy while complying with open science principles, especially in the healthcare domain. To keep pace with the rate at which researchers collect such data, there is a need for fast and flexible synthesis procedures. A popular class of such procedures is based on models of conditional distributions induced by decision trees. Such models can be combined into a synthetic distribution following two main approaches: sequential specification and fully conditional specification. However, there is currently no recommendation on which specification to use to optimize the risk-utility tradeoffs. In this work, we use both specifications to synthesize medical datasets by combining tree-based models of conditional distributions. We compare the quality of the resulting synthetic data using several metrics of risk and utility. Our results indicate no general superiority of either specification over the other. Sequential synthesis seems sensitive to the ordering of the variables, while the fully conditional specification involves more hyperparameters and incurs higher computational costs, which may limit its suitability for off-the-shelf synthesis procedures.
[LG-190] Realizing Low-Precision Storag e and Sparse Execution in a Folded FPGA Speech Enhancer ISCAS2027
链接: https://arxiv.org/abs/2610.35144
作者: Clément Laroche
类目: Audio and Speech Processing (eess.AS); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注: 5 pages, 3 figures, 2 tables. Submitted to IEEE ISCAS 2027
Abstract:Hardware-aware training often optimizes FPGA resources, latency, or bit operations, but folding changes what compression saves: shared arithmetic decouples precision, weight storage, and scheduled work. We derive training objectives from the measured costs of a folded streaming speech-enhancement accelerator. Mixed-precision weights reduce on-chip storage; structured channel sparsity removes entire operations from the engine schedule. We deploy a bit-exact implementation on the Kria KV260. The best sparse model eliminates 47.3% of convolution MACs, cutting frame time by 45% (1.86 to 1.03 ms) and dynamic energy by 60% (83 to 33 \mu J/frame) against a storage-priced model measured in the same session, with higher speech quality. Relative to the most efficient dense build measured separately, frame time falls by 42% and energy by 48%. Reducing weight width from 8 to 6 bits lowers dynamic power by 27% but changes frame time by less than 5%. Plain channel group-lasso removes more work than cycle-cost-weighted lasso. Thus, folded-accelerator training should price storage and removable work separately, rather than rely on a single bit-operation metric.
[LG-191] Macro Mondrian Conformal Risk Control
链接: https://arxiv.org/abs/2610.35035
作者: Julien Combes(SVH, Michelin),Jean-Francois Coeurjolly(SVH),Alexandre Dervill{é}(Michelin)
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Conformal Risk Control (CRC) extends conformal prediction beyond the coverage setting to the control of general monotone loss functions, in a distribution-free and modelagnostic fashion. This makes it well-suited to problems with high-dimensional targets, such as the semantic segmentation task addressed in this work. However, CRC guarantees only marginal control, meaning that the risk is controlled on average over the data distribution. Under imbalanced distributions, the risk on inputs that contain rare objects may therefore remain uncontrolled, even when the marginal guarantee holds. We propose a new method that controls the risk on average over user-defined groups of the domain, drawing on Mondrian conformal prediction and macro-coverage algorithms. The method relies on a taxonomy, that is, a function that assigns a group to each data point, and reweights each point’s risk by a coefficient that depends on its group probability, so that rare points contribute more than common ones. We illustrate the approach on two semantic segmentation datasets.
[LG-192] Reinforcement learning for constrained stochastic linear–quadratic control in continuous time
链接: https://arxiv.org/abs/2610.34995
作者: Weiping Wu,Qitong Pan,Xun Li,Peng Wu
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:We study continuous-time stochastic linear–quadratic control with a scalar state and nonnegative control constraints. With quadratic running costs, randomized exploration can change the optimal mean feedback. We introduce an auxiliary problem without entropy in which the conditional policy mean is nonnegative and the covariance is prescribed in proportion to the squared state. Moment reduction and state separation yield two regional Riccati equations and characterize the finite-horizon value function and optimal conditional mean. We give a sufficient control–cost compatibility condition under which this mean coincides with the optimal feedback of the original constrained problem. For time-invariant coefficients, we characterize the undiscounted infinite-horizon optimum and establish mean square asymptotic stability under the stated assumptions. For unknown dynamics, we develop model-free actor–critic schemes for finite and infinite horizons. Numerical experiments compare the learned policies with their analytical targets and examine the effect of exploration on the optimal mean.
[LG-193] he Noise-Transmission Burden of Diffusion Network: Why Image Prediction Can Outperform Noise Prediction
链接: https://arxiv.org/abs/2610.34834
作者: Gyeonghun Kang,Pengxi Liu,Bingshuo Qian,Xiang Cheng
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Clean-image prediction ( x -prediction) and noise prediction yield equivalent population-optimal denoisers, yet can perform very differently in finite diffusion networks. For noisy input Y_t with noise standard deviation \sigma_t , a noise predictor f_\varepsilon,t induces the clean-image estimate \widehat X_1=Y_t-\sigma_t f_\varepsilon,t(Y_t) . This reveals a noise-transmission burden: the predictor must preserve information needed to cancel the input corruption. Under Gaussian corruption of d -dimensional data, a linear input bottleneck of rank rd imposes a denoising-error floor \sigma_t^2(d-r) even with unrestricted downstream computation, while the corresponding optimal risk for image prediction remains bounded. EDM attenuates this bottleneck contribution at high noise; exact preconditioning cancels it at every noise level. Without an input bottleneck, assuming exact affine recovery of the noisy patch from the model representation, we characterize how the covariance spectrum of corrections beyond linear denoising determines the width required above the patch dimension. ImageNet experiments with JiT show that measured denoising losses reflect the predicted bottleneck floor and its attenuation or cancellation through preconditioning. Further widening yields larger relative denoising gains at low noise, where corrections beyond linear denoising spread across more directions, consistent with our spectral characterization of the required model width. Together, these results provide a denoising-based account of practical generation-quality gaps across prediction targets, preconditioning choices, and model widths.
[LG-194] Intrinsic Information Rates of Summaries in Simulation-Based Inference
链接: https://arxiv.org/abs/2610.34703
作者: Vansh Bansal,Pradip Sharma Poudel,James G. Scott
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
*备注:
Abstract:Simulation-based inference often encodes observations Y into a fixed-dimensional summary S , with the posterior conditioning only on S . While exact preservation often requires infinite information, any such summary imposes a finite information budget on inference which is rarely an explicit choice or measurement. In this paper, we make the budget explicit by modeling the summary as a lossy channel from Y to S , and ask two questions: how large must the budget be, and how best is it spent? First, we show for parametric models that this budget, measured by the mutual information I(Y;S) , grows unboundedly as the posterior concentrates, even when fixed-dimensional sufficient statistics exist. More generally, the required budget is set by the distribution of the random posterior, and we bound it from above by quantizing posteriors and from below by converses that weight each posterior state by its probability. Second, we show that a finite budget is best spent stochastically. Under a common-support condition on the simulator, covering the continuous observation spaces typical of simulators, a deterministic summary necessarily loses something: a stochastic summary achieves \emphstrictly lower distortion than every deterministic summary and every time-shared mixture of deterministic summaries at every positive budget too small for exact preservation. Motivated by these results, we insert a binary bottleneck into neural posterior estimators, measure each trained summary’s rate, and compare summaries at matched rate. Across four simulator benchmarks and several encoder families, stochastic bottlenecks achieve favorable rate–loss tradeoffs, and architecture rankings change at different information budgets.
[LG-195] Machine Learning Applications in the Solar Interior and Atmosphere
链接: https://arxiv.org/abs/2610.34687
作者: Vishal Upendran,Carlos J. Díaz Baso
类目: olar and Stellar Astrophysics (astro-ph.SR); Instrumentation and Methods for Astrophysics (astro-ph.IM); Machine Learning (cs.LG)
*备注: Review chapter submitted to the book Machine Learning Techniques for Astrophysics and Cosmology by Spring Singapore
Abstract:The solar interior and atmosphere constitute a strongly coupled, multiscale dynamical system defined by continuous temporal evolution. Predicting the state of this system requires modeling physical interactions that span disparate spatial and temporal scales, constrained by heterogeneous and sparse data. This chapter formalizes the Sun within the mathematical framework of dynamical systems theory, defining a global state vector ( \vecS ) governed by thermodynamical, kinetic, and magnetic properties. Because solar observations represent projected, multi-modal, and partially observable measurements ( \vecO[\vecS] ) of this hidden state, pure physics-based deduction is fundamentally constrained. We present a unifying thesis: the primary role of machine learning in heliophysics is to act as a structured family of mathematical operators—specifically Inference ( \mathcalI ), Emulation ( \mathcalE ), Propagation ( \mathcalP ), and Control ( \mathcalC )—that reconstruct, simulate, and couple this global state vector across disparate physical layers. By framing the vast, multi-disciplinary literature within this unified state-space taxonomy, we demonstrate how data-driven techniques bridge observational gaps to resolve hidden multiscale couplings and dramatically accelerate the simulation of non-linear processes, establishing a rigorous pathway to predict the global evolution of the Sun.
[LG-196] Pro-Active Inference
链接: https://arxiv.org/abs/2610.34681
作者: Skyler Wu,Yash Nair,Antoine Maechler,Emmanuel J. Candès
类目: Methodology (stat.ME); Machine Learning (cs.LG)
*备注: 36 pages, 8 figures
Abstract:Estimating a population mean is fundamental across scientific applications, but observing every outcome can be prohibitively expensive. Often, covariates and machine learning (ML) predictions are available for the full population, even when only a small subset of outcomes can be observed. How can we leverage this information to proactively select observations to label, update ML predictions as labels arrive, and construct statistically valid confidence intervals? We develop a general, without-replacement formulation of Pro-Active Inference (PAI), extending our earlier with-replacement approach and building on Active Statistical Inference. Rather than making fixed-order label-skip decisions, PAI proactively selects which observation to label next from the remaining covariate pool while accommodating online ML updates. We establish PAI’s finite-sample unbiasedness and asymptotic validity, develop an improved finite-sample variance estimator, and extend PAI to superpopulation inference without Bonferroni corrections or \alpha -splitting. We characterize PAI’s potential efficiency gains over Batch Active Inference in a quasi-oracle setting, highlighting the role of prediction-residual imbalance. Across eight diverse datasets, PAI achieves greater sampling efficiency than Sequential Active Inference in the vast majority of tested settings with near-nominal statistical coverage, using the same ML predictor and uncertainty score. However, comparisons with simpler baselines such as Prediction-Powered Inference and uniform sampling, as well as the closely related Prediction-Powered Active Testing, reveal when additional adaptivity helps or hurts statistical efficiency. Through interpretable, theory-motivated diagnostics and frozen-ML ablations, we investigate when and why PAI gains or loses efficiency, highlighting the initial residual-estimation problem and subsequent online learning dynamics.
[LG-197] How Much Graph Is Too Much? Graph Construction and Exposure in Spatio-Temporal Forecasting
链接: https://arxiv.org/abs/2610.34611
作者: Saj Maru,Qiyu Han,Krishna Mamidala
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Spatio-temporal graph neural networks combine a node’s history with information from graph neighbors. The central question is whether graph neighbors provide predictive information beyond each node’s own history. Forecast accuracy depends on both the choice of neighbors and the way graph messages enter the temporal model. We study these choices as graph construction and graph exposure on U.S. state unemployment, traffic speed, and electricity demand. Our diagnostic design uses two comparisons: varying exposure while holding the domain graph fixed, and varying construction while holding late gated integration fixed. A GRU using only each node’s own history and randomized graph controls help assess what the graph contributes. Under a common change forecasting protocol, the same domain graph can improve or worsen error relative to the GRU across integration designs. Late integration alone is insufficient, while adding a fusion gate yields lower mean error in all three tasks. With late gated integration fixed, a graph based on residual correlations has the lowest mean error among the tested graphs, although randomized graphs perform similarly on unemployment. The findings support evaluating graph construction and integration together, with conclusions tied to the comparisons actually performed.
[LG-198] HIFI: A Highly Efficient Framework for Statistical Inference with Foundation Models
链接: https://arxiv.org/abs/2610.34550
作者: Po-Wei Chen,Shan Yu,Chunlin Li
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Foundation models provide powerful representations for unstructured data, such as text and images, and parameter-efficient fine-tuning (PEFT) makes their adaptation to downstream tasks increasingly practical. For neural estimators that lack a tractable sampling distribution, resampling-based procedures offer a flexible route to statistical inference, but they often require hundreds of model fits, making them computationally prohibitive at foundation-model scale. We introduce HIFI, a two-stage fine-tuning framework for statistical inference on parameters such as conditional means. Stage 1 adapts a foundation model once on a larger sample, capturing the complex structure of the data and controlling estimation bias through a single fit. Stage 2 applies a resampling inference procedure to a smaller independent sample, concurrently fitting multiple lightweight PEFT estimators to the residual to quantify estimation variability without relearning the full task. HIFI accommodates a broad class of resampling procedures, including ensemble subsampling and the nonparametric bootstrap. Across vision and text experiments, HIFI scales to pretrained models ranging from tens of millions to more than ten billion parameters while providing accurate point estimates and well-calibrated confidence intervals across diverse architectures, and reduces training compute by up to 91.9% relative to direct resampling. An application to gene-expression inference illustrates the practical value of the framework for uncertainty-aware scientific inference.
[LG-199] Joint Independence Testing for Latent Position Random Graphs
链接: https://arxiv.org/abs/2610.34544
作者: Xiaoyi Wen,Zoltan Szabo
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Social and Information Networks (cs.SI)
*备注:
Abstract:Kernel techniques are among the most powerful methods in data science and statistics. One of the key properties of kernels contributing to their numerous successful applications is that the associated reproducing kernel Hilbert space is expressive enough to encode probability distributions without loss of information or to capture the joint independence of M\ge 2 random variables, under mild conditions. Computationally, these measures (i) rely on the so-called Gram matrix which contains the pairwise similarity of sample points, and (ii) converge at a rate O \left(n^-\frac12\right) , where n denotes the sample size. We focus on measuring the independence of latent position random graphs (LPRGs), a model widely-applied for instance in neuroscience, sociology, and epidemiology. In LPRGs a latent representation is associated to each node. The similarity of these representations (meant via a kernel) determines whether two nodes are connected, making the Gram matrices only implicitly observable and the estimation challenging. Despite the fundamental importance of quantifying the independence of M LPRGs, very little is known about the solution of this problem; existing techniques are restricted to M=2 components and only guarantee consistency without rates. In this work, (i) we define an independence measure called graph Hilbert-Schmidt independence criterion (GHSIC) for M\ge 2 LPRGs, (ii) we prove that it can be estimated at a rate \tildeO\left(n^-\frac13\right) relying on spectral decomposition of adjacency matrices, (iii) without the need to choose kernels and their hyperparameters. Numerical experiments demonstrate the effectiveness of the proposed GHSIC estimator both on synthetic examples and in the analysis of brain functional connectivity subnetworks in natural language perception.
[LG-200] Intrinsic Gradient Barriers in Pushforward Deep Generative Models
链接: https://arxiv.org/abs/2610.34484
作者: Edric Tam,Qingyang Liu,Barbara Engelhardt,Ray Bai
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注: 45 pages, 9 figures
Abstract:Broad classes of deep generative models (DGMs) operate by learning to map a simple source distribution, such as a standard Gaussian, to a complex high-dimensional target distribution. Despite their success, DGMs often empirically struggle to accurately represent distributions with heavy tails or well-separated modes. In this work, we show that these difficulties reflect intrinsic properties of the underlying transport problem rather than particular neural network architectures or training procedures. We establish quantitative lower bounds on the gradients and Jacobians of pushforward mappings that transport log-concave source distributions to such targets. For heavy-tailed target distributions, the required gradient magnitude grows as the target develops increasingly large moments. For multimodal target distributions, the low-probability regions between well-separated modes similarly force large gradients in the transport map. We discuss how such large derivatives relate to trade-offs between expressivity and numerical stability. Our results hold under substantially weaker regularity conditions than previous analyses and remain stable under perturbations of the target distribution. Finally, we show that these obstructions can be mitigated when the goal is approximation of a given density rather than generative modeling. Specifically, we propose simple classes of bijective reparametrizations that can alleviate the difficult geometric structure and reduce the gradient burden placed on the learned generative map. Numerical experiments support our theoretical analysis.
[LG-201] Extrapolating learned dynamics across bifurcations based on generalized synchronization
链接: https://arxiv.org/abs/2610.34471
作者: Lingfeng Chen,Masanobu Inubushi
类目: Chaotic Dynamics (nlin.CD); Machine Learning (cs.LG); Dynamical Systems (math.DS)
*备注:
Abstract:Predicting dynamics across bifurcations by extrapolating trained models is crucial in diverse fields of physics. Generalized synchronization provides a framework for understanding dynamical reconstruction by neural networks, especially in the context of reservoir computing. Building on this framework, we derive a formula combining pretrained readouts to reconstruct unseen chaotic dynamics. Using classical Lagrange interpolation theory, we derive rigorous bounds on readout errors and conjugacy residuals. Numerical examples demonstrate accurate prediction of chaotic dynamics across bifurcations.
[LG-202] End-to-End Neural Modeling of EM Response and Design Performance for Free-Form RFIC Passives NEURIPS’26
链接: https://arxiv.org/abs/2610.34462
作者: Yuhao Mao,Chenhao Chu,Martin Vechev,Hua Wang
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注: NeurIPS’26
Abstract:Passive matching networks are critical components in radio-frequency integrated circuits (RFICs), but their design is bottlenecked by expensive finite-difference-based electromagnetic (EM) simulation. We propose an end-to-end neural framework for free-form RFIC passives that predicts both EM response and downstream design performance under varying frequency and circuit contexts, combining a frequency-conditioned neural simulator with a circuit-conditioned neural solver. To improve design performance prediction, the simulator outputs a neural EM response consisting of both the explicit EM response and a complementary response, together with a dependence regularizer that encourages the two responses to capture orthogonal information. To evaluate the proposed framework, we construct a large-scale dataset of free-form passive layouts, synthesized from both expert-designed contours and random paths. Our 21M-parameter model achieves high accuracy across a broad frequency range, circuit contexts and passive matching settings, with R^20.98 for almost all targets, while providing orders-of-magnitude faster evaluation than traditional EM simulation. Code available at this https URL.
[LG-203] Non-Ergodic Markovian Mirror Descent with Unbounded Noise
链接: https://arxiv.org/abs/2610.34433
作者: Yunus Alp Biyikoglu,R. Srikant
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Probability (math.PR)
*备注: 32 pages
Abstract:We analyze stochastic mirror descent for minimizing a convex objective expressed as an expectation under the stationary distribution of a Markov chain, using noisy subgradient observations along a single sample path. Through Poisson’s equation, we decompose the Markovian noise and control its contribution under regularity and moment conditions that allow the stochastic subgradients and the Poisson solution to be unbounded across the state space. We establish an O(T^-1/2) bound on the expected suboptimality gap of the weighted-average iterate for an appropriately chosen step size. The analysis does not require aperiodicity or convergence of the chain’s marginal distributions to stationarity. For irreducible finite-state chains, we verify the required conditions on the Poisson solution and derive explicit bounds that quantify the effect of periodicity. For irreducible countable-state chains, we verify the required regularity and moment conditions under a geometric Lyapunov drift condition and suitable growth bounds on the stochastic subgradients.
[LG-204] Controlled Block Proximal Linearized Method for Nonconvex and Nonsmooth Optimization
链接: https://arxiv.org/abs/2610.34390
作者: Weifeng Yang
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:We consider multiblock nonconvex and nonsmooth optimization without assuming bounded iterates or coercivity of the original objective. We propose the Controlled Block Proximal Linearized (CBPL) method, which incorporates a control potential into block updates while retaining the original objective, variables, and constraints. Under explicit control design conditions and the KL property, the control yields boundedness, finite length, whole-sequence convergence of the original variables, and convergence of objective values. At the limit, one limiting subgradient of the original objective vanishes on all effective comparison directions. We apply CBPL to graph-regularized nonnegative matrix factorization (NMF), CP decomposition, and independent-core matrix tri-factorization (NMTF) with \ell_0 norm sparsity constraints. We characterize their reconstruction-preserving scale degeneracies, proving that scale criticality is equivalent to graph collapse and that full criticality forces such collapse. Explicit controls provide compact sublevel sets and a uniform positive lower bound on graph action, including with an independent Tucker core, while retaining genuine reconstruction-changing directions. Each block admits an explicit projection and a computable safe step parameter. For ordinary NMF and CP models, whole-sequence convergence to a full critical point follows. Experiments with direct safe updates demonstrate convergence, sustained graph action, and controlled factor and core scales.
[LG-205] MetaARIMA: Configuration-Level Transfer Learning for ARIMA Models
链接: https://arxiv.org/abs/2610.34171
作者: Vitor Cerqueira,Ricardo Inácio,Carlos Soares
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: Conditionally accepted at the International Journal of Forecasting
Abstract:Time Series Foundation Models have demonstrated that pretraining on diverse datasets enables effective transfer to new forecasting tasks. Classical methods like ARIMA, however, remain configured and trained independently for each series, discarding knowledge from previous experiments. Consequently, selecting and training an ARIMA configuration remains challenging and often relies on heuristic approaches based on in-sample goodnessof-fit measures. We propose MetaARIMA, a pretrained configuration policy based on metalearning that brings transfer learning to classical statistical forecasting at the configuration level rather than the model level. The policy learns from historical experiments which ARIMA configurations achieve top-quantile performance, and recommends promising candidates for new series while preserving interpretability and local estimation. At inference, it selects a diverse set of candidates via maximal marginal relevance and efficiently identifies the final configuration using successive halving. Experiments in a transfer learning setting, training on M4 and evaluating on M1, M3, and Tourism, show that MetaARIMA outperforms AutoARIMA and other classical methods while being more dataefficient. In a comparison with recent zero-shot foundation models, MetaARIMA remains competitive with Chronos2, Moirai2, and TimesFM. The experiments are publicly available at this https URL.
[LG-206] Synthetic Data Augmentation for Functional Data: Wasserstein Risk Bounds and Diffusion-Based Conditional Generation
链接: https://arxiv.org/abs/2610.34120
作者: Yuhao Qi,Wenlin Dai,Xuehu Zhu
类目: Methodology (stat.ME); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 53 pages, 14 figures, 6 tables
Abstract:Functional data are costly to collect; samples of tens to hundreds of curves are common. Synthetic data augmentation can help, but its theory is developed for finite-dimensional data and relies on total variation, which is uninformative for functions. We study augmentation with a class-conditional functional diffusion generator and state guarantees in the Wasserstein-1 distance. For stable downstream estimators, we bound the risk of training on synthetic curves, from above and below, by the risk of training on as many real curves plus a generation penalty, and derive the best number of synthetic curves and a condition under which synthetic data outperform real data. For allocating synthetic curves under class imbalance, we bound the excess risk of training on real and synthetic curves together, without sample splitting. In applications to temperature, hyperspectral and gait curves, augmentation improved classification, function-to-scalar regression and function-to-function regression, and directing synthetic curves to the minority class gave the largest gain under class imbalance.
[LG-207] Prior-Data Fitted Networks for In-Context Off-Policy Evaluation
链接: https://arxiv.org/abs/2610.34035
作者: Shuguang Yu,Ting Li,Fan Zhou
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Off-policy evaluation (OPE) is essential for assessing decision policies before deployment when online experimentation is costly or unsafe. Accurate estimation remains difficult when behavior-policy data are limited and the target policy favors states and actions rarely observed in those data. We propose InCOPE, a prior-data fitted network that learns a shared policy-value prediction rule across synthetic OPE tasks. We construct an OPE task prior over Markov decision processes and behavior-target policy pairs, generating behavior-policy datasets and exact target-policy values for pretraining. For a new tabular decision problem, the pretrained network uses the behavior-policy dataset, target policy, and known task settings as context to predict the policy’s expected return without task-specific parameter updates. Experiments across four tabular domains, including a semi-synthetic clinical setting, demonstrate InCOPE’s effectiveness for policy-value estimation.
[LG-208] CSP3.0: Template-Based Crystal Structure Prediction with Coefficient-Preserving Element Substitution and Structural Diversity
链接: https://arxiv.org/abs/2610.34029
作者: Ying Feng,Jeffrey Hu,Lai Wei,Liang Tao,Madeline Miklos,Dipannoy Das Gupta,Hossein Erfanshekooh,Jianjun Hu
类目: Materials Science (cond-mat.mtrl-sci); Machine Learning (cs.LG)
*备注: 23 pages
Abstract:Template-based crystal structure prediction (CSP) transfers a known crystal structure to a new composition by element substitution, so its accuracy depends on retrieving suitable templates, mapping elements and ranking the resulting candidates correctly. Auditing the original TCSP2.0 implementation on the T180 benchmark, we found that 469 of its 2,347 selected structures (20%) have the wrong composition, caused by element mappings that ignore stoichiometric coefficients and by sequential text substitution that cascades or corrupts element symbols. We propose TCSP3.0, which enforces coefficient-preserving mappings, substitutes species simultaneously on parsed structures and validates every candidate’s composition and cell. It further adds periodic-number retrieval, soft oxidation-state evidence, radius-compatibility scoring, structural deduplication, and ORB-v3 machine-learning-potential relaxation and energy ranking of its candidates, as used by deep generative CSP baselines. TCSP3.0 achieves Top-1 and Top-20 match rates of 72.2% and 90.6% on T180, compared with 28.3% and 45.6% for TCSP2.0, and a Top-20 match rate of 74.0% on a new 800-target benchmark (T800). On T180 it also outperforms PNcsp+ and deep generative models CrystalFlow and MaskGXT (Top-1 72.2% versus at most 63.9%). Even without relaxation, TCSP3.0 reaches 67.8% at Top-1 and 90.6% at Top-20 in a median of 28 s per target on two CPU cores, about ten times faster than CrystalFlow. Paired ten-arm ablations on T180 and T800 show that most of the improvement comes from the correctness fixes; structural deduplication significantly improves the Top-5 match rate, embedding-based retrieval is required for coverage, and the remaining scoring features have small or null effects.
[LG-209] Partial Modal Tensor Regression
链接: https://arxiv.org/abs/2610.34008
作者: Jintao Han,Li Zhang,Xuejun Ma
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 44 pages 6 figures
Abstract:With the rapid development of modern scientific research, tensor data have become increasingly prevalent across a wide range of fields. How to effectively characterize the relationship between tensor predictors and a response while preserving the intrinsic structure of tensor data has therefore become an important problem in high-dimensional statistical analysis. Existing tensor regression methods mainly focus on conditional means or conditional quantiles. When the response distribution is asymmetric, heavy-tailed, or contaminated by outlying observations, the conditional modal provides a robust and interpretable alternative characterization. In this paper, we propose a partial modal tensor regression (PMTR) method. The proposed method constructs the partial modal tensor covariance, which incorporates information on the conditional modal to guide tensor dimension reduction, and subsequently establishes a modal tensor regression model in the reduced tensor space. This strategy alleviates the dimensionality burden caused by direct vectorization while preserving the intrinsic structural information of tensor predictors. Under suitable regularity conditions, we establish the consistency and asymptotic properties of the proposed estimators and employ a data-driven dimension-selection procedure to determine the effective dimension along each tensor mode.
[LG-210] A Nonconvex Framework for Joint Sparse Sample Weighting and Parameter Estimation
链接: https://arxiv.org/abs/2610.33957
作者: Hassan Nojavan,Chul Moon,Miju Ahn
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 40 pages, 5 figures
Abstract:We propose a joint sparse sample weighting and parameter estimation (SWAPE) framework that jointly optimizes model parameters and sample weights to train a model while simultaneously assigning an individual weight to each training data point. Our framework accommodates several widely used loss functions for the model parameters, including convex, weakly convex, and difference-of-convex loss, and weakly convex sparsity functions to control the complexity of the sample weights. To solve the resulting bivariate problem, we propose an alternating minimization method in which the subproblems for updating the model parameters and weights are each approximated using majorize-minimize techniques. We provide several approximation methods for the loss functions considered in this paper, and show the overall algorithm’s convergence. Numerical experiments on various learning tasks and data anomaly settings demonstrate the effectiveness of the proposed framework in minimizing the impact of undesirable data points compared to benchmark methods.
[LG-211] Neural Eigenfluid Bases: Spectral Objectives Gauge and Topology
链接: https://arxiv.org/abs/2610.33925
作者: Baiming Zhang,Shiying Xiong,Zhecheng Wang
类目: Fluid Dynamics (physics.flu-dyn); Machine Learning (cs.LG)
*备注: 54 pages, 20 figures. Code and trained models: this https URL
Abstract:Reduced-order models simplify fluid simulation by representing complex flows with a small number of spatial modes. Building on Laplacian Eigenfluids, we introduce Neural Eigenfluid Bases (NEB), a neural representation whose basis functions can be evaluated anywhere in space. We enforce incompressibility and impermeable boundaries directly through differential lifts and potential boundary conditions. Geometry-local orthogonalization and Rayleigh minimization then learn an ordered basis. We connect learning in potential and velocity spaces, and derive a spectral-variance relation explaining why these objectives weight approximate modes differently, even though their exact positive spectra agree. Gauge fixing reduces the number of spatial derivatives needed for potential learning, while a Dirichlet energy loss learns independent circulation fields around obstacles. With 1024 modes, both objectives keep velocity errors below 1% in periodic vortex merger. For random flows around obstacles, potential-based learning captures more initial energy and gives lower final errors than numerical Laplacian bases of the same size. The reduced state is about a thousand times smaller than the full reference model. Three-dimensional tests separate spatial reconstruction errors from errors in coefficient evolution. Precomputed Galerkin operators allow millisecond-scale updates when the same basis is reused across simulations. This provides a physically constrained basis for continuous flow reconstruction and efficient reduced-order simulation.
[LG-212] he Geometry of Limits in ReLU Function Spaces
链接: https://arxiv.org/abs/2610.33779
作者: Kathryn A. Lindsey
类目: Functional Analysis (math.FA); Machine Learning (cs.LG); General Topology (math.GN); Metric Geometry (math.MG)
*备注: 37 pages
Abstract:We study the geometry and closure of spaces of functions realized by ReLU neural networks of fixed architecture. We first establish a general result about limits of continuous piecewise-linear (CPWL) functions: the minimum number of convex sets on which a function is affine is lower semicontinuous under pointwise convergence. The proof uses a connection to hypergraph coloring and a compactness principle, and yields quantitative bounds on the polyhedral complexity of continuous limits. We then characterize the functions realizable by shallow ReLU networks with vector-valued outputs, extending a known scalar-valued characterization. Using this and the geometry of limiting hyperplane arrangements, we prove that fixed-width shallow scalar ReLU realization spaces are closed in the weighted supremum norm on the whole input space. In contrast, we construct an explicit sequence of networks with architecture (2,3,2\mid1) whose limit is not realizable by that architecture. Non-closedness holds both globally in weighted supremum norm and under uniform convergence on compact sets with nonempty interior, disproving a conjecture of Petersen, Raslan, and Voigtlaender. Finally, we show that non-closedness for a fixed scalar-output ReLU architecture necessarily involves a \emphblow-up residue, in which diverging output coefficients amplify an emerging linear dependence among hidden features. Together, these results provide geometric and functional-analytic tools for studying the boundaries and degenerations of ReLU realization spaces.
[LG-213] Projection-Based Federated Learning on Compact Submanifolds with Geometry-Independent Differential Privacy
链接: https://arxiv.org/abs/2610.33712
作者: Thibault Pautrel,Florent Bouchard,Guillaume Ginolhac,Ammar Mian,Aurélien Bellet
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Federated learning trains a shared model across clients without exchanging their data, yet the updates they transmit still leak information about individual records. Enforcing differential privacy is harder still when the model parameters lie on a Riemannian manifold. Existing methods rebuild the noise added for privacy in the geometry of each new iterate, aggregate through intrinsic maps that often lack a closed form, and prove convergence only in restricted regimes. We introduce DP-RFedProj, a record-level differentially private federated learning method for compact submanifolds that aggregates through closed-form projections and privatises gradients in a fixed ambient geometry. Each client may run any stochastic first-order optimiser, take several local steps, and participate only intermittently. An explicit record-level f -DP guarantee holds for the entire training transcript, independently of the manifold, the aggregation scheme and the local optimiser. We establish convergence of DP-RFedProj when each client runs Riemannian SGD locally and the server aggregates by projected averaging, and show that privacy enters the rate through a term scaling with the manifold’s intrinsic dimension rather than the ambient one. Synthetic experiments on federated PCA over the Grassmann manifold corroborate these findings. On EEG motor-imagery classification with SPDNet, whose weights lie on the Stiefel manifold, the method outperforms a Euclidean baseline at every privacy budget considered. In both experimental settings, our accounting requires five to six times less noise than that of the closest prior method for the same budget, which translates into higher utility.
[LG-214] On the Error Analysis of Triangular Transport Maps for Conditional Generation via MMD-based Objectives
链接: https://arxiv.org/abs/2610.33610
作者: Michele Martino,Bamdad Hosseini,Amir Sagiv,Amirhossein Taghvaei
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注: 69 pages, 1 figure
Abstract:Conditional generative models parameterized by triangular transport maps are widely used for conditional sampling, yet theoretical guarantees on the quality of the conditional measures they produce remain scarce. We address this question for minimum divergence estimators of triangular maps. We use MMD-based objectives as a tractable setting in which the gap between joint and conditional errors can be analyzed precisely, and we measure performance through the average conditional estimation error, defined as the squared maximum mean discrepancy (MMD) between the estimated and the true conditionals, averaged over the conditioning variable. We first show that the common practice of minimizing a joint MMD may fail to recover the conditionals: writing the joint and the conditional criteria as quadratic energies of the same conditional mean embedding residual, we prove that, for parametric classes, joint minimizers are generically inconsistent in the misspecified setting. We then analyze an estimator that directly minimizes the conditional criterion through a surrogate loss computable from joint target samples and reference samples. We prove a non-asymptotic error bound that decomposes the average conditional estimation error into approximation, estimation, and reference sampling errors, with fast rates independent of the ambient dimension in the parametric setting. Finally, for joint MMDs built from concentrating Gaussian kernels on the conditioning variable, we show that the joint criterion approximates a density-reweighted version of the conditional one, and we derive an error bound for the joint minimizer whose rate suffers from a curse of dimensionality in the conditioning variable. Our results suggest that training criteria for conditional generative models should target the conditional rather than the joint error, a principle that extends to other divergences and to dynamic transport.
[LG-215] Affine-invariant Cubic Newton with Weak Learners NEURIPS2026
链接: https://arxiv.org/abs/2610.33516
作者: Nikita Zozoulenko,Daniel Falkowski,Thomas Cass,Lukas Gonon
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: Accepted to 40th Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:Modern gradient boosting algorithms are based on Newton’s method, which excels empirically but lacks global convergence guarantees. While recent work on Gradient-Regularized Newton (GRN) boosting achieved a global \mathcalO(1/\gamma^12k^2) rate, its analysis was not able to exploit the natural Hessian geometry and suffers from a heavy \gamma^12 dependence on a worst-case `weak gradient edge’ \gamma . In this paper, we tackle these theoretical drawbacks by generalizing the Affine-Invariant Cubic Newton (AICN) scheme to convex optimization with weak learners. Our analysis is based on average cosine values \overline\Theta_k along the optimization path and directional improvements \delta_k in the Hessian-induced norm. We show that the weak-AICN update reduces to a damped weak-Newton step with an explicit stepsize determined by \delta_k . Under semi-strong self-concordance, we establish a global \mathcalO(1/\overline\Theta_k^4 k^2) rate, significantly improving upon the weak-GRN bound. We further prove a local linear rate, global linear convergence for Hessian-dominated losses, cosine angle lower bounds, and present weak-AdaAICN, which adaptively tunes the cubic regularization. Finally, we demonstrate the broader applicability of this framework beyond gradient boosting by showing that Newton coordinate descent emerges naturally as a special case of optimization with weak learners.
[LG-216] Online Covariance Estimation for Averag ed Stochastic Gradient Descent on Riemannian Manifolds
链接: https://arxiv.org/abs/2610.33419
作者: Jingrong Guan,Liwei Jiang,Sen Na
类目: Machine Learning (stat.ML); Information Theory (cs.IT); Machine Learning (cs.LG); Optimization and Control (math.OC); Statistics Theory (math.ST)
*备注: 60 pqges, 7 figures, 2 tables
Abstract:We study online inference for averaged Riemannian stochastic gradient descent (RSGD) when the unknown parameter lies on a manifold. Although the averaged RSGD iterate is asymptotically normal, practical inference requires estimating the unknown covariance operator in its limiting distribution. Existing online batch-means methods cannot be applied directly because a general Riemannian manifold lacks an addition operation, and the tangent space containing the Riemannian gradient varies across iterations. Through vector transport, we construct overlapping and non-overlapping covariance estimators that adaptively combine information from successive batches while accounting for the changing tangent spaces. The proposed estimators are online, require neither Riemannian Hessian computations nor a prespecified terminal sample size, and attain the same convergence rates as their Euclidean counterparts (Zhu et al., 2023). Together with asymptotic normality, these estimators enable online statistical inference for RSGD on manifolds. We illustrate the proposed framework through a detailed study of streaming principal component analysis on Grassmann manifolds.
[LG-217] Automated screening of Parkinsons Disease using Smooth Pursuit Eye Movements
链接: https://arxiv.org/abs/2610.33359
作者: J. Camilo Puerta-Acevedo,Maria F. Alcala-Durand,Julián D. Arias-Londoño,Elisa Luque-Buzo,Francisco Grandas-Pérez,Juan I. Godino-Llorente
类目: ignal Processing (eess.SP); Machine Learning (cs.LG); Quantitative Methods (q-bio.QM)
*备注: 23 pages, 8 figures, including supplementary material
Abstract:Smooth pursuit eye movements are frequently altered in Parkinson’s disease, but previous studies have relied mainly on conventional parameters that describe the relationship between gaze and target. This study investigates whether complexity and nonlinear dynamics measures of the gaze position sequences can support the automatic screening of Parkinson’s disease. Gaze position sequences were recorded with a video eye tracker in 41 participants with early Parkinson’s disease and 47 healthy controls during 12 smooth pursuit tasks. Four experimental frameworks and four embedding delay strategies were compared. The selected experimental setting combined baseline wander removal with a fixed embedding delay. After feature selection, a support vector machine was trained for each task and axis and their outputs were combined in a mixture of experts. The pipeline was developed in 70 participants and evaluated on an external test set of 18 participants. On the external test set, the pipeline obtained an area under the receiver operating characteristic curve of 0.86. The most informative features described less persistent and more irregular gaze position sequences in Parkinson’s disease, mainly along the horizontal axis. These results support complexity and nonlinear dynamics measures of smooth pursuit as candidate markers for Parkinson’s disease screening, pending validation in larger independent cohorts.
[LG-218] When Are Nesterov Trajectories Eventually Locally Minimizing?
链接: https://arxiv.org/abs/2610.33341
作者: Peyman Morteza,Stephen Wright
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We revisit the variational characterization of Nesterov’s celebrated method and study when Nesterov trajectories are eventually locally minimizing in a variational sense, a notion that we define precisely. We show that this question is equivalent to the nonnegativity of an associated Schrödinger operator acting on variations of the trajectory. For power objectives of the form [ f(x)=\frac1p|x|^p ] (where | \cdot| denotes the Euclidean norm), we give a complete characterization: Among all exponents p\ge 2 , the Nesterov trajectory with non-zero initial position is eventually locally minimizing if and only if p=4 . We further interpret this exceptional exponent through the stability theory of semilinear elliptic equations.
[LG-219] Sharp Bounds on Continuous Treatment Effects under Unobserved Confounding
链接: https://arxiv.org/abs/2610.33278
作者: Junzhe Zhang
类目: atistics Theory (math.ST); Machine Learning (cs.LG)
*备注:
Abstract:We study partial identification of conditional treatment effects with a continuous treatment under unobserved confounding. We target the conditional average potential outcome (CAPO) E[Y_x|z] and the conditional average treatment effect (CATE) between two treatment levels. We assume a bounded outcome and Lipschitz continuity of the response functions and of the propensity scores given the latent confounder, and show that these conditions imply the smooth treatment response and smooth treatment selection conditions of Kim et al. (2018). The combined bound of Kim et al. uses only two reference points on the identified diagonal and is generally loose. Bounding through every diagonal reference point instead gives explicit bounds whose average is the sharp identified set of the CAPO at every treatment, inside or outside the observational support. Capping the difference between the upper and lower bounds gives sharp causal bounds on the CATE. These bounds are informative without monotonicity or overlap assumptions.
[LG-220] Not All Mislabels Are Equal: Efficient Task-Aware Auditing with Confidence Sequences
链接: https://arxiv.org/abs/2610.33262
作者: Fangyuan Lin,Noah Krever,Jakub Černý,Lily Xu
类目: Methodology (stat.ME); Machine Learning (cs.LG); Statistics Theory (math.ST); Applications (stat.AP); Machine Learning (stat.ML)
*备注:
Abstract:We study efficient, adaptive auditing of categorical labels in a fixed dataset, motivated by biodiversity-credit programs in which expert verification of species records is costly and downstream scores (e.g., payment) are more sensitive to errors on certain labels than on others. Our proposed ATLAS framework couples anytime-valid inference with a task-aware adaptive auditing policy. Representing label corrections through a confusion matrix, we prioritize which data points to inspect next based on both label uncertainty and the influence of potential label corrections on the downstream score function. For linear scores such as the mislabel rate, our priority-aware approach certifies the overall label error rate to \pm 2 percentage points with 24% and 11% fewer audits than random sampling on biodiversity data from Kenya and a widely used CIFAR-10N benchmark, respectively. For nonlinear scores such as the Hill-Shannon diversity, we exploit a linear-envelope representation that reduces inference to a family of linear-score audits; priority-aware sampling reaches the target precision with 36% fewer audits. When distribution-free guarantees are prohibitively costly, we develop a model-based Bayesian alternative that trades assumption-free validity for substantially fewer audits. Our biodiversity case study highlights a gap between statistical validity and practical feasibility: generic methods such as Serfling’s bound and existing confidence-sequence procedures are rigorous but can require prohibitively many expert reviews in practice. Together, these results show task-aware auditing can significantly reduce the number of expert audits needed while providing finite-sample, anytime-valid guarantees.
[LG-221] Flow Mesh: A Velocity-Regularized Flow Matching Framework for Stable Distributional Modeling of Physical Fields on Unstructured Meshes
链接: https://arxiv.org/abs/2610.33204
作者: Siddharth Verma,Jaewan Park,Jay Phil Yoo,Kazuma Kobayashi,Souvik Chakraborty,Syed Bahauddin Alam
类目: Fluid Dynamics (physics.flu-dyn); Machine Learning (cs.LG)
*备注:
Abstract:Thermal-fluid field distributions underpin the analysis and operation of energy infrastructure, from nuclear-reactor cooling channels to industrial heat exchangers, but resolving these distributions through computational fluid dynamics requires prohibitively large simulation ensembles. Their complex geometries also demand unstructured meshes that are poorly accommodated by generative models developed for regular grids. Here we introduce FlowMesh, a generative framework that combines permutation-invariant mesh encoding with deterministic latent flow matching and an l1 penalty on the learned transport velocity. FlowMesh generates full-field realizations directly on fixed unstructured meshes without auxiliary grid interpolation or a stochastic denoising chain. Across lid-driven-cavity, pressurized-water-reactor subchannel and heat-exchanger datasets, evaluated over six matched training seeds, FlowMesh achieves macro-averaged probability-density-function coverage of 91.7%-97.0%, exceeding an architecture-controlled latent diffusion model by 11.6 – 45.4 percentage points. Latent-velocity regularization increases coverage by 3.8 percentage points for reactor flows and 1.1 points for heat-exchanger flows, while reducing temperature-field Kullback–Leibler divergence by 1.7-2.4-fold. Each full-field realization requires 1.23 \pm 0.15 s, providing a 585–1902-fold wall-clock acceleration over computational fluid dynamics. FlowMesh therefore makes large-scale generative characterization of thermal-fluid states computationally feasible, providing a foundation for data-driven analysis and digital-twin development in fixed-geometry energy systems.
[LG-222] Kernelised covariance matrices dimension reduction and imbalanced spectral clustering
链接: https://arxiv.org/abs/2610.33177
作者: Xiaoyu Dong,Subhro Ghosh,Soumendu Sundar Mukherjee,Himasish Talukdar
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Probability (math.PR); Spectral Theory (math.SP)
*备注:
Abstract:The Gaussian Mixture Model (abbrv. GMM) is a classic generative model for clustered data, for which spectral clustering based on dimension reduction via PCA has been an effective framework. In this work, we investigate kernelised covariance matrices as an alternative mechanism for dimension reduction and spectral clustering, focusing on truncation kernels that select pairs of data points with suitable separation constraints. The structure of these matrices is motivated by considerations from statistical mechanics of repulsive particle systems. Our approach is particularly relevant for the setting of imbalanced clusters, where the sizes of the largest and smallest clusters have a diverging ratio with the problem size, and existing PCA-based methods turn out to be of limited effectiveness. We demonstrate efficient dimension reduction and clustering performance of these modified covariance matrices, substantiated by quantitative bounds. In particular, they achieve superior dimension reduction guarantees compared to PCA for a wide range of parameters in the regime of imbalanced clusters. For spectral clustering, our approach exhibits improvements over PCA in the imbalanced setting, in particular guaranteeing accurate clustering in a much larger regime of intra-cluster separations. Our theoretical guarantees reflected in a range of simulation experiments that exhibit these improvements for unbalanced clustering problems. Our analytical approach, based on kernel Laplacians, provide a toolbox as well as a relevant conceptual ingredient for clustering problems at large. We believe that our work will motivate the application of such kernelised covariance matrices for related classes of statistical problems, as well as spur new investigations in the theory of random matrices. Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Probability (math.PR); Spectral Theory (math.SP) MSC classes: 62H30 Cite as: arXiv:2610.33177 [stat.ML] (or arXiv:2610.33177v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.33177 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-223] Fully quantum training of parametric circuits with Quantum Hamiltonian Descent
链接: https://arxiv.org/abs/2610.33144
作者: Marco Burato,Leonardo Banchi
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注:
Abstract:Parametric quantum circuits are usually trained by classical, local optimizers, which use the quantum computer only to evaluate the cost function, or its gradient, one point at a time. Quantum Hamiltonian Descent (QHD) takes a different route: in this quantum algorithm for continuous optimization, the parameters become the position of a quantum particle, with the cost function as its potential, and two time schedules control kinetic and potential energy so that the wavefunction explores the whole landscape and tunnels towards the global minimum. QHD, however, assumes a classically computable cost function, and its digital implementation is poorly understood: standard error bounds grow exponentially with the number of qubits per parameter, and the schedules are borrowed from classical optimization. Here we propose to train parametric circuits with QHD entirely on the quantum computer, through an online step in which the qubits encoding the parameters coherently control the circuit on a second register. We prove that, for cost functions with a smooth periodic extension, including those of parametric circuits, the simulation error admits a bound independent of the grid resolution, and that for circuits the bound grows at most linearly with the number of parameters. We also show that the ideal dynamics depends on the two schedules only through a single curve, which controls the adiabaticity of the search, and that the remaining freedom can be used to minimize the gate count. Simulations on benchmark functions and parametric circuits show that, at equal cost, schedules designed in this way markedly increase the probability of finding the global minimum.
[LG-224] A Measure-Geometric Theory of Robustness in Machine Learning
链接: https://arxiv.org/abs/2610.33123
作者: Cheng-Chao Huang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:This paper develops a foundational measure-geometric theory of robustness in machine learning. Robustness is formulated as escape from a retained task state relative to a reference measure on a data-bearing manifold, a perturbation gauge, and an admissible realization mechanism. The theory distinguishes the structural possibility of escape on the reference manifold from its ambient realization under mechanism-dependent observable readouts, while placing both within a unified geometric account. It introduces a mesoscopic organization of escape geometry, through which vulnerable mass, finite-scale morphology, delivered progress, and branch coherence are characterized by explicit laws as linked but non-identical robustness phenomena. By recasting robustness as the geometric formation, organization, realization, and persistence of escape, the theory provides a structural basis for analysis beyond scalar criteria.
[LG-225] Uncertainty Quantification for Financial Foundation Models: A Survey
链接: https://arxiv.org/abs/2610.33109
作者: Zijie Zhao,Mingjun Sun,Shenbo Xu
类目: General Finance (q-fin.GN); Machine Learning (cs.LG)
*备注: Accepted at the 7th ACM International Conference on AI in Finance (ICAIF 2026). 8 pages
Abstract:Financial foundation models increasingly support workflows spanning financial question answering, disclosure analysis, forecasting, and decision-making. In these settings, uncertainty may concern whether evidence is valid and available at the relevant time, whether an interpretation is justified, which outcomes remain plausible, or which action is preferred under costs and risk constraints. Because these objects arise at different points in a workflow, uncertainty cannot be reduced to confidence in a final output. We introduce a source-aware framework that distinguishes six sources of uncertainty across evidence and context, model and inference, and outcome and action, while treating agentic workflows as the execution layer through which uncertainty can propagate, change form, or be reduced. We develop a mechanism-based taxonomy grounded in how methods construct or verify uncertainty and reliability representations, and use it to assess their applicability across financial task domains. We further map finance-specific empirical studies across the taxonomy and task domains to characterize the current research landscape. Our analysis shows that most reviewed studies address uncertainty in individual parts of financial workflows, with limited work connecting evidence and reasoning reliability to downstream prediction and constrained action. By connecting where uncertainty arises, how it is represented, and how it bears on downstream decisions, this survey provides a unified foundation for studying and evaluating uncertainty in financial foundation models.
[LG-226] ARMD-Predict: multi-site machine learning to predict antibiotic resistance before culture results
链接: https://arxiv.org/abs/2610.32998
作者: Fateme Nateghi Haredasht,Nicholas P. Marshall,Wenyuan Chen,Lauren N. Cooper,Ziming Wei,Harrison Konsker,Jacob S. Chen,Michael Klompas,Cara McKenna,Richard J. Medford,Sanjat Kanjilal,Austin J. Schoeffler,Stephen P. Ma,Niaz Banaei,Steven M. Asch,Amy Chang,Stanley C. Deresinski,Mary K. Goldstein,Robert Tibshirani,Jonathan H. Chen
类目: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG)
*备注: 31 pages, 4 figures, 2 tables, 7 Extended Data figures, 5 Extended Data tables. Research web applications: this https URL
Abstract:Clinicians select empiric antibiotics before culture susceptibility results are available. Electronic health record (EHR) models could inform that choice, but most are developed at a single hospital and transport poorly to others. We developed ARMD-Predict using 7.5 million culture-antibiotic observations from four academic medical centers. One set of models estimates susceptibility to individual antibiotics; a second, hierarchical set predicts the infecting organism and then organism-specific resistance phenotypes. Across 23 pooled susceptibility models evaluated in same-period, patient-grouped test partitions, median area under the receiver operating characteristic curve (AUROC) was 0.77 for inpatient urine and 0.70 for inpatient blood cultures; chronological retraining reduced AUROC by approximately 0.04. At an exploratory threshold with 95% positive predictive value, derived and evaluated on the same test observations, the model identified 55% of ceftriaxone-susceptible inpatient urine observations as suitable for narrower therapy and flagged 60% of ceftriaxone-resistant observations for broader coverage. Organism classifiers achieved top-3 accuracy of 51.3-83.3%; phenotype AUROCs were conditional on the observed organism. Research applications are available at this https URL. Independent calibration, threshold and end-to-end prospective validation are required before clinical use.
[LG-227] A Multimodal Machine Learning Framework Integrating Genomic Epidemiological Environmental and Clinical Imaging Data for Accelerated Measles Detection and Transmission Inference
链接: https://arxiv.org/abs/2610.32996
作者: Jiada Li
类目: Quantitative Methods (q-bio.QM); Emerging Technologies (cs.ET); Machine Learning (cs.LG)
*备注:
Abstract:Background: Measles resurged in the United States during 2025-2026, with more than 4,600 reported cases. Timely detection and transmission inference require integrated genomic, epidemiological, environmental, and clinical data. We developed a multimodal machine learning (ML) framework integrating genomic sequences, epidemiological indicators, wastewater surveillance, and clinical rash images. Methods: We evaluated a three-stage pipeline using public data. Stage 1 classified transmission linkage among 903 sequence pairs from the 2021 Ohio-Arizona-Washington outbreak using k-mer TF-IDF-SVD embeddings and pairwise single-nucleotide polymorphism (SNP) distances with leave-one-cluster-out cross-validation. Stage 2 predicted state-week outbreak risk using logistic regression and XGBoost with wastewater, mobility, and vaccination features under leave-one-year-out cross-validation. Stage 3 used BiomedCLIP ViT-B/16 embeddings and a fine-tuned multilayer perceptron for rash classification on MSLD v2.0 (755 images, 6 classes). InfoNCE contrastive learning aligned genomic and imaging embeddings. Results: Transmission-linkage classification achieved AUC 0.924 and F1 0.912 with logistic regression. Outbreak-risk prediction achieved AUC 0.707 with logistic regression across 2025-2026 leave-one-year-out folds. Contrastive alignment achieved Recall@1 = 0.985 on held-out paired data. Ablation showed genomic features dominated transmission-linkage prediction (genomic-only AUC 0.983), whereas wastewater positivity was the strongest outbreak-risk predictor (importance 0.402). Conclusions: Multimodal ML integrating genomic, epidemiological, wastewater, and clinical imaging data can support measles surveillance and transmission inference using public data. Prospective validation with paired clinical-genomic data, expanded wastewater coverage, and real-world workflows is needed before deployment.
[LG-228] Computational De Novo Design of Miniprotein Candidate Neutralizers Targeting the IL-23 Cytokine Subunit p19
链接: https://arxiv.org/abs/2610.32779
作者: Raghav Paladugu,Fardina Fathmiul Alam
类目: Biomolecules (q-bio.BM); Machine Learning (cs.LG)
*备注: Accepted as a short paper at the 2026 IEEE International Conference on Bioinformatics and Biomedicine (BIBM 2026)
Abstract:Interleukin-23 (IL-23) drives T-helper-17 inflammation underlying inflammatory bowel disease and psoriasis; anti-p19 antibodies are clinically effective but large, costly, and injectable. De novo miniproteins are an attractive alternative, yet published designs target the IL-23 receptor; the IL-23 cytokine itself has not been addressed. A further obstacle is that the standard confidence score (ipTM) tracks construct length rather than interface quality. We computationally design miniprotein binders against the IL-23R-binding face of p19 (PDB 3DUH) and score every design with ipSAE, a length-robust interface metric derived from AlphaFold-initial-guess predicted aligned error. A conventional RFdiffusion/ProteinMPNN/ColabFold pipeline yields no reliable predicted interface across 972 designs (best ipSAE 0.30), and a scoring artifact, in which single-sequence ColabFold under-rates binder pLDDT by approximately 56 points, explains an earlier misleading “binders do not fold” signal. An interface-optimized hallucination method (BindCraft), scored identically, yields 42 of 50 designs above ipSAE 0.60 (best 0.804). An AlphaFold-independent geometric check reproduces the conventional failure: no top-ranked conventional design reaches the buried area or interface hydrogen-bond count of the weakest lead. Five all-alpha-helical leads (72-104 residues) collectively engage all six targeted hotspots; two dock within the intact p40/p19 heterodimer and all show favorable Rosetta energetics. A p40-alone counter-screen returns ipSAE 0.00 for all five, indicating no spurious p40 interface. To our knowledge, these are the first computationally proposed de novo miniprotein candidates designed to occlude the IL-23R-binding face of IL-23 p19. All results require experimental validation.
附件下载
